Mechanisms and Machine Science

Volume 137

Series Editor
Marco Ceccarelli , Department of Industrial Engineering, University of Rome Tor
Vergata, Roma, Italy

Advisory Editors
Sunil K. Agrawal, Department of Mechanical Engineering, Columbia University,
New York, NY, USA
Burkhard Corves, RWTH Aachen University, Aachen, Germany
Victor Glazunov, Mechanical Engineering Research Institute, Moscow, Russia
Alfonso Hernández, University of the Basque Country, Bilbao, Spain
Tian Huang, Tianjin University, Tianjin, China
Juan Carlos Jauregui Correa , Universidad Autonoma de Queretaro, Queretaro,
Mexico
Yukio Takeda, Tokyo Institute of Technology, Tokyo, Japan
This book series establishes a well-defined forum for monographs, edited Books,
and proceedings on mechanical engineering with particular emphasis on MMS
(Mechanism and Machine Science). The final goal is the publication of research that
shows the development of mechanical engineering and particularly MMS in all
technical aspects, even in very recent assessments. Published works share an
approach by which technical details and formulation are discussed, and discuss
modern formalisms with the aim to circulate research and technical achievements
for use in professional, research, academic, and teaching activities.
This technical approach is an essential characteristic of the series. By discussing
technical details and formulations in terms of modern formalisms, the possibility is
created not only to show technical developments but also to explain achievements
for technical teaching and research activity today and for the future.
The book series is intended to collect technical views on developments of the
broad field of MMS in a unique frame that can be seen in its totality as an
Encyclopaedia of MMS but with the additional purpose of archiving and teaching
MMS achievements. Therefore, the book series will be of use not only for
researchers and teachers in Mechanical Engineering but also for professionals and
students for their formation and future work.
The series is promoted under the auspices of International Federation for the
Promotion of Mechanism and Machine Science (IFToMM).
Prospective authors and editors can contact Mr. Pierpaolo Riva (publishing
editor, Springer) at: [email protected]
Indexed by SCOPUS and Google Scholar.
Rajiv Tiwari · Y. S. Ram Mohan ·
Ashish K. Darpe · V. Arun Kumar · Mayank Tiwari
Editors

Vibration Engineering
and Technology
of Machinery, Volume I
Select Proceedings of VETOMAC XVI 2021
Editors
Rajiv Tiwari Y. S. Ram Mohan
Department of Mechanical Engineering Department of Mechanical Engineering
Indian Institute of Technology Guwahati B.M.S. College of Engineering
Guwahati, Assam, India Bengaluru, Karnataka, India

Ashish K. Darpe V. Arun Kumar

Department of Mechanical Engineering Department of Mechanical Engineering
Indian Institute of Technology Delhi B.M.S. College of Engineering
New Delhi, Delhi, India Bengaluru, Karnataka, India

Mayank Tiwari
Department of Mechanical Engineering
Indian Institute of Technology Patna
Patna, Bihar, India

ISSN 2211-0984 ISSN 2211-0992 (electronic)

Mechanisms and Machine Science
ISBN 978-981-99-4720-1 ISBN 978-981-99-4721-8 (eBook)
https://doi.org/10.1007/978-981-99-4721-8

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2023

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore

Paper in this product is recyclable.

The book is dedicated to Late Prof. J. S. Rao
Committee

The following are renowned technical experts in vibration field who contributed in
success of the conference in different capacity:

Conference Chair

Prof. Nalinaksh S. Vyas, Indian Institute of Technology Kanpur, Kanpur, India

Conference Co-chair

Dr. V. Arun Kumar, B.M.S.C.E, Bengaluru, India and The Vibration Institute of India
(TVII).

Members of Advisory Committee

Prof. Marco Ceccarelli, University of Cassino and South Latium, Italy, Past President,
IFToMM
Prof. J. M. Balthazar, University Estadual Paulista, Brazil
Prof. Joseph Mathew, Queensland University of Technology, Australia
Prof. Jyoti K. Sinha, The University of Manchester, UK
Prof. R. Rzadkowski, Polish Academy of Sciences, Poland
Prof. C. Nataraj, Villanova University, USA
Prof. Abdessattar Abdelkefi, New Mexico State University, USA
Prof. M. I. Friswell, University of Swansea, Swansea, UK
Prof. Sondipon Adhikari, University of Glasgow, Glasgow, UK
Prof. Jürg Meier, Zurich University of Applied Sciences, Winterthur, Switzerland

vii
viii Committee

Prof. Stephanos Theodossiades, Loughborough University, UK

Prof. Grzegorz Litak, Lublin University of Technology, Lublin, Poland
Prof. Diego Galar, Lulea University of Technology, Sweden
Prof. Lim Chee Wah, City University of Hong Kong, Hong Kong
Prof. A. Seshadri Sekhar, Indian Institute of Technology Madras, India
Prof. C. Sujatha, Indian Institute of Technology Madras, India
The technical committee would like to thank authors and reviewers for their contri-
bution in bringing out the book. Two of students from Indian Institute of Tech-
nology Guwahati, Mr. Pantha Pradip Das and Mr. Atul Kumar Gautam who helped
in collecting various documents from reviewers and authors in timely manner, and
they deserve a special thanks. Technical team also would like to thank a very profes-
sional team from Springer Nature, Ms. Priya Vyas, Mr. Ramamoorthy Rajangam,
and Mr. Ashok Kumar to bring out this book very efficiently.

Technical Committee

Prof. Rajiv Tiwari, Indian Institute of Technology Guwahati, Guwahati, India

Prof. Y. S. Ram Mohan, B.M.S.C.E, Bengaluru, India
Prof. Ashish K. Darpe, Indian Institute of Technology Delhi, New Delhi, India
Prof. Mayank Tiwari, Indian Institute of Technology Patna, Patna, India
Dr. V. Arun Kumar, B.M.S.C.E, Bengaluru, India
Preface

The International Conference on Vibration Engineering and Technology of

Machinery (VETOMAC) is annually held series of conferences essentially to
promote Vibration Engineering and Technology of Machinery. The first VETOMAC
was organized at IISc Bengaluru, India, in the year 2000 with the intention of encour-
aging scientific and technical cooperation and exchange across the globe. The 16th
VETOMAC was held at B.M.S. College of Engineering, Bengaluru, India, during
16 to 18 December 2021. This edition of VETOMAC was dedicated to the memory
of its founder, Prof. J. S. Rao, who passed away on July 04, 2020 and in pursuance
of his vision, the theme of this conference is kept as ‘Integrated Vehicle Health
Management (IVHM)’. This conference is the 16th in the series of annual symposia
that started in Bengaluru, India (2000), and were subsequently held in Mumbai,
India (2002), Kanpur, India (2004), Hyderabad, India (2007), Wuhan, China (2009),
New Delhi, India (2010), Hong Kong, China (2011), Vaddeswaram, India (2012),
Nanjing, China (2013), Manchester, UK (2014), Taiwan (2015), Warsaw, Poland
(2016), Queensland, Australia (2017), Lisbon, Portugal (2018), and Curitiba, Brazil
(2019).
Late Prof. Jammi Srinivasa Rao (1939–2020) popularly known as JS in the field of
Vibration Engineering was first person to acquire doctorate in Design stream from IIT
Kharagpur in independent India. He was a Distinguished Professor of rotor dynamics
and vibration engineering at IIT-Delhi, Counsellor-Science and Technology at the
Indian Embassy of USA, Washington DC, Visiting Professor at various Universities
abroad and Consultant to over 30+ industries. He is one of the 13 founding fathers of
the International Federation for the Promotion of Mechanism and Machine Science
(IFToMM) Poland in 1969. He established ‘The Vibration Institute of India (TVII)’.
An international scientific journal launched by him in 2002 has groomed itself into
yearly six volume Journal of Vibration Engineering and Technologies (JVET) co-
published with Springer. His authorship includes 200+ reputed journal and 300+
conference papers and twenty-two machinery dynamics books. Professor Rao was
instrumental in initiating, collaborating and promoting the Vibration Engineering
and Technology of Machinery (VETOMAC) conference since its inception in 2000
until 2019. B.M.S. College of Engineering has organized this prestigious conference

ix
x Preface

in the memory of Prof. J. S. Rao (who was also an INAE Distinguish Professor at
BMSCE).
VETOMAC XVI covered four main broad categories in Vibration and Technology
of Machinery fields: Vibration Analysis, Condition Monitoring Based on Vibra-
tions, Rotor Dynamics and Tribology and allied areas. Plenary lectures were given
on the theme of conference, ‘Integrated Vehicle Health Management (IVHM)’ by
Dr. V. K. Saraswat, National Institution for Transforming India (NITI) Aayog,
India and Dr. Kota Harinarayana, Chairman (Board of Governor) Indian Insti-
tute of Technology Varanasi. The technical keynote and contributed papers were
presented in hybrid mode due to prevailing COVID-19 situation all over the world
during conference period. Presenters were from across the globe, including countries
like USA, UK, France, Australia, Ireland, Italy, Poland, Portugal, Brazil, Sweden,
Mexico, China, Taiwan, India, and Nepal. The present book series have two volumes.
Volume I (present volume) contains total 38 contributed technical papers and
Volume II contains remaining 37 contributed technical papers after rigorous peer
review and revision.

Guwahati, India Rajiv Tiwari

Bengaluru, India Y. S. Ram Mohan
New Delhi, India Ashish K. Darpe
Bengaluru, India V. Arun Kumar
Patna, India Mayank Tiwari
Contents

Identification in a Magnetically Levitated Rigid Rotor System

Integrated with Misaligned Sensors and Active Magnetic Bearings . . . . . 1
Prabhat Kumar and Rajiv Tiwari
Thermo-Elastic Bending and Buckling Behavior of Functionally
Graded Shafts with Various Grading Patterns . . . . . . . . . . . . . . . . . . . . . . . . 25
Debabrata Gayen, Rajiv Tiwari, and Debabrata Chakraborty
Modeling and Analysis of Low-Pressure Steam Turbine Blade . . . . . . . . . 45
Nitin Sehra, Sushila Rani, and Vikas Rastogi
Enhanced Energy Harvesting Application of Piezoceramics (PZT)
in MEMS Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
G. Pranavanathan and S. B. Kandagal
Diesel Locomotive Alternator Bearing Damage Detection
by Adopting Comprehensive Condition Monitoring Techniques . . . . . . . . 71
Hemant M. Bari, Suhas S. Patil, and Atul A. Deshpande
Response Analysis of Inclined Edge Cracked Beam Under Moving
Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Shakti P. Jena, Naresh Kumar Sarella, Sasmita Sahu, and Dayal R. Parhi
Non-stationary Response of a Bridge Due to Moving Vehicle
with Random Arrival Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Anjaly J. Pillai and Sudip Talukdar
Multiclass Fault Diagnosis and Novelty Detection of Induction
Motor Using Deep Learning Algorithm Based on Frequency
Domain Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Aman Gupta, Rajiv Tiwari, and Dhruba Jyoti Bordoloi
Modelling and Analysis of Active Magnetic Bearing Integrated
Reaction Wheels for Satellite Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
P. A. Thashreef, Rajiv Tiwari, and Dhruba Jyoti Bordoloi

xi
xii Contents

Single Plane Balancing of Rotor-AMB System Using Virtual Trial

Unbalances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Gyan Ranjan, Rajiv Tiwari, and H. B. Nemade
Comparison of Linear and Non-linear Feedforward Algorithms
to Control Chaotic and Impulsive Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Vikas Kumar Lakhmani and Amrita Puri
Effect of Thermo-Mechanical Coupling and Large Deformation
on the Response of SMA Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Animesh Kundu and Atanu Banerjee
Performance Enhancement of Bistable Energy Harvester
with Asymmetric Potential Function Using an Elastic Magnifier . . . . . . . 191
K. Devarajan and B. Santhosh
Vibration Reduction in Ambulance Using Modified Stretcher
with Vibration Absorber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Laxmikant S. Dhamande and Vishal B. Jadhav
Simulation Studies of Low-Velocity Impact Damage in FRPS . . . . . . . . . . 219
L. Lokesh Kumar, B. S. Raviteja, S. V. Ashish, A. Arun,
A. R. Anilchandra, and C. M. Manjunatha
Experimental Identification of Unbalance and Crack Parameters
in an Internally Damped Rotor System Integrated with an Active
Magnetic Bearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Nilakshi Sarmah and Rajiv Tiwari
Vibration Response Prediction in Rotor Systems with External
Damping by Deep Learning Using Geometrical Features . . . . . . . . . . . . . . 255
R. Siva Srinivas, Rajiv Tiwari, and Amar Singh
Effect of Bluff Body on the Performance of Flutter Based Energy
Harvester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Ankit Agarwal and Ashish Purohit
A Systematic Review of Rotor Unbalance Diagnosis in Rotating
Machinery Based on Machine Learning Algorithms . . . . . . . . . . . . . . . . . . 281
Sudhar Rajagopalan, Ashish Purohit, and Jaskaran Singh
Tuning of PID Parameters for Misaligned Rotor Bearing System . . . . . . . 301
Kuppa Sampath Kumar and Mohit Lal
Machine Learning-Based Fault Prediction of Electromechanical
System with Current and Vibration Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Purushottam Gangsar, Vikas Singh, Manoj Chouksey, and Anand Parey
Identifying Condition Indicators for Artificially Intelligent Fault
Classification in Rolling Element Bearings . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Mohd Atif Jamil and Sidra Khanam
Contents xiii

Influence of Local Surface Cracks on Dynamic Parameters

of Multi-span Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
Mitesh J. Mungla and Dharmendra S. Sharma
Crack Detection in a Shaft Using Wavelet Packet Transform . . . . . . . . . . . 359
Ramnivas Kumar and Sachin K. Singh
Design and Modeling of Piezo-resistive MEMS Accelerometer
for Crash Test Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
N. Kusuma, J. Pradyumna, A. Tulasi, and Asha R. Upadhyay
Classification of Orbits of Rotor Supported on Squeeze Film
Damper by Supervised Learning Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Anurag Kumar, Mayank Tiwari, and Akhilendra Singh
Free Vibration Response of 3D Braided Reinforced Composite
Conoidal Shell Using Third-Order Shear Deformation Theory . . . . . . . . . 405
Pabitra Maji and Brighu Nath Singh
A Dynamic Model for Polymer Draft Gears . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Om Prakash Yadav and Nalinaksh S. Vyas
Modelling and Dynamic Analysis of Tapered Viscoelastic
Laminated Composite Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Krishanu Ganguly and Haraprasad Roy
Analysis of Limiting Load Capacity and Stiffness Coefficients
of Porous-Bump-Recess Foil Journal Bearing . . . . . . . . . . . . . . . . . . . . . . . . 453
S. Arokya Agustin, C. Shravankumar, A. Arul Jeya Kumar,
and T. V. V. L. N. Rao
Limiting Load Capacity and Stiffness Coefficients of Bump Recess
Foil Journal Bearing with FGM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
S. Arokya Agustin, C. Shravankumar, A. Arul Jeya Kumar,
and T. V. V. L. N. Rao
Modal Analysis of 3D Braided Composite Reinforced Twisted
Spherical Shell Using 3D Finite Element Method . . . . . . . . . . . . . . . . . . . . . 473
Sambhaji Lore, Pabitra Maji, and Brighu Nath Singh
Dynamic Analysis of Flexible-Joint and Single Flexible-Link
Manipulator by Using Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . 489
Rajesh Ranjan, Saleendra Hari Babu, and Santosha Kumar Dwivedy
Prediction of Fatigue Crack Growth Behavior Under
Cold-Turbistan Spectrum Loads Using Variable Crack Closure
Approach in GTM 720 Nickel Base Superalloy . . . . . . . . . . . . . . . . . . . . . . . 503
Sharanagouda G. Malipatil, Sandeep Kumar, Anuradha N. Majila,
D. Chandru Fernando, M. Manjuprasad, and C. M. Manjunatha
xiv Contents

Optimization of Active Vibration Absorber by Acceleration

Feedback Using Fixed Point Theory and Genetic Algorithm . . . . . . . . . . . 517
S. Mohanty and Santosha Kumar Dwivedy
Study of Time-Varying Oil Film Damping and Contact Stiffness
of Ball Bearings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
Sameera Mufazzal, S. M. Muzakkir, and Sidra Khanam
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart
Functionally Graded (SFG) Plate Subjected to Electromechanical
Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Pawan Kumar and S. P. Harsha
Influence of Squeeze Film Damper on the Rub-Impact Response
of a Dual-Rotor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
K. Prabith and Praveen Krishna

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
About the Editors

Dr. Rajiv Tiwari is currently Professor in the Department of Mechanical Engi-

neering, Indian Institute of Technology Guwahati since 1997. He worked in Regional
Engineering College, Hamirpur (Himachal Pradesh), India, for a year during 1996–
1997. He completed his Ph.D. from IIT Kanpur in the area of Rotor Dynamics in
1997. He worked as Research Officer for one year in 2001 in University of Wales,
Swansea, UK, and as DAAD Fellow in Technical University of Darmstadt Germany
in 2011 for two months. Dr. Rajiv Tiwari works in the area of rotor dynamics and
design of rolling bearings. He has more than 250 papers in international journals and
conferences. He has successfully organised national and international conferences
and short-term courses in the area of vibrations and rotor dynamics.

Dr. Y. S. Ram Mohan is currently Professor and Head of Aerospace Engineering

Department at B.M.S. College of Engineering, Bengaluru, India. Before this he
served as Professor in the Department of Mechanical Engineering at B.M.S. College
of Engineering for over 25 years. He obtained his B.E. (Mechanical) and M.Tech.
(Machine Design) from B.M.S. College of Engineering and Ph.D. from the Aerospace
Department of Indian Institute of Technology-Madras, Chennai. His research areas
of interests are fretting fatigue, finite element methods, composite materials, and
functionally graded materials. He has published over 20 research papers in reputed
international journals and proceedings. He was the reviewer for journals from ASME,
Elsevier-Materials Today, and others.

Dr. Ashish K. Darpe is currently working as Professor of Mechanical Engineering

at IIT Delhi. He obtained his Ph.D. in the area of Rotor Dynamics at IIT Delhi. He
has been pursuing his research interests in the area of vibrations, rotor dynamics,
machinery fault diagnosis and prognosis and noise control. He has published over
111 research articles and has supervised 12 Ph.D. theses. He has been the project
investigator for various R&D and consultancy projects funded by defence R&D labs,
private industry, and government funding agencies.

xv
xvi About the Editors

Dr. V. Arun Kumar currently working as Chief Technology Menor, at Dheya Tech-
nologies Ltd., Bengaluru, and as Adjunct Professor in B.M.S. College of Engineering,
Bengaluru, has obtained his B.E. Mechanical Engineering, from Bengaluru Univer-
sity, M.Sc. Engg. from Madras University, and Ph.D. from IIT, Madras. He worked
in Propulsion Division, National Aerospace Laboratories for 34 years and retired as
Director Grade Scientist/Head, Propulsion Division. His major areas of research
interests include vibration reduction in rotating machineries, advanced bearing/
dampers systems, etc. He has to his credits ten patents granted (another five under
processing) and over 100 publications. He has received National Design award from
NDRF in Mechanical Engineering and Biren Roy award from Aeronautical Society
of India, apart from outstanding performance award from NAL.

Dr. Mayank Tiwari is currently Professor in the Department of Mechanical Engi-

neering, Indian Institute of Technology Patna, since 2013. He worked in General
Electric from 2001 to 2013 in the Global Research and Development Division and the
Aviation Division in the John F. Welch Technology Centre Bengaluru. He completed
his Ph.D. from IIT Delhi in the area of nonlinear Rotor Dynamics in 1998 after which
he worked as Postdoctoral Fellow in the Acoustic and Dynamics Laboratory of the
Ohio State University USA. Dr. Mayank Tiwari works in the area of rotor dynamics
and tribology. He has more than 30 papers in international journals and patents in
USA, France, and India in the area of vacuum tribology, X-ray tube rotors, wind
turbine gear box. Dr. Mayank Tiwari also has two design registrations in the India
Patent Office.
Identification in a Magnetically Levitated
Rigid Rotor System Integrated
with Misaligned Sensors and Active
Magnetic Bearings

Prabhat Kumar and Rajiv Tiwari

Abstract Misalignment is amongst the serious faults occurring in rotating

machinery and the normal operations of machine will be crucially influenced under
this state. Therefore, there is a need for exploring the motion of a faulty rotor system
and identifying the faults for untroubled performance of machines. In this paper, a
numerical investigation has been presented on the dynamic action of an unbalanced
as well as misaligned stiff rotor system having two discs at the offset positions and
mounted on active magnetic bearings at both the shaft ends. The rotor is assumed
to be in combined misalignment with AMBs and non-contact displacement sensors.
The force arising from misaligned AMB is derived along with description of the
mathematical modelling of misaligned sensors based on the virtual trial misalign-
ment strategy. In this approach, the trial misalignment is additionally and virtually
given to the rotor by providing an additional bias current to AMBs. This helps in
creating additional misalignments in both AMBs and sensors relative to the rotor
operating axis. Dynamic equations of the rotor-sensor-AMB model is derived based
on moment equilibrium method with the consideration of inertia force, unbalance
force, gyroscopic effects and misaligned AMB forces. Further, the equations are
solved by developing and arranging SIMULINK™ blocks to obtain the time depen-
dent displacement response and AMB current signal. The foremost aim of the article
is to explain the vibrational effects on the rotor due to unbalance, eddy current prox-
imity sensors misalignment and AMBs residual misalignment. Apart from this, the
paper would also identify the initial offset of sensors.

Keywords Active magnetic bearing · Virtual trial misalignment · Unbalance ·

Rigid rotor · Gyroscopic moment · Identification

P. Kumar (B)
Department of Mechanical Engineering, NIT Manipur, Imphal West, Manipur, India
e-mail: [email protected]; [email protected]
R. Tiwari
Department of Mechanical Engineering, IIT Guwahati, Guwahati, Assam, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 1
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_1
2 P. Kumar and R. Tiwari

Nomenclature

a1 , a2 Distances of the rotor center of gravity (C.G.)

from AMB1 as well as AMB2
e1 , e2 Disc 1 and Disc 2 eccentricities
(ix 1 , iy1 ) and (ix 2 , iy2 ) Control currents at AMB 1 as well as AMB 2 in
the vertical (x) and horizontal (y) directions
(iTx 1 , iTy1 ) and (iTx 2 , iTy2 ) Trial bias currents at AMB 1 as well as AMB 2
in the vertical (x) and horizontal (y) directions
I d , I P1 , I P2 Rotational inertia of the rotor, disc 1 as well as
disc 2
(k sx 1 , k sy1 ) and (k sx 2 , k sy2 ) Stiffness (displacement) values of AMB 1 as
well as AMB 2 in the vertical (x) and horizontal
(y) directions
(k ix 1 , k iy1 ) and (k ix 2 , k iy2 ) Stiffness (current) constants of AMB 1 as well
as AMB 2 in the vertical (x) and horizontal (y)
directions
kP , kD, kI Proportional, Derivative as well as Integral
factors of PID controller
l, l 1 and l 2 Rotor length, disc 1 and disc 2 distances with
respect to the rotor C.G. point
m, md 1 and md 2 Mass of the rotor, disc 1 as well as disc 2
i 0 , s0 Bias current, Nominal AMB air gap
t Time
ux , uy Rotor displacements in the vertical (x) and
horizontal (y) directions

Greek Letter

( 1, β 2 )
β ( ) Disc 1 and Disc 2 unbalance phases
δx1
a
, δ ay1 and δx2
a
, δ ay2 Residual (without trial) misalignments of AMB
1 as well as AMB 2 in the vertical (x) and
( ) ( ) horizontal (y) directions
δx1
s
, δ sy1 and δx2
s
, δ sy2 Residual (without trial) misalignments of
sensors at AMB 1 and AMB 2 in the vertical
( ) ( ) (x) and horizontal (y) directions
Δrx1 , Δry1 and Δrx2 , Δry2 Rotor virtual trial misalignments at AMB 1 as
well as AMB 2 locations in the vertical (x) and
horizontal (y) directions
Identification in a Magnetically Levitated Rigid Rotor System … 3
( ) ( )
dΔx1 , dΔ y1 and dΔx2 , dΔ y2 Air gap between rotor and sensors (created
manually) at AMB 1 and AMB 2 in the vertical
(x) and horizontal (y) directions
ω Rotor spin speed

1 Introduction

Modern industries need a properly designed high-speed rotating machinery (i.e.,

pumps, turbines, compressors, aircraft gas turbine engines, etc.) for multiple appli-
cations along with effective as well as efficient performance. Rotating element, i.e.
rotors in the machines, in general have the support with conventional bearings,
such as the rolling element bearings, hydrodynamic journal bearings. However, due
to various operational constraints in these bearings, as such they are suitable for
lower or moderate speeds, require lubrication and chances of wear and tear [1], the
researchers in the present day are concentrating on active magnetic bearings (AMBs)
[2]. AMB is a non-contact kind of bearing device committed for overcoming the
performance restrictions of conventional bearings. The capabilities of AMB tech-
nology (consisting of electric magnets, proximity displacement sensors, controller
and current amplifiers [3]) include active control of rotor system dynamics through
adaptable damping and stiffness factors of the controller, lubrication free operation,
which leads to provide high speed rotation in the air, low power loss, etc. [4]. More-
over, the rotor in the AMB structure can be stably operated at multiple higher spin
speeds and for various system parameters. The AMB technology is also utilized for
identification and diagnosis of multiple faults in high-speed turbomachinery, such as
the centrifugal pumps and gas turbine engines [5, 6].
Malfunctions such as unbalance in rotor and misalignment in the coupled shaft
as well as supported bearings are the predominant causes for excessive vibrations
in a rotor system. Unbalance in a rotor may be caused from non-coinciding nature
of its inertia and geometric axes [7]. This occurs due to manufacturing defects,
heterogeneity of raw materials, maintenance issues, etc. [8]. On the other hand,
the misalignment fault may occur as a result of inappropriate assembly of several
machine components, system installation errors, etc. Moreover, this fault also arises
from uninterrupted functioning of the rotor setup as well as uneven temperature
based expansion and contraction of the rotor support structure [9]. There may be
misalignment in the coupled shaft (i.e., between two rotating shafts at coupling loca-
tion) as well as among the rotor and supported bearings, i.e., traditional bearings [9]
or active magnetic bearings [10] under different types of misalignment i.e., parallel,
angular and combined form. Under the influence of these rotor faults, there may be
drastic troubles in the operation of entire machine, which may minimize the effi-
ciency and productivity of a factory. Higher level of vibrations arising from faults
and sudden breaking of some components of the machine may also cause hazardous
accidents to the workers. To resolve these concerns and make risk-free life, it is
4 P. Kumar and R. Tiwari

essentially required to analyze the fault signatures and their effect on the system’s
dynamics. The suppression of the rotor vibrations based on balancing and aligning
the system or using AMB and identifying the faults quantitatively using appropriate
online monitoring techniques are also the crucial tasks.
In view of this, various literature surveys have been done through research papers.
Morton [11] presented a modal balancing technique in order to balance an elastic
shaft, in which he was not needed trial weights and values of bearing parameters.
It was observed through numerical simulation that the technique was very useful
for balancing purpose up to different critical speeds. A method was proposed by
Krodkiewski [12] for identification of magnitude and phase of residual unbalance
in a multi-bearing rotor system. He also included white noise in the displacement
signals and tested the method for checking its accuracy. Identification results were
found to be quite accurate and the method was highly robust. To evaluate the imbal-
ance state of a rotor system, a technique based on measuring pedestal vibration was
proposed by researchers [13]. The developed technique was noticed to be insensitive
in determining the unbalance characteristics against noise signals but the bearing and
mass parameters were slightly sensitive. A review on rotor dynamic nature, distinct
balancing and vibration controlling methods was presented by Zhou and Shi [14].
Later, Menshikov [15] considered a rotor system with two elastic bearing supports
and identified parameters associated with unbalance fault using the inverse method.
Researchers [16] utilized the equivalent load concept described in [17] to identify
single as well as double unbalances in a rotating machine. However, the considered
machine was very simple in terms of the number of rotating components. There-
after, Yao et al. [18] presented two distinct techniques to obtain optimized values of
the rotor unbalance parameters. The first technique was a combination of the modal
expansion method as well as optimization algorithm, whereas the second technique
was the incorporation of the inverse method into the first technique. In the recent
publication, a joint-input state estimation algorithm was utilized in identifying force
due to unbalance in a stiff rotor model having traditional bearing supports [19]. An
estimation methodology was developed for identifying unbalance characteristics in a
simple stiff rotor-AMBs test rig set up [20]. Afterwards, AMB was used as a vibration
controlling and fault identification device in an elastic shaft-discs-bearings system
[21]. Later, they have applied a virtual trial unbalance method for the purpose of
rotor balancing using AMB technology [22].
A long year back, the mathematical equations of coupled rotor misalignment force
as well as moment were developed by authors in [23]. In the paper, the distinct kinds
of coupling were also described and the effect of misalignment fault on these coupling
were studied in detail. Hori and Uematsu [9] investigated stability analysis of a system
consisting of two rotors (coupled together) with four number of journal bearings. The
misalignment (parallel as well as angular) was considered in the supported bearings.
The rotating system model was examined numerically by employing the transverse
matrix technique and Newton–Raphson method. Researchers [24, 25] investigated
the effects of coupling misalignment in the combined form (i.e., both lateral and
angular misalignments) on vibrations of rotating machinery. They observed that the
misalignment fault can be identified by looking into the vibration peak at twice
Identification in a Magnetically Levitated Rigid Rotor System … 5

the rotational speed. Afterwards, the consequence of lateral misalignment on the

system’s dynamics (two rigidly coupled Jeffcott rotors) was studied by Al-Hussain
and Redmond [26]. Later, Al-Hussain [26] incorporated the pure angular misalign-
ment between the two rigid rotors and generated the vibrational signature of the
system using both the Newmark and Newton Raphson techniques. [27]. Three theo-
retical AMB models based on the four, six and eight number of electromagnets were
proposed by Bouaziz et al. [28] to analyze the dynamic behavior of two degrees-of-
freedom misaligned rotor-AMBs support system. In the same line, Messaoud et al.
[29] presented an AMB integrated model for investigating rotor system dynamics
considering angular misalignment. The consequence of distinct misalignments was
also explored on vibrational response and stator current signature, in a misaligned
coupled rotor having fluid bearing supports [30]. The orbital response and Fourier
transform of both vibration and current were also effectively utilized along with
the vibration and current waveforms, in identifying the unique characteristics of
misalignment fault. Jang and Khonsari [31] described about misalignment in journal
bearings in his review paper. Further, the dimensionless vibrational behavior of a fully
floated misaligned rigid rotor-AMB system was analyzed numerically and found that
there was rapid increment in magnitudes of non-dimensional system responses with
a little enhancement in the disc’s eccentricity and AMB’s misalignment amount [32,
33]. Later, AMB technology was utilized by Srinivas et al. [34] in order to control
vibrations arising due to unbalance in the rotor as well as misalignment at coupling
position in a rotor-train model. They have also identified various system and fault
parameters using a suitable steering function in the rotor system modelling. Kuppa
and Lal [35] have also incorporated AMB as an active vibration mitigation in a
misaligned turbo-generator model, which was comprised of two coupled rigid rotors
with elastic bearing supports. With the help of a developed identification algorithm,
the disc unbalance parameters, conventional bearings and AMBs parameters, as well
as coupling’s damping and stiffness constants were also accurately estimated.
Through the ongoing research studies it has been observed that the publications
are quite available in analyzing the unbalance and misalignment faults for the system
diagnosis and their detection in a rotor model hold mostly by conventional bearings
and very less on AMBs support. Researchers have also been involved in reviewing
AMB technology and misalignment in journal bearings as well as couplings and
active balancing techniques. However, the dynamic investigation on the rotor system
levitated by misaligned AMBs and connected with misaligned sensors for measuring
displacement responses have not been addressed till now. Hence, the present article
explores the development of mathematical model of several system components,
and also the dynamic interaction among unbalance, AMBs misalignment as well as
sensors misalignment faults. Identification of sensor residual misalignments is also
accomplished employing a virtual trial misalignment (VTM) strategy.
In this paper, a novel VTM concept is developed for investigating the vibrational
effects of the unbalance as well as misalignment faults on a magnetically levitated
rotating system. A stiff and misaligned rotor mounted on two AMBs is mathemati-
cally modelled with innovative derivation of AMB force considering the misalign-
ment in AMBs and proximity sensors. The system’s dynamic equations are derived
6 P. Kumar and R. Tiwari

and resolved by developing a Simulink™ model to acquire the time series displace-
ment as well as controlling current data. Dynamic effect of AMBs as well as sensors
residual misalignments on the rotor performance has been presented in the paper.
Further, the sensors’ misalignment (residual amount) placed at the AMBs location
are also estimated utilizing the innovative concept of sensor air gaps for without trial
and additional (with trial) misalignments along with the amounts of rotor virtual trial
misalignments.

2 System Description

To examine the dynamic behavior and understand the effect of unbalance as well as
residual misalignment in a magnetically levitated system, a stiff rotor fastened with
two rigid discs at offset positions and mounted on two misaligned radial AMBs is
accounted as depicted in Fig. 1a. Due to the rigid behavior of the rotor throughout
the operation, the displacements are to be measured at the left and right positioned
AMBs (i.e., AMB 1 and AMB 2). For measuring the displacement response, the eddy
current proximity sensors are included in the rotor model at AMB places. The major
chance for AMB misalignment may be due to offsetting in sensors, which arise from
manufacturing, system’s assembly and test set up errors in the initial time. Therefore,
the misalignment of sensors placed at AMB positions is also taken into account (refer
Fig. 1b, c). Misalignment of the rotor with radial AMBs and sensors are supposed
to be combined (parallel and angular) misalignment. This is on account of different
amounts of x- and y-directional AMBs residual misalignment (δx1 a
, δ ay1 , δx2
a
and δ ay2 )
and sensors residual misalignment (δx1 , δ y1 , δx2 and δ y2 ) located at AMB 1 and AMB
s s s s

2 places. The center position of eddy current proximity sensors, actuators of AMB
and the rotor are, respectively, shown by O, A, and C. Both AMBs are assumed
to be anisotropic and dissimilar stiffness parameters. Modelling of the system also
incorporates the gyroscopic effect due to offset discs. Moreover, the effect of leakage
in the magnetic flux as well as loss of eddy current are also considered in the AMB
system to align with the practical situation. A proportional-integral-derivative (PID)
controller is utilized in AMBs for the rotor stability and as an agent for reducing its
undesirable vibration.

3 Mathematical Modelling of the System

This section explains the mathematical model involved in completely describing

the unbalanced as well as misaligned stiff rotor system presented in Fig. 1a. The
whole modelling includes the force model due to disc unbalance, force coming
from misaligned AMB in the presence of offsets in sensors and inertia force model.
Based on the rigid nature of the rotor during operation, its translational and angular
deflections at C.G. point can be expressed as a function of translational deflection at
Identification in a Magnetically Levitated Rigid Rotor System … 7

Fig. 1 a A rigid rotor-misaligned AMB model fixed with sensors in the x–z plane b, c residual
misalignment of AMBs (δx1a , δ a , δ a , δ a ) and sensors (δ s , δ s , δ s , δ s ) with the rotor at AMB 1
y1 x2 y2 x1 y1 x2 y2
as well as AMB 2 locations in the x–y plane

AMB 1 as well as AMB 2 sites following Fig. 2. In Fig. 2, A1 and A2 denote for
AMB 1 as well as AMB 2 locations, D1 and D2 for Disc 1 and Disc 2 positions and
G for the location of rotor center of gravity.
The relation for displacements can be expressed as

u x =a 2 u x1 + a 1 u x2 ;u y = a 2 u y1 + a 1 u y2 ; ϕ y
( )
= (−u x1 + u x2 )/l;ϕx = u y1 − u y2 /l (1)

Fig. 2 Translational as well as angular deflections of rigid shaft at various locations in a x–z plane
b y–z plane
8 P. Kumar and R. Tiwari

with
a1 a2
a1 = ; a2 =
l l
where the symbols ϕ y and ϕ x represent for the rotor angular deflections in the x–z
and y–z planes. The symbols a1 and a2 are the distances of C.G. from left AMB and
right AMB, respectively.

3.1 Modelling of Unbalance Force

The x- and y-directional unbalance force due to Disc 1 and Disc 2 can be written as

f unbx1 = m d1 e1 ω2 cos(ωt + β1 ); f unby1 = m d1 e1 ω2 sin(ωt + β1 )

f unbx2 = m d2 e2 ω2 cos(ωt + β2 ); f unby2 = m d2 e2 ω2 sin(ωt + β2 ) (2)

where md1 and md2 are, respectively, the masses of Disc 1 as well as Disc 2. The
unbalance eccentricities and phases of Disc 1 and Disc 2 are represented by (e1 , β 1 )
and (e2 , β 2 ). The symbol ω is the rotor spin speed.

3.2 Modelling of Misaligned AMB Force Considering Offsets

in Sensors

For perfect alignment case (in which the rotor, sensors and AMB axes are coinciding,
i.e. the points O, A and C are at the same point in Fig. 1b, c and an equal gap of
air is available in the lower pole as well as upper pole of AMB as well as the back
and front side poles of AMBs), Schweitzer and Maslen [36] have given the below
expression for AMB force (assuming AMB 1 and x-direction force) as
( )
(i 0 + i x1 )2 (i 0 − i x1 )2
f x1 = k1 − (3)
(s0 − u x1 )2 (s0 + u x1 )2

The linearized form of Eq. (3) is expressed as

f x1 = ksx1 u x1 + ki x1 i x1 (4)

where the stiffnesses k s1 and k i1 are indicated as

4k1 i 02 4k1 i 0 1 α
ks1 = ;ki1 = 2 ; k1 = μ0 ηN12 Aa1 cos (5)
s03 s0 4 2
Identification in a Magnetically Levitated Rigid Rotor System … 9

Here, η is the constant term indicating for consideration of magnetic flux leakage,
eddy current loss, etc. [37]. The magnetic pole area, number of turning coils and angle
between two consecutive poles in AMB are denoted by Aa1 , N 1 and α, respectively.
The symbol μ0 is the vacuum permeability of free space having 4π × 10–7 H/m
value. With different pole area and number of turning coils (i.e., Aa2 and N 2 ), the
stiffness coefficients of AMB 2 will also change. Moreover, the AMB 1 force in the
vertical (y) direction (i.e., f y1 ) will follow Eq. (4) with replacement of k sx1 by k sy1
and k ix1 by k iy1 due to its anisotropic nature, ux1 by uy1 and ix1 by iy1 .
However, in a real practice, it is very challenging to acquire a perfectly balanced
and aligned rotor system. Even if precise alignment is assured, it cannot be sustained
for an extensive period. Misalignment fault in the system can exist while machine’s
operation owing to various external aspects, such as disruption in the base support
structure, thermal expansion and contraction of machine components, etc. [38]. In a
fully floated rotor-AMB apparatus, the sensor measurement errors (due to its offset
placed position presented in Fig. 1b, c) while locating the centre position of AMB
may also cause the rotor misalignment with the supported AMB [39]. Following this,
Fig. 3 presents the cross-sectional view of rotor and misaligned sensors placed at
qth AMB site in the x–y plane, for the case of residual misalignment and additional
trial misalignment (in which the virtual trial misalignment is provided to the rotor
in addition to sensor residual misalignments). In Fig. 3b, the amounts Δrxq and Δryq
are the user-provided virtual trial misalignments to the rotor using AMB bias current
in the vertical (x) and horizontal (y) directions, respectively (this will be discussed
more briefly in Sect. 3.3.1). The points (1, 2) in Fig. 3a and points (1’, 2’) in Fig. 3a
situated at the shaft represent for the points from where the transverse translational
displacements can be measured by residually and additionally misaligned proximity
sensors.

Fig. 3 Sectional view of the shaft and misaligned sensors available at qth AMB site (where q = 1
for AMB 1, and q = 2 for AMB 2) a residual misalignment (δxq
s , δ s ) b additional trial misalignment
yq
s + Δr , δ s + Δr )
(δxq xq yq yq
10 P. Kumar and R. Tiwari

Owing to misalignment of AMB in the nth direction (where n = 1, 2 as indicated

in Fig. 3a by dashed line), the gap between the rotor and qth AMB is modified, which
gives rise to the amount (s0 − δnqa
) as the lower gap and (s0 + δnq
a
) as the upper gap,
respectively. By putting these updated gaps into Eq. (3), the force in the nth direction
from residually misaligned AMB can be formulated as
( ( ) ( ) )
m 2 m 2
i 0 + i nq i 0 − i nq
m
f nq = kq ( ) −( ) ; n = 1, 2 (6)
a − um 2
s0 − δnq a + um 2
s0 + δnq
nq nq

where ‘m’ superscript represents the condition for residually misaligned AMB. The
displacement of the rotor detected from qth misaligned (residually) sensors at the
positions ‘n’ (i.e., n = 1, 2 in Fig. 3a) is shown by u m nq . Current output of the PID
controller due to displacement u m m
nq as input is indicated with i nq . Further, considera-
tion of less vibration as compared to the air gap in the lower and upper poles, i.e. u m nq
<< (s0 − δnq a
) and u m nq << (s0 + δnq ) and negligence of the higher-order terms, such as
a

(u m
nq ) , (i nq ) , u nq (i nq ) and i nq u nq , the linearized form of Eq. (6) is given as
2 m 2 m m 2 m m

m
f nq = ksnq
m
nq + kinq i nq + f cnq
um m m m
(7)

with
( )
ksnq kinq 1 + δnq
2
f q δnq
m
ksnq =( ) ;
m
kinq = ( ) ;
m
f cnq =( ) ;
2 2
1 − δnq 2 2
1 − δnq 1 − δnq 2 2

4kq i 02 δnq
a
fq = ; δnq = (8)
s02 s0

Using Eq. (7) for (n = 1, 2) and Fig. 3a, the misaligned AMB forces ( f xq
m m
, f yq )
can be expressed in the vector form as
( ) [ ]( ) [ ]( ) ( )
m
f xq [ ] ks1q
m
0 um [ ] k m
0 i m
f m
= Sq m
1q
+ Sq i1q 1q
+ cxq
(9)
m
f yq 0 ks2q um
2q
m
0 ki2q m
i 2q m
f cyq

with
[ ] ( m1 ) ( )
[ ] cos θ1q sin θ2q f cxq [ ] f c1qm1
Sq = ; = Sq
sin θ1q cos θ2q m1
f cyq m1
f c2q

where [S q ] denotes for the transformation matrix at qth AMB site [40]. Further, with
m m m m
the help of Fig. 3a, the AMBs’ stiffness constants (ks1q , ks2q , ki1q , ki2q ) and the rotor
m m m m
displacements and currents (u 1q , u 2q , i 1q , i 2q ) can be computed with reference to
x- and y-coordinate axes as below
Identification in a Magnetically Levitated Rigid Rotor System … 11
( ) ( ) ( ) ( ) ( )
m
ks1q [ ] ksxq
m m
ki1q [ ] kimxq um
m = Rq ; m = Rq ; 1q
ks2q m
ksyq ki2q kimyq um
2q
( ) ( ) ( ) [ ]
[ ] u mxq m
i 1q [ ] i xq m [ ] cos θ1q sin θ1q
= Rq ; = Rq , Rq =
u myq m
i 2q m
i yq sin θ2q cos θ2q
(10)

On substituting the relations of Eq. (10) into Eq. (9) and afterwards on simplifi-
cation, the compact form of misaligned force vector due to qth AMB (i.e., q = 1 for
AMB 1 and q = 2 for AMB 2) is written as

Bq (t) = Ksθ q η AM Bq (t) + Kiθq icq (t) + fcq

m m m m m m
f AM (11)

with
( ) ( ) ( ) ( )
m
f xq u mxq m
f cxq im
m
f AM Bq (t)= m ; η AM Bq (t) =
m
m ; fcq =
m
m ; icq (t) = xq
m
m ;
f yq u yq f cyq i yq
[( )( )]
m
a1q ksxq + b1q ksyq
m m
b1q ksxq + c1q ksyq
m f q δxq
m
Ksθ = ( ) ( ) ; f cxq
m
=( ) ;
q m
b1q ksxq + c1q ksyq
m m
c1q ksxq + d1q ksyq
m
1 − δxq 2 2
⎡( )( )⎤
a1q kimxq + b1q kimyq b1q kimxq + c1q kimyq f q δ yq
m
Kiθ q = ⎣( )( ) ⎦; f cyq
m
=( )2 ;
b1q ki xq + c1q ki yq
m m
c1q ki xq + d1q ki yq
m m
1 − δ 2yq
a1q = cos3 θ1q + sin 3 θ2q ; b1q = sin θ1q cos2 θ1q + sin2 θ2q cos θ2q ;
c1q = sin2 θ1q cos θ1q + cos2 θ2q sin θ2q ; d1q = sin 3 θ1q + cos3 θ2q (12)
( )
ksxq ki xq 1 + δxq
2
ksyq
m
ksxq =( ) ; kimxq = ( )2 ; ksyq = (
m
)2 ;
1 − δxq2 2 1 − δxq
2 1 − δ 2yq
( )
ki yq 1 + δ 2yq δxq
a
δ ayq
kimyq = ( )2 ; δxq = ; δ yq = (13)
1 − δ 2yq s0 s0

In Eq. (11), the rotor displacement and AMB’s control current vectors at the qth
AMB site are represented by ηmAM Bq and icq m
. Further, the mathematical expression of
m
the control current vector icq , for the condition of misalignment in AMB and sensors
can be written as
[ ]
k 0 kI 0 kD 0 { m m ∫ m ∫ }T
m
icq (t) = − P u xq u yq u xq dt u myq dt u̇ mxq u̇ myq (14)
0 kP 0 kI 0 kD
12 P. Kumar and R. Tiwari

3.3 Modelling of Misaligned Sensors

This section explores the mathematical strategies for evaluating the angles (θ 1q ,
θ 2q ) for sensors residual misalignment and (θ 3q and θ 4q ) for their additional (trial)
misalignments as well as identifying their residual misalignments (δxq s
, δ syq ). Using
Fig. 3a, the following relationships i.e., Eqs. (15) and (16) can be established.
( ) )
EC = EF+FC ⇒ r =dδxq +r cosθ1q ⇒ dδxq =r 1 − cosθ1q
( ) (15)
GC = GH+HC ⇒ r =dδ yq +r cosθ2q ⇒ dδ yq =r 1 − cosθ2q
δ syq δxq
s
sin θ1q = ; sin θ2q = (16)
r r
Similarly, assisted by Fig. 3b, the exhibiting relations are
( ) ( ) ( ) ⎫
E ' C ' =E ' F ' +F ' C ' ⇒ r = dδxq + dΔxq +r cosθ3q ⇒ dδxq + dΔxq =r 1 − cosθ3q ⎬
( ) ( ) ( ) (17)
G ' C ' =G ' H ' + H ' C ' ⇒ r = dδ yq + dΔ yq +r cosθ4q ⇒ dδ yq + dΔ yq =r 1 − cosθ4q ⎭

δ syq + Δryq δxq

s
+ Δrxq
sin θ3q = ; sin θ4q = (18)
r r
Here, ‘r’ is the shaft radius. Subsequently, the subtraction of Eq. (17) from Eqs. (15)
and (18) from Eq. (16), respectively, give the next equations as Eqs. (19) and (20)

dΔxq dΔ yq
cos θ1q − cos θ3q = and cos θ2q − cos θ4q = (19)
r r
Δryq Δrxq
sinθ3q − sin θ1q = and sinθ4q − sin θ2q = (20)
r r
The distances (δ xq + dδ xq ) and (δ yq + dδ yq ) in Fig. 3a are, respectively, the vertical
and horizontal directions air gap of the sensors with the rotor without trial misalign-
ment, whilst (δ xq + dδ xq + dΔxq ) and (δ yq + dδ yq + dΔyq ) in Fig. 3b are the gaps
with consideration of trial misalignment together with sensors residual misalignment.
These air gaps are well known parameters set by a user for detecting the vibrational
displacement at qth AMB location. Subtraction of air gaps for with and without
trial misalignments would provide, respectively, the quantities dΔxq and dΔyq , that
are required in Eq. (19). Furthermore, the values of trial misalignments Δrxq and
Δryq needed in Eq. (20) are also known quantities. The method for calculating these
values will be elaborated in Sect. 3.3.1. Afterwards, the θ 1q and θ 2q magnitudes can
be acquired by solving Eqs. (19) and (20) as below
( ) (( )2 ( )2 )
−1 dΔxq −1
dΔxq + Δryq
θ1q = tan − sin and
Δryq 2r 2
Identification in a Magnetically Levitated Rigid Rotor System … 13

Apply VTM Subtract air

Use equations (21)
Measure initial sensor approach Measure final sensor gap
and (16) to identify
Start gap for residual offset gap for additional
the residual offset
of sensors offset of sensors
amounts of sensors

Fig. 4 Flow diagram describing the procedure to identify sensors residual misalignments

( ) (( )2 ( r )2 )
dΔ yq dΔ yq + Δxq
θ2q = tan−1 − sin −1
(21)
Δrxq 2r 2

Moreover, the angles θ 3q and θ 4q in Fig. 3b can be obtained by using Eq. (19).
Further, the residual misalignments of sensors (δxq s
, δ syq ) can be identified by substi-
tuting the values of θ 1q and θ 2q into Eq. (16). For more clarification in the procedures
to identify residual misalignments of sensors located at qth AMB location, the flow
chart diagram is given in Fig. 4.

3.3.1 Virtual Trial Misalignment (VTM) Concept

In this section, a novel VTM technique is explored to provide known misalignments

(i.e., Δrxq and Δryq ) to the rotor, so that the center position of rotor gets shifted from
the point (C) in Fig. 3a to the point (C’) in Fig. 3b. These trial misalignments give
rise to additional trial misalignments of sensors (located at qth AMB position) with
respect to the rotor center, i.e. the distances δxq
s
+Δrxq , δ syq +Δryq in Fig. 3b. The word
used here as virtual is due to shifting of rotor relative to AMB current position using
trial bias current. Moreover, the trial misalignments among the rotor and AMBs as
well as the rotor and sensors can also be created by physical methods as elaborated
in Kumar and Tiwari [39]. However, this physical technique is less effective, time-
consuming and more laborious in comparison to the virtual method. Therefore, it
is good to focus on the VTM concept. In this process, the vertical and horizontal
directions trial magnetic forces (i.e., f Txq and f Tyq ) derived from bias currents are
initiated at the qth AMB site to acquire the amounts of the vertical and horizontal
trial misalignments. These additional forces serving as virtual trial misalignment
(VTM) excitation forces, are developed besides controlling forces of AMBs, which
are expressed as below

f T xq = kiTxq i T xq ; f T yq = kiTyq i T yq (22)

with
( ) ( )
4kq i 0 + i T xq 4kq i 0 + i T yq
kiTxq = ; kiTyq = (23)
s02 s02
14 P. Kumar and R. Tiwari

Here, the terms kiTxq and kiTyq appeared after supplying extra bias currents (i.e., iTxq
and iTyq ) to AMB for developing trial misalignments, virtually. Moreover, the values
of trial misalignments are to be known by equating Eq. (22) with the trial constant
forces of misaligned AMB [41], as

f q Δxq f q Δ yq
f T xq = kiTxq i T xq = ( )2 ; f T yq = ki yq i T yq = (
T
)2 ;
1 − Δxq2 1 − Δ2yq
4kq i 02 Δrxq Δryq
fq = ; Δ xq = ;Δ yq = (24)
s02 s0 s0

Further, the next equation (Eq. (25)) can be obtained by substituting the terms
(kiTxq and kiTyq ) of Eq. (23) into Eq. (24) as

i 2 Δxq i 02 Δ yq
i T2 xq + i 0 i T xq + ( 0 )2 = 0; i 2
T yq + i i
0 T yq + ( )2 = 0 (25)
1 − Δ2xq 1 − Δ2yq

With the known quantities of these four parameters, such as trial bias currents (iTxq ,
iTyq ), nominal bias current and AMB air gap (i0 and s0 ), Eq. (25) helps in evaluating
the values of rotor vertical and horizontal trial misalignments (i.e., Δrxq and Δryq ).
These amounts will be useful in identifying the sensors’ residual misalignments as
elaborated in Sect. 3.3.

3.4 Dynamic Equations of the Considered Rotating System

Dynamic equations of the considered magnetically levitated rotating machine as

shown in Fig. 1a have been derived by using the moment equilibrium method [42].
For this, the moment due to unbalance forces, misaligned AMB forces considering
offsets in sensors, inertia force and the disc’s gyroscopic effect in the x–z plane and
y–z plane have taken about AMB 1 as well as AMB 2 positions. The equations of
motion (EOM) of the rotor model in the matrix form can be expressed as

MΔq̈ m (t) − ωGΔq̇m (t) = funb + f AM

m
B (26)

where the mass matrix M and gyroscopic matrix G are

⎡ ( 2 ) ( ) ⎤
ma 2 + i d 0 ma 1 a 2 − i d 0 ⎡ ⎤
⎢ ( ) ( ) ⎥ 0 ip 0 −i p
⎢ 0 ma 2 +i 0 ma a − i ⎥ ⎢ ⎥
⎢ 2 d
( ) 1 2 d ⎥ ⎢ ip 0 −i p 0 ⎥
M=⎢(
⎢ ma 1 a 2 − i d
) ⎥; G = ⎢
⎥ ⎣
⎥
⎦
(27)
⎣ 0 ma 2 +i 0 0 −i 0 i
) ⎦
d p p
( )
1 (
0 ma 1 a 2 − i d 0 ma 21 + i d −i p 0 i p 0

with
Identification in a Magnetically Levitated Rigid Rotor System … 15
( )
Id I P1 + I P2 l1 l2
id = ; iP = ; l1 = ; l2 =
l2 l2 l l

The rotor displacement vector (i.e., Δqm (t)) is given by

{ }T
Δqm (t) = u mx1 u my1 u mx2 u my2 (28)

The unbalance force vector due to Disc 1 and Disc 2 can be written as
⎧ ( ) ( )⎫
⎪
⎪ m d1 e1 ω2 cos(ωt + β1 )(a 2 + l 1 ) + m d2 e2 ω2 cos(ωt + β2 )( a 2 − l 2) ⎪⎪
⎨ ⎬
m d1 e1 ω2 sin(ωt + β1 )(a 2 + l 1 ) + m d2 e2 ω2 sin(ωt + β2 ) (a 2 − l 2 )
funb =
⎪
⎪ m e ω cos(ωt + β1 )(a 1 − l 1 ) + m d2 e2 ω cos(ωt + β2 )( a 1 + l 2) ⎪
2 2
⎪
⎩ d1 1 2 ⎭
m d1 e1 ω sin(ωt + β1 ) a 1 − l 1 + m d2 e2 ω2 sin(ωt + β2 ) a 1 + l 2
(29)

The misaligned AMB force vector considering offsets in sensors is given as

{ }T
B =
m m m m m
f AM f x1 f y1 f x2 f y2 (30)

m m m m
The force elements ( f x1 , f y1 ) and ( f x2 , f y2 ) in the vector (i.e., Eq. (30)) will follow
Eq. (11) for (q = 1, i.e. AMB 1) and (q = 2, i.e. AMB 2), respectively. Equation (26)
can be solved by making SIMULINK™ model in MATLAB software. This will
give the time dependent displacement and current signals at AMB 1 and AMB 2
locations as an output. The next section explores the numerical technique to obtain
the solution of EOM and investigate the effect of unbalance, AMBs misalignment
as well as sensors misalignment on the dynamic performance of the rotor system.
In the coming section, the values of sensors residual misalignments located at both
AMBs positions are also identified with the proposed procedure in Sect. 3.3.

4 Results Analysis and Identification

This part of the paper discusses the generation of displacement and controlling
current signals of the faulty rigid rotor model (follow Fig. 1a) at both AMB locations.
Utilizing the developed SIMULINK™ model as shown in Fig. 5, Eq. (26) has been
computationally solved to obtain the rotor displacement response (i.e., u mx1 and u my1 in
the x- and y-directions at AMB 1 position as well as u mx2 and u my2 at AMB 2 position)
m m
and controlling current of the PID controller (i.e., i x1 and i y1 at AMB 1 position as
m m
well as i x2 and i y2 at AMB 2 position). Figure 5 presents several blocks which perform
various functions in which the clock is the first block. This block executes the time
taken in simulation. The next triangular type block is the gain block, which is used for
the multiplication purpose. Following this, the add block performs the summation of
16 P. Kumar and R. Tiwari

Fig. 5 SIMULINK™ model of the considered rigid rotor model

unbalance force vector (i.e., Eq. (29)), misaligned AMB force vector (i.e., Eq. (30))
and force vector due to gyroscopic couple effect (i.e., ωGΔq̇m (t) part of Eq. (26)).
The two integrator blocks are placed one after another to obtain the velocity and
displacement of the rotor model. Further, taking the displacement response as input,
the PID block is used for obtaining the controlling current response of the system.
Lastly, the workspace block (i.e., Displacement block and Current block) preserves
the time domain displacement and current signal data for plotting multiple plots as
required by a user [7, 43].
The numerical simulation was executed for 5 s using the fault and system char-
acteristics displayed in Table 1. The displacement and current signals have been
generated at AMB 1 and AMB 2 locations (i.e., solution of Eq. (26)) with the help of
Runge–Kutta 4th order method with fixed step size of time 0.0001 s. The time-domain
plots for these system responses in the vertical (x) and horizontal (y) directions at
35 Hz angular frequency of the shaft are shown in Figs. 6 and 7, in which Fig. 6
represents for AMB 1 location and Fig. 7 illustrates for AMB 2 location. The nature
of the displacement and current signals are noted to be sinusoidal nature. It is due to
the sinusoidal nature of unbalance force f unb (i.e., Eq. (29)) on the right-hand side of
Eq. (26). However, there is a shift of signals from zero-mean position in the vertical
(x) and horizontal (y) directions current responses in Figs. 6c, d and Figs. 7c, d, which
is not the case for the displacement responses either at AMB 1 location or AMB 2
location. This happens because of the vertical (x) and horizontal (y) directions AMB
m m
constant forces, f cxq and f cyq in misaligned AMB force (i.e., Eq. (11)), which have
appeared in the force derivation considering misalignment in AMBs and sensors.
These constant forces also cause to move the displacement signals; however, these
signals get compensated from k I component of the PID. This integral factor (k I )
brings them to the mean position by generating a non-zero mean biased controlling
currents.
Further, Fig. 8 has been plotted to explore the dynamic effect of AMBs and sensors
residual misalignments on the vibrating phenomena of the rigid rotor. This figure
contains three kinds of orbital plots in each subplot i.e., Figs. 8a–d. Figures 8a, b
represent for the rotor displacement and current orbits at AMB 1 location, whereas the
displacement as well as the current orbits at AMB 2 location are shown by Fig. 8c, d,
Identification in a Magnetically Levitated Rigid Rotor System … 17

Table 1 Assumed values of parameters related to faulty rigid rotor model

Shaft, disc and unbalance fault parameters Values
Mass of the rotor, m 3.91 kg
Disc mass, md 1 , md 2 1.065 kg, 2.081 kg
Unbalance eccentricity, e1 , e2 50 μm, 80 μm
Unbalance phase, β 1 , β 2 10 deg, 30 deg
Distances of disc 1 and disc 2 from center of 0.126 m, 0.074 m
gravity of the rotor, l 1 , l 2
Distances of AMB 1 and AMB 2 from center of 0.226 m, 0.174 m
gravity of the rotor, a1 , a2
Diameter of the shaft, d 17.6 mm
Length of shaft, l 0.4 m
Diametral mass moment of inertia of the rotor, 0.0455 kg-m2
Id
Polar mass moment of inertia of discs, I p1 , I p2 0.0019 kg-m2 , 0.0059 kg-m2
Parameters of AMBs and controller Residual and trial misalignments, additional
air gaps and trial bias currents
k sx 1 156,735 N/m kP 6000 A/m δx1
a 0.140 mm dΔx 1 0.050 mm
k sy1 175,500 N/m kD 3 A-s/m δ ay1 0.160 mm dΔy1 0.070 mm
k sx 2 329,139 N/m kI 3000 A/m-s δx2
a 0.150 mm dΔx 2 0.060 mm
k sy2 344,700 N/m s0 0.400 mm δ ay2 0.145 mm dΔy2 0.055 mm
k ix 1 31.35 N/A i0 2A Δrx1 0.100 mm iTx 1 0.462 A
k iy1 32.58 N/A N1 100 Δry1 0.130 mm iTy1 0.620 A
k ix 2 65.83 N/A N2 120 Δrx2 0.120 mm iTx 2 0.565 A
k iy2 67.86 N/A α 45 deg Δry2 0.110 mm iTy2 0.512 A
* System characteristics of Table 1 are detailed in Nomenclature

respectively. Here, the black colour plots depict for displacement and current orbits
considering misalignments in the supported AMBs as well as proximity sensors.
Similarly, the red colour displacement as well as controlling current orbital responses
exhibit for the misalignment in AMBs and without sensors’ misalignment. Whilst
the displacement and current orbits at both AMBs places in the perfect alignment
state (i.e., neither AMBs nor sensors are misaligned with the rotor) is illustrated by
blue colour plots. For a better explanation of Fig. 8, it would be good to label the
condition of misalignments, i.e. the perfectly aligned rotor-sensor-AMB system as
the first level, alignment in sensors and misalignment in AMBs as the second level,
and misalignment in AMBs as well as sensors as the third level. It is observable from
Fig. 8 that the displacement as well as controlling current orbits at both AMBs places
are noticed to be quite elliptical in the availability of both disc unbalance forces
and residual misalignments of AMBs as well as proximity sensors, i.e. the third
level of misalignment. This manifests due to the overshadowing nature of combined
18 P. Kumar and R. Tiwari

Fig. 6 Time domain signals at 35 Hz angular frequency a, b vertical (x) and horizontal (y) directions
displacement responses at AMB 1 c, d vertical (x) and horizontal (y) directions current responses
at AMB 1

Fig. 7 Time domain signals at 35 Hz angular frequency a, b) vertical (x) and horizontal (y) direc-
tions displacement responses at AMB 2 c, d vertical (x) and horizontal (y) directions current
responses at AMB 2

misalignment fault (i.e., amalgamation of lateral and angular misalignments) in the

supported AMBs and sensors upon the unbalance force.
Moreover, the orbital responses for displacement as well as current in Figs. 8a–d
are also not pure circular for the first misalignment level. It is due to the anisotropic
nature of AMBs and the different stiffness constants of both AMBs. For the first
level, the plots have been obtained by putting zero in place of the values of θ 1q and
θ 2q in Eq. (12) (this makes perfect alignment in sensors) and taking AMBs residual
misalignments (δxq a
, δ ayq where q = 1, 2) equal to zero. Orbital plots for the second
Identification in a Magnetically Levitated Rigid Rotor System … 19

Fig. 8 Orbital plots for exploring the effect of residual misalignments of AMB and sensor at 35
Hz frequency a, b displacement and controlling current orbits at AMB 1 c, d displacement and
controlling current orbits at AMB 2

level of misalignment are generated by taking only the values of θ 1q and θ 2q equal
to zero. The size of the orbits is also observed to be enhanced as the misalignment
value elevates from the first condition to the third condition. Furthermore, all orbit
plots are perceived to be stable as their size is not increasing with respect to time.
Apart from these discussions of Fig. 8, Tables 2 and 3 have been provided which
explore the maximum values of the vertical (x) and horizontal (y) directions rotor
displacements and currents at AMB 1 and AMB 2 positions for all three levels of
misalignment. The percentage increase in these amplitudes in the third misalignment
condition relative to the second and first levels are also displayed in Tables 2 and
3. At 35 Hz of rotor angular frequency, the maximum values of the vertical (x) and
horizontal (y) directions rotor displacement for the first, second and third levels at
20 P. Kumar and R. Tiwari

Table 2 Maximum values of displacement and their percentage increase at AMB 1 as well as AMB
2 sites for the different misalignment levels
Position Maximum 1st level 2nd level 3rd level % increase in 3rd level % increase
amplitude in misalignment misalignment misalignment from 2nd level (%) in 3rd level
the time from 1st
domain level (%)
AMB 1 x-displacement 3.92 × 10–5 6.06 × 10–5 7.47 × 10–5 23.27 90.56
(m)
y-displacement 3.57 × 10–5 5.21 × 10–5 8.70 × 10–5 66.99 143.70
(m)
AMB 2 x-displacement 1.90 × 10–5 3.02 × 10–5 9.19 × 10–5 204.30 383.68
(m)
y-displacement 1.66 × 10–5 3.15 × 10–5 5.62 × 10–5 78.41 238.55
(m)

AMB 1 are obtained to be (3.92 × 10–5 m, 3.57 × 10–5 m), (6.06 × 10–5 m, 5.21
× 10–5 m) and (7.47 × 10–5 m in the x-direction, 8.70 × 10–5 m in the y-direction),
respectively. Similarly, at AMB 2 position, the peak values of the rotor displacement
in the x- and y-directions for all three conditions (i.e., first to third) are, respectively,
(1.90 × 10–5 m, 1.66 × 10–5 m), (3.02 × 10–5 m, 3.15 × 10–5 m) and (9.19 × 10–5 m,
5.62 × 10–5 m). The percentage increase in the x-directional displacements at AMB
1 and AMB 2 positions for the last two conditions i.e., second and third levels
are (54.59%, 90.56%) and (58.95%, 383.68%) relative to perfectly aligned state,
whereas (143.70%, 45.93%) and (238.55%, 89.76%) are the percentage increase in
the y-direction displacements for the third and second levels with respect to the first
level, at AMBs locations. However, in comparison to AMBs residual misalignment
only (i.e., no sensors offset), the x-displacement enhances by (23.27% at AMB 1,
204.30% at AMB 2) and the y-displacement enhances by (66.99%, 78.41%) for the
residual misalignment in AMBs and sensors (i.e., the third condition), as displayed
in Table 2.
Similarly, from Table 3, it can be depicted that the maximum values of the
vertical (x) and horizontal (y) directions controlling current at AMB 1 position are
perceived to be (0.24 A, 0.21 A) for the first condition, (0.61 A, 0.58 A) for the second
condition and (0.70 A, 1.35 A) for the last condition of misalignment. At AMB 2
location, the absolute peak values of the current response in the x- and y-directions
for the first, second and third misalignment levels are found to be (0.11 A, 0.10 A),

Table 3 Highest values of controlling current and their percentage increase at AMB 1 as well as
AMB 2 sites for the different misalignment levels
Position Maximum 1st level 2nd level 3rd level % increase % increase
amplitude in the misalignment misalignment misalignment in 3rd level from in 3rd level from
time domain 2nd level 1st level
AMB 1 x-current (A) 0.24 A 0.61 A 0.70 A 14.75% 191.67%
y-current (A) 0.21 A 0.58 A 1.35A 132.76% 542.86%
AMB 2 x-current (A) 0.11 A 0.44 A 0.95 A 115.91% 763.64%
y-current (A) 0.10 A 0.45 A 0.48 A 6.67% 380.00%
Identification in a Magnetically Levitated Rigid Rotor System … 21

(0.44 A, 0.45 A) and (0.95 A, 0.48 A), respectively. Moreover, the maximum value
of percentage increase in the controlling current is observed to be 763.64% when
the x-directional current response at AMB 2 position for the third level is calculated
relative to the perfectly aligned condition. One of the remarkable observation can
be made from Tables 2 and 3 that the rotor displacement and current peak values
in the vertical (x) and horizontal (y) directions at AMB 1 and AMB 2 locations
are in the same order. It means that the higher value of x-displacement at AMB 1
(i.e., 7.47 × 10–5 m) gives the maximum value of x-current (i.e., 0.70 A) at AMB
1 position. In the same way, the higher value of displacement at AMB 2 (i.e., 9.19
× 10–5 ) gives the high value of controlling current (i.e., 0.95 A) at the same AMB
location. The similar pattern is being followed for other misalignment conditions as
well as in the y-direction at both AMBs locations. This is due to straightway calcula-
tion of the controlling current from the rotor displacement as presented in Eq. (14).
These enhancements in the rotor displacement response and more controlling current
consumption in the time domain plots as well as the orbital form represent the severity
of AMBs and sensors’ residual misalignment fault in a magnetically levitated rigid
rotor-AMB system. The discussed signatures in the displacement as well as current
signals will be able to ascertain the possible condition of rotating machinery equipped
with AMBs, as every fault has a specific signature in the measured signal [1].
Apart from investigating the dynamic effect of unbalance and AMBs as well as
sensors residual misalignments on the rotor system, the residual misalignments of
sensors located at AMB 1 and AMB 2 positions for measuring the rotor displace-
ments are also identified using Eqs. (21) and (16) (as elaborated in the flow chart of
Fig. 4). The identification of sensors residual misalignments uses the data of differ-
ence between measured sensors gaps for residual and additional trial misalignment
conditions (i.e., dΔx1 and dΔy1 at AMB 1 location, dΔx2 and dΔy2 at AMB 2 location)
and rotor virtual trial misalignments (Δrx1 , Δry1 , Δrx2 and Δry2 ) as given in Table 1. The
estimated values of the vertical (x) and horizontal (y) directions residual misalign-
ments of sensors available at both AMBs locations are presented in Table 4. It would
be very useful for a user to know these misalignments so that the proper alignment
can be performed in sensors while installation and time-to-time for measuring the
correct response signals at the desired spin speed of the rotor [44–46].

Table 4 Identified values of residual misalignments of sensors located at AMB 1 and AMB 2
positions
Parameters related to residually misaligned sensors Identified values
Vertical (x) direction residual misalignment of sensor at AMB 1 position, δx1
s 4.50 × 10–3 m
Horizontal (y) direction residual misalignment of sensor at AMB 1 position, 2.80 × 10–3 m
δ sy1
Vertical (x) direction residual misalignment of sensor at AMB 2 position, δx2
s 3.20 × 10–3 m
Horizontal (y) direction residual misalignment of sensor at AMB 2 position, 3.70 × 10–3 m
δ sy2
22 P. Kumar and R. Tiwari

Hence, the novelty of the work is in the developed virtual trial misalignment
(VTM) strategy provided to the rotor using AMB. Through that concept, this paper
is capable to study the interaction between the rotor unbalance fault and misalignment
fault in the supported AMBs and proximity sensors. Moreover, the identification of
sensors residual misalignments is also possible assisted by VTM technique.

5 Conclusions

In the present work, a virtual trial misalignment (VTM) approach has been proposed
to analyse the effects of unbalance and AMBs and sensors residual misalignments
on a rotating machinery integrated with active magnetic bearings. For executing
this motive, a rigid rotor linked with two discs (at offset positions) mounted on two
misaligned AMBs at the end locations are considered and mathematically modelled.
Derivation for the AMB force in the presence of its own misalignment and sensors
misalignment is innovatively performed. Further, the dynamic equations of motion
of the complete rotor model are obtained utilizing the moment equilibrium method,
which consists of the moment due to unbalance force, misaligned AMB force, inertia
force and moment as well as gyroscopic moments coming from both discs tied to
the shaft at offset positions. Following this, a Simulink model (in MATLAB™) has
been built to determine the solutions of EOMs and generate the time domain system
responses i.e., displacement as well as controlling current at both the supported
AMB locations. Dynamic effect of AMB and sensor residual misalignment on the
vibrating nature of the rotor is also presented considering three misalignment levels
in which the first level includes a perfectly aligned state, the second is only AMB
residual misalignment but no sensors offsets and the third condition is where both
AMB and sensor are residually misaligned. The orbital size of the rotor displacement
and controlling current signals at both AMB locations are observed to be decreased
with reduction in the misalignment level from the third level (i.e., highest level of
misalignment) to the first level. Afterwards, the residual misalignments of the eddy
current proximity sensors are also identified with the help of mathematical modelling
of misaligned sensors and virtual trial misalignments of the rotor. The experimental
investigation of the presented analysis and identification derived from VTM based
misaligned AMBs and sensors methodology would be a good future scope of work.

References

1. Tiwari R. Rotor Systems: analysis and Identification. Boca Raton: CRC Press; 2017.
2. Schweitzer G. Active magnetic bearings-chances and limitations. In: 6th International
conference on rotor dynamics; 2002. pp. 1–14.
3. Srinivas RS, Tiwari R, Kannababu C. Application of active magnetic bearings in flexible
rotordynamic systems–a state-of-the-art review. Mech Syst Signal Process. 2018;106:537–72.
Identification in a Magnetically Levitated Rigid Rotor System … 23

4. Swanson E, Heshmat H. Oil-free foil bearings as a reliable, high performance backup bearing
for active magnetic bearings. In: ASME Turbo Expo 2002: Power for Land, Sea, and Air; 2002.
pp. 589–598.
5. Aenis M, Knopf E, Nordmann R. Active magnetic bearings for the identification and fault
diagnosis in turbomachinery. Mechatronics. 2002;12:1011–21.
6. Clark DJ, Jansen MJ, Montague GT. An overview of magnetic bearing technology for gas
turbine engines. USA 213177; 2004.
7. Kumar P, Arambam C, Singh LD, Singh ND. Analysing the dynamic interaction between
unbalance and crack responses in a jeffcott rotor supported by foil bearings: a numerical study.
In: Emerging trends in mechanical and industrial engineering: select proceedings of ICETMIE
2022, ed: Springer; 2023. pp. 971–980.
8. Edwards S, Lees A, Friswell M. Estimating rotor unbalance from a single run-down. In: imeche
conference transactions; 2000. pp. 323–334.
9. Hori Y, Uematsu R. Influence of misalignment of support journal bearings on stability of a
multi-rotor system. Tribol Int. 1980;13:249–52.
10. Kumar P, Tiwari R. Dynamic response analysis of an unbalanced and misaligned rotor supported
on active magnetic bearings and touchdown bearings. In: Proceedings of the 6th national
symposium on rotor dynamics; 2020. pp. 407–418.
11. Morton P. Modal balancing of flexible shafts without trial weights. Proc Inst Mech Eng C J
Mech Eng Sci. 1985;199:71–8.
12. Krodkiewski J, Ding J, Zhang N. Identification of unbalance change using a non-linear
mathematical model for multi-bearing rotor systems. J Sound Vib. 1994;169:685–98.
13. Shih Y-P, Lee A-C. Identification of the unbalance distribution in flexible rotors. Int J Mech
Sci. 1997;39:841–57.
14. Zhou S, Shi J. Active balancing and vibration control of rotating machinery: a survey. Shock
Vib Digest. 2001;33:361–71.
15. Menshikov Y, Identification of rotor unbalance as inverse problem of measurement. Adv Pure
Math 2013.
16. Chatzisavvas I, Dohnal F. Unbalance identification using the least angle regression technique.
Mech Syst Signal Process. 2015;50:706–17.
17. Markert R, Platz R, Seidler M. Model based fault identification in rotor systems by least squares
fitting. Int J Rotating Mach. 2001;7:311–21.
18. Yao J, Liu L, Yang F, Scarpa F, Gao J. Identification and optimization of unbalance parameters
in rotor-bearing systems. J Sound Vib. 2018;431:54–69.
19. Shrivastava A, Mohanty AR. Identification of unbalance in a rotor system using a joint input-
state estimation technique. J Sound Vib. 2019;442:414–27.
20. Kumar P, Kumar V, Kumar K, Meena LS. Unbalance and dynamic parameters estimation
in a rigid rotor mounted on active magnetic bearings. In: Advances in applied mechanical
engineering, ed: Springer; 2020, pp. 363–371.
21. Ranjan G, Tiwari R. Application of active magnetic bearings for in situ flexible rotor residual
balancing using a novel generalized influence coefficient method. Inverse Problems Sci Eng.
2019;27:943–68.
22. Ranjan G, Tiwari R. On-site high-speed balancing of flexible rotor-bearing system using virtual
trial unbalances at slow run. Int J Mech Sci. 2020; 105786.
23. Gibbons C, Coupling misalignment forces. In: Proceedings of the 5th turbomachinery
symposium; 1976.
24. Sekhar A, Prabhu B. Effects of coupling misalignment on vibrations of rotating machinery. J
Sound Vib. 1995;185:655–71.
25. Rao AS, Sekhar A. Vibration analysis of rotor-coupling-bearing system with misaligned shafts.
In: ASME 1996 International gas turbine and aeroengine congress and exhibition; 1996, pp.
V005T14A001-V005T14A001.
26. Al-Hussain K, Redmond I. Dynamic response of two rotors connected by rigid mechanical
coupling with parallel misalignment. J Sound Vib. 2002;249:483–98.
24 P. Kumar and R. Tiwari

27. Al-Hussain K. Dynamic stability of two rigid rotors connected by a flexible coupling with
angular misalignment. J Sound Vib. 2003;266:217–34.
28. Bouaziz S, Messaoud NB, Mataar M, Fakhfakh T, Haddar M. A theoretical model for analyzing
the dynamic behavior of a misaligned rotor with active magnetic bearings. Mechatronics.
2011;21:899–907.
29. Messaoud NB, Bouaziz S, Fakhfakh T, Maatar M, Haddar M, Dynamic behavior of active
magnetic bearings in presence of angular misalignment defect. Int J Appl Mech. 2011;3:491–
505.
30. Verma AK, Sarangi S, Kolekar M. Experimental investigation of misalignment effects on rotor
shaft vibration and on stator current signature. J Fail Anal Prev. 2014;14:125–38.
31. Jang JY, Khonsari MM. On the characteristics of misaligned journal bearings. Lubricants.
2015;3:27–53.
32. Kumar P, Tiwari R. A numerical study on the effect of unbalance and misalignment fault
parameters in a rigid rotor levitated by active magnetic bearings. In: ASME 2019 gas turbine
India conference; 2019.
33. Kumar P, Tiwari R. Effects of unbalance and AMB misalignment in a rigid rotor with an offset
disc levitated by active magnetic bearings: a numerical investigation. In: 12th International
conference on vibrations in rotating machinery; 2020, pp. 151–168.
34. Srinivas RS, Tiwari R, Kannababu C. Model based analysis and identification of multiple fault
parameters in coupled rotor systems with offset discs in the presence of angular misalignment
and integrated with an active magnetic bearing. J Sound Vib. 2019;450:109–40.
35. Kuppa SK, Lal M. Dynamic behaviour analysis of coupled rotor active magnetic bearing system
in the supercritical frequency range. Mechanism and machine theory; 2020, p. 103915.
36. Schweitzer G, Maslen EH. Magnetic bearings: theory, design, and application to rotating
machinery, vol. 1; 2009. Springer, Berlin.
37. Lijesh K, Hirani H. Optimization of eight pole radial active magnetic bearing. J Tribol.
2015;137.
38. Reddy MCS, Sekhar A. Detection and monitoring of coupling misalignment in rotors using
torque measurements. Measurement. 2015;61:111–22.
39. Kumar P, Tiwari R. Finite element modelling, analysis and identification using novel trial
misalignment approach in an unbalanced and misaligned flexible rotor system levitated by
active magnetic bearings. Mech Syst Signal Process. 2021;152: 107454.
40. Tiwari R, Kumar P. An innovative virtual trial misalignment approach for identification of
unbalance, sensor and active magnetic bearing misalignment along with its stiffness parameters
in a magnetically levitated flexible rotor system. Mech Syst Signal Process. 2022;167: 108540.
41. Kumar P, Tiwari R. Development of a novel approach for quantitative estimation of rotor
unbalance and misalignment in a rotor system levitated by active magnetic bearings. Iranian J
Sci Technol Trans Mech Eng. 2020;45:769–86.
42. Kumar P, Tiwari R. Dynamic analysis and identification of unbalance and misalignment in a
rigid rotor with two offset discs levitated by active magnetic bearings: a novel trial misalignment
approach. Propulsion Power Res. 2020;10:58–82.
43. Kumar P, Model based analysis and identification of unbalance and misalignment in rotor
systems levitated by active magnetic bearings PhD, Mechanical Engineering, Indian Institute
of Technology Guwahati, Guwahati; 2021.
44. Kumar P, Tiwari R. A review: multiplicative faults and model-based condition monitoring
strategies for fault diagnosis in rotary machines. J Brazil Soc Mech Sci Eng. 2023;45(5).
https://doi.org/10.1007/s40430-023-04203-z
45. Kumar P, Swain BP. Recent advances in materials select. In: Proceedings of ICSTE 2023
condition monitoring of mechanical and electrical faults in stationary and rotating equipments.
A Review Springer Nature Singapore, Singapore; 2023, pp. 297–306.
46. Kumar P, Sanket M, Srivastav S, Madav TD, Swain BP (2023) Recent advances in mate-
rials Select. In: Proceedings of ICSTE 2023 vibrational nature of an unbalanced rigid rotor
system with three discs secured by two active magnetic bearings. Springer Nature Singapore,
Singapore; 2023, pp. 335–345.
Thermo-Elastic Bending and Buckling
Behavior of Functionally Graded Shafts
with Various Grading Patterns

Debabrata Gayen, Rajiv Tiwari, and Debabrata Chakraborty

Abstract Improvement of new engineering materials has a prime challenge and an

important research topic for material scientists and engineers. Functionally graded
materials (FGMs) are developed as multifunctional, advanced heterogeneous mate-
rials with the smooth and continuous variation of thermo-elastic properties along a
preferred direction(s) following various grading patterns, and potential uses in high-
temperature applications. In the present work, an analytical solution is presented for
bending and buckling behavior of FG shaft (FGS) using Timoshenko beam theory
(TBT). Here, silicon nitride (Si3 N4 ) and Ti-6Al-4 V are considered as ceramic and
metallic constituents for the FGS respectively, assuming that Ti-6Al-4 V content
decreases gradually from the inner diameter towards the outer diameter. First,
temperature-dependent and radially graded properties are obtained according to
linear/power/exponential laws. Next, using 1-D Fourier heat conduction equation,
radial variations of temperature distributions are obtained using different grading
patterns for achieving desired temperature-dependent material properties. Equations
of motions are then derived using Hamilton’s principle for bending and buckling
behavior of FGS. Based on the analytical formulation, a MATLAB code is developed
and various numerical problems are solved for the bending and buckling behavior of
FGS. Finally, numerical results are produced to examine the influences of material
grading patterns, temperature gradients, boundary conditions, and slenderness ratio
on the thermo-elastic bending and critical buckling temperature of the FGS. Results
show that the material grading patterns and temperature gradients have significant
impact on thermo-elastic responses of the FGS. The present results will help in devel-
oping more accurate analytical formulations of dynamics of the FGS subjected to
different thermal loads.

D. Gayen (B)
Department of Mechanical Engineering, NIT Durgapur, Durgapur, West Bengal, India
e-mail: [email protected]
R. Tiwari · D. Chakraborty
Department of Mechanical Engineering, IIT Guwahati, Assam, India
e-mail: [email protected]
D. Chakraborty
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 25
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_2
26 D. Gayen et al.

Keywords FGM · FGS · TBT · Bending · Buckling · Material grading patterns

1 Introduction

Functionally graded materials (FGMs) are introduced as multifunctional, non-

homogeneous, advanced composite materials. FGMs are basically formed with
smooth variation of volume fraction of the metal and ceramic along a specific direc-
tion using gradation laws for achieving desired temperature-dependent material prop-
erties. Traditional composites and laminated composites materials are replacements
for FGMs to mitigate some major drawbacks such as delamination problems, high
residual stresses, high thermo-mechanical stresses, high interlaminar stresses, high
stress concentrations and poor bonding strength. FGMs are generally used in mechan-
ical, automotive, aerospace and biomedical domains because of their lightweight,
high strength and stiffness, moderate to high fracture toughness, good fatigue resis-
tance and temperature gradient. However, real life applications of FGMs are limited
because of the challenge in mass productions, and manufacturing cost. Historically,
material researchers [1, 2] first reported the gradation concept in composite and poly-
meric materials in 1972. The first FGM was conceptualized by a group of material
scientists at Japan in 1984 [3]. The number of book [4] and review articles [5–9] are
published on the modelling, analysis and optimization of structures (beams, shafts,
rotors, discs, pipes, etc.) made of FGMs.
Bending and buckling analysis are important for structural members subjected to
transverse loads and axial loads, respectively. The study of influence of shear defor-
mation is more pronounced for bending and buckling analysis in the laminated and
traditional composites and FGMs structure subjected to transverse loads. Here, few of
the important works are discussed briefly. Kiani and Eslami [10] modelled FG beams
based on Euler–Bernoulli beam theory and show the influences of different end condi-
tions and thermal loading on thermal buckling of such FG beams. Wattanasakulpong
and Ungbhakorn [11] employed higher order shear deformation theory for bending,
buckling and vibration of carbon nanotube-reinforced composite beams and exam-
ined the importance of spring constants, thickness ratios, volume fractions and type
of carbon nanotube on those analyses. Li et al. [12] derived analytical solutions for
deflection and rotational angle, bending moment, and shear forces of FGM Timo-
shenko beams in terms of homogenous Euler–Bernoulli beams with same geom-
etry and end conditions under arbitrary transverse loads, and examined the effects
of material inhomogeneity and shear deformation on the solutions. Ansari et al.
[13] modelled a size-dependent FGM Timoshenko beam using strain gradient elas-
ticity theory and obtained bending, buckling and free vibration characteristics of
FGM microbeams using modified strain gradient theory and generalized differen-
tial quadrature method. Finally, influences of dimensionless length scale parameter,
material property gradient index, and length to thickness aspect ratio, on the mechan-
ical behaviors of FGM microbeams, were examined. Simsek et al. [14] studied the
analytical solution for static bending of FG microbeams with simply supported end
Thermo-Elastic Bending and Buckling Behavior of Functionally … 27

condition based on the modified couple stress theory (MCST), and classical beam
theory. The influences of volume fraction index, length scale parameter, aspect ratio
and Poisson effect on the static bending behavior examined and reported that the
deflections of the microbeam by the classical beam theory are always larger than those
by the modified couple stress theory. Akgöz and Civalek [15] studied the thermo-
mechanical buckling response of simply supported FG microbeams using MCST and
performed a parametric study to examine the effects of aspect ratio, material gradient
index, and temperature on critical buckling loads of the FG microbeams. Ebrahimi
and Salari [16] carried out thermal buckling and thermal vibrational behavior of FG
nanobeams subjected to thickness wise linear temperature distribution using nonlocal
elasticity theory. Finally, a parametric study was performed to examine the impor-
tance of small scale parameter, material gradient index, mode number, temperature
and beam thickness on critical buckling temperature and fundamental frequencies
for FG nanobeams. It was reported that nonlocality leads to reduction in natural
frequency and buckling temperature for the FG beams. Jing et al. [17] studied static
bending and free vibration characteristics of FG Timoshenko beam by considering
three shear correction factors using the finite volume method. They examined the
influences of different end conditions, span-depth ratios and volume fraction expo-
nents on transverse displacement and reported that the third shear correction factor
was more suitable for the thick FG beam. Sayyad and Ghugal [18] studied closed
form solution for bending, buckling, and free vibration of FG beams with power and
exponential functions of material distribution and different end conditions, using
modified exponential shear deformation theory. Chen et al. [19] carried out thermal
buckling analysis for FG Euler–Bernoulli beam with power law function of mate-
rial distribution, different end conditions and different temperature changes. The
influences of material gradient index, slenderness ratios, temperature distributions
and end conditions were examined on thermal buckling behaviors of FG beams
and reported that these parameters significantly affected the obtained critical buck-
ling temperature. Gayen [20] performed static behaviors of FG components and
obtained temperature, displacement and stress fields for different material gradation
patterns. It was reported that for static performances gradient indices and temper-
atures played a counting role and hence high temperature applications FGMs are
more suitable as compared to conventional materials. Gayen et al. [21–24] carried
out statics and dynamics of cracked FG shaft using the finite element method and
considered power law function for radially graded thermo-elastic properties. They
have studied the importance of geometric, material and crack parameters for natural
whirling frequencies, critical speed and stability threshold speed and reported that a
proper and suitable material gradient index was required to design FG shaft which
may operate with higher spin speed during the service.
The abovementioned studies show that thermo-elastic analyses are essentially
required for understanding structural performance of FGM components under
thermal environment. Various analytical and numerical methods have been devel-
oped based on different couple stress theory, and classical beam theories for bending
and buckling behavior of FG beams and shafts subjected to thermal loads during
28 D. Gayen et al.

the last few years. However, in the comprehensive literature survey most of the arti-
cles are available exclusively on FG beams subjected to uniform temperature and
power law of material distributions. For shafts as opposed to other structural compo-
nents, temperature distributions were not developed in particular, and the thermal
behaviors of such structures have not been thoroughly investigated. As far as the
author concerns, there has been no documented work for thermo-elastic bending
and bucking behavior of FG shafts subjected to different temperature distribution
(linear temperature distribution (LTD), non-linear temperature distribution (NLTD),
exponential temperature distribution (ETD) and uniform temperature distribution
(UTD)), different material grading patterns (exponential law of material gradation
(ELMG), linear law of material gradation (LLMG) and power law of material grada-
tion (PLMG)), and temperature-dependent properties. Therefore, the present analyt-
ical thermal study aims to compute the bending behavior and critical bucking temper-
ature of FG shafts with different end conditions, different temperature distribution
and different material grading patterns.

2 Mathematical Modeling for Bending and Buckling

Analysis of FG Shafts

2.1 Thermo-Elastic Material Properties of FGMs

FGMs were originally conceived with possible uses requiring great temperature
endurance in consideration. As a result, estimates of the characteristics of effective
thermo-elastic properties are a fundamental necessity and are estimated using both
power and exponential functions. The thin temperature-dependent FG hollow circular
shaft depicted in Fig. 1 is taken into consideration in this instance. As well as invoking
the planar stress condition, the thickness of the shaft is expected to be suitably
minimal in comparison to its diameter.
According to research of Reddy and Chin [25], the material properties P
(mass density ρ (in Kg/m3 ), Young’s modulus E (in Pa), thermal conductivity

Fig. 1 Cross-sectional view

of a thick-walled FG shaft
Thermo-Elastic Bending and Buckling Behavior of Functionally … 29

K (in W/m K), and Poisson’s ratio ν) in a radially graded FG shaft for varying
temperature T (in K) are expressed as

P(r, T ) = Pi (T ) + {Po (T ) − Pi (T )}Vo (r ) (1)

Here, the inner and outer surfaces are denoted by the subscripts i and o, respectively.
According to the PLMG, the volume fraction of the outer phase material Vo and the
volume fraction of the inner phase material Vi are represented for a hollow shaft with
a circular cross section as
{ / }k
Vo (r ) = (r − Ri ) (Ro − Ri ) , Vi (r )= 1 − Vo (r ),Ro ≤ r ≤ Ri , 0 ≤ k ≤ ∞
(2)

where, as illustrated in Fig. 1, Ro and Ri are the shaft’s outer and inner radius,
respectively. For hollow beams and shafts, the two extremes of perfectly homoge-
neous outer and inner sides are represented by k = 0 and k = ∞, respectively, by
the power law of material gradient index (PLMGI) k. For k = 1, Eq. (2) designates
the LLMG, whereas for other values of k, Eq. (2) displays the PLMG. For k = 1,
k < 1 and k > 1 represent how the material behaves in linear, concave, and convex
patterns, respectively.
Variations in material properties from the exponential law of material gradation
(ELMG) are described by
{( / )/ }
P(r ) = Pi eλ(r −Ri ) ,λ = ln Po Pi (Ro − Ri ) , Ri ≤ r ≤ Ro (3)

2.2 Derivation of Temperature Field

For any components made of FGMs, material properties as function of temperature

are expressed using the work of Touloukian [26] as
( )
P j (T ) = Coj C−1 j T −1 + 1 + C1 j T + C2 j T 2 + C3 j T 3 (4)

where j = o or i, T indicates temperature in Kelvin and

Coj , C−1 j , C1 j , C2 j , and C3 j are unique constants.
For cylindrical/circular structures especially shafts, the “steady-state one dimen-
sional Fourier equation of heat conduction law without heat generation” is given
as
[ ]
d dT
K (r ) = 0 with T (Ri ) = Ti at r = Ri and T (Ro ) = To at r = Ro (5)
dr dr
30 D. Gayen et al.

where K is chosen by using Eqs. (1) and (3) and the temperature fields (called as
LTD) for radially graded thin FG shaft are expressed using LLMG as
{ / }
T (r ) = Ti + (To − Ti ) (r − Ri ) (Ro − Ri ) , Ri ≤ r ≤ Ro (6)

Similarly for PLMG, the NLTD for radially graded thin FG shaft is expressed as
⎡ ( ⎤ ⎡ ⎤
) ) ) ) jk+1 ) / ∑ ) )
( ) ∑5
(−1) j K o − K i j r − Ri
5
(−1) j K o − K i j ⎦
T (r ) = Ti + To − Ti ⎣ ⎦ ⎣ (7)
jk + 1 Ki R o − Ri jk + 1 Ki
j=0 j=0

Here, convergence study is fulfilled by using first seven terms. Similarly for ELMG,
the ETD for radially graded thin FG shaft is expressed as
( ) ( ) ( / )
(To − Ti ) (To − Ti ) ln K o K i
T (r ) = Ti − −λ(R −R ) + −λ(R −R ) e−λ(r −Ri ) where λ =
e o i −1 e o i −1 (Ro − Ri )
(8)

2.3 Problem Formulation and Basic Assumptions

Figure 2 shows a FG circular shaft having simply supported ends with a uniform
length L and diameter d. The FG shaft is modelled based on TBT. Here the Cartesian
coordinate system x yz is used such that at z = 0, the x y plane coincides with the
mid surface of the FG beam. Here, the basic assumptions are considered as (a) the
Cartesian coordinate system is taken on the neutral axis of the FG beam (b) The
displacements are small in comparison with the diameter of the FG beam thus the
associated deformations are infinitesimal and (c) displacements u in the x-direction
consist of extension and bending components.

Fig. 2 A shaft made of FGMs: a simply supported ends with coordinate system b beam element
subjected to internal and external forces
Thermo-Elastic Bending and Buckling Behavior of Functionally … 31

2.4 Kinematic and Constitutive Equation

Based on the assumptions of TBT, the displacements of an arbitrary point of the FG

beam along the x-, y- and z-directions are expressed as

u(x, z, t) = u 0 (x, t) + zψ(x, t) and w(x, z, t) = w0 (x, t) (9)

where u 0 is the axial displacement and w0 is the transverse

/ displacement of mid
surface and those are functions of x and t, ψ = ∂w0 (x, t) ∂ x is the rotation of the
FG beam cross-section and t denotes time.
The strain–displacement relations of a FG shaft is expressed as

/ / / ( / / )/ /
εx x = ∂u ∂ x =∂u 0 ∂ x + z∂ψ ∂ x and γx z = ∂u ∂z + ∂w ∂ x 2 = ψ + ∂w0 ∂ x
(10)

where εx x and γx z are the axial and shear strain respectively.

As the shaft is made of FGMs hence the constitutive equations for FG shafts,
using the linear thermo-elasticity is expressed in x z plane as
/
σx x = E(r, T )[εx x − α(r, T ){T (r ) − T0 }] and τx z = E(r, T )γx z 2{1 + ν(r, T )}
(11)

where σx x and τx z are the axial and shear stress in x z plane, respectively. T0 is the refer-
ence temperature i.e. ambient temperature (T0 = 300 K) and T is the temperature
distribution profile along radial direction of the shaft.
By substituting Eq. (10) into Eq. (11) and taking integration over the cross-section
of shaft, the resultant normal force N x a, resultant bending moment Mx and resultant
transverse shear force Q x z is expressed as

∫ ∫ ∫Ro ∫ ∫
Nx = σx x d A = σx x dz, Mx = zσx x d A
A Ri A

∫Ro ∫ ∫ ∫Ro
= zσx x dz and Q x z = κs τ x z d A = κs τx z dz (12)
Ri A Ri

where the shear correction factor or shape coefficient κs for circular c/s shaft is
expressed as
/
κs (r, T ) = 6{1 + ν(r, T )} {7 + 6ν(r, T )} (13)
32 D. Gayen et al.

By substituting Eqs. (10), (11) and (13) into Eq. (12), the resultant normal force
N x , the resultant bending moment Mx and resultant transverse shear force Q x z are
expressed as
⎫
∫Ro ) ) ∫Ro ) ) ⎪
⎪
∂u 0 ∂ψ ∂u 0 ∂ψ ⎪
Nx = σx x dz = A1 + A2 − N xT , Mx = zσx x dz = A2 + A3 − Mx ⎪
T ⎪
⎪
∂x ∂x ∂x ∂x ⎪
⎪
⎪
⎬
Ri Ri
⎪
(14)
∫Ro ) ) ⎪
⎪
A1 κs (r, T ) ∂w0 ⎪
⎪
and Q x z = κs (r, T )τx z dz = ψ+ ⎪
⎪
2{1 + ν(r, T )} ∂x ⎪
⎪
⎭
Ri

where A1 , A2 , and A3 are stretching stiffness, stretching-bending coupling stiffness,

and bending stiffness, respectively, N xT and MxT are resultant thermal forces and
resultant thermal moments, which are obtained using the following relations
∫
( )
(A1 , A2 and A3 ) = 1, z, z 2 E(r, T )d A (15)
A
∫
( )
N xT and Mx = T
(1, z)E(r, T ) α(r, T ) {T (r ) − T0 }d A (16)
A

2.5 Derivation of Governing Equation of Motion

By employing the Hamilton’s principle, the governing equilibrium equations of

motion of an FG shaft is derived as

∫t
δ(U + T − W )dt = 0 (17)
0

where δ is the vibrational operator, U is the strain energy, T is the kinetic energy and
W is the work done by external forces. The virtual strain energy is given as
∫
δU = (σx x δεx x + κs τx z δγx z ) d V (18)
V

With the aid of the stress resultant definition defined in Eq. (11), the Eq. (18) is
rewritten as

∫L ( ) ) ) ) ) ))
∂u 0 ∂ψ ∂w0
δU = Nx δ + Mx δ + Qxz δ + δψ dx (19)
∂x ∂x ∂x
0
Thermo-Elastic Bending and Buckling Behavior of Functionally … 33

The kinetic energy T and virtual kinetic energy δT for Timoshenko beam are
written as

∫L ∫ () ) ) ) )
1 ∂u 2 ∂w 2
T = ρ(r, T ) + d Ad x (20)
2 ∂t ∂t
0 A
∫L [ ) ) ) ) ) )]
∂u 0 ∂δu 0 ∂w0 ∂δw0 ∂ψ ∂δu 0 ∂u 0 ∂δψ ∂ψ ∂δψ
δT = I0 + + I1 + + I2 dx (21)
∂t ∂t ∂t ∂t ∂t ∂t ∂t ∂t ∂t ∂t
0

where (I0 , I1 and I2 ) are the mass moment of inertias and defined as
∫
( )
(I0 , I1 and I2 ) = 1, z, z 2 ρ(r, T )d A (22)
A

Since, FGMs are mainly used for high temperature applications. Hence, first
variation of the work done corresponding to the temperature gradient is given by

∫L
( / )( / )
δV = N xT ∂w0 ∂ x ∂δw0 ∂ x d x (23)
0

By substituting Eqs. (18), (21) and (23) into Eq. (17) and setting the coefficients
δu 0 , δw0 , and δψ equal to zero, the governing equations of motion are obtained as
/ ( / ) ( / )
d N x d x = I0 ∂ 2 u 0 ∂t 2 + I1 ∂ 2 ψ ∂t 2 (24a)

( / ) ( / ) ( / )
d Mx d x − Q x z = I1 ∂ 2 u 0 ∂t 2 + I2 ∂ 2 ψ ∂t 2 (24b)

( / ) ( / ) ( / )
d Q x z d x − N xT ∂ 2 w0 ∂ x 2 = I0 ∂ 2 w0 ∂t 2 (24c)

By using Eq. (14) and (24a–24c), the equations of motion of FG beam/shaft in

terms of the displacements are obtained as
( / ) ( / ) ( / ) ( / ) ( / )
A1 ∂ 2 u 0 ∂ x 2 + A2 ∂ 2 ψ ∂ x 2 − ∂ N xT ∂ x − I0 ∂ 2 u 0 ∂t 2 − I1 ∂ 2 ψ ∂t 2 = 0
(25a)
( / ) ( / ) ( / ) ) )
A1 κs (r, T ) ∂w0
A2 ∂ u 0 ∂ x + A3 ∂ ψ ∂ x −
2 2 2 2
∂ MxT ∂x − ψ+
2{1 + ν(r, T )} ∂x
( / ) ( / )
− I1 ∂ 2 u 0 ∂t 2 − I2 ∂ 2 ψ ∂t 2 = 0 (25b)
) ) ( / ) ( / )
A1 κs (r, T ) ∂ψ ∂ w0 2
+ − N xT ∂ 2 w0 ∂ x 2 − I 0 ∂ 2 w0 ∂t 2 = 0 (25c)
2{1 + ν(r, T )} ∂ x ∂x2
34 D. Gayen et al.

Table 1 Boundary
B.Cs B.Cs at x = 0 B.Cs at x = L
conditions (BCs) for FGM
Timoshenko beams S–S w = N x = Mx = 0 w = N x = Mx = 0
C–C u=w=ψ =0 u=w=ψ =0
C–S u=w=ψ =0 w = N x = Mx = 0
/
C–R u=w=ψ =0 u = ψ = d Mx d x − N xT ψ = 0
/
S–R w = N x = Mx = 0 u = ψ = d Mx d x − N xT ψ = 0

Here, five types of boundary conditions (S–S, C–C, C–S, C–R and S–R) are assumed
for the FG beams with the combination of the roller (R), simply supported (S), and
clamped (C) edges and the boundary conditions in each case are tabulated in Table 1.

2.6 Analytical Solution Procedure

An analytical solution of the governing equations for bending and buckling of a

simply supported FG beam/shaft is presented based on Navier’s solution procedure.
The displacement functions are expressed as products of undetermined coefficients
and known trigonometric functions to satisfy the governing equations Eq. (24) along
with the conditions at x = 0 and L. Therefore, the displacement fields are presented
in the form of
∞
∑
{ }T r { }T r
u 0 (x, t) w0 (x, t) ψ(x, t) = Um cos αx Wm sin αx ψm cos αx eiωt
m=1
(26)
√
where i = −1, ω is the frequency for free vibration, Um , Wm and ψm are the
unknown Fourier coefficients to be determined for each mode m, and α = mπ/L,
and superscript T r denotes the transpose of a matrix.
For the static bending problem, all time derivatives and the external force are set
to zero. The applied transverse load q(x) is written in the trigonometric form as

∑
N ∫L
2
q(x) = Q n sin αx, where Q n = q(x) sin αx d x (27)
n=1
L
0

where Q n are the Fourier coefficients, which are given for uniform and point loads
as
( /
( 4q/0 mπ,m
) = 1, 3, 5...... → for uniform load
Qn = (28)
2P L sin mx p ,m = 1, 2, 3...... → for point load
Thermo-Elastic Bending and Buckling Behavior of Functionally … 35

where q0 is the intensity of the uniformly distributed load, P is the point load, and x p
is the position of the applied point load. By substituting Eq. (26) into Eqs. (25a–25c)
respectively, leads to Eqs. (29–31)
( ) ( )
−A1 α 2 +I0 ω2 Um + −A2 α 2 +I1 ω2 ψm = 0 (29)
( ) ) )
A1 κs (r, T ) A1 κs (r, T )
−A2 α 2 +I1 ω2 Um − αWm + −A3 α 2 − + I2 ω2 ψm = 0 (30)
2{1 + ν(r, T )} 2{1 + ν(r, T )}
) )
A1 κs (r, T ) 2 2 T 2 A1 κs (r, T )
− α +I0 ω +N x α Wm − αψm = 0 (31)
2{1 + ν(r, T )} 2{1 + ν(r, T )}

By substituting Eq. (26) and the load function of Eq. (27) into the equations of
motion in Eq. (25a–25c) in order to obtain the analytical solutions and the results of
the substitution arranged into the following matrix form
⎧( )⎫⎧ ⎫ ⎧ ⎫ ⎧⎛⎡ ⎤ ⎡ ⎤⎞
⎪
⎪ [K ] ⎪
⎪⎪ ⎪ ⎪ ⎪ ⎪
⎨ [ ] ⎬⎨ Um ⎬ ⎨ 0 ⎬ ⎨ a11 a12 a13 b11 b12 b13
+ΔT K T ⎜⎢ ⎥ ⎢ ⎥⎟
Wm = Q m ⇒ ⎝⎣ a21 a22 a23 ⎦ + ΔT ⎣ b21 b22 b23 ⎦⎠
⎪
⎪ ⎪
⎪ ⎪
⎩ ⎭ ⎪
⎪ ⎩ ⎪
⎭ ⎪
⎩
⎩ ⎭ ψm
− ω2 [M] 0 a31 a32 a33 b31 b32 b33
⎡ ⎤⎫ ⎧ ⎫ ⎧ ⎫⎫
c11 c12 c13 ⎪⎬⎪ ⎬ ⎪
⎨ Um ⎪ ⎨ 0 ⎪⎬⎪
⎬
⎢ ⎥
− ω2 ⎣ c21 c22 c23 ⎦ Wm = Q m (32)
⎪
⎭⎩⎪ ⎪
⎭ ⎩ ⎪ ⎭⎪
⎪⎭
c31 c32 c33 ψm 0

where [K ] and [K T ] are the stiffness matrix and the coefficient matrix of the temper-
ature gradient ΔT , respectively, and [M] is the mass matrix. The elements of [K ],
[K T ] and [M] matrices in Eq. (32) is written as
⎫
A1 κs (r, T ) ⎪
a11 = −A1 α 2 , a12 = 0, a13 = −A2 α 2 , a21 = −A2 α 2 , a22 = − α, ⎪
⎬
2{1 + ν(r, T )}
A1 κs (r, T ) A1 κs (r, T ) A1 κs (r, T ) ⎪
a23 = −A3 α 2 − , a31 = 0, a32 = − α 2 , a33 = − α⎪
⎭
2{1 + ν(r, T )} 2{1 + ν(r, T )} 2{1 + ν(r, T )}
(33)
b11 = b12 = b13 = b21 = b23 = b31 = b32 = b33 = 0 and b22 = N xT α 2 (34)

c11 = I0 , c12 = 0, c13 = I1 , c21 = I1 , c22 = 0, c23 = I2 , c31 = 0, c32 = I0 , c33 = 0 (35)

For bending analysis, the temperature gradient (ΔT ) and natural frequency (ω) in
Eq. (36) are set to zero. The resulting simultaneous equations can be solved for
displacements and stresses of the bending problems. So for bending analysis for
TBT, the unknown Fourier coefficients Um , Wm and ψm are written as
/{ ( / ) ( / )}
Um = Q m a21 + a11 a22 a33 a13 a32 − a11 a23 a13 (36)

( / )/ { ( / ) ( / )}
Wm = Q m a11 a33 a13 a32 a21 + a11 a22 a33 a13 a32 − a11 a23 a13 (37)

( / )/ { ( / ) ( / )}
ψm = Q m −a11 a13 a21 + a11 a22 a33 a13 a32 − a11 a23 a13 (38)
36 D. Gayen et al.

The complete solutions for the displacements are known from Eq. (26). The
analytical expression of bending results for simply supported supports and for point
load are presented in the dimensionless forms which are written as
) )
Em I L
w = 100 w0 ,0 (39)
P L3 2

Similarly for simply supported and distributed loads, the dimensionless quantities
are written as
) )
Em I L
w = 100 w ,0 (40)
q0 L 4 0 2

where I = π4 (Ro4 − Ri4 ) is the moment of inertia of a hollow circular section.

For buckling analysis, the determinant of (the coefficient
) matrix of Eq. (32) is set
to zero for predicting the axial thermal force N xT and critical buckling temperature
(ΔTcr ). All analytical results are presented in the dimensionless forms which are
written as follows. Based on the work of Kiani and Eslami [10], the critical axial
thermal force/load of the beam for all cases of boundary conditions can be written
in the form
{( )/ }
T
N xcr =C A1 A3 − A22 A1 L 2 (41)

where C is a constant and depends on the type of boundary conditions and is 39.47842
for C–C beams, 9.86960 for S–S and C–R beams, 2.46740 for S–R beams, and
20.19077 for C–S beams.
Now, for radially graded FG shaft/beam and for uniform( temperature
) rise (TR +
ΔT , TR is reference temperature), the axial thermal load N xT is predicted using
( )
Eq. (16) and the critical buckling temperature ΔT Ucrni is predicted by using Eqs.
(41) and (42) as
) )
E oi αi + E i αoi E oi αoi
N xT = ΔT (Ro − Ri ) E i αi + + (42)
k+1 2k + 1
) )2
C R o − Ri M(k, ξ )
ΔTcrU ni = (43)
αi L N (k, ξ, ζ )

where, ξ = E oi /E i , ζ = αoi /αi and the functions M(k, ξ ) and N (k, ξ, ζ ) are
expressed as
( )
1 ξ k 2 +k + 2
M(k, ξ ) = +
12 4(k + 1)(k + 2)(k + 3)
ξ 2 k2
− , N (k, ξ, ζ )
4(k + 1)(k + 2)2 (k + 1 + ξ )
Thermo-Elastic Bending and Buckling Behavior of Functionally … 37

ξ +ζ ξζ
=1+ + (44)
k+1 2k + 1

( Similarly
) for radially graded FG shaft/beam with LTD, the axial thermal load
N xT is evaluated using Eqs. (6) and (16) and the critical buckling temperature
( )
ΔT crLT D
is predicted as
))
Ro − Ri 2 M(k, ξ )
C N (k, ξ, ζ )
ΔTcrL T D = − (Ti − TR ) where
αi
L H (k, ξ, ζ ) H (k, ξ, ζ )
1 ξ +ζ ξζ
H (k, ξ, ζ ) = + + (45)
2 k+2 2k + 2

( Similarly
) for radially graded FG shaft/beam with NLTD, the axial thermal load
N xT is evaluated using Eqs. (7) and (16) and the critical buckling temperature
( )
ΔT crN LT D
is predicted as
) )2
C R o − Ri M(k, ξ ) N (k, ξ, ζ ) /
ΔTcrN L T D = − (Ti − TR ) , γ = K oi K i
αi L I (k, ξ, ζ, γ ) I (k, ξ, ζ, γ )
(46)

where
⎡ ⎤
∑
5 [ ]/ ∑5 ) )
(−γ )i 1 ξ +ζ ξζ ⎣ (−1) j K oi j ⎦
I (k, ξ, ζ, γ ) = + + (47)
ik + 1 ik + 2 ik + k + 2 ik + 2k + 2 jk + 1 K i
i=0 j=0

( Similarly
) for radially graded FG shaft/beam with ETD, the axial thermal load
N xT is evaluated using Eqs. (8) and (16) and the critical buckling temperature
( )
ΔT crET D
is predicted as[
C ( )
ΔTcrE T D = 2 λ21 Ro Ri − 2 λ21 Ri2 + 1
L 2 λ31
( ) ]( ( ) )
Ti − T R { λ ( Ro −R ) ( )} e−λ Ro −Ri −1
− e 2 i − e−λ1 Ro −Ri /AA (48)
λ1 + λ2 αi

where
( / )/ ( / )/ ( / )/
λ1 = ln E o E i (Ro − Ri ), λ2 =ln αo αi (Ro − Ri ),λ = ln K o K i (Ro − Ri )
(49)

and AA is expressed as

(λ1 + λ2 ) e(λ2 −λ)(Ro −Ri ) −(λ1 + λ2 − λ) eλ2 (Ro −Ri ) −λ e−λ(Ro −Ri )
AA = (50)
(λ1 + λ2 − λ)(λ1 + λ2 )
38 D. Gayen et al.

3 Results and Discussion

Here, various numerical examples are presented to understand the bending and buck-
ling behaviour of radially graded FG circular shaft with different material grading
patterns (LLMG, PLMG, and ELMG), end conditions (C–C ends, S–S ends, C–R
ends, S–R ends, and C-S ends), length to thickness ratios and temperature distribu-
tion (UTD, LTD, NLTD and ETD). The FG shaft considered here is composed of a
ceramic (Si3 N4 ) and metal (Ti-6Al-4 V) with decreasing metallic content at the outer
diameter. For FG (Si3 N4 /Ti-6Al-4 V) shaft, temperature dependent material proper-
ties are varied smoothly and continuously from the ceramic phase to the metal phase
based on material gradation laws (see Eqs. (1–4)) and corresponding coefficients
of bi-material FGMs are considered as reported in Touloukian [26]. The geometric
parameters for FG (Si3 N4 /Ti-6Al-4 V) shaft are assumed as follows: Ri = 0.05 m,
Ro = 0.1 m and different length to diameter ratios. The temperature at the inner
surface is always consedered as Ti = 300 K and at the outer surface it is varied for
considering the temperature gradient ΔT = To − Ti .

3.1 Influences of Material Gradient Index and Temperature

Gradient on w for S–S FG Shaft

Figure 3a plots the dimensionless transverse deflections of a simply supported FG

shaft subjected to uniformly applied load for different values of the material gradient
index k and for slenderness ratio L/(Ro − Ri ) = 20 and temperature gradient ΔT =
0 K. The numerical results are depicted that the deflection increases with the increase
in the magnitudes of k. This is due to the fact that the larger k, means the shaft has a
smaller amount of ceramic composition and larger amount of metallic composition
and hence the stiffness is reduced.

Fig. 3 Dimensionless transverse deflection of S–S FG shaft subjected to uniform load for different
a k and b ΔT
Thermo-Elastic Bending and Buckling Behavior of Functionally … 39

Figure 3b plots the dimensionless transverse deflections of a simply supported

FG shaft subjected to a uniformly applied load for different values of temperature
gradient ΔT and for slenderness ratio L/(Ro − Ri ) = 20 and material gradient index
k = 0.5. The numerical results are depicted that with increase in the magnitudes of
ΔT , deflection decreases. This is because with the increase in ΔT materials become
softer at high magnitudes of ΔT and hence stiffness is reduced.

/
3.2 Influences of L (Ro − Ri ) and End Conditions on ΔTcr
for FG Shaft with UTD, LTD, NLTD and ETD

Figure 4a, b show the influences of slenderness ratio L/(Ro − Ri ) and various end
conditions (C–C ends, S–S and C-R ends, S-R ends, and C-S ends) on the crit-
ical bucking temperature of radially graded FG (Ti-6Al-4 V/Si3 N4 ) shaft subjected
to UTD, LTD, and NLTD. Here, temperature gradient (ΔT ) is considered as 0 K
and PLMGI (k) is considered as 0.5 and, respectively. It is observed that the crit-
ical bucking temperature decreases as the ratio L/(Ro − Ri ) increases for all the
described temperature distribution. Also, it is observed that the critical buckling
temperature for S–S and C-R ends FG shaft are identical and lower than C–C and
C-S ends FG shaft, but larger than the FG shaft supported with S-R ends.
Figure 5 shows the influences of slenderness ratio L/(Ro − Ri ) and various end
conditions (C–C ends, S–S ends, C-R ends, S-R ends, and C-S ends) on the critical
bucking temperature of radially graded FG (Ti-6Al-4 V/Si3 N4 ) shaft subjected to
ETD for temperature gradient ΔT = 0 K, respectively. It is observed that the critical
bucking temperature decreases as the ratio L/(Ro − Ri ) increases for all the temper-
ature distribution. Also, it is observed that the critical buckling temperature for S–S
and C-R ends FG shaft are identical and lower than C–C and C-S ends FG shaft, but
larger than the FG shaft supported with S-R ends. Figure 4a–c and Fig. 5 show the
critical buckling temperature is higher for ETD than UTD, LTD and NLTD which
are obtained based on the material gradation laws (PLMG and ELMG).

/
3.3 Influences of L (Ro − Ri ) and Material Gradient Index
on ΔTcr for FG Shaft with UTD, LTD, and NLTD

The influences of slenderness ratio L/(Ro − Ri ) and material gradient index k on crit-
ical buckling temperature of simply supported FG (Ti-6Al-4 V/Si3 N4 ) shaft subjected
to UTD, LTD, and NLTD for ΔT = 0K, are presented in Fig. 6a, b respectively. It
is noticed from Fig. 6a that the critical bucking temperature decreases as the ratio
L/(Ro − Ri ) increases for all the described temperature distribution and the variation
is closer to each other for NLTD and LTD. However, for UTD, the critical buckling
temperature is quite lesser than the NLTD and LTD. It is noticed from Fig. 6b that
40 D. Gayen et al.

/
Fig. 4 Variation of a ΔTcrU ni b ΔTcrLT D and c ΔTcrN LT D with L (Ro − Ri ) and end conditions
for k = 0.5 and ΔT = 0K

/
Fig. 5 Variation of ΔTcrE T D with different L (Ro − Ri ) and end conditions for FG shaft for
ΔT = 0K
Thermo-Elastic Bending and Buckling Behavior of Functionally … 41

/ of ΔTcr for S–S FG shaft having UTD,/ LTD, and NLTD for ΔT = 0 K with
Fig. 6 Variation
a different L (Ro − Ri ) and k = 0.5 b different k and L (Ro − Ri ) = 40

the critical buckling temperature decreases as the material gradient index increases
around k = 10.0 for NLTD and LTD and then with the increase in k, the variation
of the critical buckling temperature is almost linear. However, in the case of UTD,
the critical buckling temperature decreases for k < 3, and then linear variation is
observed for 3 < k ≤ 20. It is also observed that the variation of the critical buckling
temperature is higher for NLTD and LTD than for UTD.

3.4 Influences of ΔT and Material Gradient Index on ΔTcr

for FG Shaft Subjected to UTD and NLTD

Figure 7a, b show the influences of material gradient indices (k) on the critical buck-
ling temperature (ΔTcr ) for radially graded FG (Ti-6Al-4 V/Si3 N4 ) shaft subjected
to UTD and NLTD for different values of temperature gradients (ΔT ) and for
L/(Ro − Ri ) = 40. It is observed that for UTD, the critical buckling temperature

Fig. 7 Variation of a ΔTcrU ni and b ΔTcrN LT D as functions of ΔT and k for S–S FG shaft subjected
to UTD, and NLTD
42 D. Gayen et al.

decreases for k < 3, and then linear variation is observed for 3 < k ≤ 20, and
also the critical bucking temperature decreases up to k = 10.0 for NLTD. It is also
observed that with the increase in temperature gradient (ΔT ) the critical buckling
temperature decreases as the materials become softer at higher magnitudes of ΔT
for both UTD and NLTD.

4 Conclusions

The aim of the present work is to derive analytical expressions for the bending and
buckling behaviour of radially graded FG (Ti-6Al-4 V/Si3 N4 ) shaft subjected to UTD,
LTD, NLTD and ETD. The whole derivation is based on Timoshenko beam theory and
thermo-elastic material properties are assumed following different material gradation
laws (LLMG, ELMG and PLMG) for the FG (Ti-6Al-4 V/Si3 N4 ) shaft. To this end,
influences of slenderness ratio L/(Ro − Ri ), material gradient index (k), temperature
gradient (ΔT ) and end conditions (C–C ends, S–S ends, C-R ends, S-R ends, and C-S
ends) on the bending and critical buckling temperature are presented and discussed.
The following points are summarized as main conclusions from the present work
based on various numerical examples:
• Numerical results reveal that bending behaviour is influenced by material gradient
index and temperature gradient, significantly, besides other parameters.
• The critical buckling temperature for the FG shaft is decreased with the increase
in slenderness ratio L/(Ro − Ri ) for all the temperature distributions and end
conditions.
• The critical bucking temperature is decreased with the increase in material gradient
index k as metallic content decreases with the increase in k for all the temperature
distributions and end conditions. It is observed that the critical buckling temper-
ature is lowest for a completely metallic shaft and is highest for a completely
ceramic shaft.
• The critical buckling temperature for S–S and C-R ends FG shaft are identical and
lower than C–C and C-S ends FG shaft, but larger than the FG shaft supported
with S-R ends, for all the temperature distributions.
• In the case of UTD, the critical buckling temperature is decreased for k < 3, and
then linear variation is observed for 3 < k ≤ 20. The critical bucking temperature
is decreased up to k = 10.0 and then linear variations are observed for NLTD and
LTD.
• The critical buckling temperature is decreased with the increase in temperature
gradient (ΔT ) as the material becomes softer at higher magnitudes of ΔT for
both UTD and NLTD.

Acknowledgements Debabrata Gayen would like to thank Prof. Rajiv Tiwari and Prof. Debabrata
Chakraborty for their encouragement and constant support for conducting the present research work.
Thermo-Elastic Bending and Buckling Behavior of Functionally … 43

References

1. Bever MB, Duwez PE. Gradients in composite materials. Mater Sci Eng. 1972;10:1–8.
2. Shen M, Bever MB. Gradients in polymeric materials. J Mater Sci. 1972;7:741–6.
3. Koizumi M. The concept of FGM. Ceram Trans. 1993;34:3–10.
4. Miyamoto Y, Kaysser WA, Rabin BH, Kawasaki A, Ford RG. Functionally graded materials:
design. London: Processing and Application. Kluwer Academic; 1999.
5. Birman V, Byrd LW. Modeling and analysis of functionally graded materials and structures.
Appl Mech Rev. 2007;60:195–216.
6. Swaminathan KD, Naveenkumar T, Zenkour AM, Carrera E. Stress, vibration and buckling
analyses of FGM plates—a state-of-the-art review. Compos Struct. 2015;120:10–31.
7. Eltaher MA, Khater ME, Emam SA. A review on nonlocal elastic models for bending, buckling,
vibrations, and wave propagation of nanoscale beams. Appl Math Model. 2016;40:4109–28.
8. Sayyad AS, Ghugal YM. Bending, buckling and free vibration of laminated composite and
sandwich beams: a critical review of literature. Compos Struct. 2017;171:486–504.
9. Gayen D, Tiwari R, Chakraborty D. Static and dynamic analyses of cracked functionally graded
structural components: a review. Compos B Eng. 2019;173: 106982.
10. Kiani Y, Eslami MR. Thermal buckling analysis of functionally graded material beams. Int J
Mech Mater Des. 2010;6:229–38.
11. Wattanasakulpong N, Ungbhakorn V. Analytical solutions for bending, buckling and vibra-
tion responses of carbon nanotube-reinforced composite beams resting on elastic foundation.
Comput Mater Sci. 2013;71:201–8.
12. Shi-Rong L, Da-Fu v, Ze-Qing W. Bending solutions of FGM Timoshenko beams from those
of the homogenous Euler-Bernoulli beams. Appl Math Model. 2013;37:7077–7085.
13. Ansari R, Gholami R, Shojaei MF, Mohammadi V, Sahmani S. Size-dependent bending, buck-
ling and free vibration of functionally graded Timoshenko microbeams based on the most
general strain gradient theory. Compos Struct. 2013;100:385–97.
14. Simsek M, Kocatürk T, Akbas SD. Static bending of a functionally graded microscale
Timoshenko beam based on the modified couple stress theory. Compos Struct. 2013;95:740–7.
15. Akgöz B, Civalek Ö. Thermo-mechanical buckling behavior of functionally graded microbeams
embedded in elastic medium. Int J Eng Sci. 2014;85:90–104.
16. Ebrahimi F, Salari E. Thermal buckling and free vibration analysis of size dependent
Timoshenko FG nanobeams in thermal environments. Compos Struct. 2015;128:363–80.
17. Jing L-L, Ming P-J, Zhang W-P, Fu L-R, Cao Y-P. Static and free vibration analysis of func-
tionally graded beams by combination Timoshenko theory and finite volume method. Compos
Struct. 2016;138:192–213.
18. Sayyad AS, Ghugal YM. Analytical solutions for bending, buckling, and vibration analyses of
exponential functionally graded higher order beams. Asian J Civil Eng. 2018;19:607–23.
19. Chen W-R, Chen C-S, Chang H. Thermal buckling analysis of functionally graded Euler-
Bernoulli beams with temperature-dependent properties. J Appl Comput Mech. 2020;6(3):457–
70.
20. Gayen D. Analysis of temperature, displacement, and stress in shafts made of functionally
graded materials with various grading laws. Adv Eng Mater. 2021;24(5):2101328.
21. Gayen D, Tiwari R, Chakraborty D. Thermo-mechanical analysis of a rotor-bearing system
having a functionally graded shaft with transverse breathing cracks. In: Proceedings of the 6th
national symposium on rotor dynamics. Springer; 2021. pp. 93–104.
22. Gayen D, Chakraborty D, Tiwari R, Determination of local flexibility coefficients of a function-
ally graded shaft with breathing crack. In: Advances in rotor dynamics, control, and structural
health monitoring. Springer; 2020. pp. 171–187.
23. Gayen D, Chakraborty D, Tiwari R. Stability behavior of two-crack functionally graded shaft
in a rotor-disc system: finite element approach. Mater Today Proc. 2020;24:432–41.
24. Gayen D, Chakraborty D, Tiwari R. Transverse vibration and stability of a cracked function-
ally graded rotating shaft system. In: Advances in materials and manufacturing engineering.
Springer; 2020. pp. 625–631.
44 D. Gayen et al.

25. Reddy JN, Chin CD. Thermoelastical analysis of functionally graded cylinders and plates. J
Therm Stresses. 1998;21:593–626.
26. Touloukian YS. Thermophysical properties of high temperature solid materials. New York:
Macmillan; 1967.
Modeling and Analysis of Low-Pressure
Steam Turbine Blade

Nitin Sehra, Sushila Rani, and Vikas Rastogi

Abstract The rotor blades are very crucial component in the steam turbine, and any
failures that occur, particularly in the final stages of the low-pressure (LP) blades,
can significantly reduce the lifespan of the turbine. The LP blades are operated at
severe dynamic conditions inside a steam turbine as they are exposed to significant
levels of stress and vibrations that can result in their failure. In order to optimize the
performance and longevity of steam turbines, it is imperative to investigate the fail-
ures of turbine blades. The current study focuses on utilizing computational analysis
to identify the factors responsible for blade failure, utilizing a real-life case study
as the basis for an investigation. This study employs ANSYS® software to conduct
static and dynamic analysis of a real-life L-0 stage low-pressure steam turbine blade.
The static structural analysis measures the stress distribution and determines the
critical location where the blade could fail during operation. An investigation into
the dynamic behavior of the LP steam turbine blade was conducted, and resonant
frequencies were determined. During start-up and shut-down operations, turbine
blades have to pass through many natural frequencies. As a result, the turbine blades
will be subjected to resonance and cumulative damage, so determining critical speeds
is vital. Campbell diagram is used to determine the critical speed at the intersecting
fundamental frequency with the harmonic’s lines, and through this, resonance can
be avoided.

Keywords Steam turbine blade · Static analysis · Dynamic analysis · Campbell

diagram

N. Sehra · S. Rani · V. Rastogi (B)

Department of Mechanical, Production, Industrial and Automobile Engineering,
Delhi Technological University, New Delhi, India
e-mail: [email protected]
S. Rani
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 45
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_3
46 N. Sehra et al.

1 Introduction

For many years, the power generation industry has faced a serious problem with the
breakdown of low-pressure steam turbine blades. According to the latest data from the
Ministry of Power, Government of India, thermal power plants contribute to around
62% of the total electricity generated in India. Due to the sudden failure of turbine
blades, the steam turbine stops working; such failure results in huge economic loss
[1]. One of the most critical challenges in power plant operations is the failure of rotor
blades, which causes premature failure of steam turbines and other internal compo-
nents. The Steam turbine comes under contact with severe dynamic forces subjected
to unsteady flow and centrifugal forces [2]. Low-pressure steam turbine blades are
more vulnerable to failure than high-pressure (HP) and intermediate-pressure (IP)
because LP blades derive a large portion of energy from the passing stream flow [3].
Low-pressure (LP) turbine blade failures can occur due to various factors, with a
significant proportion of failures associated with fatigue, stress corrosion cracking
(SCC), and corrosion fatigue [4]. One of the prime factors of failure in the steam
turbine blade is vibration persuasion fatigue [5–7] the fluid flow in the steam turbine
blade excites the blade vibrations, and it becomes dangerous when resonance occurs
as it starts to vibrate with large amplitude and result in catastrophic failure of blade
[8]. Multiple investigators have performed vibration analysis to scrutinize the cause
of failure in low-pressure steam turbine blades.
Modal analysis and harmonic analysis are identified as efficient techniques for
examining the vibration or dynamic characteristics of blades [9]. A steam turbine
blade has to pass several critical speeds at distinct operating and dynamic conditions.
These blades are designed to operate above and pass through from different critical
speeds safely. The blade natural frequency investigation is a big concern for the safety
of the steam turbine. The Campbell diagram is used to identify potential resonance
conditions, which occur when the machine’s rotational speed matches with one of
its natural frequencies [10]. Bhat et al. [11] have determined the natural frequencies
of the fourth and fifth stage of a low-pressure (LP), 235 MW steam turbine blade,
two-blade disk system, and three-blade disk system using the Finite element method,
dual-channel FFT analyzer, Campbell diagram. They have found that the frequencies
decrease when the number of blades increases. Also, their conclusion indicates that
a model with a single-blade disk provides a more precise prediction of the natural
frequencies. Rani et al. [12] identified the failure mode and the exact location of
the crack on the first-stage gas turbine blade. The investigative process involved
the utilization of two methods, experimental modal analysis using OROS ® vibra-
tion analyzer, and also done numerical analysis which employs finite element anal-
ysis (FEA) methodology by using ANSYS ® software. The comparison of the results
obtained from both techniques led to the conclusion that the findings of experimental
modal analysis aligned with those of the computational analysis. Additionally, the
study revealed that resonance at the sixth modal frequency was the underlying cause
of the blade failure. Prabhunandan et al. [13] have performed the dynamic anal-
ysis of a single-blade made of titanium material and also for the assembly of rotor
Modeling and Analysis of Low-Pressure Steam Turbine Blade 47

blades at two different speeds, 8500 rpm, and 19,000 rpm. They have observed that
at 8500 rpm, the blade is safe, but critical speed is obtained when the assembly of
blades runs at 19,000 rpm. Based on their findings, it was concluded that in order to
prevent resonance, the turbine must operate at a speed below the critical threshold.
This research paper focuses on the comprehensive analysis of the final stage low-
pressure steam turbine blade. The study involves conducting static structural and
dynamic analysis using finite element analysis, employing the ANSYS® software
as the primary tool.
This paper aims to locate the stress concentration regions responsible for low-
pressure steam turbine blade failure. Furthermore, the resonant frequencies of the
steam turbine blade are ascertained by subjecting the blade to a modal analysis. The
operational condition of the steam turbine may induce resonant vibration in the blade,
and it can be predicted with the help of the Campbell diagram (Table 1).

2 Computation Analysis of an LP Steam Turbine Blade

The blade is collected from the 29th stage of 210 MW thermal power plant. The
low-pressure steam turbine blade is made up of X10CrNiMoV12-2-2 alloy. Visual
examination shows that the blade has a hairline crack in the transverse direction at a
distance of 350 mm from the blade’s tip, as shown in Fig. 1.
To obtain the precise shape and dimensions of the blade, a 3D laser scanning
process was conducted to create a three-dimensional CAD model of the blade using
a laser light line. Scanning was done with the latest blue light scanning technology by
Solutionix C500 structure blue light scanner. By utilizing the static structure analysis
toolbox of ANSYS ® 19, the solid model of the scanned blade was created through the
transformation of its 3D CAD model as shown in Fig. 2.
Figure 3 illustrates, the mesh model of the blade depicting a triangular surface
mesh created initially, followed by volumetric meshing using the auto meshing capa-
bilities of ANSYS ® software. The generated finite element modal has a total number
of nodes as 50,478 and a total number of elements as 28,996. The fixed support is
provided at the fir-tree part of the blade as its boundary condition, as shown in Figs. 4
and 5. The turbine blade has fixed supports at the root. The operating speed of the
turbine blade is taken as 3000 rpm. The rotation of the blade induces centrifugal
forces that result in a stiffening impact on the blade.
The centrifugal force acting on the blade can be calculated as:
Where,
( )/
F(centrifugal force) = ρAω2 R22 − R21 2 (1)

ρ is Material density = 7850 kg/m3

A is the cross-section of turbine blade = 0.00097m2
R1 equal to (R2 —Length of the turbine blade) = (1.6–0.9) m
48 N. Sehra et al.

Table 1 Comparison of the causes of low-pressure steam turbine blade failure

S. no Author Title Methodology Failure reason
1 Zhao et al. Vibration analysis Analyzed dynamic Found that the blade failure
[10] for failure behavior both is the result of the combined
detection in LP theoretically and effects of fretting wear and
steam turbine experimentally. 3D high cycle fatigue.
blades in nuclear FEM method, According to the Campbell
power plant Campbell diagram diagram, the 2nd vibration
mode of the blade has a
higher chance of resonance
with the 9th harmonic when
the rotational speed is at
2828 rpm
2 Rani et al. Vibration analysis Global Rational Found that the experimental
[12] for detecting Fraction Polynomial model’s frequencies were
failure mode and Method (GRFP), lower and resonance at the
crack location in Frequency Response sixth modal frequency was
first-stage gas Function, Scanning discovered to be the cause of
turbine blade Electron Microscopy the blade failure. SEM
(SEM) images revealed that high
cycle fatigue is the reason
for gas turbine blade failure
3 Bhagi et al. Dynamic Stress The author used Found that both dynamic
[5] Analysis of mathematical and stresses and centrifugal
L-1Low-Pressure computational stresses were the reason for
Steam Turbine modeling to examine fatigue failure in
Blade: the impact of stresses low-pressure steam turbine
Mathematical on the low-pressure blades
Modeling and steam turbine blade’s
Finite Element dynamic response
Method
4 Present Modeling and Performed static A maximum deformation of
work Analysis of structural and dynamic 47 mm was observed near
Low-pressure analysis of the 3D cad the blade’s tip and found that
steam turbine model of LP steam tip of the blade is critical for
blade turbine blade using the failure. Also identified, the
FEM approach critical speeds with the help
(ANSYS ® 19 software) of the Campbell diagram, as
to determine the the Campbell diagram helps
deformation, mode the operator to find out the
shape, and natural regions of safe limits where
frequencies to prevent resonance can be avoided
resonance

R2 is blade tip radius of the blade = 1.6 m

ω is rotational speed of the turbine blade = 314.2 rad/s
The calculated centrifugal force is 778 KN/blade.
Modeling and Analysis of Low-Pressure Steam Turbine Blade 49

Fig. 1 Shows the crack location and actual condition of the LP steam turbine blade

Fig. 2 3D Cad Model of LP steam turbine blade

Fig. 3 Mesh Model of LP steam turbine blade

50 N. Sehra et al.

Fig. 4 Blade is fixed at the fir-tree part of the blade

Fig. 5 Centrifugal force applied on the blade

2.1 Chemical Composition

For this research, a real case of the last stage of the LP steam turbine blade is taken.
The blade is made of X10CrNiMoV12-2–2 alloy. The chemical composition of the
blade material is tabulated in Table 2.

Table 2 [14] Chemical composition of X10CRNIMOV 12–2-2 alloy

Element C Cr Ni Mo Mn V Si
Weight-% 0.117 11.4 2.70 1.64 0.70 0.31 0.23
Modeling and Analysis of Low-Pressure Steam Turbine Blade 51

Table 3 Mechanical properties of blade material

Young Tensile Ultimate Elongation Contraction Vickers Density Poisons
Modulus Yield Tensile at fracture at fracture Hardness (Kg/m3 ) ratio
(GPa) Strength strength (HV)
(MPa) (MPa)
213 843 1001 17.7% 58% 334 7850 0.3

2.2 Mechanical Properties

The mechanical properties of the blade material (X10CrNiMoV12-2–2 alloy) are

tabulated in Table 3.

2.3 Static Structural Analysis of Blade

Static structural analysis has been performed to identify the deformation and stress
distribution in a Low-pressure steam turbine blade. This analysis allows us to identify
the potential risks and areas that are critical and could fail during operation. The
study is conducted with a numerical approach, and the software used is ANSYS®
19. The maximum deformation of 47 mm was observed near the blade’s tip (Fig. 6).
This shows that the area near the tip or at the blade’s tip is critical for failure. The
Maximum Principal Stress, Maximum Principal Strain and Shear stress induced in
the blade have a maximum value of 713.81 MPa, 0.00318 MPa, and 83.203 MPa,
respectively (Figs. 7, 8 and 9). The Von-Mises stress induced in the blade is near the
concave side of the root and has a maximum value of 678 MPa and a minimum value
of 0.465 MPa (Fig. 10).

Fig. 6 Maximum deformation of the LP steam turbine blade

52 N. Sehra et al.

Fig. 7 Maximum Principal Stress of the LP steam turbine blade

Fig. 8 Maximum Principal Elastic Strain of the LP steam turbine blade

Fig. 9 Maximum Shear Stress of the LP steam turbine blade

2.4 Dynamic Analysis of LP Steam Turbine Blade

A finite element approach is employed to perform dynamic analysis of an LP steam

turbine blade. The three-dimensional model of the blade is created and analyzed
Modeling and Analysis of Low-Pressure Steam Turbine Blade 53

Fig. 10 Maximum Von-Mises Stress of the LP steam turbine blade

on ANSYS® software. The first six modal frequencies are obtained as 45.984 Hz,
108.62 Hz, 192.75 Hz, 240.2 Hz, 318.54 Hz and 394.54 Hz. The first six natural
frequencies and their respective mode shapes are shown in Figs. 11, 12, 13, 14, 15
and 16.

Fig. 11 Mode 1 (Bending) of the blade at frequency 45.984 Hz

Fig. 12 Mode 2 (Bending-Torsion) of the blade at 108.62 Hz

54 N. Sehra et al.

Fig. 13 Mode 3 (Torsion) of the blade at 192.75 Hz

Fig. 14 Mode 4 (Bending-Torsion) of the blade at 240.2 Hz

Fig. 15 Mode 5 (Bending-Torsion) of the blade at 318.54 Hz

2.5 Campbell Diagram of Lp Steam Turbine Blade

The Campbell diagram is a graphical representation that displays the relationship

between blade rotational speed and frequency and is utilized to predict potential
resonant conditions during blade operation. It becomes necessary to determine the
Modeling and Analysis of Low-Pressure Steam Turbine Blade 55

Fig. 16 Mode 6 (Bending-Torsion) of the blade at 394.54 Hz

Table 4 Natural frequencies of the blade at the different rotational speed

Mode 0 rpm 500 rpm 1000 rpm 1500 rpm 2000 rpm 2500 rpm 3000 rpm
(Hz) (Hz) (Hz) (Hz) (Hz) (Hz)
First 45.984 39.788 30.006 22.884 18.173 14.964 12.676
Natural
frequency
Second 108.625 102.642 96.375 91.390 86.275 80.810 75.211
Natural
frequency
Third 192.752 208.534 201.284 186.963 174.446 164.892 157.960
Natural
frequency
Fourth 240.199 255.954 294.068 302.082 304.015 304.776 305.157
Natural
frequency
Fifth 318.538 324.219 372.317 465.675 548.790 625.429 703.505
Natural
frequency
Sixth 394.538 411.251 453.053 528.410 636.922 764.685 898.966
Natural
frequency

component’s critical speed at the intersecting fundamental frequency. The Campbell

diagram is drawn on the Graph Pad software to draw harmonic lines. Three harmonic
lines were graphed on Graph Pad software, with each line starting from the zero point
and sloping upwards toward the right of the vertical axis. Within the operational speed
range, the point where the natural frequency lines and the harmonic lines meet signi-
fies the likelihood of blade resonances. According to the manufacturer’s operation
guide, the operating speed range for low-pressure steam turbines is 2880 rpm to
3000 rpm. The first six natural frequencies of the blade at different rotational speeds
are obtained by dynamic analysis of the blade and tabulated in Table 4 (Fig. 17).
56 N. Sehra et al.

Fig. 17 Campbell Diagram of LP steam turbine blade

3 Results and Discussion

This research analyzes a real case LP steam turbine blade made of X10CrNiMoV12-
2-2 alloy to identify the critical areas and failure regions or zones during operation.
When a load of 778 kN/blade was applied to the blade, then maximum deformation
of 47 mm was observed near the tip of the blade. This shows that the tip of the blade
is critical for failure. Other stresses such as Von-Mises stress, maximum principal
stress, maximum principal strain, and shear stress of the LP steam turbine blade are
also evaluated computationally to identify the blade areas prone to failure.
Dynamic analysis is also performed to evaluate the natural frequencies of the
blade so that resonance conditions can be avoided during operation. Further, the
first six natural frequencies at 0 rpm, 500 rpm, 1000 rpm, 1500 rpm, 2000 rpm,
2500 rpm, and 3000 rpm have been evaluated. The low-pressure steam turbine runs
at 3000 rpm and 50 Hz, and when the turbine starts from zero rpm to reach 3000 rpm,
it has to pass through many critical speeds. The critical speeds are obtained by the
intersection of the harmonics line with the natural frequencies line, as shown in the
Campbell diagram in Fig. 16. The operator needs to prevent a long run on these critical
speeds and should move in the safe zone as soon as possible to avoid resonance. So,
it has been necessary to provide safe zones to the operator for machines such as
steam turbine blades with different natural frequencies. Campbell diagram helps the
operator find the regions of safe limits where resonance can be avoided.
The intersection of the harmonics line with the frequency line obtains the critical
speeds. Critical speeds at the 7th, 15th and 18th harmonic lines are shown in Table 5.
Modeling and Analysis of Low-Pressure Steam Turbine Blade 57

Table 5 Critical Speed at 7th, 15th, 18th harmonic lines

7th harmonic line (rpm) 15th harmonic line (rpm) 18th harmonic line (rpm)
356.3 175.25 147.2
842.94 414.65 348.19
1584.4 815.82 686.12
2613.1 1188.3 973.49
2504.6 1617.2
2967.1

4 Conclusion

Computational analysis of the LP steam turbine blade was carried out to detect the
areas on the blade that are critical and prone to failure. The following conclusions
were drawn:
• Computational analysis has been carried out on a failed low-pressure steam turbine
blade of the final stage. A 3D scanning process is employed to create a precise and
efficient representation of the blade’s dimensions, contours, and profiles, resulting
in a three-dimensional CAD model
• Deformation, Von-Mises stress, maximum principal stress, maximum principal
strain, and shear stress of the LP steam turbine blade are analyzed through static
structural analysis using ANSYS® Software. The stress distribution pattern reveals
that the concave root portion of the blade’s trailing edge bears the maximum stress
of 678 Mpa, which is caused by steam flow and centrifugal load. Hence, it will be
the vulnerable zone for crack initiation and propagation during severe dynamic
conditions. However, the blade encounters stress levels that do not exceed its
yield strength, so the blade is safe at this load but owing to other dynamic running
conditions and steam force; these may be elevated.
• To perform dynamic analysis of a turbine blade at various speeds, it is necessary to
run the blade through variable speed cycles. The blade’s first six natural frequen-
cies are obtained as 45.984 Hz, 108.62 Hz, 192.75 Hz, 240.2 Hz, 318.54 Hz and
394.54 Hz, and their respective mode shapes have been drawn so that resonance
conditions could be avoided during operation.
• Due to substantial fluctuations in the natural frequencies of the blade, these fluc-
tuations may affect the critical speed of components which may fail within the
operational range. So, the Campbell diagram analysis is required to find out
the critical speed of the blade in that condition. The Campbell diagram shows
three harmonic lines, i.e., 7th, 15th and 18th, intersecting with the 4th, 5th, and
6th modes of the low-pressure steam turbine blade at 2613 rpm, 2504 rpm, and
2967 rpm, respectively.
58 N. Sehra et al.

References

1. Rani S, Agrawal AK, Rastogi V. Failure analysis of a first stage IN738 gas turbine blade tip
cracking in a thermal power plant. Case Studies Eng Failure Anal. 2017;8:1–10.
2. Měšťánek P. Low cycle fatigue analysis of a last stage steam turbine blade. Appl Comput Mech.
2008;2:71–82.
3. He H, Zheng Z, Yang Z, Wang X, Wu Y. Failure analysis of steam turbine blade roots. Eng
Failure Anal. 2020;115(May):1–11.
4. Mukhopadhyay NK, Ghosh Chowdhury S, Das G, Chattoraj I, Das SK, Bhattacharya DK. An
investigation of the failure of low-pressure steam turbine blades. 2001;5(3). Elsevier Science
Ltd.
5. Bhagi LK, Rastogi V, Gupta P, Pradhan S. Dynamic stress analysis of L-1 low-pressure
steam turbine blade: mathematical modelling and finite element method. Mater Today Proc.
2018;5(14):28117–26.
6. Tsai GC. Rotating vibration behavior of the turbine blades with different groups of blades. J
Sound Vib. 2004;271(3–5):547–75.
7. Mohan RS, Sarkar A, Sekhar AS. Vibration analysis of a steam turbine blade. INTERNOISE
2014—43rd Int. Congr. Noise Control Eng. Improv. World Through Noise Control, 2014;
pp. 1–10.
8. Sanvito M, Pesatori E, Bachschmid N, Chatterton S. Analysis of LP steam turbine blade
vibrations: experimental results and numerical simulations. Wood head Publishing Limited,
2012.
9. Shukla A, Harsha SP. Vibration response analysis of last stage LP turbine blades for variable
size of crack in root. Procedia Technol. 2016;23:232–9.
10. Zhao Q, Li Y, Xue M, Wang P, Jiang J. Vibration analysis for failure detection in low pressure
steam turbine blades in nuclear power plant. Eng Fail Anal. 2018;84:11–24.
11. Bhat MM, Ramamurti V, Sujatha C. Studies on the determination of natural frequencies of
industrial turbine blades. J Sound Vib. 1996;196(5):681–703.
12. Rani S, Agrawal AK, Rastogi V. Vibration analysis for detecting failure mode and crack location
in first stage gas turbine blade. J Mech Sci Technol. 2019;33(1):1–10.
13. Prabhunandan GS, Byregowda HV. Dynamic analysis of a steam turbine with numerical
approach. Material Today Proc. 2018;5(2):5414–20.
14. Kovacs S, Beck T, Singheiser L. Influence of mean stresses on fatigue life and damage of a
turbine blade steel in the VHCF-regime. Int J Fatigue. 2013;49:90–9. https://doi.org/10.1016/
j.ijfatigue.2012.12.012.
Enhanced Energy Harvesting
Application of Piezoceramics (PZT)
in MEMS Devices

G. Pranavanathan and S. B. Kandagal

Abstract Piezoceramic (PZT) layers are more widely used as vibration energy
harvesters. These materials have found a greater application both in actuation and
sensing. In this study, the characterization of energy output from optimally located
PZT under base excitation and force excitation is investigated and experimentally
correlated. SDOF (Single Degree of Freedom) mathematical modeling is commonly
used to estimate the voltage generated across the PZT which may result in inaccu-
rate output. The cantilever beam with PZT is modeled as a continuous system with
discrete PZT patches at high strain zones depending on the mode under consideration.
The output voltage from PZT can be enhanced significantly using tip mass. This has
better applications in MEMS devices. A detailed simulation study and experimental
verification have been investigated on a cantilever beam with tip mass. The space
constraints in MEMS devices call for using a tip mass to generate higher voltage for a
smaller beam length. The influence of rotary inertia of tip mass is primarily focused.
The sensitivity parameters consider the beam to PZT length ratio, Young’s modulus
ratio and tip mass. Optimization criteria to achieve maximum energy generation
has been studied using the mode superposition method to find the optimal voltage
generated across the PZT. Electrodynamic shaker, B&K accelerometer, conditioning
amplifiers and National Instrument data acquisition are used during the experimental
investigation. The energy thus harvested from a beam with tip mass and PZT can
encourage the use of power generated for power supply to the MEMS device. The
results are encouraging to implement in MEMS devices.

Keywords Energy harvesting · Base excitation · Rotary Inertia · Piezoceramic

(PZT) · Mathematical modeling · Mode superposition · Damping · MEMS
devices · Weight optimization

G. Pranavanathan (B) · S. B. Kandagal

Department of Aerospace Engineering, Indian Institute of Science, Bengaluru 560012, Karnataka,
India
e-mail: [email protected]
S. B. Kandagal
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 59
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_4
60 G. Pranavanathan and S. B. Kandagal

1 Introduction

The utilization of piezoelectric materials (PZTs) for energy harvesting has garnered
significant prominence owing to their convenient application. Environmental vibra-
tions inherently possess a certain amount of kinetic energy, which can be efficiently
converted into electrical energy through the use of piezoelectric materials. This
technique proves highly advantageous for low-power electronic devices, including
wireless sensors, IoT devices, sensor-controlled structural condition health moni-
toring systems, and implantable medical devices, due to the broad frequency range
and adaptability it offers. Vibrations and other forms of kinetic energy abound in
various manifestations all around us, such as machine vibrations, flowing water,
wind, human motion, and shock waves. Moreover, the manufacturing of complete
electronic devices with a power consumption of less than 10 mWs (10 mJ) can be
accomplished in a cost-effective manner.
The pioneering contributions of Hagood and VonFlotow [1] shaped the land-
scape of energy harvesting by devising a compelling experimental demonstration
and model. Their innovative approach harnessed the potential of shunt circuits to
achieve damping, while tuned mass dampers were utilized for effective vibration
control. Through their meticulous analysis of the latest literature, Steven and Henry
[2] provide a comprehensive overview of the field of power harvesting. Their review
delves into the current advancements, applications, and future prospects of this tech-
nology, particularly in the realm of self-powered devices. Eurterk and Inman [3] have
conducted a comprehensive analysis on the dynamic characterization and modeling
of cantilever-based piezoelectric vibration energy harvesters. Their study focuses
on understanding the intricate dynamics involved in these devices and developing
accurate mathematical models to represent their behaviour. Specifically, they inves-
tigate the response of cantilever structures integrated with piezoelectric transducers
(PZTs) to vibrations, considering factors such as translational and rotational motions
at the base of the beam under base excitation. By carefully examining and quanti-
fying the mechanical properties and performance of these harvesters, the research
contributes to the advancement of efficient energy harvesting technologies. Chen
et al. [4] have put forth an influential proposition, advocating for the integration
of PZT ceramics and novel circuit design. Their research introduces an ingenious
power circuit design that specifically targets efficient power generation. The key high-
light of their work is the introduction of a hybrid direct current (DC-DC) system,
which seamlessly combines a switched capacitor DC-DC converter and a low dropout
(LDO) linear voltage regulator. Vatansever et al. [5] conducted an extensive study
on energy harvesting from renewable sources using PVDF and PZT patches. The
research focused on evaluating the voltage responses of ceramic-based piezoelectric
fiber composite structures (PHCs) and polymer-based piezoelectric strips (PVDF)
under varying wind speeds and water droplet conditions. The investigation aimed
to explore the potential of these natural renewable energy sources for generating
energy suitable for low-power electronic devices. Shukla et al. [6] propose an uncon-
ventional energy harvesting concept that targets the low-frequency movements of
Enhanced Energy Harvesting Application of Piezoceramics (PZT) … 61

the human waistline. By harnessing the kinetic energy derived from activities like
walking and jogging through a precisely designed pendulum mechanism, this concept
offers an efficient means of capturing and converting the harvested energy into elec-
trical power using a compliant piezoelectric unimorph. Abu Raihan et al. [7] have
conducted an extensive and comprehensive review focused on vibration-based power
generators employing electromagnetic and piezoelectric transducer mechanisms. The
study delves into the utilization of electromagnetic, piezoelectric, and electrostatic
transducers, which effectively convert the inherent kinetic energy from mechanical
vibrations into electrical energy. Tang and Wang [8] delve into the impact of tip
mass size on the performance of a cantilevered piezoelectric energy harvester with a
dynamic magnifier. The research highlights the observation that even a minor alter-
ation in tip mass can have a substantial influence on the accuracy of the analysis.
Kanno [9] has conducted an investigation aimed at developing innovative microde-
vices with commercial potential, including gyro sensors and micropumps for inkjet
printer heads. The research focuses on creating functional microdevices that offer
promising commercial opportunities.

2 Methodology

The cantilever beam considered for the study is shown in Fig. 1. Euler–Bernoulli
beam theory is used. The base excitation translation g(t) is considered for unimorph
bender having two layers. The substructure layer (at the base) and PZT layer (at
the top) as proposed by Eurturk and Inman [3] is used in the theoretical approach
(Fig. 2).
The absolute displacement of the beam at any point x along the beam axis in the
transverse direction ( ie in y-direction) is denoted by w(x, t). The equation of motion
for free vibrations (without damping) can be expressed as

Fig. 1 Aluminum beam

with PZT patch (Unimorph)
under base excitation
62 G. Pranavanathan and S. B. Kandagal

Fig. 2 Geometric details of

the subsystem

∂ 4 w(x, t) ∂ 2 w(x, t)
EI +m =0 (1)
∂x 4 ∂t 2
where EI is the flexural stiffness (E is Young’s modulus and I is the area moment
of Inertia) and m is the mass per unit length of beam. The beam is excited with
base motion at the fixed end mounted on the electrodynamic shaker. Following the
standard modal expansion method, the solution of Eq. (1) can be represented by an
absolutely and uniformly convergent series of the eigen functions as
∞
∑
wrel (x, t) = φr (x)ηr (t) (2)
r =1

where ϕr and ηr are the mass normalized eigen function and the modal coordi-
nate of the clamped-free beam for the rth mode respectively. Considering the stan-
dard boundary conditions of a fixed-free beam, the resulting mass normalized eigen
function of the rth mode is
√ [ ( )]
1 λr λr λr λr
φr (x) = cosh x − cos x −σr sinh x − sin x (3)
mL L L L L

Simplification of Eq. (2) with Eq. (3) with damping coefficient ξ and modal forcing
function Nr the response of the beam can be expressed as
∞
∑ ∫
φr (x) t
wrel (x, t) = Nr (τ )e−ζr ωr (t−τ ) sinωr d (t − τ )dτ (4)
r =1
ωr d τ =0

The open-circuit voltage across the PZT at a certain instant of the vibration can be
obtained based on the self-sensing actuator concept [3]. The PZT is used as a sensor,
the electric displacement can be obtained.

D3 = d31 T1 (5)
Enhanced Energy Harvesting Application of Piezoceramics (PZT) … 63

In order to obtain the electric displacement, it is required to find the axial bending
strain at a point of interest is a function of the radius of curvature S1 (x, t) of the
neutral axis and the distance hc of that point from the neutral axis

hc
S1 (x, t) = − (6)
ρ(x, t)

where the radius of curvature is related to the transverse deflection by

1 ∂ 2 wrel (x, t)
= (7)
ρ(x, t) ∂x2

Using Eq. (3) and Hooke’s law along with Eqs. (6) and (7) the electric displacement
can be rewritten as

∂ 2 wrel (x, t)
D3 (x, t) = −d31 Yp h c (8)
∂x2
where Yp is the Young’s Modulus of the PZT layer and hc is the distance between
the center of the PZT layer (in thickness direction) and the neutral axis of the beam.
Using general form of the relative displacement can be obtained after evaluating the
spatial integral under the assumption that relative slope at t the root of the PZT, the
charge on the electrodes can be expressed as
| ∫
dφr (x) ||
∞
∑ t
1
q(t) = − d31 Yp h c b | × Nr (τ )e−ζr ωr (t−τ ) sinωrd (t − τ )dτ
r =1
dx | ωr d τ =0
x=L
(9)

The voltage v(t) across the electrodes of the PZT is obtained by dividing the charge
on the electrodes by capacitance Cp of the PZT patch x1 and x2 are the locations of
the PZT patch location details
|x=x2 ∫
d31 Yp h c b ∑ dφr (x) ||
∞ t
1
v(t) = − | × Nr (τ )e−ζr ωr (t−τ ) sinωr d (t − τ )dτ
Cp r =1
dx | ωrd τ =0
x=x1
(10)

Since the beam with PZT is under harmonic base excitation in the transverse
direction, the maximum steady state voltage at resonance (ω = ωn ) is written for the
rth mode as
|x=x
γrw d31 Yp h c bmY0 ei (ωr t−π/2) dφr (x) || 2
v(t) = − (11)
2ζr Cp d x |x=x1

The results of the above theoretical approach are shown in Fig. 6.

64 G. Pranavanathan and S. B. Kandagal

The ANSYS APDL software was employed to conduct finite element analysis,
aiming to establish a correlation and simulate the voltage generated across the PZT
structure, as depicted in Fig. 3. For the piezo patch, the SOLID 5 element type
was assigned. Within the ANSYS framework, SOLID5 elements possess compre-
hensive capabilities in 3D magnetic, thermal, electric, piezoelectric, and structural
fields, with some limitations in the coupling between these fields. Each SOLID5
element consists of eight nodes, wherein each node exhibits up to six degrees of
freedom. Remarkably, these elements exhibit remarkable capabilities in handling
large deflections and stress stiffening phenomena, rendering them highly suitable
for structural and piezoelectric analyzes. On the other hand, the SOLID 45 Element
was designated for the substructure. In ANSYS, SOLID 45 elements are regarded as
the optimal choice for isotropic structures. Furthermore, the tip mass was assigned
the 3D MASS21 element. MASS21 elements in ANSYS represent point elements
that allow for up to six degrees of freedom. These degrees of freedom encompass
translational movement along the nodal x, y, and z directions, as well as rotations
around the nodal x, y, and z axes. It is possible to assign distinct masses and rotary
inertias to each coordinate direction. In this particular study, the Mass21 element
was assigned to the tip mass along the y-axis.
The Finite Element Analysis (FEA) is meticulously conducted with the objective
of acquiring essential information through two distinct analyzes i.e. Modal Analysis
and Harmonic Analysis. Modal Analysis, a crucial component of the study, facil-
itates the determination of the system’s inherent frequencies. By delving into this
analysis, one can gain valuable insights into the natural vibrational characteristics of
the system under investigation. Harmonic Analysis, on the other hand, is undertaken
to discern the system’s response concerning both voltage and displacement in rela-
tion to the varying frequency inputs. This endeavor can be approached through two

Fig. 3 FEA model in Ansys APDL

Enhanced Energy Harvesting Application of Piezoceramics (PZT) … 65

potential methodologies i.e. the Full Method and the Modal Superposition Method.
However, it is worth noting that, for the present simulations, the Full Method has been
employed consistently. Thus, through a judicious amalgamation of Modal Analysis
and Harmonic Analysis conducted via the Full Method, the FEA analysis endeavors
to unveil a comprehensive understanding of the system’s natural frequencies and its
response to frequency-induced variations in voltage and displacement. The outcomes
derived from the finite element method are visually presented in Fig. 7.

3 Experimental Details

Experimental tests were conducted on an aluminum cantilever beam specimen with

surface bonded piezoceramic patches. One end of the aluminum beam is clamped at
the electrodynamics shaker to simulate clamped boundary condition, PZT is attached
on one side at 23 mm from the fixed end. The experimental setup is shown in Figs. 4
and 5.
Material properties are given in Table 1. The electrodynamics shaker (Derritron
VP 4/4B) is used.

Base accelerometer
Charge Amplifier
Charge Amplifier type B & K 2635
type B & K 2635
Voltage

Beam
Piezoceramic
(Pzt 5-H)

Tip accelerometer
Derritron
electro magnetic
shaker
Type Vp4/4B
Signal
generator Data Acquisition
/Power system
amplifier

Fig. 4 Experimental setup

Fig. 5 Beam with PZT on

electrodynamic shaker
66 G. Pranavanathan and S. B. Kandagal

Fig. 6 Theoretical (Euler–Bernoulli) simulation for tip amplitude and voltage generated for beam
with PZT

Fig. 7 Voltage generated across PZT for the beam with PZT for different tip masses [FEM]

Table 1 Beam and PZT

Subsystem Parameter Value
material properties
PZT Length (Lp ) 50.8 mm
Width (b) 25.4 mm
Thickness (tp ) 0.5 mm
Young’s modulus (Yp ) 69 GPa
Piezo-strain constant (d31) 265e-12 m/V
Aluminium Beam Length (L0 ) 254 mm
Width (b0 ) 26 mm
Thickness (t0 ) 0.5 mm
Young’s modulus (Y0 ) 70 GPa

The base acceleration is measured by a 4335 B&K (30gms, sensitivity 71.37 pc-
g−1 ) type accelerometer and the tip acceleration is measured by 4344 B&K (2.1gms,
sensitivity 2.87 pc-g−1 ) type accelerometer. Both the sensors are calibrated by B&K
4294 calibrator. The accelerometers are connected to two separate charged amplifiers
Enhanced Energy Harvesting Application of Piezoceramics (PZT) … 67

Fig. 8 Experimental voltage

generated across PZT for
beam (no mass)

Fig. 9 Experimental voltage

generated across PZT for
beam with 10 gm mass at tip

B&K 2635. The output signal is connected to data acquisition system NI PCI 6259
to acquire the response of the beam. Tip amplitude of 2.5 mm is maintained as the
reference for all the cases. The resultant voltage across the PZT in the beam is shown
in Figs. 8 and 9.
The experimental setup used here is similar to Gurubrahmam et al.’s study on
amplitude variation with air damping [10].

4 Results and Discussion

The Euler–Bernoulli beam theory [3] is used to estimate the voltage across the PZT.
The tip amplitude and the voltage generated across the PZT are shown in Fig. 6 for
the case without tip mass. The results are verified with FEM software as shown in
Fig. 7 and further correlated with experiments as observed in Figs. 8 and 9. Here
damping is measured with logarithmic decrement method and found to have damping
68 G. Pranavanathan and S. B. Kandagal

Table 2 Beam with PZT and tip mass: modal parameters and power generated
Subsystem Natural frequency (Hz) Voltage generated (V) Power
generated
(micro Watt)
Method Theory FEM Expt Theory FEM Expt FEM
Beam (no 27.6 27.6 27.5 11.4 11.4 11.18 1354
mass)
Beam (tip 5 22.13 22.13 22.10 14.9 14.88 14.81 1858
gm)
Beam (tip 10 18.95 18.95 18.75 18.4 18.4 17.9 2414
gm)
Beam (tip 15 16.83 16.83 16.79 22.0 22.0 21.9 3069
gm)
* Note The power generated was determined by the FEM method

coefficient of 1.3%. The experimentally measured damping is used in FEM and in

Eq. (11) to estimate the voltage generated across the PZT for various cases. The
agreement for the theoretical, FEM and experimental results is found to be good and
the simulation results are listed in Table 2. The tip mass at the cantilever beam is
5 gms, 10 gms and 15 gms which is a fraction of the beam weight of 35 gms. The
PZT covered is about 20% of the beam length. The voltage generated across the PZT
can be enhanced by 60% and power by 80% with the addition of 30% weight of
the beam as mass at the tip. The voltage generated across the PZT in theoretical and
Finite element simulation reflects 0-peak values in contrast to peak-peak values as
observed in experimental investigation as shown in Figs. 7 and 8.

5 Conclusions

The voltage generated across the PZT can be effectively enhanced by optimally
locating the PZT patch in the 30% area of beam near to root. The power gener-
ated can be effectively used to power the MEMS devices. The concept tried in this
paper is the optimization of the PZT patch covering the full-length versus the 20%
coverage near the root. The tip mass can be 30% of the total beam weight to over-
come the weight penalty of PZT and less strain as the PZT is moved towards the tip.
The classical Euler–Bernoulli’s beam theory and FEM are used to verify the voltage
across the PZT and correlated and compared with experimental results, good corre-
lation is observed. The results of this optimization can be used in ultra-low-power
applications, particularly in MEMS and wireless sensor power applications.
Enhanced Energy Harvesting Application of Piezoceramics (PZT) … 69

References

1. Anderson EH, Hagood NW. Simultaneous piezoelectric sensing/actuation: analysis and

application to controlled structures. J Sound Vib. 1994;174(5):617–39.
2. Anton SR, Sadano HA. A review of power harvesting using piezoelectric materials. Smart
Mater Struct. 2007;16:R1–R21.
3. Erturk A, Inman DJ. On Mechanical modeling of cantilevered piezoelectric vibration energy
harvestors. Jl Intell Mater Syst Struct. 2008;19:1311–25.
4. Chen H, Jia C et al. Power harvesting with ceramics and circuit design. Analog Integer Circ
Sign Process. 2010:263–268.
5. Vatansever D, Hadimani RL, Shah T, Siores E. An investigation of energy harvesting from
renewable sources with PVDF and PZT. Smart Mater Struct. 2011;20:055019.
6. Shukla R, Bell AJ. PENDEXE: a novel energy harvesting concept for low frequency human
waistline. Sensors Actuat. 2015;222:39–49.
7. Abu Raihan MS, Mahmud S et al. A comprehensive review on vibration based power genera-
tors using electromagnetic and piezoelectric transducer mechanisms. Energy Conver Manag.
2015;106:728–747.
8. Tang L, Wang J. Size effect of tip mass on performance of cantilevered piezoelectric energy
harvestor with dynamic magnifier. Acta Mech. 2017;228:3997–4015.
9. Kanno I. Piezoelctics MEMS for energy harvesting. J Phys Conf Ser. 2015;660: 012001.
10. Gurubrahmam K, Ramamohan Rao T, Chandra Sekhar Reddy M. Study on added damping
due to resistive shunting on cantilever, simply supported beam, and effect on the variation of
amplitude at resonance. Proc Inst Mech Eng Part C J Mech Eng Sci. 2023;237(7):1613–1630.
Diesel Locomotive Alternator Bearing
Damage Detection by Adopting
Comprehensive Condition Monitoring
Techniques

Hemant M. Bari, Suhas S. Patil, and Atul A. Deshpande

Abstract This paper shares a success story of the implementation of condition

monitoring techniques, wherein imminent alternator bearing deterioration of diesel
locomotive was diagnosed. Coordinated condition monitoring techniques vibration,
acceleration, noise, and ultrasonic analysis helped in diagnosing the exact root cause
of abnormity of alternator knocking sound and deterioration of bearings at notch 3 and
4 speed during load condition at early stage and thus avoided the catastrophic failure
of alternator. During maintenance work execution, thorough inspection of alternator
rotor, stator winding, and exciter was carried out in the presence of original equipment
manufacturer (OEM) expert. No damage or scoring marks were observed. This has
saved the plant from huge losses in terms of demurrage risk and coal availability
for electricity generation. This case study has clearly demonstrated the importance
of acceleration as a valuable tool in rotating machinery analysis and is especially
useful in detecting early and advanced bearing degradation even if vibration levels
are on the lower side. Noise measurement and ultrasonic analysis played a major
supportive role to confirm the source of abnormal knocking sound from alternator. It
is a very powerful diagnostic tool. Every vibration analyst should use it to improve
vibration analysis accuracy which can help maintenance to become more proactive
and accurate.

Keywords Vibration · Bearing · Acceleration · Noise · Ultrasonic · Frequency

analysis

H. M. Bari · S. S. Patil · A. A. Deshpande (B)

Department of Maintenance Planning, Adani DTPS, Dahanu, Maharashtra, India
e-mail: [email protected]
H. M. Bari
e-mail: [email protected]
S. S. Patil
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 71
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_5
72 H. M. Bari et al.

1 Introduction

Following is a success story of the implementation of condition monitoring tech-

niques at Dahanu thermal power station (DTPS), wherein alternator non-drive end
(NDE) and drive end (DE) bearing deterioration of diesel locomotive was diagnosed.
Initially, maintenance team was focusing on abnormal knocking sound reported by
loco pilot & possible maintenance actions were discussed with operation and main-
tenance team. Then, the issue referred to the condition monitoring cell and source
of knocking sound was detected using coordinated condition monitoring techniques
vibration, noise, and ultrasonic monitoring. This technique analysis helped to reach
the root cause of abnormality well in advance guiding the maintenance team on
what exact actions to be executed to avoid any catastrophic failure of alternator and
secondary damage to other rotating parts of the locomotive [1, 2].

2 Methodology for Detection of Alternator Bearing

Detoriation

Methodology discusses regarding how alternator non-drive end (NDE) and drive end
(DE) bearing deterioration is detected in a systematic way. Here in Sect. 2.1, the pre-
maintenance history is analyzed, and a complete sequence of events (Refer Fig. 1)
is listed to know the problem that arise and how the maintenance team resolved the
same with the help of condition monitoring data analysis.

Fig. 1 Acceleration spectrum of alternator non-drive end (NDE) bearing

Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 73

2.1 Pre-maintenance History

With reference to Table 1, on 27.06.2015, an abnormal knocking sound was noticed

from alternator side at the time of start and stop. Vibration and acceleration readings
were taken; all readings were found well within limits. The noise level measured
for reference was 98.5 decibel (dB). The abnormal sound at the time of start and
stop was constant for the next 45 days. No variation in vibration and acceleration
value was noticed. Meanwhile, ultrasonic monitoring started alternator drive end
(DE) bearing side and database generated. On 02/08/2015, the loco pilot observed
that, abnormal continuous knocking sound comes from the alternator after 4 h of
continuous operation at notch 03 and 04 speed. On 04/08/2015, decoupled trial of
diesel engine taken, and no knocking sound was observed which confirmed that the
source of knocking sound is from alternator only. Hence, it was decided to keep a
close watch on all condition monitoring parameters. On 10/08/2015, readings were
taken after 04 h of operation when the knocking sound was continuous at load.
On 11/08/2015, detailed vibration analysis, noise, and ultrasonic monitoring report
along with various probable causes discussed in plant meeting. Then, it was decided
to grease alternator bearings and collect one more set of readings. The alternator
bearings greasing was done and on 12/08/2015 @10.30 am, noise level checked in the
presence of maintenance team and observed that all readings are still on higher side.
On 12/08/2015, in view of steep increasing trend of alternator bearings acceleration
value, noise level, and ultrasonic level, it was decided to stop loco operations for
replacement of alternator bearings. Detailed diagnosis analysis was carried out as
follows.

Table 1 Sequence of events

Date Event occurred/activity done
27/06/2015 Abnormal knocking sound noticed from Alternator only at the time of start & stop.
Vibration & acceleration reading taken, all readings found well within limits
02/08/2015 Loco pilot observed that, abnormal continuous knocking sound comes from
alternator side after 4 h of continuous operation at notch 03 & 04 speed
04/08/2015 Decoupled trial of diesel engine taken & no knocking sound observed which
confirmed that source of knocking sound is from alternator only
10/08/2015 Alternator drive end (DE)bearing acceleration value fluctuates from 0.3 to 1.3 g
&non-drive end (NDE) from 0.4 to 0.9 g. Noise level fluctuates from 101 to
110 dB..ultrasonic value deviated from 5 to 9 leds
11/08/2015 Detailed vibration analysis, noise & ultrasonic monitoring report along with
various probable causes discussed in daily hod meeting
12/08/2015 Both alternator bearings greasing done & noise level checked in presence of hods
@ 10.30am & observed that all readings are still on higher side
12/08/2015 In view of steep increasing trend of alternator bearings acceleration value, noise &
@ 11.30am ultrasonic level, decided to stop loco operation for replacement of alternator
bearings
74 H. M. Bari et al.

3 Experimental Details

Experimental details have four different Sects. 3.1–3.4 which discusses complete
analysis for condition monitoring data and based on that execution of maintenance
work is done. Section 3.1 discusses about vibration and acceleration data analysis
with the use of supportive tools like noise and ultrasound analysis. Section 3.2 iden-
tifies the exact cause for condition monitoring data deviation and confirms the pres-
ence of bearing deterioration for alternator non-drive end (NDE) and drive end (DE)
bearing. Based on analysis, observations during maintenance work are discussed in
Sect. 3.3 and maintenance work execution i.e. replacement of both alternator bearings
is illustrated in Sect. 3.4.

3.1 Vibration Diagnosis

On 12/08/2015, alternator non-drive end (NDE) and drive end (DE) bearing accelera-
tion values were on higher side (0.9 g and 1.3 g respectively). Acceleration spectrum
(Refer Figs. 1 and 2) analysis revealed that there is increase in higher frequencies and
the contribution of non- synchronous energy (any vibration that is greater than the 1x
frequency but is not a whole number multiple of it) is at a higher side which is mainly
due to bearing defects. From vibration diagnosis, it was suspected deterioration of
alternator bearings. To confirm the fault, ultrasonic analysis and noise monitoring
were carried out. With reference to Figs. 3 and 4, higher noise and ultrasonic values
confirmed the deterioration of alternator bearings.

Fig. 2 Acceleration spectrum of Alternator drive end (DE) bearing

Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 75

Fig. 3 Ultrasonic level after greasing

Fig. 4 Noise level after greasing

3.2 Frequency Analysis

As discussed in Sect. 3.1, acceleration spectrum (Refer Figs. 1 and 2) analysis

revealed that there is an increase in higher frequencies and the contribution of non-
synchronous energy. These high frequency stress wave events occur at repeatable
time intervals corresponding to a bearing defect frequency, then it is likely that a
bearing defect is present. These frequency stress waves event occur at repeatable
time intervals corresponding to a bearing defect frequency, then it is likely that a
bearing defect is present (Refer Table 2).
Hence, it was concluded that acceleration value of both alternator bearings
increased due to early stage deterioration which in turn increased the noise and ultra-
sonic level of alternator at higher rpms. Therefore, it was decided to stop locomotive
operations for replacement of alternator bearings.
76 H. M. Bari et al.

Table 2 Frequency analysis chart

Synchronous Non-synchronous Sub-synchronous
Probable causes
Mass unbalance Bearing defects Nearby machines and components
Looseness Cavitation Severe looseness
Worn or damaged gears Electrical fault Rubbing of rotors and shafts
Eccentric rotation Nearby machines and components Belt vibrations
Misalignment System resonances
Reciprocating motion
Vane pass

Fig. 5 Alternator pulley

3.3 Observations During Maintenance Work

Condition of alternator pulley (Refer Fig. 5), non-drive end (NDE) and drive end
(DE) bearing housings and cooling fan found okay (Refer Fig. 6). Minor scoring
marks were observed on both bearing rollers (Refer Fig. 7) and inner race (Refer
Fig. 8). Through inspection of alternator stator winding, rotor and exciter carried out
in presence of original equipment manufacturer (OEM) expert. No scoring marks or
abnormality was observed (Refer Figs. 9 and 10).

3.4 Maintenance Work Carried Out

Both alternator bearings were removed (Refer Fig. 11) and replaced with new ones
(Refer Fig. 12). Meanwhile, all critical associated parts such as alternator rotor (Refer
Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 77

Fig. 6 Alternator bearing

housing

Fig. 7 Scoring marks on

bearing roller

Fig. 13) and stator (Refer Fig. 14) cleaning and overhauling are done. Alternator was
put back in position (Refer Fig. 15) and aligned with respect to the diesel engine
and exciter (Refer Fig. 16). Trial was taken and no abnormal knocking sound was
observed from alternator side at the time of start and stop. Locomotive boxed-up and
made available for operation. Full load trial (with 16 loaded wagons) was taken, and
all condition monitoring parameters were found well within limits.
78 H. M. Bari et al.

Fig. 8 Scoring marks on

bearing inner race

Fig. 9 Alternator rotor

4 Results and Discussion

Post maintenance condition monitoring data is well within limits after the execu-
tion of maintenance work. (i.e., after replacing alternator bearings) (Refer Table 3
and Fig. 17). Presently, the plant has 02 locomotives for hauling the coal wagons
to wagon tippler for unloading. These locomotives have been in service for more
than 21 years. Considering the coal unloading requirement of plant, the availability
of these locomotives cannot be compromised. Non-availability of locomotive will
greatly affect the unloading of coal wagons within the stipulated time allotted by
Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 79

Fig. 10 Alternator stator

Fig. 11 Removal of old

alternator bearing

railways and any delay will lead to heavy demurrage risk of rupees 150 per wagon
per hour causing financial loss of rupees 9000 per hour for 60 wagons hampering
the coal availability for electricity generation. With reference to Fig. 18, leanings are
implemented successfully to detect future deviations well in advance.
80 H. M. Bari et al.

Fig. 12 Installation of new

alternator bearing

Fig. 13 Overhauling of
alternator rotor
Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 81

Fig. 14 Overhauling of
alternator stator

Fig. 15 Installation of
alternator with pulley

Fig. 16 Alignment of
alternator with rest of drive
82 H. M. Bari et al.

Table 3 Post Maintenance vibration and acceleration data

Vibration report of locomotive
Point Position At load and notch 04 speed
Initial readings After replacing alternator
bearings
12/08/2015 17/08/2015
Velocity Acceleration Velocity Acceleration
(mm/sec) value (g) pk (mm/sec) value (g) pk
RMS RMS
Alternator (NDE) H 2.3 0.96 1.6 0.32
V 2.1 – 1.5 –
A 1.9 – 1.6 –
Alternator (DE) H 2.5 1.35 1.9 0.39
V 2.6 – 2.0 –
A 2.1 – 1.7 –
Diesel engine (DE) H 3.0 0.8 2.0 0.7
V 2.1 – 1.9 –
A 2.3 – 1.3 –
Ultrasonic level LEDs 09 05
Noise level dB 110.2 102.6

Fig. 17 Post maintenance noise and ultrasonic data

Diesel Locomotive Alternator Bearing Damage Detection by Adopting … 83

Fig. 18 Online vibration monitoring system made operational for loco

5 Conclusion

Coordinated condition monitoring techniques vibration, noise and ultrasonic analysis

helped in diagnosing the exact root cause of abnormity of alternator knocking sound
and deterioration of bearings at notch 3 and 4 speed during load condition at early
stage and thus avoided the catastrophic failure of alternator this case study has clearly
demonstrated that acceleration is a valuable tool in rotating machinery analysis and is
especially useful in detecting early and advanced bearing degradation even vibration
levels are on lower side. Noise measurement and ultrasonic analysis played a major
supportive role to pinpoint and confirm the damage of alternator bearing. It is a very
powerful diagnostic tool every vibration analyst should use it to improve vibration
analysis accuracy as ultrasonic analysis can help maintenance organization to become
more proactive and accurate.

References

1. Rao JS. Vibratory condition monitoring of machines. New Delhi: Narosa Publishing House;
2000.
2. Harris TA, Kotzalas MN. Essential concepts of bearing technology. 5th ed. Boca Raton, Florida:
Taylor and Francis; 2007.
Response Analysis of Inclined Edge
Cracked Beam Under Moving Mass

Shakti P. Jena, Naresh Kumar Sarella, Sasmita Sahu, and Dayal R. Parhi

Abstract The dynamic behavior of structure with the presence of inclined edge
crack is an emerging topic in the era of structural dynamics. In the present analogy,
a novel method is focused to investigate the dynamic response of an inclined edge
cracked simply supported structure subjected to moving mass. The responses of
the beam structure subjected to traversing load are calculated with different crack
inclination angles and crack depth using Duhamel integral approach. For the authen-
tication of the numerical approach, the finite element analyzes (FEA) is accom-
plished to corroborate the exactness of computational approach (Duhamel integral
approach). The FEA approach has been conducted using the commercial ANSYS
WORKBENCH software. The significance of the crack parameters (crack inclina-
tion angle and crack depth) and moving mass parameters (moving speed and magni-
tude of moving mass) on the responses of the beam are investigated. The numerical
analyzes followed by FEA approaches are illustrated for different crack angles, crack
depth, moving speed and moving mass. The results which are obtained from both
the numerical method as well as FEA approach are compared with each other. As
per the results are concerned, the applied numerical analogy converges well towards
the FEA approach. So the Duhamel integral method is found to be convergent and
can be useful to obtain the response of structure under transit mass.

Keywords Crack inclination angle · Crack depth · Moving mass

S. P. Jena (B)
Department of Mechanical Engineering, Nalanda Institute of Technology, Bhubaneswar, India
e-mail: [email protected]
N. K. Sarella
Department of Mechanical Engineering, JNTU Kakinada, Vardhaman College of Engineering,
Hyderabad, India
e-mail: [email protected]
S. Sahu
School of Mechanical Engineering, KIIT University, Bhubaneswar, India
e-mail: [email protected]
D. R. Parhi
Department of Mechanical Engineering, National Institute of Technology, Rourkela, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 85
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_6
86 S. P. Jena et al.

Nomenclature

M Moving or Transit mass magnitude,

v Transit velocity
L Beam length
d Crack depth
b Beam width
H Beam thickness
δ Dirac Delta function
λ Eigen function
Tn (t) Amplitude function
Y(x) Shape function
y(x, t) Response function
n Number of modes
α Relative depth of crack at the intact positions
l Crack length at the crack opening
β = vt Position of the transit mass
ωn Frequency at nth mode of the beam
EI Flexural rigidity
m̄= ρA Beam mass per unit length
ρ Density of the structure
A Area of cross-section of the beam
θ Crack inclination angle

1 Introduction

The vibrant behavior of structure under transit mass is an emerging topic in the era
of structural dynamics. The behavior of the structure will alter in a different way in
dynamic loading condition as comparison to static loading condition. During real
operating condition or mechanized progression, defects may arise in the structure.
Those defects may be in the form of cracks which are inducing due to natural or arti-
ficial effects. The cracks also may be in various forms. Due to the existence of cracks
(effects of severity and localities) and consequence of moving load parameters, the
response of the structure is significantly affected. Again due to the development
of various types of cracks on the structure, the response will be exhibited differ-
ently. Researchers and engineers have carried out numerous works on moving load
dynamics problem with the mechanism of various types of cracks.
Reis and Pala [1, 2] have explored the vibration based amplitude of cracked struc-
tures (cantilever and simply supported) under transit mass. The impacts or effects of
centripetal, vertical and Coriolis’s accelerations of the moving mass on the behavior
of the vibrating of the structure are included and its significances are also discussed.
Jena and Parhi [3–5] also performed several works (Numerical, FEA, experiments) to
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 87

find out the vibration based amplitudes of different types of beams under traversing
load. They have also investigated the possessions of crack constraints on the responses
of the different structures. Viola et al. [6] investigated the effect of interface of crack
on flutter and divergence stabilities subjected to considerable forces. The nature of
the force is triangularly distributed in their analyzes. They have obtained the stability
maps from the analysis of Eigen values which predicts the domain of the divergence
and flutter. Attara et al. [7] formulated a numerical model to analyze the possessions
of surface defects on dynamic behavior of beam under traversing mass. The formu-
lation has been carried out on a Timoshenko beam under viscoelastic foundation.
The lattice spring model concepts were utilized to ascertain the connection between
the relating springs and resilient properties of the material.
Rjoub and Hamad [8] developed an analytical solution to characterize the forced
vibration responses of axially loaded damaged Euler–Bernoulli’ and Timoshenko
beams along with two transit loads traveling at constant speed. They have found out
the special effects of the existence of number crack and its depth on the response of
the structure. They have also obtained the significance of the two transit load’s speed
and axially loading on the responses of the structure. Sarvestan et al. [9] obtained the
spectral finite element model to scrutinize the dynamics of moving mass problem.
They have taken into account a Timoshenko beam for the analysis and carried out the
solution in both time and frequency domain. They have also discussed the significance
of intensity and locality of cracks on the vibrant response of structure. They have
also emphasized the supremacy of spectral FEA as compared to the general finite
element method. Kim and Lee [10] proposed the frequency domain spectral element
technique to obtain the vibrant response of a multi-span structure under transit force.
The cause of resonance vibrations which are induced due to the impact of a series of
moving forces are also investigated in this study.
Rajasekaran and Khaniki [11] presented the general material variation for the free
vibration analysis of single and multi-fractured beam using FEM. The significance
of bidirectional material variation on the beam structure was investigated. They have
also derived the stiffness matrix for the bidirectional functionally graded cracked
Euler–Bernoulli’s beam. They have carried out a parametric study to analyze the char-
acteristics of material variation and cracks. Nikhoo and Sharifinejad [12] inspected
the consequence of cracks on the inertial effects of transit forces on thin beam. The
output of their study revealed that the reduction of stiffness because of the occurrence
of crack has an insignificant impact in the case of moving mass problem as compar-
ison to moving load problem. The study has also revealed that for high speed moving
forces, the retort of the structure due to traversing mass is not at all times superior
than that of traversing load even at certain depth of crack. At high speed, the inertial
effects have less impact with increasing crack depth. Guler [13] proposed a mathe-
matical model by considering a cracked axially rotating functionally graded beam to
analyze the free vibration analysis of the structure. The Rayleigh–Ritz method was
implemented to carry out the analysis of the centrifugally stiffened Euler–Bernoulli’s
beam.
88 S. P. Jena et al.

Kharazan et al. [14] inspected the nonlinear performance of beam with the pres-
ence of numerous breathing cracks. They have applied both the continuous polyno-
mial function and the Weierstrass estimation to replicate the bilinear performance
of those types of cracks. The compassion of the nonlinear response of the beam
with multiple breathing cracks was also investigated. Panigrahi and Pohit [15] have
performed the analysis of nonlinear modeling of cracked Timoshenko functionally
graded beam in the time domain. Initially, the crack modeling was done for open
crack, later the method was extended to breathing crack. They have also developed
an innovative multi-degree of freedom discrete model to analyze the performance
of breathing crack. Their study was focused on the significance of various cracks,
material properties and loading parameters on the behavior of the structure. Gupta
et al. [16] presented a methodical analysis to learn the pulsation features of fractured
plate structures at varying thicknesses. The influences of crack spot on the elemental
frequency of the structure are discussed and the nonlinear response curve of the plate
with varying thickness is too shown in their investigation. They have observed that
the crack and crack location have a significance impact on the primary frequency of
tapered plate.
Eftekhari and Jafari [17] approached the mixed method (Combination of FEM,
Differential Quadrature Method (DQM), and Integral Quadrature Method (IQM) to
revise the transitory performance of multi-span rectangular plates under traversing
locomotives. They have applied the FEM technique with advanced order exclama-
tion functions to separate the spatial fractional derivatives relating to a co-ordinate
direction of the plate followed by DQM and IQM to analyze the differential equations
of the structure. The resulting equation was obtained from the mixed method, after-
ward resolved by Newmark integration approach. The method developed by them can
produce very accurate results with little quantity of finite elements and DQ sampling
units. Tan et al. [18] formulated a technique on the free vibration investigation of
incessant beam with indiscriminate number of cracks and spring mass system. They
have also used this formed free vibration analysis approach for damage detection.
The effects of the crack parameters and spring mass structural system on the natural
frequencies of the proposed beam are probed. Chatterjee and Vaidya [19] focused
their studies (Theoretical and Experimental) to examine the effect of speed of the
vehicle on the dynamic response of a bridge structure under transit load. They have
developed a laboratory test model (small scale) to obtain the deflection and accel-
eration of the proposed structure. They have surveyed that amid enlarging speed,
there will be an increase in vibration amplitude but the transient fluctuation will
be decreasing. Akbas [20] has studied the propagation of wave in an edge cracked
FGM beam under a sudden force. The sudden force was modulated by harmonic
motion with transverse triangular impulse force. The detailed belongings of crack
deepness & place and various material distributions on the propagation of wave of
the proposed FGM structure are investigated.
As per the literatures are concerned, on the way to the preeminent of the authors’
acquaintance, the vibration analysis of inclined edge cracked beam under transit
mass is a little. Again, the retort of the inclined edge cracked structure under transit
loading condition using Duhamel integral approach along with FEA verification
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 89

is also very inadequate. Here, the proposed formulation is intended to find out the
vibration amplitudes of inclined edge cracked beam at various crack inclination angle
and transit speed.

2 Problem Statements

In the present problem statement, the dynamic behavior of an inclined edge cracked
beam (simply supported) under transit mass is spotlighted. As shown in Fig. 1, a
transit mass having magnitude ‘M’ with speed ‘v’ is moving. The crack is situated
at a distance of ‘L 1 ’ from the left support end. The entire analysis of the problem is
to obey Euler–Bernoulli’s postulation.
The prevailing equation of this problem can be expressed as:

∂4 y ∂2 y
EI + m = F(x, t)
∂x4 ∂t 2
[ ( 2 2 )]
∂ y ∂2 y 2∂ y
W her e F(x, t) = Mg − M + 2v + v δ(x − vt) (1)
∂t 2 ∂ x∂ x ∂x2

If β = vt, then the above equation can be written as:

[ ( )2 ]
∂ 4 y(x, t) ∂ 2 y(x, t) ∂ ∂
EI +m = Mg − M + v ) y(β, t) δ(x − β) (2)
∂x4 ∂t 2 ∂t ∂β

The resolution of the prevailing Eq. (1) may be marked in series form i.e.
∞
∑
y(x, t) = Yn (x) Tn (t) (3)
n=1

Fig. 1 Schematic view of inclined edge cracked simply supported beam under moving mass
90 S. P. Jena et al.

Fig. 2 Enhances view of the

inclined cracked zone

To appraise shape function i.e. Yn (x), Eq. (3) can be formed as

Yniv (x) − λ4n Yn (x) = 0 (4)

ω2
where λ4n = m E nI , n = 5 = No of modes of natural vibration
The detailed geometry of the crack portion of the structure is shown in Fig. 2.
Because of the continuation of inclined edge crack on the beam, the crack depth
is varying on the entire crack segment. The detailed geometrical analysis of the
proposed crack section is presented in Fig. 2. α = d/H , L 2 − L 1 = crack length,
d = Crack depth at the intact position (constant)., d = Maximum depth at the crack
intact position. The crack zone length (L 2 − L 1 ) now divided into ‘N’ no of equal
parts i.e. ‘r ’ = 1, 2, 3…….N. From the geometrical analysis (Fig./2.), the dimensions
of the inclined crack section are/determined i.e. L 2 − L 1 = d tan θ , In the crack
portion length of each part, = d N tan θ . Length of ‘r th’ part of the crack section
/
= r d N tanθ . The ) segment below the inclined crack division =
( width/of the initial
H-AA' = H − L (2 − L 1 N/ cot θ . The ) r th part of the width beneath the inclined
segment = H − r L 2 − L 1 N cot θ .
α ' = d ' /H = Relative crack depth in the inclined crack segment which is changing
as the depth of the crack (d ' ) in the crack zone is varying right through the crack
subdivision.
Similarly, the thicknesses of the other inclined sectors are calculated by consid-
ering the geometry.
Due to the occurrence of the inclined edge crack, the structure will now exhibit like
a beam with multiple cracks with the properties of open slanting and inner cracks. So
the structure is now stimulated with different beam fragments. Considering Euler–
Bernoulli’s beam theory to each segment of the beam, the common solution of Eq. (3)
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 91

for transverse deflection of each one fragment of the beam can be expressed as:

Yn (x) =A1 sin(λn x) + B1 cos(λn x) + C1 sinh(λn x)

+ D1 cosh(λn x), 0 ≤ x < L 1 (5a)

Yn (x) =A2 sin(λn x) + B2 cos(λn x) + C2 sinh(λn x)

( )
L2 − L1
+ D2 cosh(λn x),L 1 ≤ x < L 1 + r (5b)
N

Yn (x) =A3 sin(λn x) + B3 cos (λn x) + C3 sin(λn x)

+ D3 cosh(λn x),L 2 ≤ x < L (5c)

A, B, C and D are the different constants which describe the amplitude and shape
of the structure and are determined from the boundary conditions of the simply
supported structure [21].
Substituting Eq. (3) in (2), multiplying Yn (x) and integrating the newly formed
equation with the total length (L) of beam, the then equation may be stated as:

N [
∑ ] ∫L Mg
∫L
Tn,tt + ωn2 Tn Ym (x)Yn (x)d x = δ(x − β)Ym (x)d x
m
n=1 0 0
⎡ ⎤
∑
N ∫L
M⎢ ⎥
− ⎣ Tm,tt Yn (x)Ym (x)δ(x − β)d x ⎦
m
m=1 0
⎡ ⎤
∫L
M ⎢∑ 2
N
⎥
− ⎣ v Tm Yn (x)Ym (x)δ(x − β)d x ⎦
m
m=1 0
⎡ ⎤
∑
N ∫L
M⎢ ⎥
− ⎣ 2vTm,t Yn' (x)Ym (x)δ(x − β)d x ⎦
m
m=1 0
(6)

Tn,t and Tn,tt symbolize the primary and secondary derivatives of Tn (t) respec-
' ''
tively, while, Yn,x and Yn,x symbolize the primary and secondary derivatives of Yn (x)
respectively.
Applying the mechanism of orthogonality principle, Eq. (6) can be written as:
⎧ ⎡ ⎤ ⎡ ⎤⎫
M ⎨ ∑
N ∑
N ∑
N ⎬
Tn,tt + ωn2 Tn = Yn (β) g − ⎣ Tm,tt Ym (β)⎦ − v 2 Tm Ym'' (β)] − ⎣2v Tm,t Ym' (β)⎦ (7)
m ⎩ ⎭
m=1 m=1 m=1

Because of the orthogonal properties of Eigen functions, so it yields:

92 S. P. Jena et al.
( )
L 2 −L 1
L 1 +r
∫L ∫L ∫L 1 ∫ N
∫L
Ym2 d x = Yn2 d x = 2
Yn1 dx + 2
Yn2 dx + 2
Yn3 dx = k (8)
0 0 0 L1 L2

Applying Picard’s scheme (iterative approach) to Eq. (7) and considering the
initial phrase on the R.H.S of that equation, the solution of equation can be written
by this form [2]:
⎧ ⎫
⎪ Mg L1 ⎪
⎪
⎪ φ (β), 0 <t ≤ ⎪
⎪
⎪
⎪ m
n1
v ⎪
⎪
⎪
⎨ Mg ( )⎪
⎬
L 1 L 2 − L 1
Tn,tt + ωn Tn =
2
φn2 (β), < t ≤ L1 + r (9)
⎪
⎪ m v N ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ Mg L L ⎪
⎪
⎩ φn3 (β),
2
<t ≤ ⎭
m v v

The resolution for the homogenous division of Eq. (7) for t ≤ L1

v
is

(Tn1 )h = a1 sin ωn t + a2 cos ωn t (10)

where a1 and a2 are the invariable terms to be calculated.

Considering the correct resolution of Eq. (7) in this type i.e.

(Tn1 ) p =An1 sin Ωn t + B n1 cos Ωn t + C n1 sinh Ωn t

+ D n1 cosh Ωn t ,where Ωn = λn v (11)

Applying the values from Eq. (11) in Eq. (9), it gives:

Mg An1 Mg Bn1
An1 = ( ) ,B n1 = ( )
m ωn − Ω2n
2 m ωn − Ω2n
2

Mg Cn1 Mg Dn1
C n1 = ( ) ,D n1 = ( ) (12)
m ωn + Ω2n
2 m ωn + Ω2n
2

The common solution of Eq. (7) can be expressed, when t ≤ L1

v
t ≤ L1
v

Tn1 (t) = a1 sin ωn t + a2 cos ωn t + (Tn1 ) p (13)

'
Commencing the preliminary condition Tn1 (0) = 0 and Tn1 (0) = 0 and after
generalization, one can calculate

Ωn
a1 = − (An1 + C n1 ) ,a2 = (−B n1 − D n1 ) (14)
ωn
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 93

The last form of Tn1 (t) can be expressed as

Tn1 (t) =Tm1 (t) = (Tn1 )h + (Tn1 ) p = a1 sin ωn t + a2 cos ωn t

+ An1 sin Ωn t + B n1 cos Ωn t
+ C n1 sinh Ωn t + D n1 cosh Ωn t (15)

Substituting Eq. (15) in Eq. (7), when t ≤ L1

v
, then the Eq. (9) can be expressed
as:

Tn1,tt + ωn2 Tn1

⎧ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎫
2M ⎨ ∑
N ∑
N ∑
N ⎬
= Y (β) g − ⎣ Tm1,tt Ym1 (β)⎦ − v ⎣
2 ''
Tm1 Ym1 (β)⎦ − 2v ⎣ '
Tm1,t Ym1 (β)⎦ = Q n1 (16)
m L n1 ⎩ ⎭
m=1 m=1 m=1

The values of Tm1 , Tm1,t and Tm1,tt are to be calculated using Eq. (15).
Again substituting Eq. (15) in the R.H.S of Eq. (16) and then;

L1
Tn,tt + ωn2 Tn = Q n1 ,t ≤ (17)
v
Commencing from the initial states, the elucidation of Eq. (16) can be articulated
as:

∫t
1 L1
Tn (t) = Q n1 (τ ) sin ωn (t − τ )dτ ,t ≤ (18)
ωn v
0

( )
L 2 −L 1
L 1 +r
L1
<t ≤
N
Similarly for v v
, one can develop the equation like;
( )
L 2 −L 1
L 1 +r
∫1 / v
L ∫ N
1 1
Tn (t) = Q n1 (τ ) sin ωn (t − τ )dτ + Q n2 (τ ) sin ωn (t − τ )dτ
ωn ωn
0 L1/ v
(19)

For the last part of the beam segment;

∫1 / v
L
L2 L 1
< t ≤ Tn (t) = Q n1 (τ ) sin ωn (t − τ )dτ
v v ωn
0
( )
L 2 −L 1
L 1 +r
∫ N
1
+ Q n2 (τ ) sin ωn (t − τ )dτ
ωn
L1/ v
94 S. P. Jena et al.

L
∫ /v
1
+ Q n3 (τ ) sin ωn (t − τ )dτ (20)
ωn
L2/ v

Equation (20) which is the final solution equation has been explained by
employing Duhamel integral method and the response of the structure has been
obtained.

3 Finite Element Analysis (FEA) and Numerical

Formulation

The equation of motion for linear elastic system can be used to solve the Eigen value
problem by reducing the dimension of the system. In the present work, the proposed
numerical method has been authenticated using ANSYS WORKBENCH2015 soft-
ware. The dynamic equation of motion for initial value problem by ignoring the
damping effect of the structure can be given as:

M[ü t ] + K [u t ] = F(t) (21)

where M[ü t ] = Inertial force, K [u t ] = Internal or Stiffness force, F(t) = External

force to the transit mass.
The present FEA analysis includes Newmark’s integration method. The linear
acceleration with implicit type method has been adopted here to determine the
vibration amplitudes of the structures.
The crack modeling has been done in FEA which is shown in Fig. 3b. The
responses of the structure under different parameters of cracks and moving load
are determined compared with those obtained from numerical analyzes. For the
comparison of results between numerical and FEA analyzes, numerical examples
are formulated by considering a mild steel beam. For the numerical analysis the
following parameters are considered; Beam dimensions = (140 × 5 × 0.5) cm,
crack angle (θ ) = 22°, 32° and 42°, Transit mass (M) = 2 & 3 kg, Moving velocity
(v = )5.68 & 7.17 m/s), d = 0.25 cm are considered. The results obtained from FEA
are elaborated in the results and discussion part.

4 Results and Discussion

The present problem formulation has been focused to find out the responses or
vibration amplitudes of the inclined edge cracked beam under transit mass. The
concluding equation of motion of the proposed problem has been solved by Duhamel
integral approach followed by FEA authentication. The responses of beam due to the
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 95

Fig. 3 a FEA model of the moving mass structure, b FEA model of crack at θ = 42°

movement of transit mass are determined at different crack inclination angles, transit
masses and transit speeds. The dynamic response or vibration amplitudes of beam
which are obtained from the numerical approach are represented in Figs. 4 and 5.
The major focus of the proposed analysis is to know the significance impacts of
crack angles on the vibration amplitude of the structures. From the representation
of Figs. 4 and 5, it has been brought to the notice that with the enhancement of
crack inclination angle (θ ), there is a reduction in crack length which results in the
enhancement of the beam deflection. But if ‘θ ’ reduces, then crack length enhances
but the corresponding vibration amplitude decreases. So the crack inclination angle
in edge cracked structure has also a momentous impact on the vibration amplitude
of the structure.

Fig. 4 Deflection versus Time for v = 5.68 m/s

96 S. P. Jena et al.

Fig. 5 Deflection versus Time for v = 7.17 m/s

Apart from the influence of the crack inclination angle, the transit mass and speed
have also influenced the response of the structure. There will be an increase in the
vibration amplitude of the beam subjected to an increase in the magnitude of the
traversing mass. Thus it has been also obtained; the vibrated amplitude of the beam
is increasing up to the central part of the span and after that declines towards the
right supported end of the structure. The amplitude of the beam is increasing with the
enhancement of the transit speed. The comparison of vibration amplitudes obtained
from both the numerical approach and FEA is represented in Table 1. From the
representation of results in Table 1, the deviation of results between the two methods
is varying nearly about 3.02% which is a convergent one. So the FEA method is
converged well to the Duhamel integral approach.

Table 1 Comparison of results between numerical and FEA methods for Deflection of beam (cm)
versus Time (sec)
M = 2 kg, v = 5.67 m/s
Time(sec) Numerical FEA Numerical FEA Numerical FEA
(θ = 22˚) (θ = 22˚) (θ = 32˚) (θ = 32˚) (θ = 42˚) (θ = 42˚)
0.141 0.9313 0.9611 0.9591 0.989 0.9905 1.021
0.159 0.915 0.9401 0.955 0.9826 0.9901 1.019
0.1904 0.5827 0.601 0.6264 0.646 0.674 0.695
0.2148 0.1695 0.1744 0.1974 0.2029 0.2249 0.2313
Average percentage of 3.017 2.992 3.06
error
Average percentage error = 3.02%
Response Analysis of Inclined Edge Cracked Beam Under Moving Mass 97

5 Conclusions

In the present problem statement, the response analysis of an inclined edge cracked
structure subjected to traversing mass has been executed. The vibration amplitudes
of the cracked beam caused by the movement of the traversing load or mass are
determined at diverse crack inclination angles, traversing mass and speeds. The
impact of crack inclination angles, transit mass and speeds on the vibration amplitude
of the structure are discussed in this work. It has been found that with the enhancement
of the crack leaning angle and transit load, the amplitudes of the structure are also
amplifying whereas with the enhancement of the speed, the amplitudes are raising
up to a certain extent (middle of the beam), then start decreasing.

References

1. Reis M, Pala Y. Vibration of a cracked cantilever beam under a moving mass load. J Civil Eng
Manag. 2012;18(1):106–13.
2. Pala Y, Reis M. Dynamic response of a cracked beam under a moving mass load. J Eng Mech.
2013;139(9):1229–38.
3. Jena SP, Parhi DR, parametric study on the response of cracked structure subjected to moving
mass. J Vib Eng Technol. 2017;5(1):11–19.
4. Jena SP, Parhi DR. Dynamic and experimental analysis on response of multiple cracked
structures carrying transit mass. J Risk Reliab. 2017;231(1):25–35.
5. Jena SP, Parhi DR. Dynamic response and analysis of cracked beam subjected to transit mass.
Int J Dyn Control. 2018;6(1):961–72.
6. Erasmo V, Alessandro M, Nicholas F. Interaction effect of cracks on flutter and divergence
instabilities of cracked beams under subtangential forces. Eng Fract Mech. 2016;151:109–29.
7. Attara M, Karrechb K, Regenauer-Liebc A. A lattice spring model for dynamic analysis of
damaged beam-type structures under moving loads. Europ J Mech A/Solids. 2016;60:196–207.
8. Rjoub YS Al, Hamad AG. Forced vibration of axially-loaded, multi-cracked Euler-Bernoulli
and Timoshenko beams. Structures. 2020;25:370–385.
9. Hamid S, Mirdamadi VR, Ghayour M. Vibration analysis of cracked Timoshenko beam under
moving load with constant velocity and acceleration by spectral finite element method. Int J
Mech Sci. 2017;122:318–30.
10. Kim T, Lee U. Dynamic analysis of a multi-span beam subjected to a moving force using the
frequency domain spectral element method. Comput Struct. 2017;192:181–95.
11. Rajasekar S, Khaniki HB. Free vibration analysis of bi-directional functionally graded single/
multi-cracked beams. Int J Mech Sci. 2018;144:341–56.
12. Nikkhoo A, Sharifinejad M. The impact of a crack existence on the inertial effects of moving
forces in thin beams. Mech Res Commun. 2020;107: 103562.
13. Guler S. Free vibration analysis of a rotating single edge cracked axially functionally graded
beam for flap-wise and chord-wise modes. Eng Struct. 2021;242: 112564.
14. Kharazana M, Irani S, Noorian M Ali, Salimi MR. Nonlinear vibration analysis of a cantilever
beam with multiple breathing edge cracks. Int J Non-Linear Mech. 2021.https://doi.org/10.
1016/j.ijnonlinmec.2021.103774
15. Panigrahi B, Pohit G. Nonlinear dynamic response of open and breathing cracked functionally
graded beam under single and multi-frequency excitation. Eng Struct. 2021.https://doi.org/10.
1016/j.engstruct.2021.112437
98 S. P. Jena et al.

16. Gupta A, Jain NK, Salhotra A, Joshi PV. Effect of crack location on vibration analysis of
partially cracked isotropic and FGM micro-plate with non-uniform thickness: an analytical
approach. Int J Mech Sci. 2018;145:410–29.
17. Eftekhari SA, Jafari AA. A mixed method for forced vibration of multi-span rectangular plates
carrying moving masses. Arab J Sci Eng. 2014;39:3225–50.
18. Tan G, Zhu Z, Wang W, Cheng Y. Free vibration analysis of a uniform continuous beam with
an arbitrary number of cracks and spring-mass systems. Arab J Sci Eng. 2018;43:4619–34.
19. Chatterjee A, Vaidya T. Experimental investigations: dynamic analysis of a beam under the
moving mass to characterize the crack presence. J Vib Eng Technol. 2019;7:217–26.
20. Akbas SD. Wave propagation in edge cracked functionally graded beams under impact force.
J Vib Control. 2014;22(10):2443–57.
21. Fryba L. Vibration of solids and structures under moving loads. 3rd ed. Czech Republic: Thomas
Telford Ltd.; 1999.
Non-stationary Response of a Bridge Due
to Moving Vehicle with Random Arrival
Rate

Anjaly J. Pillai and Sudip Talukdar

Abstract Moving vehicle excites the bridge with dynamic force which is realised
as a stationary process when the vehicle velocity is constant. However, this condi-
tion is not always true when the vehicle speed varies with time while travelling over
the bridge. In this paper, the bridge response to non-stationary excitation has been
studied considering speed variation, uneven pavement and also random arrival rate
of the vehicle. The bridge vehicle interaction has been modelled using continuum
approach and the solution has been obtained using orthogonal polynomial expan-
sion method. The generalised co-ordinates of the system response are expressed in
terms of orthogonal polynomial series, which offered certain advantages to arrive
at the expression of first and second order statistics of system response using the
properties of the polynomial. The movement of multiple vehicles has been consid-
ered in different time windows assuming their arrival rate follows a Poisson process.
Response statistics- mean and standard deviation has been studied for a single cell
box girder section of single span bridge in different time windows to observe the
effect of vehicle arrival rate, vehicle speed and acceleration and pavement uneven-
ness. The amplification of maximum static flexural stress due to dynamic effect has
been obtained incorporating the standard error of the mean. Sequence of accelerating
vehicles is found to cause higher stress in a bridge with poor maintenance of surface.
The segment of response history in an optimal time window is found to decrease the
computational cost since the presence of total number of vehicles over the bridge
were dependent on the vehicle speed and their arrival rate.

Keywords Non-stationary · Arrival time · Continuum approach · Road surface

roughness · Time windows

A. J. Pillai (B) · S. Talukdar

Department of Civil Engineering, IIT Guwahati, Assam, India
e-mail: [email protected]
S. Talukdar
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 99
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_7
100 A. J. Pillai and S. Talukdar

Nomenclature

DAF Dynamic Amplification Factor

SEM Standard Error of the Mean
DI Dynamic Increment
As Amplitude of cosine wave
cs Suspension damping
cw Tyre damping
C Damping matrix
Cmean Mean values of damping matrix
F Force vector
Fdynamic Maximum dynamic response on the bridge
Fmean Mean values of force vector
Fstatic Maximum static response of the bridge
h(x̃) Bridge deck profile
hmean (x̃) Deterministic mean surface profile
hroad (x̃) Random road roughness of the pavement
ks Suspension stiffness
kw Trye stiffness
K Ztiffness matrix
Kmean Mean values of stiffness matrix
L ( ) Span of the bridge
∼
L ln λ tn Orthogonal function considered
ms Sprung mass
mw Unsprung mass
M Mass matrix
n Shape parameter of Gamma distribution and represents number of
vehicle arrivals
nd Number of degrees of freedom
N Number of terms used to construct the road surface roughness
Ns Number of samples
∼
N1 Number of basic functions with respect to λ tn
ptn (t) Probability density function of the arrival time
Qil (t) Time variation of displacement
x̃ Spatial distance
tn Vehicle arrival time on the bridge
v Velocity of vehicle
y(x̃, t) Displacement of the bridge at time instant, t at location, x̃
z1 Displacement of sprung mass
z2 Displacement of unsprung mass
δlk Kronecker delta function
G Gamma function
∼
λ Mean arrival rate
Non-stationary Response of a Bridge Due to Moving Vehicle … 101

∼
μ(λ tn ) Mean arrival time
μf (x̃, t) Mean of bridge response
σf (x̃, t) Standard deviation of bridge response
θs Independent random phase angle uniformly distributed from 0 to 2π
ΩL Lower cut off frequencies of spatial unevenness
Ωs Spatial frequency (c/m)
ΩU Upper cut off frequencies of spatial unevenness

1 Introduction

The dynamic response of bridge considering vehicular movement has been studied
by various researchers. Most of them have idealised bridge as a beam which can be
simply supported or continuous [1]. In all of these models, the vehicle loads have
been idealised as several concentrated loads [1]. However, in these works the effect
of interaction between bridge and vehicle is not considered. In order to take the effect
of bridge vehicle interaction, the vehicle has been modelled as spring mass dashpot
system [2, 3] and the same has been considered in obtaining the bridge dynamic
response [2, 4, 5]. In the above works, the effect of road surface roughness has not
been considered, which also plays an important role in the dynamic response of the
bridge. The effect of road surface roughness and bridge vehicle interaction has been
considered to evaluate the dynamic response of the bridge and it has been observed
from the studies that the dynamic response of the bridge is affected by the road
surface roughness [6, 7].
In the above studies, the bridge response was obtained by taking deterministic
vehicular loads [6–9]. However, the vehicular loads arriving on the bridge in terms
of number of vehicles, axle weight, axle interval and vehicle velocity are random
in nature. Hence, some of the researchers have studied the bridge dynamic response
due to the random nature of the vehicles passing on the bridge assuming vehicle
arrival to be a random variable [10, 11]. It was also observed from the studies
that the response of vehicle induced by pavement irregularity becomes a stationary
random process in time domain when the vehicle velocity is assumed to be constant.
The response of vehicle induced by pavement irregularity becomes a non-stationary
random process in time domain when the vehicle velocity is varying with time [12–
16]. The non-stationary response of a vehicle travelling on homogeneous road surface
has been analysed using state space approaches [12, 13]. The non-stationary response
of vehicle was also obtained in which the equations of motion were first established
in space domain and then covariance of the response was computed in time domain
[14]. A Monte Carlo simulation technique was used to simulate the deck profile for
generating input samples in numerical integration of the system equations. This was
used to obtain the non-stationary response of the vehicle [16].
Thus it is evident from the studies conducted that very less work has been done
to evaluate the non-stationary response of the bridge when the vehicle travels at
102 A. J. Pillai and S. Talukdar

variable velocity. In addition, it has been observed that most of the vehicular loads
on the bridge are taken as deterministic in nature and have considered the effect of
single vehicular movement. Hence, this paper presents a methodology to evaluate the
bridge response considering random vehicular loads, multiple vehicular movement
and variable velocity. Further, DAF has been evaluated for different bridge and vehicle
parameters.

2 Methodology

In the present work, the vehicle arrival time is assumed as a random variable
following Poisson process. The bridge response considering random vehicle arrival
time and bridge vehicle interaction is obtained using Orthogonal Polynomial Expan-
sion Method. The bridge considered in the study is a simply supported bridge with
single span. It is idealised as a Euler–Bernoulli beam with uniform cross section. The
vehicle is modelled as a quarter car model. The road surface roughness is assumed to
be Gaussian process with zero mean and is represented by a power spectral density
function. In addition to the road surface roughness, the mean surface of the bridge
deck has been considered as a half sine wave. This represents the pre-chamber of the
bridge.

2.1 Theoretical Formulation

The multiple vehicle movement modelled as quarter car model on the single span
bridge has been shown in Fig. 1.
The deck profile is represented by,

h(x̃) = h mean (x̃) + h r oad (x̃) (1)

Fig. 1 Vehicle movement on bridge

Non-stationary Response of a Bridge Due to Moving Vehicle … 103

The position of vehicle from the reference station along the bridge span at any
time instant is given by

∑
m
x̃(t) = apt p (2)
p=0

In Eq. (2), the coefficients ap represent different conditions of the vehicle motions.
The coefficients a0 and a1 have non-zero values while the remaining coefficients are
zero when the vehicle velocity is constant. The velocity of the vehicle at any time
instant is obtained by taking the first derivative of Eq. (2) with respect to x.
A series of cosine terms with random phase angles and a certain probability density
function have been used to calculate the road surface roughness [17], which is given
by

∑
N
h r (x̃) = As cos(2π Ωs x̃ + θs ) (3)
s=1

The parameters As and Ω s are taken from Yin et al. [17].

The governing differential equation is written after expanding C, K and F using
Taylor series [18] as
[ { ( )}] [ { ( )}]
M x̃¨ + Cmean + Cλ̃tn λ̃tn − μ λ̃tn x̃˙ + Kmean + Kλ̃tn λ̃tn − μ λ̃tn x̃ =
{ ( )}
Fmean (t) + Fλ̃tn λ̃tn − μ λ̃tn (4)

where, C, K and F with subscript λt n represent the differentiation of the variable

with respect to λ̃tn , which is computed at t = λ̃tn . M, C and K are the mass matrix
consisting of bridge and vehicle mass, damping matrix consisting of bridge and
vehicle damping and stiffness matrix consisting of bridge and vehicle stiffness. The
response X̃ is expressed as the summation of product of transformed time dependent
coordinate and orthogonal polynomial function as,

∑
N1 ( )
X̃ j (t) = Q jl (t)L ln λ̃tn ; j = 1, 2, ..., n do f (5)
l=0

Since, the vehicle arrival time follows Poisson process, the distribution of arrival
time is Gamma distribution. The probability density function of λ̃tn is given as [19]
( ) t n−1 exp(−t)
pλ̃tn λ̃t = (6)
[(n)

Since the distribution of arrival time is Gamma distribution, Associated Laguerre

Polynomial is the orthogonal polynomial function considered [20] in the present
104 A. J. Pillai and S. Talukdar

study which is found to satisfy the orthogonality condition. The response statistics
have been found using the recurrence relationships [20] shown in Eq. (7).
( ) ( ) ( )
λ̃tn L ln λ̃tn = αl−1 L ln λ̃tn + βl L ln λ̃tn +
( )
γl+1 L ln λ̃tn
αl−1 = −(l − 1 + n)
γl+1 = −(l + 1 + 1)
βl = (2l) + n + 1 (7)

Substitute Eq. (5) in Eq. (4),

N1
∑ ( ) ( { ( )}) ∑
N1 ( )
M Q̈ l (t)L ln λ̃tn + Cmean + Cλ̃t λ̃tn − μ λ̃tn Q̇l (t)L ln λ̃tn
n
l=0 l=0
( { ( )}) ∑
N1 ( ) { ( )}
+ Kmean + Kλ̃t λ̃tn − μ λ̃tn Ql (t)L ln λ̃tn = F(t)mean + F(t)λ̃t λ̃tn − μ λ̃tn (8)
n n
l=0

In order
( )to find the response statistics of the bridge, Eq. (8) has to be multiplied
by L k λ̃tn . Further, the recurrence relation shown in Eq. (7) has to be used and the
n
( )
resulting equation has to be multiplied by probability density function pλ̃tn λ̃t and
integrated in the domain of the random variable using the orthogonality property of
the polynomial considered. The resulting equation is given as,

N1
∑
M Q̈ l (t)dl2 δlk
l=0
N1
∑ [ ( ) ( ) ]
+ Q̇l (t) Cmean dl2 δlk + Cλ̃t αl−1 dl−1
2 δ 2 2 2
l−1k + βl dl δlk + γl+1 dl+1 δl+1k − Cλ̃tn μ λ̃tn dl δlk
n
l=0
N1
∑ [ ( ) ( ) ]
+ Ql (t) Kmean dl2 δlk + Kλ̃t αl−1 dl−1
2 δ 2 2 2
l−1k + βl dl δlk + γl+1 dl+1 δl+1k − Kλ̃t μ λ̃tn dl δlk
n n
l=0
( ) ( )
= F(t)mean d02 δ0k + F(t)λ̃t αk−1 d02 δ0k−1 + β0 d02 δ0k + γ0+1 d0+1
2 δ 2
0+1k − F(t)λ̃tn μ λ̃tn d0 δ0k (9)
n

where,

[(l + n + 1)
dl2 = (10)
[(l + 1)[(n + 1)

In Eq. (10), the value of k changes from 0 to N 1 . Newmark’s Method [21] has
been used to solve Eq. (10) to obtain the time variation of displacement Q.
Non-stationary Response of a Bridge Due to Moving Vehicle … 105

2.1.1 Expectation of the Response Vector

Using the property of orthogonal polynomial function, the expectation of the response
has been evaluated and is shown below,
[ ]
E X̃ i (t) = Q i0 (t)d02 ; i = 1, 2, ..., n do f (11)

2.1.2 Standard Deviation of the Elements of the Response Vector

The covariance of the elements of the response vector is written as,

( ) [ ] [ ] [ ]
Cov X̃ i (t1 ), X̃ i (t2 ) = E X̃ i (t1 ) X̃ i (t2 ) − E X̃ i (t1 ) E X̃ i (t2 ) ; i = 1, 2, ..., n do f
(12)

In Eq. (12),

[ ] ∑N1 ∑
N1 [ ( ) ( )]
j j
E X̃ i (t1 ) X̃ i (t2 ) = Q il (t1 )Q im (t2 )E L l λ̃t j L m λ̃t j ; i = 1, 2, ..., n do f
l=0 m=0
(13)

Equation (13) is simplified using the property of orthogonal polynomials and is

given as

[ ] ∑N1 ∑
N1
E X̃ i (t1 ) X̃ i (t2 ) = Q il (t1 )Q im (t2 )dl2 δlm ; i = 1, 2, ..., n do f (14)
l=0 m=0

The covariance of the elements of the response vector is written by substituting

Eqs. (18) and (15) in Eq. (16) which is shown below,

∑
N1 ∑
N1
Cov(X i (t1 ), X i (t2 )) = Q il (t1 )Q im (t2 )dl2 δlm −Q i0 (t1 )Q i0 (t2 )d04 ; i = 1, 2, ..., n d
l=0 m=0
(15)

Due to the presence of Kronecker delta, Eq. (15) is simplified further and is given
as,

( ) ∑N1
Cov X̃ i (t1 ), X̃ i (t2 ) = Q il (t1 )Q il (t2 )dl2 − Q i0 (t1 )Q i0 (t2 )d04 ; i = 1, 2, ..., n do f
l=o
(16)
106 A. J. Pillai and S. Talukdar

The variance of the elements of the response vector is evaluated by substituting

t1 = t2 = t in Eq. (16), and is given as

[ ] ∑N1
V ar X̃ i (t) = [Q il (t)]2 dl2 − [Q i0 ]2 d04 ; i = 1, 2, ..., n do f (17)
l=0

DAF is evaluated using the obtained response statistics of the bridge, which is the
mean and standard deviation.

3 Dynamic Amplification Factor (DAF)

In bridge design, dynamic analysis is not considered. This is because several codes
have suggested DAF to amplify the static effect. It is also observed that the DAF
suggested in various codes only depends on the span length. However, parameters
such as vehicle velocity, road surface roughness and vehicle acceleration have not
been considered in the evaluation of DAF [22]. From the previous study, it is observed
that DAF depends on the above parameters [23]. In the present work, the effect of
DAF on the mentioned parameters has been studied.
The DAF is evaluated considering the effect of mean and standard deviation of
flexural stresses, which is given as,

Fstatic + Fdynamic Fdynamic

D AF = =1+ = 1 + DI (18)
Fstatic Fstatic

The static response is obtained by traversing the vehicles at 5 km/hr. The maximum
dynamic response is given as,
| |
Fdynamic = |μ f (x̃, t) + E f (x̃, t)| (19)

where, E f (x, t) is the standard error of the mean (SEM) defined as [23]

σ f (x, t)
E f (x, t) = √ / (20)
Ns 4

The maximum dynamic response in Eq. (19) takes into account the effect of mean
and standard deviation of flexural stresses.
Non-stationary Response of a Bridge Due to Moving Vehicle … 107

4 Numerical Study

The bridge considered for study is a single span box girder of span length 30 m with
twin cell cross section. The ratio of vehicle suspension and tyre stiffness ratio is 4.0;
fundamental natural frequency of the bridge is 4.5 Hz; vehicle weight is 40 tonnes;
for the road surface roughness, Ω L is taken as 0.1 cycle/m and Ω U is taken as 2
cycle/m [24]. The mean profile is assumed to be sinusoidal with an amplitude of
0.01 m. The response is evaluated when the vehicle is at the mid-span of the bridge.
The road roughness coefficient considered for good and very poor road case is 32 ×
10–6 m2 /cycle/m and 1024 × 10–6 m2 /cycle/m respectively [25].
To account for multiple vehicular movement on the bridge, the flexural stresses
are obtained for different time windows. The vehicle velocity considered is 20 km/hr,
40 km/hr and 60 km/hr. The window kept for vehicle movement is changed from 10
to 25 s. The comparison of mean flexural stresses for the time windows is shown in
Table 1 for good and very poor road conditions. The comparison of standard deviation
of flexural stresses for the above-mentioned time windows is shown in Table 2 for
good and very poor road conditions. The numerical analysis has been done using
MATLAB.

Table 1 Comparison of mean flexural stresses for different time windows

Vehicle velocity Road roughness coefficient (m2 /cycle/m Mean flexural stress (MPa)
(km/hr) × 10–6 ) 10 15 20 25
secs secs secs secs
20 32 11.8 14.4 15.6 15.6
1024 16 22.5 28 28
40 32 8.2 9 10 10
1024 12 15 18 18
60 32 5.8 5.8 5.8 5.8
1024 9 9 9 9

Table 2 Comparison of standard deviation of flexural stresses for different time windows
Vehicle velocity Road roughness coefficient (m2 /cycle/m Standard deviation of flexural
(km/hr) × 10–6 ) stress (MPa)
10 15 20 25
secs secs secs secs
20 32 0.2 0.2 0.2 0.2
1024 0.8 0.8 0.8 0.8
40 32 1.18 1.18 1.18 1.18
1024 3.2 3.2 3.2 3.2
60 32 1.64 1.64 1.64 1.64
1024 4.2 4.2 4.2 4.2
108 A. J. Pillai and S. Talukdar

From Tables 1 and 2, it is observed that the mean and standard deviation of the
flexural stresses do not vary significantly as time window changes from 20 to 25
secs. The time window for evaluating DAF may be reasonably taken not above 20
secs. The DAF varying the arrival rate of the vehicle, road surface roughness, vehicle
velocity and acceleration of the vehicle is obtained.

4.1 Parametric Variations

The factors that are varied to observe the effect on Dynamic Amplification Factors
are:
1. Variable vehicle velocity
2. Arrival rate of the vehicle
3. Road surface roughness.

4.1.1 Variable Vehicle Velocity

The variable vehicle velocity is obtained using Eq. (2). The vehicle is made to accel-
erate at 0.5 m/s2 , 1 m/s2 and 1.5 m/s2 . The mean flexural stresses obtained after
varying the acceleration are then compared with the mean flexural stress for a constant
vehicle velocity. The comparison plot for 20 km/hr vehicle velocity and very poor
road condition is shown in Fig. 2. The arrival rate of the vehicle is considered as 2
vehicles per second.
The maximum value of mean stress when there is no acceleration is 19.3 MPa,
0.5 m/s2 is 20 MPa, 1 m/s2 is 23 MPa and 1.5 m/s2 is 28 MPa. The increase in
acceleration values increases the mean flexural stresses. This is because it increases
the longitudinal vibration of the bridge.

Fig. 2 Comparison of mean

stress for different values of
acceleration for velocity
20 km/hr and very poor road
surface

1.5 m/s2 0.5 m/s2 * 1 m/s2 0

Non-stationary Response of a Bridge Due to Moving Vehicle … 109

The mean flexural stresses and the standard deviation of flexural stresses for good
and very poor for acceleration of the vehicle 1.5 m/s2 and varying the vehicle velocity
are shown in Figs. 3, 4, 5 and 6.
It is observed from Figs. 3 and 4 that the mean flexural stresses are higher for
lower velocity since the duration of loading is more for lower velocity as compared
to higher velocity. The standard deviation of the flexural stresses does not follow the
same trend as shown in Figs. 5 and 6. The DAF for varying the acceleration of the
vehicle for different initial vehicle velocities and arrival rate of 2 vehicles per second
for very poor road case is shown in Fig. 7.
It is observed from Fig. 7 that the mean flexural stresses are higher for lower
velocity since the duration of loading is more for lower velocity as compared to

Fig. 3 Comparison of mean

stress for varying velocity
with acceleration 1.5 m/s2
and good road surface

20 km/hr 40 km/hr 60 km/hr

Fig. 4 Comparison of mean

stress for varying velocity
with acceleration 1.5 m/s2
and very poor road surface

20 km/hr 40 km/hr 60 km/hr

110 A. J. Pillai and S. Talukdar

Fig. 5 Comparison of
standard deviation of flexural
stress for varying velocity
with acceleration 1.5 m/s2
and very good road surface

20 km/hr 40 km/hr 60 km/hr

Fig. 6 Comparison of
standard deviation of flexural
stress for varying velocity
with acceleration 1.5 m/s2
and very poor road surface

20 km/hr 40 km/hr 60 km/h r

higher velocity leading to higher DAF for lower vehicle velocity. Also, there is a
significant change in the DAF when the vehicle velocity is considered constant and
when the vehicle velocity is variable as seen from Fig. 7.

4.1.2 Arrival Rate of the Vehicles

The arrival rate of the vehicles is varied from 1 vehicle per second to 3 vehicles per
second for initial vehicle velocity of 20 km/hr and vehicle acceleration of 1.5 m/s2 .
The road condition considered is very poor. The mean and standard deviation of the
flexural stresses for different arrival rates are shown in Figs. 8 and 9.
Non-stationary Response of a Bridge Due to Moving Vehicle … 111

2.5

2
DAF

1.5

0.5

0
0 0.5 1 1.5
Acceleration of the vehicle (m/s2)
20 km/hr 40 km/hr 60 km/hr

Fig. 7 Comparison of DAF with acceleration of the vehicle for different initial vehicle velocities
and arrival rate of 2 vehicles per second for very poor road conditions

Fig. 8 Comparison of mean

stress for varying arrival rate

3 vehicles 2 vehicles 1 vehicle

per second per second per second

It is observed from Fig. 8 that higher arrival rates signify more vehicular movement
on the bridge leading to higher dynamic stresses. The standard deviation of the
flexural stresses follows the same pattern as the mean flexural stresses for varying
arrival rates as shown in Fig. 9. The DAF for varying arrival rates for initial vehicular
velocities 20 km/hr, 40 km/hr and 60 km/hr is shown in Fig. 10.
It can be observed from Fig. 10 that the DAF increases upto 2 vehicles per second
and decreases till arrival rate 3 vehicles per second. The increase in DAF can be due
to an increase in the dynamic forces. However, as the arrival rate increases from 2
vehicles per second, the static forces also increase leading to a decrease in the DAF.
112 A. J. Pillai and S. Talukdar

3 vehicles 2 vehicles 1 vehicle

per second per second per second

Fig. 9 Comparison of standard deviation of flexural stress for varying arrival rate

2.5

2
DAF

1.5

0.5

0
1 1.5 2 2.5 3
Arrival rate (vehicles per second)
20 km/hr 40 km/hr 60 km/hr

Fig. 10 Comparison of DAF with arrival rate of the vehicle for very poor road conditions

4.1.3 Road Surface Roughness

The deterioration in road surface condition leads to higher flexural stresses as the
dynamic forces due to the vibratory motion of the vehicle on the bridge increases as
shown in Figs. 3, 4, 5 and 6. The DAF for different road surface roughness for an
initial vehicle velocity of 20 km/hr and arrival rate of 2 vehicles per second is shown
in Fig. 11.
Non-stationary Response of a Bridge Due to Moving Vehicle … 113

2.7
2.65
2.6
2.55
DAF

2.5
2.45
2.4
2.35
0 0.5 1 1.5
Acceleration of the vehicle (m/s2)
Good Road Very Poor Road

Fig. 11 Comparison of DAF with acceleration of the vehicle for initial vehicle velocity 20 km/hr
and arrival rate of 2 vehicles per second for good and very poor road conditions

It is observed from Fig. 11 that the DAF increases with deteriorating road surface
conditions. This is because bridge experiences higher dynamic forces due to very
poor road surface condition.

5 Conclusions

The present study outlines an approach for the evaluation of response statistics of
a single span bridge for random arrival time of the vehicles based on orthogonal
polynomial expansion method. The DAF is evaluated using the response statistics
obtained from the method. Parameters such as vehicle velocity, arrival rate of the
vehicles, acceleration of the vehicles and road surface roughness are considered to
study the effect on DAF. Based on the results obtained from the study described
above, the main conclusions are as follows:
1 The optimum time window to evaluate the DAF is considered as 20 secs.
2 The mean flexural stresses are higher for lower vehicle velocity.
3 The DAF increases with the road surface irregularity and decreases with increase
in vehicle velocity for a single span bridge.
4 The DAF increases for arrival rate up to 2 vehicles per second and then decreases
further for a single span bridge.
5 The DAF increases for higher acceleration of the vehicles. Vehicle acceleration
may contribute significantly to bridge response and the resulting DAF may exceed
those adopted in current design codes.
114 A. J. Pillai and S. Talukdar

References

1. Fryba L. Dynamics of railway bridges. Academia Praha; 1996.

2. Blejwas TE, Feng CC, Ayre RS. Dynamic interaction of moving vehicles and structures. J
Sound Vib. 1979;67(4):513–521. https://doi.org/10.1016/0022-460X(79)90442-5
3. Green MF, Cebon D. Dynamic interaction between heavy vehicles and highway bridges.
Comput Struct. 1997;62(2):253–264. https://doi.org/10.1016/S0045-7949(96)00198-8
4. Yang YB, Lin CW. Vehicle-bridge interaction dynamics and potential applications. World
Scientific Publishing Co. Pte. Ltd; 2004. https://doi.org/10.1016/j.jsv.2004.06.032
5. Zeng Q, Stoura CD, Dimitrakopoulos EG. A localized lagrange multipliers approach for the
problem of vehicle-bridge-interaction. Eng Struct. 2018;168:82–92. https://doi.org/10.1016/j.
engstruct.2018.04.040.
6. Coussy O, Said M, van Hoove J-P. The influence of random surface irregularities on the dynamic
response of bridges under suspended moving loads. J Sound Vib. 1989;130(2):313–320. https://
doi.org/10.1016/0022-460X(89)90556-7
7. Frýba L. Vibration of solids and structures under moving loads. London: Thomas Telford Ltd.;
1972.
8. Pesterev AV, Bergman LA, Tan CA, Tsao TC, Yang B. On asymptotics of the solution of
the moving oscillator problem. J Sound Vib. 2003;260(3):519–536. https://doi.org/10.1016/
S0022-460X(02)00953-7
9. Frýba L. Non-stationary response of a beam to a moving random force. J Sound Vib.
1976;46(3):323–338. https://doi.org/10.1016/0022-460X(76)90857-9
10. Sniady P. Vibration of of a beam due to a random stream of moving forces with random velocity.
J Sound Vib. 1984;97(1):23–33.
11. Turner JD, Pretlove AJ. A study of the spectrum of traffic-induced bridge vibration. J Sound
Vib. 1988;122(1):31–42. https://doi.org/10.1016/S0022-460X(88)80004-X
12. Virchis VJ, Robson JD, Response of an accelerating vehicle to random road undulations.
Journal of Sound and Vibration; 1971. 18(3): 423–71.
13. Hammond JK, Harrison RF. Non-stationary response of vehicles on rough ground. J Dyn Syst
Meas Control Trans ASME. 1981;103(3):245–50.
14. Nigam NC, Yadav D. Dynamic response of accelerating vehicles to ground roughness. In:
Proceedings Noise shock and veibration conference. Monas University; 1974. pp. 280–5.
15. Hwang JH, Kim JS. On the approximate solution aircraft landing gear under non-stationary
random excitations. KSME Int J. 2000;14(9):968–77.
16. Sasidhar MN, Talukdar S. Non-stationary response of bridge due to eccentrically moving
vehicles at non-uniform velocity. Adv Struct Eng. 2003;6(4):309–24.
17. Yin X, Fang Z, Cai CS, Deng L. Non-stationary random vibration of bridges under vehicles
with variable speed. Eng Struct. 2010;32(8):2166–2174. https://doi.org/10.1016/j.engstruct.
2010.03.019
18. Li J, Chen J. Stochastic dynamics of structures. Wiley (Asia) Pte Ltd; 2009.
19. Nigam NC. Introduction to random vibrations. Masachusetts, London, England: The MIT Press
Cambridge; 1983.
20. Schoutens W. Lecture notes in statistics. In: Bickel P, Diggle P, Fienberg S, Krickeberg K, Olkin
I, Wermuth N, Zeger S (Eds.), Stochastic processes and Orthogonal polynomials. Springer;
2000. https://doi.org/10.1007/978-1-4612-1470-0
21. Chopra AK. Dynamic of structures theory and applications to earthquake engineering. (W. J.
Hall, Ed.) (Fourth). Prentice Hall; 2011.
22. IRC 6. Standard specifications and code of practice for road bridges, section-II Loads and
stresses. Indian Road Congress, New Delhi, 2017.
23. Humar JL, Kashif AM. Dynamic response of bridges under travelling loads. Canadian J Civil
Eng. 1993;20(2):287–298. https://doi.org/10.1139/l93-033
24. DeCoursey W. Statistics and probability for engineering applications. Elsevier; 2003.
25. Sun L, Kennedy TW. Spectral analysis and parametric study of stochastic pavement loads. J Eng
Mech. 2002;128(3):318–327. https://doi.org/10.1061/(ASCE)0733-9399(2002)128:3(318)
Multiclass Fault Diagnosis and Novelty
Detection of Induction Motor Using Deep
Learning Algorithm Based on Frequency
Domain Signal

Aman Gupta, Rajiv Tiwari, and Dhruba Jyoti Bordoloi

Abstract This paper presents the development of a deep learning model for effec-
tive fault diagnosis of the Induction Motor (IM). For prompt detection of these
faults, vibration and current frequency domain signals are investigated. The proposed
methodology predicts the sensor fault condition and based on that IM faults were
predicted so that corrupt sensor data cannot manipulate the output. The vibration
and current signals were acquired after multiple experiments of varying operating
speed and loading conditions. The ten distinct health conditions that encountered in
IM (one no defect condition, four mechanical fault and five electrical fault condi-
tions) have been considered. Fault predictions have been performed by extracting
different frequency domain features of the vibration and current signal. The training
of deep learning neural network model was performed at various operating condi-
tions and tested for same conditions. The result is obtained in terms of accuracy of
fault prediction and found to be better than the previous literatures.

Keywords Induction motor (IM) · Vibration and current signal · Frequency

domain data · Sensors faults · Multi-fault classification · Deep learning

1 Introduction

Induction motors are the most frequently used rotatory machinery in the industries.
Therefore, IMs are exposed to various stresses which cause faults in IMs. If these
faults are not discovered in the early stages it will develop further and cause the
failure of those components and also nearby components of IM. This will lead to the

A. Gupta (B) · R. Tiwari · D. J. Bordoloi

Department of Mechanical Engineering, IIT Guwahati, Assam, India
e-mail: [email protected]
R. Tiwari
e-mail: [email protected]
D. J. Bordoloi
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 115
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_8
116 A. Gupta et al.

catastrophic failure of the IM and the shutdown of the system. Hence to evade such
circumstances condition based maintenance is performed. It is a health monitoring
technique, some system parameters are continuously monitored and changes in the
system performance can be detected by the change in these parameters. Hence fault
can be detected in its initial stages and total shutdown of the system can be prevented.
The faults found in the induction motor are categorised as two types [1]:
(i) Mechanical faults-These types of faults include
(a) Bearing Faults (BF)
(b) Unbalance Rotor (UR)
(c) Bowed Rotor (BR)
(d) Rotor Misalignment (RM)
(ii) Electrical faults—These types of faults include
(a) Broken Rotor Bar (BRB)
(b) Phase Unbalance and Single Phasing Fault with high resistance (MPUSPF)
(c) Phase Unbalance and Single Phasing Fault with low resistance (PUSPF)
(d) Stator winding faults with high resistance (MSWF)
(e) Stator winding faults with low resistance (SWF)
All these faults were predicted by analyzing the signals acquired from the condi-
tion monitoring. The signals can be analyzed in three different forms—time domain,
frequency domain and time–frequency domain. Since physical characteristics of
the signal are frequency dependent, therefore frequency signal domain analysis is
performed in the present work.
Artificial intelligence methods like fuzzy logic, support vector machine (SVM),
neural networks (NN) and Deep learning (DL) have been recognized for condition
health monitoring in recent years [2–8]. Researchers used these methods for the fault
diagnosis of several mechanical systems like pumps, bearing, gears and induction
motors.
Karmakar et al. [1] studied the faults that occur in the induction motor. Tran et al.
[2] performed frequency domain analysis for IM fault diagnosis and presented a
methodology combination of fuzzy logic and ANN. Glowacz et al. [3] considered
acoustic signals for fault detection on DC motors. Fast Fourier Transform (FFT)
was used as a preprocessing method on sound signals for converting into frequency
domain data and rotor faults were classified by using the SVM and Nearest Neighbor
classifiers. Schoen et al. [4] preprocessed raw signal data using FFT and developed
an unsupervised model to predict the BRB fault in IM. Gangsar et al. [5] devel-
oped the SVM model based on frequency domain data for predicting IM faults.
Rodriguez et al. [6] predicted the BRB and air-gap eccentricity of the IM using fuzzy
logic method from FFT of the signal data. Rapur et al. [7] extracted 18 statistical
features from the frequency domain data of vibration and current signal data of a
centrifugal pump and performed fault diagnosis using SVM. Shao et al. [8] used
vibration signal data for extracting 7 time-domain features and developed a deep
neural network for training and fault diagnosis of IM. Wong et al. [9] investigated
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 117

vibration signal data for novelty detection using self-organized mapping. Sadooghi
and Khadem [10] performed statistical feature and nonlinear feature extraction for
novelty detection using One-class SVM of rotary system.
From the literature review, it is found that fault diagnosis is extremely important.
Deep Learning is a new field having a lot of potential and very few attempts have been
made. In this paper, the multiclass classification is presented for the fault prediction
of IM. A deep learning model is developed and vibration and current signatures were
monitored for fault detection. Frequency domain feature extraction is performed to
obtain the information from frequency domain data. The novelty detection of the
sensor fault is also performed. It is necessary to perform novelty detection and detect
abnormality conditions in data. If a sensor is having a fault like sensor damaged
or disconnected and giving wrong default values and it is not discovered, then the
corrupt signal will be sent as input to the model and due to this incorrect input data,
the result obtained will also be incorrect. And the model will be misguided and
the actual fault cannot be identified. Therefore, in the present work, sensor faults
diagnosis is also performed and faulty sensors data is excluded from further analysis
and sensor fault condition is alarmed.

2 Experimental Setup Details

The experimental test-rig is shown in Fig. 1. It consists of the following sub

components- a machine fault simulator (MFS), a data acquisition system, a tri-axial
accelerometer, three AC current probes, a tachometer, a constant DC power source,
and a signal monitor. The test IM has specifications as Power −0.37 kW, Nominal
frequency −50 Hz, and Speed −3450 rpm. The shaft end was coupled with IM using a
flexible coupling and another end was connected to the gearbox by pulley-belt mech-
anism. External load is applied on the magnetic brake attached to the gearbox. The
motor speed was controlled using a variable frequency drive (VFD). The tachometer
was employed at the end of the motor shaft to measure the shaft rotational speed.
The vibration isolators were attached to the baseplate of the setup.
Vibration signals were obtained from accelerometer sensors. Current signals were
obtained from current probes and a tachometer was used to obtain frequency signal.
Different motors having faults present inside them were available with us. And
the faulty sensor data was collected by detaching the sensors from their respective
position so that only environmental noise is received.
Data from all the seven sensors i.e. three accelerometers, three current probes,
and one tachometer were collected. National Instruments data-acquisition system
and LabVIEW software were used for the collection of data. Table 1 shows the
specification of data collected in time domain form.
118 A. Gupta et al.

Fig. 1 The experimental test-setup used in the laboratory

Table 1 Dataset specifications

Data collection Values
Sample size 2000
Time interval between two data points 5 × 10–5 s
Time interval for one data set 0.1 s
Total time duration of data collection 30 s
Fault condition 1 No fault + 9 faults
Speed 10 Hz, 15 Hz, 20 Hz, 25 Hz, 30 Hz, 35 Hz, 40 Hz
Loading condition (i) No load −0% of rated torque (0 Nm)
(ii) Light load −0.11% of rated torque (0.113 Nm)
(iii) High load −0.55% of rated torque (0.565 Nm)

3 Fault Classification Methodology

In this section, the transformation of time domain data into frequency domain data is
performed and various statistical features of the frequency domain data are obstained.
Further, the structure of the Deep Learning model for the fault classification is
discussed.

3.1 Frequency Domain Data

The frequency-domain graph explains the number of signals that exist within each
given frequency band over a range of frequencies. Discrete time signals are trans-
formed into sines and cosines series. This representation is called the discrete
Fourier transform (DFT). The Fast Fourier transform (FFT) algorithm was used
for converting the time domain data into frequency domain data. The FFT algorithm
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 119

calculates the DFT of a sequence. FFT is a fast version of the DFT, it uses the fact
that the Fourier transform does various identical multiplications repeatedly. The FFT
takes advantage of periodicities of sines to perform the calculation, eliminates redun-
dant computations and saves a significant amount of calculation. For any signal tn, the
discrete Fourier transform is calculated by Eq. (1) [5].

1 ∑ N −1 − j2 pnk/N
Fn = tk e (n = 0, 1, 2, ........., N − 1) (1)
N k=0

3.2 Statistical Feature Extraction

The input features are the most crucial things in Artificial Neural Network (ANN) to
predict the output. Since raw data contains redundant information and noise hence
it cannot be directly adopted as an input feature. Therefore, to obtain significant
information and get rid of very large data 25 statistical features were extracted from
the frequency domain data of vibration and current signals using MATLAB software.
Overall, 300 feature data sets were extracted out from the frequency domain dataset
at each and every condition. The various statistical features used are given in Table 2.

3.3 Deep Learning Model

In the present work, Python software is employed and open-source Keras and Scikit
learn deep learning library is used. Two Deep Neural Network (DNN) models were
developed. First DNN predicts the fault present in sensors. If a fault in any sensor is
predicted then that faulty sensor data is wiped out and correct sensor data is passed as
an input to the second for prediction of the fault in the IM. The statistical features were
extracted from frequency domain data to obtain significant information. And these
extracted features were used as input features for our deep learning model. The four-
layer neural network is adopted to evade the overfitting and underfitting conditions
and get the best fit for data. The categorical cross-entropy is used as the cost function
and Adam optimizer is used with back propagation algorithm to update the weight
and bias values. Since a large dataset was available for different fault conditions
at different speeds. So data was divided as 75:25 into a training and testing dataset
respectively. And to track the fitting of our deep learning model this training data was
further divided as 85:15 into training and validation datasets respectively. Network
parameters used to get the best performance of the classification model are shown in
Tables 3 and 4 [11].
Figure 2 exhibits the flow chart for the prediction of the fault of IM using deep
neural networks.
120 A. Gupta et al.

Table 2 Frequency domain statistical features

∑
n
1. T1 = xi /n [mean]
i=1
∑
n
2. T2 = |xi |/n [absolute mean]
i=1
( n √
)2
∑
3. T3 = |xi |/n
i=1
√
∑
n
4. T4 = xi2 /n [RMS]
i=1

∑
n
5. T5 = xi4 /n
i=1
∑
n
6. T6 = xi3 /n
i=1
√
∑
n
7. T7 = (xi − T 1)2 /n [Standard Deviation]
i=1

8. T8 = min(xi ) [Maximun]
9. T9 = max(xi ) [Minimum]
10. T10 = T9/T8 [Minimum to Maximum]
11. T11 = T9/T2 [Minimum to absolute Mean]
12. T12 = T9/T3 [Minimum to Mean]
13. T13 = T9/T4 [Minimum to RMS]
( √ )4
14. T14 = T 5/ T 7

15. T15 = T 5/(T 3)2

∑
n
16. T16 = (|xi |)2 /n [Energy]
i=1

1 ∑
n
n (xi −T 1)4
17. T17 = i=1
T 74
[Kurtosis]

1 ∑
n
n (xi −T 1)3
18. T18 = i=1
T 73
[Skewness]
∑
n
19. T19 = (xi − T 1)2 /n [Variance]
i=1

20. T20 = T9-T8 [Range]

21. T21 = max(T9,T8)/T4 [peak to RMS]
(continued)
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 121

Table 2 (continued)
∑
n
22. T22 = − Pi × log2 Pi [Entropy]
i=1
P is probability calculated by plotting histogram
∑
n
(Ai −T1( A))(Bi −T1(B))
23. T23 = √( i=1
)( ) [Correlation]
∑
n ∑
n
(Ai −T1(A))2 (Bi −T1(B))2
i=1 i=1

∑
n √
24. T24 = (xi − T 1)3 /n (T 7)3
i=1
∑
n
25. T25 = (xi − T 1)4 /n(T 7)2
i=1

Table 3 Network parameters

Layer Type Node Activation function
used in 1st DNN
1 Input layer – –
2 Hidden layer 100 ReLU
3 Hidden layer 100 ReLU
4 Output layer 1 Sigmoid

Table 4 Network parameters

Layer Type Node Activation function
used in 2nd DNN
1 Input layer – –
2 Hidden layer 100 ReLU
3 Hidden layer 100 ReLU
4 Output layer 10 Softmax

4 Results and Discussion

In this section, the analysis on different DNNs is performed using different combina-
tions of statistical features and signals and the best combinations of statistical features
and signal is obtained. Finally, the fault diagnosis is performed on the combined DNN
model at different sensor fault conditions.

4.1 First DNN for the Sensors Fault Prediction

This model checks every sensor individually and gives binary output whether the
sensor is working or not. And if the sensor is predicted faulty its data is wiped out
and the remaining sensor data is sent as input to the second DNN. The mean (T2)
122 A. Gupta et al.

Time domain /
Frequency domain /
Time Frequency domain
Data

Feature
Extraction

Check weather all

sensors are working 1st ANN

Yes No

Consider the Features of

Consider the Features of all
working Sensors only as
Sensors as Input Features
Input Features

Training of Model using 2nd ANN

given Input Features

Trained Deep Neural

Network

Output: Fault
classification of IM

Fig. 2 Flow chart for deep learning model

and RMS (T4) were considered together for the statistical features extraction from
the frequency domain data and 100% accuracy is achieved for the prediction of all
seven sensors.
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 123

4.2 Second DNN for the IM Faults Prediction

4.2.1 Performance of Individual Statistical Feature

In this classification of fault with individual statistical features are performed. The
accuracy of fault prediction using the individual statistical feature is shown in Fig. 2.
It is discovered that some features are giving very good results (accuracy above 90%)
when considered individually while some features are giving poor results (accuracy
below 40%). Now in the next section try to analyze the situation for considering the
various combinations of statistical features.

4.2.2 Performance Over Combination of Statistical Feature

In this classification of fault with combination of statistical feature is performed. The

feature combinations are divided into different classes.
i. Features having accuracy above 95%.
ii. Features having accuracy above 89%.
iii. Features having accuracy above 60%.
iv. Features having accuracy above 40%.
v. Features having accuracy above 89% and below 40%.
vi. All Features
The accuracy of fault prediction using different combinations of statistical feature
is shown in Fig. 3. It is found that considering statistical features having individual
accuracies above 60% are giving better results as compared to the other classes.
Hence, considering statistical features above 60% accuracy gives the best accuracy.

4.2.3 Performance Over Combination of Sensors

In this classification of fault with combination of sensors is performed. The sensor

combinations are divided into different classes.
i. Vibration signal data.
ii. Current signal data.
iii. Vibration and current signal data.
iv. All sensors data.
The accuracy of fault prediction using different combinations of sensors is shown
in Fig. 4. The accuracy is above 86% by using the vibration signal data and current
signal data, individually. But, the model is giving best results on using the vibration,
current and tachometer data together. Hence, considering all the seven sensors data
gives the best result.
124 A. Gupta et al.

Fig. 3 Bar Graph for Accuracy of predicting IM faults with individual statistical features

Fig. 4 Bar Graph for accuracy of prediction of IM fault using different statistical combination

4.2.4 Performance Over Different Frequencies

This classification of fault is performed by considering all the sensors data and
all the statistical features at different frequencies. Bar graph plotted with different
frequencies and their accuracy is shown in Fig. 5. Hence, from these results it is
found that considering all sensors and statistical features having individual accuracy
above 60% are giving high accuracies at all the different frequencies.
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 125

Fig. 5 Bar Graph for Accuracy of prediction of IM faults using different sensors combination

4.2.5 Performance Over Different Loading Conditions

In this DNN model, the input features are used such that it is giving best results which
are extracted from the above work. Since considering all the seven sensor data with
statistical data above 60% is giving the best results. Here three different loading data
are considered.
i. No Load
ii. Light Load
iii. High Load
Figure 6 shows the accuracy of fault prediction at different loading conditions. It
can be seen that the accuracy of 99.9% was achieved at no load and light load condi-
tions, also for high load conditions 99.94% accuracy was achieved. These accuracies
were achieved when the speed is not used as the input feature of the DNN model.
Table 5 represents the confusion matrix and report at different loading conditions.
In confusion matrix ‘P’ represents the predicted fault label, ‘A’ represents the actual
fault label and ‘NF’ represents no fault condition.
126 A. Gupta et al.

Fig. 6 Bar Graph for Accuracy of prediction of IM faults at different frequency

4.3 Combined DNN

The main DNN is a combination of first DNN and second DNN. Its accuracy at
different conditions is discussed in the following sub-section.

4.3.1 Performance When One Sensor is Having Fault

The classification of IM faults was performed when one sensor is faulty. Figure 7
shows the accuracy of test data for predicting the IM faults at different loading
conditions, when faulty sensor input features were excluded. The ‘1’, ‘2’ and ‘3’
are the vibration sensors, whereas ‘4’, ‘5’ and ‘6’ are current sensors and ‘7’ is the
tachometer. It is found that fault prediction accuracy is above 99.0% when one sensor
is having fault.

4.3.2 Performance When Two Sensors Are Having Fault

The classification of IMs faults was performed when two sensors are faulty. Figure 8
represents the average accuracy of faulty sensor with combinations of other faults.
The accuracy of fault prediction for two faulty sensors is above 98.60% (Fig. 9).
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 127

Table 5 The confusion matrix at different loading conditions

(a) No load
A P
NF BF UR BR RM BRB MPUSPF PUSPF MSWF SWF
NF 2094 0 0 0 0 6 0 0 0 0
BF 0 2100 0 0 0 0 0 0 0 0
UR 0 0 2099 0 1 0 0 0 0 0
RM 0 0 0 2097 0 3 0 0 0 0
BRB 0 0 0 0 2100 0 0 0 0 0
BRB 0 0 0 0 1 2098 1 0 0 0
MPUSPF 0 0 0 1 0 0 2099 0 0 0
PUSPF 0 0 0 0 0 0 0 2100 0 0
MSWF 0 0 0 0 0 0 0 0 2100 0
SWF 0 0 0 0 0 0 0 0 0 2100
(b) Light load
NF 2098 0 0 1 0 1 0 0 0 0
BF 0 2100 0 0 0 0 0 0 0 0
UR 1 0 2094 5 0 0 0 0 0 0
RM 1 0 0 2098 0 1 0 0 0 0
BRB 0 0 0 3 2097 0 0 0 0 0
BRB 0 0 0 9 0 2091 0 0 0 0
MPUSPF 0 0 0 0 0 0 2100 0 0 0
PUSPF 0 0 0 0 0 0 0 2100 0 0
MSWF 0 0 0 0 0 0 0 0 2100 0
SWF 0 0 0 0 0 0 0 0 0 2100
(c) Heavy load
NF 2099 0 0 0 0 1 0 0 0 0
BF 0 2100 0 0 0 0 0 0 0 0
UR 0 0 2100 0 0 0 0 0 0 0
RM 1 0 0 2099 0 0 0 0 0 0
BRB 0 0 0 8 2100 0 0 0 0 0
BRB 2 0 0 0 0 2098 0 0 0 0
MPUSPF 0 0 0 0 0 0 2100 0 0 0
PUSPF 0 0 0 0 0 0 1 2099 0 0
MSWF 0 0 0 0 0 0 0 0 2099 1
SWF 0 0 0 0 0 0 0 0 2 2098
128 A. Gupta et al.

Fig. 7 Bar Graph for Accuracy of prediction of IM faults at different loading conditions

Fig. 8 Bar Graph showing accuracy when one sensor is having fault

5 Conclusions

In the present study, frequency-domain data is used for fault diagnosis of induction
motors. The feature extraction was done and based on that the deep learning model
was trained. The deep learning model consists of two neural networks for predicting
the IMs faults. The novelty condition was examined in the first DNN and faulty
sensors were predicted. The mean and RMS statistical features were used for the
Multiclass Fault Diagnosis and Novelty Detection of Induction Motor … 129

Fig. 9 Bar Graph showing accuracy when two sensors are having faults

feature extraction. The second DNN predicts the IM faults and 25 frequency domain
features were extracted. These statistical features were tested individually and based
on their accuracy their combinations were made and the model was tested. And
considering statistical features having individual accuracies above 60% provides the
best accuracy. Also considering all the sensor data (vibration and current signal data
and tachometer data) was giving the best results compared to considering individual
sensor data.
For prediction of the sensor fault, 100% accuracy is obtained. And for prediction
of IMs faults, when all sensors were working, 99.90% accuracy was achieved in
no load and light load conditions and 99.94% accuracy was achieved in high load
conditions. When one sensor was having fault the accuracy of prediction of fault
was found to be above 99.0%. And when two sensors were having fault accuracy of
prediction of fault above 98.65% was achieved. Whereas Gangsar et al. [5] obtained
the 98%, 97.71% and 98% accuracies for no load, light load and high load condition
respectively. Hence, In the present work, the results obtained were found to be better
as compared to the previous existing literatures.

References

1. Karmakar S, Chattopadhyay S, Mitra M, Sengupta S. Induction motor and faults. Springer,

Singapore, 2016.
2. Tran VT, Althobiani F, Ball A, Choi B. Expert systems with applications an application to
transient current signal based induction motor fault diagnosis of Fourier—Bessel expansion
and simplified fuzzy ARTMAP. Expert Syst Appl. 2013;40(13):5372–84.
3. Glowacz A. Fault diagnostics of DC motor using acoustic signals and MSAF-RATIO30-
EXPANDED. Arch Electr Eng. 2016;65:733–44.
130 A. Gupta et al.

4. Schoen RR, Lin BK, Habetler TG, Schlag JH, Farag S. An unsupervised, on-line system
for induction motor fault detection using stator current monitoring. IEEE Trans Ind Appl.
1995;31(6):1280–6.
5. Gangsar P, Tiwari R. Online diagnostics of mechanical and electrical faults in induction motor
using multiclass support vector machine algorithms based on frequency domain vibration and
current signals. ASCE-ASME J Risk nd Uncertainty in Eng Syst Part B Mech Eng. 2019;5(3):
031001.
6. Rodríguez PVJ, Belahcen A, Arkkio A, Laiho A, Antonino JA. Air-gap force distribution and
vibration pattern of induction motors under dynamic eccentricity. Electr Eng. 2008;90(3):209–
18.
7. Rapur JS, Tiwari R. Automation of multi-fault diagnosing of centrifugal pumps using multi-
class support vector machine with vibration and motor current signals in frequency domain. J
Braz Soc Mech Sci Eng. 2018;40:278.
8. Shao SY, Sun WJ, Yan RQ, Wang P, Gao RX. A deep learning approach for fault diagnosis of
induction motors in manufacturing. Chinese J Mech Eng (English Edition). 2017;30(6):1347–
56.
9. Wong MLD, Jack LB, Nandi AK. Modified self-organising map for automated novelty detection
applied to vibration signal monitoring. Mech Syst Signal Process. 2006;20:593–610.
10. Sadooghi MS, Khadem SE. Improving one class support vector machine novelty detection
scheme using nonlinear features. Pattern Recognit. 2018;83:14–33.
11. Kumar D, Dewangan A, Tiwari R, Bordoloi DJ. Identification of inlet pipe blockage level in
centrifugal pump over a range of speed by deep learning algorithm using multi-source data.
Measurement. 2021;186:110–46.
Modelling and Analysis of Active
Magnetic Bearing Integrated Reaction
Wheels for Satellite Applications

P. A. Thashreef, Rajiv Tiwari, and Dhruba Jyoti Bordoloi

Abstract Reaction wheel assemblies are widely used in satellite manoeuvring appli-
cations because of the advantage of precision in manoeuvring without use of the fuel.
Since the micro-vibrations of the reaction wheel can highly affect the performance
of critical equipment in the satellite, it is crucial to control those vibrations effec-
tively using vibration isolation systems. The dynamic disturbance model is required
to predict the effects of disturbance sources and for designing the vibration isolation
systems. This work focuses on the dynamic modelling of the reaction wheel inte-
grated with active magnetic bearings to predict the micro-vibrations of the system.
A complete dynamic model of the reaction wheel systems levitated by the active
magnetic bearings is developed by considering rotor dynamics, magnetic bearing
control systems, and rotor unbalance as the disturbance source. Magnetic bear-
ings are controlled using a PD controller or the Linear Quadratic Regulator (LQR)
controller, which is tuned to obtain the critical speed of the system greater than
the operational speed. The vibration characteristics of the reaction wheel system are
predicted from the developed model by numerical simulation. The model is simulated
at different operating speeds using MATLAB SIMULINK to obtain the character-
istics of the force and moment induced on the satellite structure from the reaction
wheel system. The influence of the disturbance sources in the dynamic responses is
discussed using those simulation results. The obtained results can be used to design
the vibration control and isolation systems for reaction wheel assemblies of satellites
and micro-vibration suppression algorithms.
Keywords Magnetic bearing reaction wheels · Reaction wheel assemblies ·
Active magnetic bearing · Micro-vibration analysis · Dynamic modelling of the
reaction wheel

P. A. Thashreef (B) · R. Tiwari · D. J. Bordoloi

Department of Mechanical Engineering, IIT, Guwahati, Assam, India
e-mail: [email protected]
R. Tiwari
e-mail: [email protected]
D. J. Bordoloi
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 131
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_9
132 P. A. Thashreef et al.

Nomenclature

e Radial eccentricity of the wheel

ez Axial eccentricity
f Force
f Force vector
G Gyroscopic matrix
h State vector
I Mass moment of inertia
k Stiffness
K Stiffness matrix
l Length of shaft
m Mass of the wheel
M Moment
M Mass matrix
L State observer matrix
r Radius
T Transformation matrix
ϕx Angular displacement about x axis
ϕy Angular displacement about y axis
η Displacement vector
ω Angular velocity

Superscripts

c Control
m Magnetic
unb Unbalance

Subscripts

a Axial
b Bearing actuator centre
D Derivative factor
d Diametric
i Force- current factor
m Magnetic bearing
P Proportional factor
p Polar
r Radial
s Sensor position
δ Force displacement factor
Modelling and Analysis of Active Magnetic Bearing Integrated … 133

1 Introduction

Reaction wheel systems are used as spacecraft attitude controllers and large angle
slewing manoeuvres for years. These reaction wheels can provide enough torque
by accelerating and decelerating the wheel with the help of a brushless DC motor
powered by electricity from the solar panel [1]. The required torque for controlling
the attitude of the satellite is provided by the momentum exchange from reaction
wheels. Conservation of angular momentum of the closed system is the basic prin-
ciple behind the working of the reaction wheel [2]. The satellite manoeuvring is
performed using the torque and force generated internally. Therefore, the angular
momentum accumulated on the wheel is identical in magnitude and reverse in direc-
tion to that associated with the spinning of the satellite. A usual reaction wheel
system comprises of a brushless DC motor driven flywheel on ball bearings [3].
An alternative to this configuration is to substitute the ball bearings with Active
magnetic bearings (AMB). AMB is a mechatronic device with many advantages
like no mechanical wearing, low vibration, precise control, high rotational speed, no
lubrication, less servicing, and a clean environment. Thus it makes a better substi-
tute for conventional ball bearing [4]. The precision and accuracy of the satellite
equipment can be affected by the vibration forces and moments [5, 6].
Micro-vibrations in the satellites are mechanical disturbances of low level caused
by the operation of the mechanical equipment on the satellite [7]. Reaction Wheel
systems are one of the primary sources of micro-vibrations in satellites. These vibra-
tion forces and moments get amplified by the satellite structure while transmitted
to the sensitive equipment, affecting the equipment’s performance. So it is essen-
tial to study the micro-vibration sources and the ways to reduce the effect of this
micro-vibration.
In general, the disturbance sources in the reaction wheels suspended on ball bear-
ings are flywheel unbalance, ball bearing disturbances and torque ripple. The two
types of flywheel unbalance are static and dynamic. These unbalances result from
uneven distribution of the mass around the axis of rotation [6]. Bearing disturbances
are caused by irregularities in the balls, cages, or races [8]. On the other hand,
the magnetic bearing reaction wheel system has dominant disturbances from sensor
runout and unbalance of the wheel [9, 10]. Addari et al. [11] investigated the coupled
micro-vibration dynamics of the reaction wheel numerically and experimentally.
The developed model is capable of measuring the disturbance induced in the hard-
mounted configuration of the reaction wheel. They also validated the result with the
coupled wheel structure disturbance model. Wang et al. [12] developed a reaction
wheel model with non-linear rolling bearings. They considered the mass imbalance
of the flywheel, rolling bearing force and flexibility of the supporting structure.
Bailke and Hansell proposed that most of the space mission failures are due to reac-
tion wheel failures, and ball bearing failure causes the failure of the reaction wheel
[13]. They explained that the root cause of ball bearing failures is space charging envi-
ronments. After tests and analysis, they concluded that the space environment and the
likely space charging caused high friction behaviour in the ball bearings resulting in
134 P. A. Thashreef et al.

failure. A dynamic model for the measurement of the micro-vibrations in the magnet-
ically suspended flywheel is proposed by Peng et al. [14]. The model is developed
by considering rotor dynamics, magnetic bearing control system, micro-vibration
sources such as rotor mass imbalance, and sensor runout. Gallego et al. [15] worked
on a space actuator made of permanent magnetic bearing for the development of
an online method for micro-vibration measurement and identifying the forces. The
bearing forces are created by permanent magnets and active electromagnets. The
micro-vibration sources they considered were unbalanced driven disturbances in the
wheel [16].
From the literature review, it is found that there is a lot of scope for the study
of the reaction wheel used in satellite systems and its micro-vibration-induced. The
reduction of failure time is essential due to the use of conventional bearing. The
objective of the present work is to model AMB integrated reaction wheel system
and study the micro-vibrations it produces. As part of it, a reaction wheel system is
modelled as a cantilever type with a Radial Active Magnetic Bearing (RAMB) and
Thrust Active Magnetic Bearing (TAMB). The AMBs are controlled using either of
PD controller or Linear Quadratic Regulator (LQR) controller.

2 Modelling of the Reaction Wheel System

The reaction wheel system considered in the present work is suspended on a 2-DOF
radial active magnetic bearing (RAMB) and a 3-DOF thrust active magnetic bearing
(TAMB), as shown in Fig. 1a as the upward cantilever configuration. The Z-axis of
the system is considered, such that it is coinciding with the shaft axis. Moreover,
the origin of the coordinate system is made in such a way that it coincides with the
geometrical centre of the disc. The axial translational motion of the system is taken
care by the thrust AMB, and the radial translational motion is taken care of by the
radial AMB. The TAMBs will make a moment about x and y axes along with the
moment caused by the forces of RAMB because the position of the RAMB is offset
to the wheel axis. Thus the radial and thrust AMBs together are responsible for the
rotor system’s radial tilt (about the x and y axes).
Six displacement sensors are arranged in such a way that two sensors measure
the radial translational displacement (along the x- and y-directions) of shaft at a
distance l 1 from the wheel and four other sensors are arranged radially on the
disc at a distance r s from the axis of the shaft to measure the axial translational
displacements (along z direction). The Fig. 1b shows the axial view of the reaction
wheel system. The TAMB actuators are arranged and numbered (as TAMB1 to
TAMB4), as shown in the diagram. The radial distance of TAMB actuator centre
from the z-axis is shown by r b . Similarly, displacement sensors are also arranged
and numbered as S1 to S4 as depicted in Fig. 1b, and the radial distance of sensors
from the axial axis z is marked as r s .
Modelling and Analysis of Active Magnetic Bearing Integrated … 135

Fig. 1 a Schematic diagram of the reaction wheel system with AMBs, b Axial view of the reaction
wheel system

To develop the dynamic model of the reaction wheel system, the disc displacement
 T
vector is taken as x, y, ϕ y , ϕx with x and y are the linear displacements of the
wheel while ϕx and ϕ y are angular displacements of the disc in the z-y and z-x planes,
respectively, as shown in Fig. 1a.
Considering force equilibrium and moment conditions of the wheel, the dynamic
model of the system is,

M η̈− ωG η̇ = f unb − f c (1)

where M is the mass matrix, G is the gyroscopic matrix, f c is the control force vector,
f unb is the unbalance force vector and η is the wheel displacement vector. They can
be detailed as
⎡ ⎤
0 0 0
0
⎢0 0 0⎥
0
M = diag[m, m, Id , Id ] ; G = ⎢
⎣0
⎥
0 Ip ⎦
0
0 0 −I p
0
 unb unb 
unb T
 c c c T

f unb
= f x , f y , Mzx , M yz
unb
; f = f x , f y , Mzx
c c
, M yz
 T
η = x, y, ϕ y , ϕx (2)

where, m is reaction wheel mass, f iunb is static unbalance force in the ith direction,
f ic is the control forces generated by the radial AMB in the ith direction (with i = x,
y), ω is the angular velocity, I p is polar mass moment of inertia, Id is the diametric
mass moment of inertia, Miunb j is the dynamic unbalance moment on plane ij (with ij
136 P. A. Thashreef et al.

= z-x, y–z) and Micj is the restoring or control moment caused by the AMB’s control
force on ij plane.
The unbalance force vector, which is the disturbance source in the reaction wheel
is given by,
⎧ ⎫ ⎧ ⎫
⎪
⎪ f xunb ⎪
⎪ ⎪ meω2 cos ωt ⎪
⎨ f unb ⎪
⎪ ⎬ ⎪ ⎨ ⎪
⎬
meω2 sin ωt
f unb = y
= (3)
⎪ Mzx ⎪
⎪ unb
⎪ ⎪ ⎪ meω ez cos ωt ⎪
2
⎪
⎩ M unb ⎪
⎪ ⎭ ⎩ meω2 e sin ωt ⎭
yz z

where, e is the radial eccentricity and ez is the axial eccentricity of the disc. The
control force vector f c in (1) can be written in terms of magnetic force vector f m by
coordinate transformation as

f c = Tm f m (4)
 T
where, f m =  f x , f y , f z1 , f z2 , f z3 , f z4 and T m ∈ R4×6 is the force transfor-
mation matrix. f x , f y are the force in x- and y-direction by the radial AMB and
( f z1 , f z2 , f z3 , f z4 ) are the thrust AMB forces by each actuator. Transformation matrix
Tm can be written in detail as,
⎡ ⎤
1 0 0 0 0 0
⎢ 0 1 0 0 0 0⎥
Tm = ⎢
⎣ −(l1 + l2 )
⎥ (5)
0 0 −rb 0 rb ⎦
0 (l1 + l2 ) −rb 0 rb 0

Since the displacements are obtained at the sensor locations, coordinate transfor-
mation from sensor displacement to wheel displacement and displacements at bearing
centres. The transformation between sensor coordinate and the wheel displacement
is obtained as

η = T s ηs (6)

where, ηs = {xs , ys , z s1 , z s2 , z s3 , z s4 }T and T s ∈ R4×6 is the transformation matrix

from sensor coordinates to wheel coordinates. The pseudo-inverse of the transfor-
 −1
mation matrix can be written as T sT T s T sT . (xs , ys ) are radial displacements
measured by the sensors at l2 distance from RAMB and z si is the axial displacement
measured using ith sensor (i = 1 to 4), which is arranged at a radial distance of r s
from the axis of rotation as shown in Fig. 1b. The transformation matrix Ts can be
written as
Modelling and Analysis of Active Magnetic Bearing Integrated … 137
⎡ ⎤
1 0 0 − 2rl1s 0 2rl1s
⎢0 1 2rl1s 0 − 2rl1s 0 ⎥
⎢ ⎥
Ts = ⎢ ⎥ (7)
⎣0 0 0 − 2r1s 0 2r1s ⎦
0 0 − 2r1s 0 1
2rs
0

Similarly, the coordinate transformation from sensor coordinates to bearing centre

coordinates can be written as

ηb = T b ηs (8)

where, ηb = {xb , yb , z b1 , z b2 , z b3 , z b4 }T and T b ∈ R6×6 is the transformation matrix.

(xb , yb ) are radial displacements at RAMB and z bi is the axial displacement at the ith
TAMB actuator centre (i = 1 to 4) which is arranged at a radial distance of r b from
the axis of rotation as shown in Fig. 1b. The transformation matrix Tb is,
⎡ ⎤
1 0 0 l2
2rs
0 − 2rl2s
⎢0 1 − 2rl2s 0 l2 ⎥ 0
⎢    2rs ⎥
⎢ ⎥
⎢0 0 2+1 rb
0 1
− 0 rb ⎥
⎢ 
2rs  2  2rs ⎥
Tb = ⎢
⎢0
⎥ (9)
⎢ 0 0 1
+ 2rrbs 0 1
− 2rrbs ⎥⎥
⎢   2   2
⎥
⎢0 0 21 − 2rrbs 0 1
+ rb
0 ⎥
⎣   2 2rs  ⎦
00 0 1
2
− 2rrbs 0 1
2
+ 2rrbs

The magnetic force provided by the AMB is proportional to the control current
flowing through the actuator coil and displacement. AMB force vector f m in the
Eq. (4) can be found by,

f m = K i i − K δ ηb (10)
 T
where, i = i x , i y , i z1 , i z2 , i z3 , i z4 is the control current vector and the matrix Ki
and K δ are given by

K i = diag[kir , kir , kia , kia , kia , kia ]T and

K δ = diag[kδr , kδr , kδa , kδa , kδa , kδa ,]T (11)

where, k ir and k ia are the force-current factors or force-current constant of the radial
and thrust bearing actuators, respectively. kδr and kδa are the displacement force
factors of the radial and thrust bearing actuator, respectively. Substituting Eqs. (4),
(10) and (8) into (1) gives the complete dynamic model of the system as
138 P. A. Thashreef et al.
 
M η̈ − ωG η̇ = f unb − T m K i i − K δ T b ηs (12)

Control current vector i can be obtained based on the proper controller used.

2.1 Control Current

The control current to the magnetic bearing plays an important role in minimising
the vibration amplitude of the system. Therefore, selection and tuning of the proper
controller play an essential role in the rotor system having AMB. In this model, we
used Proportional Derivative (PD) controller and Linear Quadratic Regulator (LQR)
controller separately.

2.1.1 PD Controller

The control current vector i is obtained by using a proper controller. In this case,
Proportional Derivative (PD) controller is used.

i = K P ηb + K D η̇b (13)

The matrices KP and KD are given by,

K P = diag{k Pr , k Pr , k Pa k Pa , k Pa k Pa }T (14)

K D = diag{k Dr , k Dr , k Da k Da , k Da k Da }T (15)

where, k Pr and k Pa are the proportional factors for the radial and axial bearings, k Da
and k Dr are the derivative factors for the axial and radial bearings of the PD controller.
Substituting Eqs. (13) and (8) in (12), the complete micro-vibration dynamic model
of the reaction wheel system with PD control can be written as

M η̈ − ωG η̇ = f unb − T m K i K P Tb ηs − T m K i K D Tb η̇s +T m K δ T b ηs (16)

The proportional factor of the PD controller has been selected so that it should be
greater than the ratio of the displacement force factor to the force-current constant
of the magnetic bearing [4]. Also, PD controller parameters are tuned for getting the
critical speed of the system greater than the maximum working spin speed of the
system, which is 5000 rpm. The Campbell diagram gives the critical speed of the
system by plotting the natural whirl frequency against the spin speed of the system.
The Eigenvalue problem of the system for free vibration gives the natural frequency.
The equation of motion of the reaction wheel system with PD controller for free
vibration from (16) with f unb = 0 and using Eq. (6) obtained as
Modelling and Analysis of Active Magnetic Bearing Integrated … 139
  −1 
M η̈+ T m K i K D Tb T sT T s T sT − ωG η̇
  −1  −1 
+ T m K i K P Tb T sT T s T sT − T m K δ Tb T sT T s T sT η=0 (17)

The above equation can be written as a state-space form from which the whirl
frequency of the system can be obtained

ḣ = Dh (18)

with,
 
η
h= (19)
η̇
 
0 I
D= (20)
−M −1 K −M −1 (C−ωG)

where,
 −1  −1
K = T m K i K P Tb T sT T s T sT − T m K δ Tb T sT T s T sT (21)

 −1
C = T m K i K D Tb T sT T s T sT (22)

Solving the state-space form in Eq. (18) as a standard Eigenvalue problem, the
imaginary part of the complex Eigenvalue gives the natural whirl frequency of the
system. The PD Controller parameters are tabulated in Table 1, along with the
system’s physical properties. Figure 2 shows the Campbell diagram of the system
with the PD controller. Solid lines represent the system’s natural whirl frequency,
and the dashed line represents whirl frequency equal to spin speed. The magnified
view of the intersection area is shown as an inset plot. The first two critical speeds
of the system are obtained as 584.37 rad/sec and 647.82 rad/sec.

2.1.2 LQR Controller

The control current can be obtained using LQR Controller. The design process of
the LQR controller is described in detail in the previous chapter. Using Eq. (6) into
(12) the reaction wheel system dynamics can be written as
 −1
M η̈ − ωG η̇ = f unb − T m K i i + T m K δ T b T sT T s T sT η (23)

The output of the system is obtained in terms of sensor displacement ηs . The

above equation can be written in state-space form as
140 P. A. Thashreef et al.

Table 1 Physical properties of the system

Properties Values
Mass of the wheel, m 1.3 kg
Static mass unbalance, me 5 × 10−6 kgm
Dynamic mass unbalance, meez 4 × 10−6 kgm2
Polar mass moment of inertia, I p 2.34 × 10−3 kgm2
Diametrical mass moment of inertia, I d 1.17 × 10−3 kgm2
Distance between wheel and sensors in shaft, l 1 6 cm
Distance between sensor and RAMB, l 2 3.5 cm
Distance from RAMB to base, l 3 15 cm
Radial distance of axial displacement sensors, r s 6 cm
Radial distance of centre of TAMB actuators, r b 4 cm
Proportional factor for the radial bearings k Pr 70000A/m
Proportional factor for the axial bearings k Pa 90000A/m
Derivative factor for the radial bearings k Dr 20As/m
Derivative factor for the axial bearings k Da 25As/m

Fig. 2 Campbell Diagram of the system with PD controller

unb
ḣ = Ah + Bi + E f
ηs = C h (24)

where, the system matrix A, control input matrix B, output matrix C and unbalance
force input matrix E can be written in detail as
Modelling and Analysis of Active Magnetic Bearing Integrated … 141
   
0 I 0
A=  −1 ; B =
M −1 T m K δ T b T sT T s T sT ωM −1 G −M −1 T m K i
    
 0
C = T sT T s T sT −1 0 ; E = (25)
M −1

The LQR problem rests upon the assumption that the plant dynamics and all the
state variables are available [17]. The LQR problem is to find the optimal control
vector of i, which minimises the cost function given by.
 ∞  
J= h T Qh + i T Ri dt (26)
0

where Q and R are the diagonal positive definite weighing matrices that penalise the
states and control effort, respectively. The matrices Q and R determine the relative
importance of the error and the expenditure of this energy. These matrices are the
controlling tuning parameters. The weighing matrix Q should be chosen accordingly
to how we want to penalise certain states. Similarly, the R matrix can also be selected.
The quadratic performance of the cost function gives the mathematical solution to
the system as an optimal control law, which is given by

i = −K c h (27)

where Kc is the controller gain. Substituting Eq. (27) in Eq. (24) gives,

ḣ = ( A − B K c )h (28)

The minimisation of the cost function gives the optimal control matrix Kc as,

K c = R−1 B T X (29)

where X = X T is the unique positive semi-definite solution of the Algebraic Reccati

Equation (ARE),

AT X + X A − X B R−1 B T X + Q = 0 (30)

The process of minimising the cost function involves solving this equation.
MATLAB can be used to solve this minimisation problem using lqr command,
which returns the controller gain matrix.
The design of the LQR requires all the state variables to be measurable for the
feedback control. However, in practical conditions, all the state variables may not be
measurable because of practical constraints. This makes the state observer or state
estimator requirement which can reconstruct the complete state information based
on the measured output. The state observer can be used to estimate the full state
vector as
142 P. A. Thashreef et al.
 
˙
ĥ = A ĥ + Bi + L ηs − C ĥ + E f unb (31)

where, ĥ is the estimate of the state vector h. Introducing an error vector as

e = h − ĥ (32)

Differentiating Eq. (32) and substituting Eqs. (24) and (31) will give

ė = (A − LC)e (33)

The Eq. (33) governs the estimation error. We can control the estimated state
to approach the actual state or equivalently the rate of error approaches zero if all
eigenvalues can be assigned arbitrarily [18]. According
 T toTthe linear system theory,
if a pair (A, C) is observable, then a dual
 pair A ,
 C is controllable. Since
(A − LC) have the same eigenvalues as AT − CT LT , the observer gain matrix L
can be solved using the pole placement method. The Pole placement method can be
solved using MATLAB by using the place command.
Optimal state observer gain matrix can also be obtained using the LQR method.
The cost function for the state observer problem is obtained as
 ∞  
Jo = eT Q o e + uoT Ro uo dt (34)
0

where Qo and Ro diagonal positive definite matrices which taken as the identity
matrices, uo = y − C ĥ. State observer gain matrix is then given by,

L = Y C T R−1
o (35)

where Y = Y T is the unique positive semi-definite solution of the ARE given by,

Y AT + AY − Y C T R−1
o CY + Q o = 0 (36)

From the above derivation state, the observer matrix can also be found using the
same function lqr in MATLAB. The transpose of matrices A and C should be given
as the input to the function, and the output to the function will be the transpose of
the required state observer.
Using Eq. (27) in Eq. (24) the system dynamics in the form of state-space is,
unb
ḣ = Ah − B K c ĥ + E f (37)

The state-space form of the system for free vibration if all states are obtained as
output is given by

ḣ = ( A − B K c )h (38)
Modelling and Analysis of Active Magnetic Bearing Integrated … 143

Fig. 3 Campbell diagram of the system with LQR controller

The Eq. (38) can be reduced to the standard Eigenvalue problem. The eigenvalue
λ obtained is of complex form and the imaginary part gives the damped natural whirl
frequency of the system with LQR controller. The positive definite weighing matrix
R is taken as the identity matrix, and Q is tuned in such a way that the system’s
critical speed is greater than the operating speed to avoid resonance in the normal
operation. The matrix Q is a diagonal matrix with Q ii = 6.5 × 109 for i = 1 to 4
which corresponds to the penalties for displacement, and Q ii = 1 for i = 5 to 8,
which corresponds to the velocity part in the state vector. The Campbell diagram of
the system is shown in Fig. 3. The dashed line represents the spin speed equal to the
natural whirl frequency, and the solid lines represent the natural whirl frequency of
the system’s corresponding spin speed. The area where the first and second natural
whirl curve intersects with the curve’s natural whirl frequency equal to spin speed is
magnified and shown in the inset plot. The first two critical speed of the system is
obtained as 612.13 rad/s and 612.21 rad/s respectively.

2.2 Micro-vibration Measurement

Micro-vibrations in the satellites are mechanical disturbances of low level caused

by the operation of the mechanical equipment on the satellite [7]. Reaction Wheel
systems are one of the primary sources of micro-vibrations in satellites. The force
generated by the AMBs in our reaction wheel system will produce force and moment
sources for micro-vibration in the satellite structure. Therefore, measurement of these
micro-vibrations is necessary to design the proper vibration damping structure for
the satellite.
144 P. A. Thashreef et al.

Fig. 4 Simplified model of

reaction wheel system Axial Magnetic
cb kb Bearing
Reaction wheel

l1+l2
kb
Radial Magnetic
Bearing
cb
Stator/ casing
l3
cs ks

Base plate

The micro-vibrations produced by the reaction wheel system are measured in

terms of the force and moment induced on the satellite. A simplified view of the
cantilever type reaction wheel system is shown in Fig. 4. All force transmission
points in the structure are modelled as a spring-damper equivalent even though there
is no physical spring and damper present. The connection between the stator and the
base plate of the reaction wheel system is direct, and there is no physical spring and
damper present for this case.
The dynamic reaction force developed by the AMBs will generate force and
moment in the satellite base through the stator of the reaction wheel system. The
disturbance force vector obtained as,

F dis = T base f m (39)

 T
where, f dis = Fx , Fy , Mzx , M yz and T base ∈ R4×6 is the force transformation
matrix from bearing locations to base. The transformation matrix is written in detail
as,
⎡ ⎤
1 0 0 0 0 0
⎢ 0 1 0 0 0 0⎥
T base =⎢
⎣ −l3
⎥ (40)
0 0 −rb 0 rb ⎦
0 l3 −rb 0 rb 0

where, l 3 is the distance from RAMB to the base plate. In this space, the motor for
the reaction wheel system is placed. To reduce the micro-vibration moment induced
on the base plate motor is selected such that the length l3 is as small as possible, and
in our system, the value is taken as 15 cm.
Modelling and Analysis of Active Magnetic Bearing Integrated … 145

3 Numerical Simulation

The numerical simulation is carried out for the developed dynamic model of the
reaction wheel system described in Eq. (12) using Matlab Simulink. The controller
for the system is taken as PD and LQR, so the controller part will be changed
accordingly for the system. The input to the LQR controller is the displacement at the
sensor points itself and for the PD controller, the bearing displacements. Therefore,
sensor displacement should be transformed to bearing displacement before taking
into the PD controller. The time steps for the simulation is taken as 10−4 s, and
fourth-order Runge–Kutta (RK4) is taken as the solver. The operational speed of the
system is set from 0 to 5000 rpm with an interval of 250 rpm for the simulation. At
each time interval, the required outputs from the Simulink are noted for the system
with both PD and LQR controller separately.
Simulation for the model with both the controller is done at each interval, and
force acting on the base plate is extracted. The half spectrum of the force amplitude
corresponding to different spin is obtained by carrying out the Fast Fourier Transform
(FFT) of the time domain signal of the force at the base plate.
The simulation results of micro-vibration are visualised as a waterfall plot, i.e.,
the amplitude of force or momentum vs frequency vs spin speed of the system. The
waterfall plot is a widely used tool to identify the disturbance trends in the rotational
system. The waterfall plot of the micro-vibration force in x- and y-direction and
micro-vibration moment about y axis in z-x and about x axis in y–z plane with PD
controller is shown in Figs. 5 and 6, respectively. Similarly, the micro-vibration
force and moments with LQR controller is shown as waterfall plot in Figs. 7 and 8,
respectively.

Fig. 5 Waterfall plot of micro-vibration force with PD controller a x-direction b y-direction

146 P. A. Thashreef et al.

Fig. 6 Waterfall plot of the micro-vibration moment with PD controller a z-x plane b y–z plane

Fig. 7 Waterfall plot of the micro-vibration force with LQR controller a x-direction b y-direction

It is visible from the plots there is a harmonic component in the waterfall plot.
Since we consider unbalance of the rotor as the only disturbance source in the reaction
wheel system, the harmonic must be caused by the mass unbalance of the wheel and
is dependent on the speed of rotation. From the simulation results, we can see that
the micro-vibration force acting on the satellite structure will be having a higher
amplitude for the system with PD controller at a higher spin speed than the system
with LQR controller. At lower spin speed, the micro-vibration forces generated by
the system with the PD controller are negligible. Micro-vibration moments generated
Modelling and Analysis of Active Magnetic Bearing Integrated … 147

Fig. 8 Waterfall plot of the micro-vibration moment with LQR controller a z-x plane b y–z plane

by the system with the PD controller are lower than the LQR controller system. The
objective of using an LQR controller is to check whether it has any advantages over
a conventional PD controller. The results show that using an LQR controller helps
reduce the micro-vibration forces and amplitude of wheel displacements at higher
operating speeds while the amplitude of micro-vibration moments increases.

4 Conclusion

In this work, the dynamic model of the reaction wheel system integrated with AMB is
developed and studied the system micro-vibration pattern. The reaction wheel system
is modelled as a cantilever type with a RAMB and a TAMB. Only the unbalance
force in the wheel is considered for the input disturbance to the system for modelling
the reaction wheel system. The system is modelled with both the PD controller and
LQR controller. According to the proposed model, simulations are conducted and
predict the micro-vibration characteristics of the AMB integrated reaction wheel
system. The results from the simulation show that the unbalance force affects the
micro-vibration amplitude proportionally. So the best way to suppress the micro-
vibration in the reaction wheel system is to minimise the unbalance in the system. In
the present model, the only source of disturbance is wheel unbalance. The influence
of other disturbance sources such as sensor runout and motor induced disturbances
can be added to the system. A proper micro-vibration suppression system can be
developed using the reaction wheel’s developed dynamic micro-vibration model.
148 P. A. Thashreef et al.

References

1. Schaub H, Junkins JL. Analytical mechanics of aerospace systems. Amer Inst Aeronaut
Astronautics United States; 2003.
2. Mehrjardi MF, Sanusi H, Ali MAM. Developing a proposed satellite reaction wheel model
with current mode control. In: International conference on space science and communication
(IconSpace), Langkaw, Malaysia; 2015. https://doi.org/10.1109/IconSpace.2015.7283839
3. Markley FL, Crassidis JL. Fundamentals of spacecraft attitude determination and control.
Newyork: Springer; 2014.
4. Tiwari R. Rotor systems: analysis and identification. Boca Raton: CRC Press; 2017.
5. Davis L, Wilson J, Jewel R, Rodon J. Hubble space telescope reaction wheel assembly vibration
isolation system. NASA Marshall Space Flight center; 1986. pp. 1–37.
6. Le MP. Micro-disturbances in Reaction Wheels [PhD Thesis]. Technische Universiteit
Eindhoven, Eindhoven; 2017.
7. European Space Agency, European Cooperation for Space Standardization (ESA ECSS),
Space Engineering. Spacecraft mechanical loads analysis handbook. ESA Requirements and
Standards Division, Noordwijk; 2013.
8. Li B, Goddu G, Chow MY. Detection of common motor bearing faults using frequency-domain
vibration signals and a neural network based approach. In: Proceedings of the American control
of conference; 1998. pp. 2032–2036. https://doi.org/10.1109/ACC.1998.702983
9. Setiawan JD, Mukherjee R, Maslen EH, Song G. Adaptive compensation of sensor runout
and mass unbalance in magnetic bearing systems. IEEE/ASME international conference on
advanced intelligent mechatronics; 1999. pp. 800–805. https://doi.org/10.1109/AIM.1999.
803275
10. Setiawan JD, Mukherjee R, Maslen EH. Synchronous sensor runout and unbalance compen-
sation in active magnetic bearings using bias current excitation. J Dyn Syst Meas Contr.
2002;124:14–24. https://doi.org/10.1115/1.1435644.
11. Addari D, Aglietti GS, Remedia M. Experimental and numerical investigation of coupled
microvibration dynamics for satellite reaction wheels. J Sound Vib. 2017;386:225–41. https://
doi.org/10.1016/j.jsv.2016.10.003.
12. Wang H, Han Q, Ruizhi L, Qing T. Dynamic modeling of moment wheel assemblies with
nonlinear rolling bearing supports. J Sound Vib. 2017;406:124–45. https://doi.org/10.1016/j.
jsv.2017.06.019.
13. Bialke W, Hansell E. A newly discovered branch of the fault tree explaining systemic reaction
wheel failuires and anomalies. In: Proceedings 17th European space mechanisms and tribology
symposium; 2017. pp. 1–8.
14. Peng C, Fang J, Cui P. Dynamics modeling and measurement of the microvibrations for a
magnetically suspended flywheel. IEEE Trans Instrum Meas. 2015;64:3239–52. https://doi.
org/10.1109/TIM.2015.2459491.
15. Gallego GB, Rossini L, Onillon E, Achtnich T, Zwyssig C, Seiler R, Araujo MD, Perriard Y. On-
line micro-vibration measurement method for lorentz-type magnetic-bearing space actuators.
Mechatronics. 64;2019. https://doi.org/10.1016/j.mechatronics.2019.102283
16. Gallego GB, Rossini L, Onillon E, Achtnich T, Zwyssig C, Seiler R, Araujo MD, Perriard Y.
Magnetic bearing reaction wheel micro-vibration signature prediction. In: Proceedings of the
18th European space mechanisms and tribology symposium (ESMATS 2019); 2019.
17. Ogata K. Modern control engieering. 4th ed. Tehran: Pearson Education International; 2002.
18. Chen CT. Linear system theory and design. 3rd ed. New York: Oxford University Press; 1999.
Single Plane Balancing of Rotor-AMB
System Using Virtual Trial Unbalances

Gyan Ranjan, Rajiv Tiwari, and H. B. Nemade

Abstract The present work utilizes the influence coefficients and virtual trial unbal-
ances as magnetic excitation to predict the residual unbalances in the rotating
machinery. The virtual trial unbalances (VTUs) are generated at the Active Magnetic
Bearing (AMB) position by adding excitation current to the control current. The
forces generated by additional excitation current imitate the manually added trial
unbalance without interfering with the stability of the rotor system. The experiments
are carried out on the RK-4 rotor test rig supported on two journal bearings and
integrated with AMB. The PID controller implemented in the AMB is designed in
the SIMULINK platform with Data Acquisition System (DAQ). The half-spectrum
amplitudes and phases of the displacement responses are utilized with and without the
addition of VTUs to estimate the residual unbalances. The rotor system is balanced
as per the estimation results, and the displacement response of the system is found
out to be reduced.

Keywords Active magnetic bearing · Balancing · Rigid rotor · PID controller ·

Virtual trial unbalances

Nomenclature

e Eccentricity of unbalance
fd Unbalance force at disc location
fP Static load

G. Ranjan (B) · R. Tiwari

Department of Mechanical Engineering, IIT Guwahati, Guwahati, India
e-mail: [email protected]
R. Tiwari
e-mail: [email protected]
H. B. Nemade
Department of Electronics and Electrical Engineering, IIT Guwahati, Guwahati, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 149
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_10
150 G. Ranjan et al.

fr AMB force
fT Trial excitation force
ic and Control and current
iT Excitation current
ip Static current
m Mass of the unbalance
rd Displacement response at disc location
rm Displacement response at AMB location
α Influence coefficient related to unbalance forces at the displacement
locations
AMB Active Magnetic Bearing
A/D Analog to Digital
D/A Digital to Analog
ICM Influence Coefficient Method
PID Proportional-Integral-Derivative

1 Introduction

The rotating machinery is subjected to various types of faults such as residual unbal-
ances, bow, misalignment, crack, and mechanical looseness of the components. The
residual unbalances are the most common fault that arises in the rotating components.
The intensity of unbalances increases with increase in the operating speed of the rotor
system that causes excessive vibration of the rotating components. Due to the unde-
sirable vibration in the rotor system, the life of the rotating components is reduced
and also affects the productivity of the machinery. Therefore, it is necessary that the
rotor system gets balanced periodically to control the level of vibration in the system.
With the advancement of technology, the AMB is utilized in the condition monitoring
of the machinery. The vibrational amplitude of the rotor system is controlled within
desirable limits with the help of AMB and also allows the prediction of faults in the
rotating machinery. Various researchers have developed identification algorithms to
identify the residual unbalances existing in the AMB controller rotor system [1].
Sahinkaya et al. [2] controlled the forces and vibration transmitted to the base of
the rotor system with the application of AMBs using an adaptive multi-objective
algorithm. Jiang and Zhu [3] utilized adaptive finite duration impulse response to
suppress the multiple frequency vibrational responses of the rotor setup utilizing
AMB as a controller. Hutterer et al. [4] developed an unbalance controller with a
self-sensing algorithm to decrease the fluctuation arising in the control current due
to unbalance. Saeed and Kamel [5] controlled the nonlinear vibration of the Jeffcott
rotor system using a control algorithm with tuned positive position feedback. They
achieved a sufficient reduction in vibration with the designed controller. Ranjan and
Tiwari [6] discussed a generalized influence coefficients based methodology to iden-
tify the residual unbalance existing in the rotating machinery with magnetic bearing
Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 151

control. They also identified the appropriate location of the AMB to perform the
balancing experiment with maximum reduction in vibration. Fang et al. [7] devel-
oped an algorithm for balancing rotating machinery supported on AMBs without
the application of trial masses. They utilized the current in the coils of actuator to
identify the balancing masses of the rotor-AMB system. Wang et al. [8] discussed
a balancing methodology utilizing virtual trial weights in the flexible rotor system
supported on the AMBs. They successfully identified the balance masses present in
the rotor-AMB system. Ranjan and Tiwari [9] designed an identification algorithm
to identify the balancing masses for balancing the rotor setup at higher speeds by
operating the rotor system at low speed. They utilized influence coefficient method
and virtual trial unbalances to predict the residual unbalance in the rotor system
supported on flexible bearings and integrated with AMB. Kejian et al. [10] designed
an algorithm to identify the orientation of the unbalance with the generation of control
signal based on the orientation of the unbalances in real time. They also discussed
that the orientation and magnitude of unbalances are not required to be calculated as
they are independent of the operating speed. Other researchers utilize different algo-
rithm to predict unbalances in the rotor-AMB system. Tiwari and Chakravarthy [11]
described a methodology to predict the residual unbalances and bearing parameters
in a rotor having multi degree of freedoms. Tiwari and Chougale [12] described an
identification procedure based on developed FEM model to predict the bearing prop-
erties and unbalance existing in the rotor setup supported on the AMBs. They also
performed validation of the developed identification algorithm through experiments.
Xu et al. [13] estimated the stiffness parameters of the magnetic bearings in the rotor
system having residual unbalances.
The control strategies developed to reduce the vibrational response are compli-
cated and require extensive computational effort. Also, the model-based identification
procedure requires a lot of information about the physical properties of the rotating
machinery. The addition of trial unbalance manually in the ICM requires mechanical
effort and time to identify the residual unbalances. The present work utilizes the
simple PID control to control the rotor vibrations and replaces manually added trial
unbalances by generating virtual trial unbalances as magnetic excitations. The ICM
is utilized to predict the residual unbalances based on the vibrational response of the
rotor-test rig and the amplitude and phase of the excitation forces with minimum
information of the rotor-test rig.

2 Implementation of ICM Using Virtual Trial Unbalances

The AMB is implemented in the rotor system to limit the undesirable vibrations and
also to excite virtual trial unbalances as magnetic excitation. The differential driving
mode control is utilized for AMB operation in the two perpendicular axes to the axis
of the shaft. The control current is subtracted and added to the respective pair of poles
of the actuator to generate a magnetic force that keeps the rotor in its equilibrium
position. The magnetic force ( f r ) generated by the AMB is given as
152 G. Ranjan et al.

fr = −ks rm + ki i c (1)

fr = −ks rm + ki (k P rm (t) + k I r m (t) + k D ṙm (t)) (2)

where, k s and k i are the displacement stiffness factor and current stiffness factor of the
magnetic bearing force; ic and r m are the control current and displacement responses
at AMB location respectively. (.) and (−) represents the derivative and integration of
displacement response. The PID control parameters are represented as integral gain
(k I ), derivative gain (k D ) and proportional gain (k P ) in Eq. (2). The AMB is integrated
in the rotor setup as presented in Fig. 1. To estimate the residual unbalances in the
system using ICM, the rotor is operated under its 1st critical frequency. In the first
run, the displacement responses at the disc position with AMB control are obtained
as shown in Fig. 1a. The displacement responses related to influence coefficient and
the unbalance force are given as
 
r1 = α1 α2 f1 (3)
 T  T  T  T
with, r1 = rd rm ; α1 = αdd αmd ; α1 = αdm αmm ; f1 = f d 0
where, f d and r d are the residual unbalance force and vibrational displacement at the
disc position. α is the influence coefficient related to the forces and the corresponding
displacement responses. In the second run, a VTU is generated in the rotor as magnetic
excitation at the AMB location (Fig. 1b). The additional magnetic force ( f T ) is given
as

f T = meω2 ej(ωt+φ) (4)

 
ki i T = meω2 ej(ωt+φ) ; i T = meω2 ej(ωt+φ) /ki (5)

Fig. 1 Unbalance forces and displacement responses in the rotor-AMB system in a 1st run and
b 2nd run with additional excitation force
Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 153

where, e, φ and m are the eccentricity, phase and mass of the trial unbalance generated
as magnetic excitation. ω is the operating frequency of the system. iT is the trial
current added with the control current to generate the control and excitation current
simultaneously as

fr = −ks rm + ki (k P rm (t) + k I r m (t) + k D ṙm (t) − i T ) (6)

Now, the new displacement responses are related to influence coefficient and the
trial and residual excitation forces as
 
r2 = α1 α2 f2 (7)
 T  T
with, r2 = rd rm ; f1 = f d f T .
The displacement responses obtained in Eq. (7) are subtracted from Eq. (3) to
obtain the influence coefficient as
 
αmd = αdm = rd − rd / f T (8)

In order to estimate the influence coefficient as shown in Eq. (8), the value of
current stiffness factor (k i ) is necessary to be estimated. In the present setup as
shown in Fig. 2a, a static load is added at AMB location. The static load ( f p )
displaces the shaft from the equilibrium location, which is then brought back to
the equilibrium location by the generation of static control current (ip ) in the AMB
system as described in Fig. 2b. The relationship between static load and magnetic
force generated to balance the load is given as.

f m, p = f p ; ki i P = f p (9)

ki = f p /i P (10)

Now, from Eq. (3), the residual unbalance is estimated as

Fig. 2 a Displacement of shaft with the addition of static load and b restoring AMB force shifting
the shaft to the equilibrium position
154 G. Ranjan et al.

rm = αmd f d (11)

f d = rm /αmd (12)

The residual unbalance estimated at the disc location is added to the balancing
plane with opposite orientation to verify the reduction in displacement response with
balancing.

3 Estimation of Unbalances with Magnetic Excitation

The rotor-AMB test setup is integrated at the Vibration and Acoustics Lab of IIT
Guwahati to perform the balancing experiment based on ICM with the generation of
virtual trial unbalances. The test rig RK-4 rotor system is supported on a bush bearing
with balancing discs as shown in Fig. 3. The 8-pole AMB actuator is integrated in the
test rig to generate magnetic forces in the flexible shaft of the rotor system. The actu-
ator is manufactured from thin sheets of CRGO silicon steel sheets. The physical
parameters of the test setup are provided in Table 1. The displacement measure-
ment is obtained with the help of Eddy current sensors. The additional components
utilized in the AMB system are 4 amplifiers, DAQ system with SIMULINK plat-
form, Digital to Analog (D/A) and Analog to Digital (A/D) converters. The reference
location is considered at the slot (key phasor) near flexible coupling (motor side) as
shown in Fig. 3b. The motor control unit can be utilized to rotate the system at an
operating frequency of 5 Hz to 167 Hz. In the test rig described in Fig. 3, trans-
lational displacements are measured at disc in right side and AMB location. Since
the experimentation is for rigid rotor case so for symmetrically placed other disc in
left side and no measurement was taken. Even the correction masses were equally
divided at the same phase on both left and right discs to maintain rigid balancing. The
physical connection of the rotor-AMB setup is displayed in Fig. 4. The vibrational
response generated at the AMB location is measured with Eddy current sensors.
The measured signals are then transferred to the A/D panel of the DAQ. The PID
controller designed in the DAQ-SIMULINK platform generates the required control
current to minimize the vibration. The current signal is then transmitted to the ampli-
fier through D/A converters to generate the required amount of control current in the
actuator integrated in the rotor system.

3.1 Control Structure

The PID control structure of the AMB system is developed in DAQ-SIMULINK

platform as presented in Fig. 5. The AMB system is operated in the differential
driving mode control utilizing both control and bias current through PID controller
Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 155

Fig. 3 a A rotor-bearing setup with AMB integrated as controller and exciter, b Reference slot in
rotor system

Table 1 Physical parameters

Properties Values
of the rotor-AMB setup
Length of shaft 0.51 m
Diameter of shaft 0.01 m
Mass of disc 0.8 kg
Actuator leg width 0.0045 m
Actuator stator length 0.0265 m
Actuator stator diameter 0.053 m
Air gap 4 × 10–4 m
AMB current stiffness factor (k i ) 9.16 N/A

Fig. 4 Physical connection of rotor-AMB test setup

156 G. Ranjan et al.

Saturation
ix+iT,x
Vx x ix
Left Pole
Disp/Volts PID controller Volts/Amp
ADD Right Pole DAQ
(Control+Excitation) (D/A)
Vy current iy+iT,y
y iy Upper Pole panel
DAQ
(A/D) Disp/Volts PID controller Volts/Amp
panel Lower Pole

iT,x iT,y

Cosine
wave
Clock Freq
Sine
wave

Fig. 5 A PID controller model designed at DAQ-SIMULINK platform

in The AMB system. The vibration responses at the disc and AMB locations are
obtained by Eddy current sensors as voltage signals. The DAQ (A/D) block as shown
in Fig. 5, then converts the voltage signals to the digital signals. The voltage signals
(V x and V y ) are then converted into displacement signals (x and y) by multiplying
with a sensitivity factor (1/7874 m/V) as provided in the catalog of Eddy current
sensors. The displacement signals in both x (horizontal) and y (vertical) directions
are then sent to the PID controller to generate the control currents (ix and iy ) in the
respective directions. The control currents are then converted into voltage as per
the setting in the amplifier. The voltage signal is then added with the bias voltage
to generate the required amount of bias current along with the control currents in
AMB actuator. The bias voltage is added in such a way that the control current gets
subtracted to the bias current in the direction of the displacement and gets added in
the opposite direction and are supplied to the relevant poles of the electromagnet.
To generate the additional magnetic excitation, an excitation current is generated in
the DAQ-SIMULINK model as depicted in Fig. 5. In the run time of the rotor, the
excitation currents (iT,x and iT,y ) are generated with a frequency equivalent to rotor’s
operating speed. The excitation currents are added to the respective control currents
as shown in Fig. 5 to generate the excitation and control current simultaneously.

3.2 PID Control

The PID control structure is implemented to control the vibrational displacement of

rotor setup within appreciable limit. The derivative gain (k D ) influences the damping
property of rotor setup, the proportional gain (k P ) influences the overall stiffness
of rotor setup, and the integral gain (k I ) keeps the rotor system in its equilibrium
position by removing the steady state error in the response. In order to implement
the differential mode, control a bias current of 0.5 A is added to the control current
of the AMB system and it is ensured that the control current remains in the limit of
the bias currents. The PID gains are changed from k D = 0 to 10 A-s/m and k P =
0 to 10,000 A/m. The test setup was stable for control parameters k P > 3000 A/m
Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 157

and k D > 3 A-s/m. So, for the current problem, the controller gains are considered as
k P = 5000 A/m, k D = 5 A-s/m, and k I = 2000 A/m-s.
To perform rigid rotor balancing of the system, the system is to be operated
below its critical speeds. Therefore, the critical speed of the rotor system is identified
with the help of impulse test with the considered PID gain parameters. First, the
impulse test is performed without AMB control and the displacement response with
the application of impulsive force in time domain is presented in Fig. 6a. The time
domain signal is then converted to frequency domain by FFT as presented in Fig. 6b.
The natural frequency without AMB control is found out to be 25.6 Hz. Now, the
impulse test is performed with control action of AMB and the displacement response
is presented in Fig. 6c, d respectively. The natural frequency with AMB control is
found out to be 40 Hz. The shift in the natural frequencies is seen with the addition
of stiffness and damping property by the PID controller gains. So, the rotor system
is required to be rotated below 40 Hz to perform the present experiment.
To investigate the reduction in amplitudes of displacement responses achieved
with the application of AMB in the rotor system, the setup is operated at 15 Hz i.e.,
below its 1st critical speed. The time domain vibrational responses of the system at
the AMB location are presented in Fig. 7. The amplitude of responses at 1X harmonic
are estimated be 8.32 × 10–5 m without AMB control and 3.14 × 10–5 m with AMB
control, that gives 62.3% reduction. The reduction in vibrational amplitude shows
the effective working of the AMB control structure. Now, the balancing procedure is
discussed to estimate the unbalances present in the experimental test rig with virtual
trial unbalances.

-4
x 10 -4
2 10
Without AMB X: 25.61
X: 96.03 Without AMB
Displacement (m)

Y: 9.521e-05
1 Y: 2.619e-06
Amplitude (m)

-6
10
0

-1 -8
10
-2
0 0.2 0.4 0.6 0.8 1 0 200 400 600 800 1000 1200 1400
Time (s) Frequency (Hz)
(a) (b)
-5
x 10 -4 X: 96
10 Y: 1.69e-05
With AMB With AMB
5 X: 40
Displacement (m)

Y: 1.259e-06
-6
10
Amplitude (m)

0
-8
10
-5
-10
10
0 0.2 0.4 0.6 0.8 1 0 200 400 600 800 1000 1200 1400
Time (s) Frequency (Hz)
(c) (d)

Fig. 6 Impulse test response plot for the rotor system. a, b Displacement versus time plot,
Displacement amplitude (log10 ) versus frequency plot c, without and d with control action of
AMB
158 G. Ranjan et al.

Fig. 7 Displacement (m) versus time (s) plot at 15 Hz operating frequency without and with AMB
control

3.3 Balancing Procedure

The steps involved in the present balancing methodology with magnetic excitation
through AMB are given as
1. To identify the orientation of the unbalance, the overall procedure is performed
as per a reference position on the system as shown in Fig. 3. An Eddy current
probe is used to measure the reference signal as shown in Fig. 8. The balancing
discs (Fig. 3) have holes for the addition of masses at an orientation of 22.5°. To
get a common reference position the zero-degree marking on the disc is made
coincident with the slot orientation.

Fig. 8 The reference signal generated for an operating speed of 15 Hz

Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 159

Fig. 9 Frequency domain plot of rotor-AMB setup without and with the addition of trial current
a, b at 16.7 Hz, and c, d at 15 Hz

2. The phase compensation of the half-spectrum responses is carried out as per the
reference signal shown in Fig. 8. As the system has residual bow in the shaft, the
slow run (5 Hz) responses at measurement locations are required to be removed
vectorially from the 1X response of the setup at higher operating frequency.
3. In the 1st run, the rotor-AMB system is operated at 15 Hz frequency and the
vibrational responses of the rotor at disc and magnetic bearing locations are
measured. The FFT of the displacement signals are done and the displacement
amplitude at 1X harmonics in the half spectrum is used to estimate the unbalances.
4. In the second run, an additional magnetic excitation is generated at the AMB
location from the SIMULINK model presented in Fig. 5. The changed vibrational
responses amplitude and phases are obtained related to the 1X harmonics as
shown in Fig. 9.
5. The current stiffness factor is identified for the present system by the addi-
tion of point load at the AMB location. For the addition of 40 g of
weight at the AMB location, the static current generated is found out to be
0.043 A. The value of the current stiffness factor is estimated as per Eq. (10) and
is found out to be 9.16 N/A.
6. The influence coefficient of the system is calculated as per Eq. (8) from the
displacement obtained in the first and second run and the residual unbalance is
estimated based on influence coefficient and vibration responses at AMB location
in the first run.
160 G. Ranjan et al.

Fig. 10 Displacement versus time plot after addition of balance masses a 4 g and b 5.5 g without
AMB control

4 Results and Discussions

The residual unbalances are identified for different operating frequencies i.e., 15
and 16.7 Hz utilizing different trial currents at the AMB as shown in Table 2. To
perform balancing of the present test rig, a balance mass of 4 gm is added at the
balancing plane at 247.5° excluding AMB control. The vibrational responses are
compared at 15 Hz after balancing of the rotor as presented in Fig. 10. The balancing
is also verified by considering replacing the balance mass of 4 gm with 5 gm. After
balancing, 1X amplitude is estimated to be 4.84 × 10–5 m and 4.00 × 10–5 m for 4
gm and 5.5 gm respectively, with a decrement of 41.89% and 51.78% in vibrational
response, as shown in Table 3. Now, with AMB control, the displacement responses
are compared after balance masses are applied as shown in Fig. 11. After balancing,
1X amplitude is found out to be 3.14 × 10−5 m and 1.83 × 10−5 m and 1.43 × 10−5
m for 4 gm and 5.5 gm respectively, with a decrement of 41.38% and 54.12%.
As shown in Table 3, the displacement responses are reduced by 82.7% after
balancing with AMB control. The developed balancing algorithm (IICM) correctly
predicts balance masses using magnetic excitations and the rotor-AMB system can
be balanced efficiently at low speed with less manual and computational effort.

5 Conclusions

The estimation of the residual unbalances can be performed efficiently by replacing

the manually added trial unbalances in influence coefficient method. The VTUs
generated as additional excitation force through AMB can be implemented along
Single Plane Balancing of Rotor-AMB System Using Virtual Trial … 161

Fig. 11 Displacement versus time plot after addition of balance masses a 4 g and b 5.5 g with
AMB control

Table 2 Estimated balance masses utilizing IICM at different speeds

Spin speed (Hz) Trial current Balance mass
Amplitude Phase Magnitude Phase
(A) (Degree) (kg-m) × 10–4 (Degree)
15.0 0.10 21.34° 1.6456 244.18°
15.0 0.05 139.74° 1.6269 243.84°
16.7 0.10 301.20° 1.6544 246.47°
16.7 0.08 260.42° 1.6624 246.16°
16.7 0.05 267.11° 1.6504 246.42°

Table 3 Percentage reduction in vibrational response after balancing at 15 Hz

Balance mass Displacement
Magnitude Phase Amplitude % Reduction
(grams) (m) × 10–5
0 0 8.32 0
Without AMB 4 247.5 4.83 41.9
5.5 247.5 4.01 51.8
With AMB 0 0 3.14 62.3
4 247.5 1.84 77.9
5.5 247.5 1.44 82.7
162 G. Ranjan et al.

with the control force without affecting the stability of rotor setup. The balancing
performed in the experimental test rig using IICM shows the accuracy and appli-
cability of the algorithm in reducing the vibrations of the rotating machinery. The
method can also be advantageously implemented to the multiplane balancing of rotor
systems with virtual excitations as magnetic force through AMBs.

References

1. Tiwari R. Rotor systems: analysis and identification. Boca Raton: CRC Press; 2017.
2. Sahinkaya MN, Abulrub AHG, Burrows CR. An adaptive multi-objective controller for flexible
rotor and magnetic bearing systems. J Dyn Syst Meas Contr. 2011;133(3): 031003.
3. Jiang K, Zhu C. Multi-frequency periodic vibration suppressing in active magnetic bearing-
rotor systems via response matching in frequency domain. Mech Syst Signal Process.
2011;25(4):1417–29.
4. Hutterer M, Hofer M, Schrödl M. Selfsensing unbalance rejection and reduction of the gyro-
scopic effect for an active magnetic bearing system. 10th Asian control conference (ASCC).
IEEE; 2015. pp. 1–6.
5. Saeed N, Kamel M. Active magnetic bearing-based tuned controller to suppress lateral
vibrations of a nonlinear Jeffcott rotor system. Nonlinear Dyn. 2017;90(1):457–78.
6. Ranjan G, Tiwari R. Application of active magnetic bearings for in situ flexible rotor residual
balancing using a novel generalized influence coefficient method. Inverse Problems Sci Eng.
2018; 1–26.
7. Fang J, Wang Y, Han B, Zheng S. Field balancing of magnetically levitated rotors without trial
weights. Sensors. 2013;13(12):16000–22.
8. Wang Y, Fang J, Zheng S. A field balancing technique based on virtual trial-weights method
for a magnetically levitated flexible rotor. J Eng Gas Turbines Power. 2014;136(9): 092502.
9. Ranjan G, Tiwari R. On-site high-speed balancing of flexible rotor-bearing system using virtual
trial unbalances at slow run. Int J Mech Sci. 2020;105786.
10. Kejian J, Changsheng Z, Liangliang C. Unbalance compensation by recursive seeking unbal-
ance mass position in active magnetic bearing-rotor system. IEEE Trans Industr Electron.
2015;62(9):5655–64.
11. Tiwari R, Chakravarthy V. Simultaneous identification of residual unbalances and bearing
dynamic parameters from impulse responses of rotor–bearing systems. Mech Syst Signal
Process. 2006;20(7):1590–614.
12. Tiwari R, Chougale A. Identification of bearing dynamic parameters and unbalance states
in a flexible rotor system fully levitated on active magnetic bearings. Mechatronics.
2014;24(3):274–86.
13. Xu Y, Zhou J, Lin Z, Jin C. Identification of dynamic parameters of active magnetic bearings
in a flexible rotor system considering residual unbalances. Mechatronics. 2018;49:46–55.
Comparison of Linear and Non-linear
Feedforward Algorithms to Control
Chaotic and Impulsive Noise

Vikas Kumar Lakhmani and Amrita Puri

Abstract Linear feed-forward adaptive algorithms are widely used for creating a
quiet zone inside an enclosure for different noise signals like tonal, multi-tonal, broad-
band, impulsive and chaotic. However, these control strategies do not give optimal
results under the presence of non-linearities. Non-linearities can occur in an active
noise control system due to the following reasons: (a) Non-linear noise like noise
generated due to vacuum cleaner, compressor, mixer and grinder, (b) Non-linear
behavior of actuators and sensors and (c) A very high sound pressure level of noise,
e.g., noise generated during rocket launch. Conventional ANC control strategies, like
FxLMS algorithm, give poor performance and even sometimes fail to control this type
of noise signal. Therefore, mitigation of these non-linear noise signal, in which adap-
tive algorithms give better results for creating a quiet zone inside the enclosure, need
to be identified. In this study, we explore linear and non-linear feedforward control
strategies to control chaotic and impulsive noise. Linear algorithms studied in this
paper are (a) Filtered-x least mean square (FxLMS) and (b) Filtered-x recursive least
square (FxRLS), and non-linear algorithm is (a) Filtered-x back propagation neural
network (FxBPNN). The performance of these algorithms in terms of convergence
time, steady state noise, stability (or robustness) and computational complexity to
reduce chaotic and impulsive noise in a vibro-acoustic cavity is discussed.

Keywords Active noise control (ANC) · FxLMS · FxRLS · FxBPNN · Impulsive

noise · Chaotic noise · Vibro-acoustic cavity

V. K. Lakhmani · A. Puri (B)

Department of Mechanical Engineering, IIT Jodhpur, Jodhpur, Rajasthan, India
e-mail: [email protected]
V. K. Lakhmani
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 163
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_11
164 V. K. Lakhmani and A. Puri

1 Introduction

In modern times the world is becoming technologically smart. No doubt, these tech-
nological advancements bring comfort to human life, but on the other hand, they also
bring so many new challenges which negatively impact our environment. Noise in
acoustics can be defined as an undesired sound, which when perceived by human ears
can cause partial or total loss of the human auditory system. After the industrial revo-
lution, noise-related problems are becoming more prominent, and the revolution led
to an increment in the number of equipment like compressors, blowers, fans etc. Noise
created due to mechanical vibrations also creates unhealthy environment as seen in
the manufacturing industry, transportation and sometimes in household appliances
also. Noise barriers are one of the most common approaches that have been used for
noise control. They are mainly used around highways, airports, and construction sites.
The high construction expenditure and the vast size make it difficult to be applicable
in densely populated urban areas where low-frequency urban noises are problematic.
Another passive approach is noise insulation; here the sound-absorbing material is
used to isolate the receiver from the noise environment. This method achieves good
noise absorption because it is kept close to the receiver and absorbs a large amount
of noise produced. But due to the large size and heavy investment, this method is
also ineffective.
All the above-stated methods are applicable for noise having a high frequency. An
alternate method is used for a noise with low frequency, where the space requirement
is relatively minimal compared to passive noise control techniques (PNC). That
method is called Active Noise Control (ANC) technique.
In many engineering applications, an acoustic region and non-rigid walls are
coupled, and a structural–acoustic system is formed, as shown in Fig. 4. Various
applications like noise control in automobile cabins, office cubicles, aircraft etc.,
are of utmost importance, where a sound knowledge of the interconnection between
structural vibration and noise inside the cavity is required. For the last few decades,
a rectangular cavity has been rigorously investigated and focused on the coupled
systems where active control is the main focus. Among these studies, the modal
coupling theory approach is widespread, where the structural modes in a vacuum and
the acoustic cavity modes with rigid walls are supposed to be evaluated, thereafter,
these two modes are combined with the help of coupling coefficients and the coupled
system response is found out.
It is mandatory to make the ANC system adaptive due to changes in the charac-
teristics of noise with respect to time. So different adaptive learning algorithms are
used in ANC. The basic and most important algorithm utilized in adaptive filters is
the least mean-square (LMS) algorithm. Two microphones are: reference and error
microphones which provide reference signal and residual signal, respectively, as input
to the adaptive algorithm, which adjusts the adaptive filter coefficients to approximate
the acoustic-channel effects. Widrow et al. [1] presented LMS based noise cancelling
adaptive algorithms. In order to improvise the performance of ANC, modifications
Comparison of Linear and Non-linear Feedforward Algorithms … 165

were done in the FxLMS (Filtered-x least mean square) algorithm and all the algo-
rithms that the researchers developed have their advantages and limitations with
respect to the convergence speed, residual noise, computational complexity, stability
etc.
Several studies have been conducted in the field of ANC, where different algo-
rithms are implemented to control various types of noise. In 2010, Ardekani and
Abdulla [2] implemented the FxLMS algorithm to control broadband noise; where
they did the theoretical analysis of the algorithm and assumed the secondary path to
be a moving average process and the input signal to be a stochastic one and this way,
they tried to overcome the drawback of the deterministic input signal. In 2011, Reddy
et al. [3] proposed a hybrid adaptive algorithm that utilized the FxRLS algorithm’s
fast convergence and the low residual error property of the FxNLMS algorithm. The
FxRLS algorithm was used at the initial convergence stage; after that, once it stopped
converging, the given algorithm switched to the Fx normalized LMS algorithm. In
2012, Chan and Chu [4] analyzed the convergence behavior of the FxLMS algo-
rithm; they also performed online modeling of the secondary path. Equations have
been derived which describe the convergence behavior of the adaptive algorithms.
They used the difference equations, analyzed the system’s stability and, at last, solved
the equations at the steady state to obtain the MSE.
In 2011, Xiao [5] proposed a new narrowband ANC system which reduced the
computational cost of the system. In 2013, Chang and Kuo [6] developed a parallel
narrowband ANC system that uses delay-less band-pass filter bank to split the
measured error signal and utilizes individual error signals to update the adaptive
filters. An algorithm was derived with the help of a modified cost function for a
parallel narrowband ANC. In 2014, Wang and Gan [7] performed a stochastic anal-
ysis of an FxLMS algorithm. They developed a mathematical model for the adapta-
tion process of the same system and also derived a step-size for achieving stability
and fastest convergence. In 2019, Aslam et al. [8] proposed a filtered-x optimally
weighted RLS (FxOWRLS) algorithm.
In 2006, Sun et al. [9] developed a new ANC algorithm to control impulsive
noise. Simulations were conducted and it was found that the proposed algorithm
had a better convergence and stability as compared to FXLMS algorithm. In 2016,
Mirza et al. [10] developed a variant of FxRLS i.e., filtered-x state-space recursive
least square (FxSSRLS), a state-space RLS based adaptive algorithm for ANC. They
performed various simulations in order to confirm the improved performance of
the FxSSRLS algorithm. In 2017, Zeb et al. [11] proposed a new algorithm for
impulsive noise, which was based on active noise control, where they did a few
improvisations, which eventually improved the performance of FxRLS algorithm.
Simulations were performed initially and it was observed that the algorithm provides
a better rate of convergence, as compared to FxLMS family algorithms but lacks
strength in the presence of impulsive noise. Two modifications were suggested by
this paper to improve the robustness. They are threshold based and modified gain
FxRLS algorithms, respectively. The modified algorithms showed better stability and
robustness, when compared with FxRLS algorithm.
166 V. K. Lakhmani and A. Puri

In 2004, Ya Li Zhou and Li [12] studied a feed-forward ANC using neural network.
Both FEBPNN and FXBPNN are derived, and then a comparison has been made
among the two algorithms. It has been concluded that FEBPNN has computational
advantage over FXBPNN. Various simulations and experiments were performed
and it was concluded from the results that for non-linear problems, FEBPNN algo-
rithm gives improved performance as compared to the FXLMS algorithm. From
experiments, it can be inferred that more work is required to improve the perfor-
mance of the N.Net controller based on the FEBPNN algorithm. In 2017, Akraminia
et al. [13] developed an ANC scheme, where they used different wavelet frames
with a new structure. Dynamic part of the network utilizes IIR filter with non-
linear static mapping. In order to fit into the network parameters, online dynamic
back propagation algorithms (gradient descent method based) are used. FxLMS
and FxBPNN are used for simulations along with the proposed method. Experi-
ments were performed to assess the performance of the proposed algorithms. Results
obtained from simulations as well as experiments give better results for the developed
algorithms.
It is clear from the literature survey that there is a need to mitigate non-linear
noises like chaotic and impulsive noises. Hence, this study presents a comparison
of two linear and one non-linear feed-forward algorithm to control the non-linear
noises inside a vibro-acoustic cavity. We have generated different series of logistic
chaotic and impulsive noises using MATLAB to create an environment of non-linear
noise (like noise coming from the vacuum cleaner, compressor, mixer and grinder). A
vibro-acoustic cavity has been considered, whose top surface is a vibrating plate, and
all other sides of the cavity are rigid walls. Performance analysis of these adaptive
algorithms is tested under the presence of each type of non-linear noise inside this
cavity. The filtered-x back propagation neural network (FxBPNN) algorithm shows
fast convergence and reduces the computational complexity compared to the FxLMS
algorithm.
This paper is organized as follows: Sect. 2 explains linear algorithms such as
FxLMS and FxRLS and Sect. 3 explains non-linear adaptive algorithm i.e., FxBPNN.
Section 4 presents a numerical study to understand the performance on the basis of
stability (or robustness) and computational complexity of different adaptive algo-
rithms for different noise signals. Section 6 presents conclusions drawn from these
studies.

2 Linear Algorithms

The active noise control (ANC) technique had been stagnating for a long time due to
its complexity in implementation before the adaptive filter was proposed by Widrow
et al. [1]. Despite the fact that many researchers developed a signal-channel ANC
system with analog circuits during this period, it is difficult to convince the public
Comparison of Linear and Non-linear Feedforward Algorithms … 167

that ANC systems are practical and realistic due to its poor noise reduction and
uncertainty [14]. Still, with the continuous development of adaptive filtering theories,
many adaptive control methods were developed and implemented in ANC. In addition
to this, the availability of a robust microprocessor stimulates the use of ANC systems
[15, 16].

2.1 FxLMS Algorithm

FxLMS algorithm is among the most practical adaptive algorithms proposed to

compensate for the influence of the secondary path in an ANC system. In this algo-
rithm, two paths are considered i.e., “Primary path”—which is from the reference
sensor to the output of the error sensor and “Secondary path”- from input of the
secondary speaker (control source) to output of the error sensor. The schematic
diagram of FxLMS algorithm is illustrated in Fig. 1. In the figure, P(z) denotes the
primary path impulse response from the reference microphone to the error micro-
phone; W (z) represents the control filter; S(z) stands for secondary path impulse 

response from the secondary source to the error microphone, and its estimation is S (z)
[17]. The reference signal vector xr (n) is [xr (n), xr (n − 1), . . . . . . , xr (n − N + 1)]T
and N is length of the control filter.
The error signal at nth sample can be written as follows,

e(n) = p(n) + ys (n) (1)

Primary noise, ( )
Noise
Primary Path, P(z)
Source Secondary
noise, ( )
Reference signal,
( )
Adaptive Filter, Secondary Path,
W(z) S(z)

Secondary Path
Model, (z)

Filtered reference
signal, ( ) Residual noise, ( )
LMS

Fig. 1 Schematic diagram of feedforward FxLMS algorithm [Green colour zone denotes the
acoustic field and blue colour zone denotes the digital field]
168 V. K. Lakhmani and A. Puri

Table 1 Pseudocode of the

Steps Pseudocode of FxLMS
FxLMS algorithm
Step: 1 Choose w and μ
Step: 2 For n ≥ 0
ys (n) = s(n) ∗ w(n) ∗ xr (n)
e(n) = p(n) + ys (n)
Step: 3 w(n + 1) = w(n) − μe(n)x f r (n)

Superposition of primary noise, coming from the noise source via primary path,
occurs with the secondary noise at the error sensor in order to obtain a silence zone.
The pseudocode of FxLMS algorithm is given in Table 1.

2.2 FxRLS Algorithm

When dealing with broadband noise, the conventional FxLMS algorithm usually
faces the slow convergence issue. In contrast, the filtered-x recursive least square
(FxRLS) algorithm shows faster convergence, that is not sensitive to the eigenvalue
ratio of the input auto-correlation matrix, and it also achieves a better steady-state
error [18]. The schematic diagram of FxRLS algorithm is illustrated in Fig. 2 and
the weight update equations are given in Table 2.

Fig. 2 Schematic diagram of FxRLS Algorithm

Comparison of Linear and Non-linear Feedforward Algorithms … 169

Table 2 Pseudocode of the

Steps Pseudocode of FxRLS
Fx-RLS algorithm
Step: 1 Choose w(0) = 0
C(0) = δ −1 I
C(n−1)x f r (n)
Step: 2 g(n) =
x Tf r (n)C(n−1)x f r (n)+λ

e(n|n − 1) = p(n) + s(n) ∗ w(n − 1) ∗ xr (n)

Step: 3 C(n) = λ−1 C(n − 1) − λ−1 g(n)x Tf r (n)C(n − 1)
w(n) = w(n − 1) − g(n)e(n|n − 1)

3 Non-linear Algorithm

This section presents non-linear filtered-x back propagation neural network

(FxBPNN) algorithm. It comprises of an input layer, an output layer and a hidden
layer. For adjusting the weights of the hidden layer and output layer gradient descent
algorithm has been used. FxLMS algorithm is extended to form a new algorithm i.e.,
FxBPNN, in which the input layer has multi nodes whereas output layer has only
one single node. [12, 13]
Figure 3 represents the schematic diagram of feedforward ANC system using,
Fx-BP neural network algorithm. In Fig. 3a, primary path is denoted by P(z) and
secondary path is represented by S(z). To minimize the objective function, FxBPNN
has been divided into two steps,

1. Forward Propagation
2. Backward Propagation

Both the parts of NNet are explained below:

1. Forward Propagation:
Step (I): At time n, reference signal xr (n) and the error signal e(n) is introduced,
where reference signal vector is represented by xr (n) =
[xr (n)xr (n − 1) . . . . . . . . . xr (n − I + 1)T ]
Step (II): Compute output of the hidden layer q j (n)
 
q j (n) = f j net j (n) = tanh(net j (n)) (2)
 I −1
where, net j (n) = i=0 ai j xr (n − i)
⎡ ⎤
a00 a01 · · · a0 J −1
⎢ a10 a11 · · · a1J −1 ⎥
A I,J =⎢
⎣ ···
⎥ (3)
··· ai j ··· ⎦
a I −10 a I −11 · · · a I −1J −1

A I,J is the hidden layer weights matrix, in which I denotes the number of inputs,
J represents the total number of hidden layer neurons and q j (n) denotes output of
170 V. K. Lakhmani and A. Puri

(a)

( ) ( ) 0( )

( − 1)
( )

1( )
( )

( )

( − + 1)

−1 ( )

(b)

Fig. 3 a The schematic diagram of ANC system using NNet (Neural Network) controller and
b Structure of NNet controller

the jth hidden layer neuron. In this study, tangent hyperbolic function is used as an
activation function and is denoted by f (.).
Step (III): Now, compute output layer y(n) values at time (n),

J −1
y(n) = b j (n)q j (n) (4)
j=0
Comparison of Linear and Non-linear Feedforward Algorithms … 171

In above Eq. 4. b j (n), represents jth weight in the output layer.

Step (IV): In last step, compute output function of the system ys (n) :

M−1
ys (n) = s j y(n − j ) (5)
j=0

where s j is the impulse response of the secondary path and length of the secondary
path is represented by M.
2. Back Propagation
In order to minimize the MSE, weights of the hidden layer and output layer have
to be updated by using back propagation. Basic gradient descent algorithm is used
to adjust the weights of the hidden and also for output layer. The weight update
algorithm can be given by [12]
M−1
b j (n + 1) = b j (n) − μe(n) si q j (n − i) (6)
i=0

Also, weight updating equation in the hidden layer is [12]:

M−1  
ai j (n + 1) = ai j (n) − μe(n) st b j (n − t) f j  net j (n − t) x(n − i − t) (7)
t=0

where i = 0 . . . . . . . . . I − 1, and j = 0 . . . . . . . . . J − 1.

4 Numerical Study

4.1 Vibro Acoustic Cavity

In this study, as shown in Fig. 4, a vibro-acoustic cavity, whose top surface is a

thin aluminum plate and all other sides of the cavity are rigid wall has been consid-
ered. Based on FEM, a MATLAB code has been written to determine the natural
frequencies and different mode shapes of the vibro-acoustic cavity. Dimensions of
an aluminum plate and cavity are given in Table 3 and Table 4 presents results
obtained from numerical simulation for in-vacuo plate, rigid wall cavity and coupled
plate-cavity.
Active noise control with different linear and non-linear adaptive algorithm in a
vibro-acoustic cavity is studied. Noise at two locations other than the location of
the error microphone is also studied. Figure 4 shows the locations of primary source
(P), location of secondary source (S), location of three microphones, E1, E2, and
E3 inside the vibro-acoustic cavity.
172 V. K. Lakhmani and A. Puri

Fig. 4 a Vibro-acoustic cavity and b Locations of the primary source (P), secondary source (S),
error microphone (E1), and other microphones (E2) and (E3) in the vibro-acoustic cavity

Table 3 Plate and rigid wall cavity dimensions and basic parameters [19]
Parameters Plate values Rigid wall cavity values
Density (ρ) 2.7 × 103 kg/m3 –
Young Modulus 7.102 × 1010 N/m2 –
Air Density (ρ ai r ) – 1.210 kg/m3
Length of plate (L x ) 0.414 m –
Width of plate (L y ) 0.314 m –
Thickness of plate (t) 1 mm –
Length of cavity (L x ) – 0.414 m
Width of cavity (L y ) – 0.314 m
Height of cavity (L z ) – 0.360 m
Speed of sound (in air) (C) – 343 m/s

4.2 Chaotic Noise

The noise signal coming from airfoils, fans etc. are chaotic signal. In this study,
the logistic chaotic noise is generated using MATLAB and utilized for performance
analysis of linear and non-linear algorithms. Logistic chaotic (LC) noise can be
represented as follows [20]:

y(m) = βy(m − 1)[1 − y(m − 1)] (8)

with β = 4 and y(0) = 0.9, m = 1, 2, 3, . . . . . .

It can be seen from the above relation that the next state of the signal depends
upon only on its current state. The signal generated from the Eq. 8 is referred as
logistic chaotic noise of type-1.
Similarly, there are different types of chaotic noise that can be generated where
the next state of the signal depends upon previous state. Therefore, Eq. 8 can be
modified to generate different other types of chaotic signals.
Comparison of Linear and Non-linear Feedforward Algorithms … 173

Table 4 Natural modes and frequencies of in-vacuo plate, rigid-wall cavity and coupled plate-
cavity
In-vacuo plate Rigid-walled cavity Coupled plate-cavity
[0.414 × 0.314 m2 ] [0.414 × 0.314 × 0.360 m3 ] [0.414 × 0.314 × 0.360 m3 ]
Modes Frequency (Hz) Modes Frequency (Hz) Modes Frequency (Hz)
(1,1) 71 (0,0,0) 0 (1,1)–(0,0,0) 80.9
(2,1) 118.9 (2,1)–(0,0,0) 116.9
(1,2) 166.2 (1,2)–(0,0,0) 163.9
(3,1) 199.1 (3,1)–(0,0,0) 197.9
(2,2) 205.5 (2,2)–(0,0,0) 203.4
(3,2) 275.5 (3,2)–(0,0,0) 273.3
(4,1) 310.1 (4,1)–(0,0,0) 307.3
(1,3) 312.1 (1,3)–(0,0,0) 309.5
(2,3) 344.0 (2,3)–(0,0,0) 340.7
(4,2) 377.6 (4,2)–(0,0,0) 375.2
(4,2)–(1,0,0) 416.1
(3,3) 429.2 (1,0,0) 413.3 (3,3)–(0,0,1) 416.6
(5,1) 453.2 (5,1)–(0,0,1) 456.2

1. Modified logistic chaotic noise type-2 is

y(m) = βy(m − 2)[1 − y(m − 2)] (9)

With β = 4 and y(m) = 0.9 for 0 ≤ m ≤ 1

2. Modified logistic chaotic noise type-3 is

y(m) = βy(m − 3)[1 − y(m − 3)] (10)

With β = 4 and y(m) = 0.9 for 0 ≤ m ≤ 2

3. Modified logistic chaotic noise type-4 is

y(m) = βy(m − 4)[1 − y(m − 4)] (11)

With β = 4 and y(m) = 0.9 for 0 ≤ m ≤ 3

4. Modified logistic chaotic noise type-5 is

y(m) = βy(m − 5)[1 − y(m − 5)] (12)

With β = 4 and y(m) = 0.90 for 0 ≤ m ≤ 4

5. Modified logistic chaotic noise type-6 is

y(m) = βy(m − 6)[1 − y(m − 6)] (13)

174 V. K. Lakhmani and A. Puri

Fig. 5 Phase plots of different LC noise (i) LC noise Type-1, (ii) modified LC noise Type-2, (iii)
modified LC noise Type-3, (iv) modified LC noise Type-4, (v) modified LC noise Type-5 and (vi)
modified LC noise Type-6. [20]

With, β = 4 and y(m) = 0.9 for 0 ≤ m ≤ 5

Figure 5 shows different types of chaotic noise generated using MATLAB for
performance analysis of liner and non-linear algorithms. Figures 6 and 7 show that
FxBPNN algorithm is capable to control the chaotic noise when the error micro-
phones are placed at location E1 and E2. Figure 8 shows that FxRLS algorithm
is also capable to control the chaotic noise when the error microphones are placed
at location E1. Simulation results obtained when parametric values are μ = 0.0004
and δ = 100000 for FxBPNN algorithm and FxRLS algorithm, respectively. Results
depicted that for chaotic noise, FxBPNN algorithm provides faster convergence as
compared to FxRLS algorithm.

4.3 Impulsive Noise

In this section, control of impulsive noise is studied. Stable distributions can be

expressed by their characteristic function in the form [11]:

φ(t) = ex p{ j δt − γ |t|α [1 + jβsign(t)ω(t, α)]} (14)

Comparison of Linear and Non-linear Feedforward Algorithms … 175

Fig. 6 Plot of primary noise, secondary noise and error noise at location E1 in the vibro-acoustic
cavity using FxBPNN algorithm for six different chaotic noise signals. (In Figure, Red color:
Primary Noise, Blue color: Control (secondary) Noise and Black color: Error (residual) Noise.)

Fig. 7 Plot of primary noise, secondary noise and error noise at location E2 in the vibro-acoustic
cavity using FxBPNN algorithm for six different chaotic noise signals. (In Figure, Red color:
Primary Noise, Blue color: Control (secondary) Noise and Black color: Error (residual) Noise.)

The above-stated Eq. 14 can be used to demonstrate the impulsive noise by

symmetric alpha stable SαS distribution, [11] when β = 0 and δ = 0, which is
given by:
α
|φ(t)| = e−γ |t| (15)

where α values vary from 0 to 2. α is known as characteristic exponent and it controls

the heaviness of its tail. γ is called as dispersion. When γ is equal to 1, then the
176 V. K. Lakhmani and A. Puri

Fig. 8 Plot of primary noise, secondary noise and error noise at location E1 in the vibro-acoustic
cavity using FxRLS algorithm for six different chaotic noise signals. (In Figure, Red color: Primary
Noise, Blue color: Control (secondary) Noise and Black color: Error (residual) Noise.)

distribution is referred as standard SαS distribution. Figure 9 shows the effect of

variation in α between the specified range.
1. If α ≈ 0 then distribution represents highly impulsive noise [i.e., very heavy
tail].
2. If α = 2 distribution will be gaussian distribution.
3. If α = 1 distribution will be cauchy distribution.
An impulsive noise is generated using MATLAB when α = 1.8 for performance
analysis of the linear algorithms. Probability density function of α stable process
for different values of α has been plotted by using “makedist” and “pdf” function
in MATLAB. The probability density functions of symmetric α stable process for

Fig. 9 a PDFs of α stable process for various values of α [10] and b Impulsive noise in time domain
when α = 1.8
Comparison of Linear and Non-linear Feedforward Algorithms … 177

different values of α and impulsive noise in time domain for α = 1.8 and β =
0, γ = 1 and δ = 0 by using Eq. 14 are shown in Fig. 9. Impulsive noise can also
be generated using the expression mentioned in section 4 of [21].
Simulation results were obtained when parametric values are μ =
0.00001 and δ = 1000 for FxLMS algorithm and FxRLS algorithm, respectively.
On comparison of results obtained from FxLMS (Fig. 11) and FxRLS algorithm
[Fig. 10], it is inferred that both FxRLS and FxLMS algorithms are capable to control
the impulsive noise, but FxRLS algorithm provides faster convergence as compared
to FxLMS algorithm at both error microphone locations E1 and E2. On the other
hand, computational complexity of FxRLS is higher as compared to FxLMS as given
in Table 5, where M represents length of secondary path impulse response and N
represents the number of control filter weights.

Fig. 10 Plot of primary noise, secondary noise and error noise at location E1 and location E2 in
the vibro-acoustic cavity by using FxRLS Algorithm for Impulsive noise when α = 1.8. (In Figure,
Red color: Primary Noise, Blue color: Control (secondary) Noise and Black color: Error (residual)
Noise.)

Fig. 11 Plot of primary noise, secondary noise and error noise at location E1 and location E2 in
the vibro- acoustic cavity using FxLMS algorithm for Impulsive noise when α = 1.8. (In Figure,
Red color: Primary Noise, Blue color: Control (secondary) Noise and Black color: Error (residual)
Noise.)
178 V. K. Lakhmani and A. Puri

Table 5 Comparison of computational complexity of FxLMS and FxRLS algorithms [11]

Algorithms Additions Multiplications
FxLMS 2N + M − 2 2N + M + 1
FxRLS 6N 2 +3N +M − 1 3(N − 1)2 + N 2 + 3N + M

5 Conclusion

In this paper, a comparative study of different linear and non-linear feed-forward

adaptive algorithms has been done. Numerical simulations for different noise signals,
like chaotic and impulsive, for creating the silent zone inside a vibro-acoustic cavity at
different error microphone locations have been performed. In case of chaotic noise,
simulation results show that FxBPNN and FxRLS algorithms both are capable to
control this type of noise signal. Simulation results depicted that for chaotic noise,
FxBPNN algorithm provides faster convergence and better stability as compared
to FxRLS algorithm. In case of impulsive noise, when α = 1.8, it was concluded
that both FxRLS and FxLMS are capable to control the impulsive noise, but FxRLS
algorithm provides faster convergence and better stability as compared to the FxLMS
algorithm at both error microphone locations (E1 and E2) inside the vibro-acoustic
cavity.

References

1. Widrow B, McCool J, Glover J, Zeidler J, Williams C. Adaptive noise cancelling: principle

and application. Proc IEE. 1975;6312:1694–716.
2. Ardekani IT, Abdulla WH. Theoretical convergence analysis of FxLMS algorithm. Signal
Process. 2010;90(12):3046–55. https://doi.org/10.1016/j.sigpro.2010.05.009.
3. Reddy RM, Panahi IMS, Briggs R. Hybrid FxRLS-FxNLMS adaptive algorithm for active noise
control in fMRI application. IEEE Trans Control Syst Technol. 2011;19(2):474–80. https://doi.
org/10.1109/TCST.2010.2042599.
4. Chan SC, Chu Y. Performance analysis and design of FxLMS algorithm in broadband ANC
system with online secondary-path modeling. IEEE Trans Audio Speech Lang Process.
2012;20(3):982–93. https://doi.org/10.1109/TASL.2011.2169789.
5. Xiao Y. A new efficient narrowband active noise control system and its performance analysis.
IEEE Trans Audio Speech Lang Process. 2011;19(7):1865–74. https://doi.org/10.1109/TASL.
2010.2101598.
6. Chang CY, Kuo SM. Complete parallel narrowband active noise control systems. IEEE Trans
Audio Speech Lang Process. 2013;21(9):1979–86. https://doi.org/10.1109/TASL.2013.226
3143.
7. Wang T, Gan WS. Stochastic analysis of FXLMS-based internal model control feedback active
noise control systems. Signal Process. 2014;101:121–33. https://doi.org/10.1016/j.sigpro.2014.
01.025.
8. Aslam MS, Shi P, Lim CC. Weighted least squares approach. 2019;66(10):3955–3967.
9. Sun X, Kuo SM, Meng G. Adaptive algorithm for active control of impulsive noise. J Sound
Vib. 2006;291(1–2):516–22. https://doi.org/10.1016/j.jsv.2005.06.011.
10. Mirza A, Zeb A, Sheikh SA. Robust adaptive algorithm for active control of impulsive noise.
2016;(1). https://doi.org/10.1186/s13634-016-0344-0
Comparison of Linear and Non-linear Feedforward Algorithms … 179

11. Zeb A, Mirza A, Khan QU, Sheikh SA. Improving performance of FxRLS algorithm for active
noise control of impulsive noise. Appl Acoust. 2017;116:364–74. https://doi.org/10.1016/j.apa
coust.2016.10.011.
12. Ya Li Zhou QZZ, Li ZD. Analysis and DSP Implementation of ANC system using a Filter
error neural network. J Sound Vib.
13. Akraminia M, Mahjoob MJ, Niazi AH. Feedforward active noise control using wavelet frames:
Simulation and experimental results. JVCJ Vib Control. 2017;23(4):555–73. https://doi.org/
10.1177/1077546315581939.
14. Hansen CH. Understanding active noise cancellation. SPON Press, 2001.
15. Elliott SJ, Nelson PA. The active control of sound. Electron Commun Eng J. 1990;2(4):127–36.
https://doi.org/10.1049/ecej:19900032.
16. George NV, Panda G. Advances in active noise control: a survey, with emphasis on recent
nonlinear techniques. Signal Process. 2013;932:363–77.
17. Puri A, Modak SV, Gupta K. Study of FxLMS algorithm for active noise control inside a cavity.
J Mech Des. 2015;01(01):9–22.
18. Dongyuan S. Algorithms and Implementations to overcome pratical issues in active noise
control systems. 2020.
19. Fahy F, Gardonio P. Sound and structural vibration. Second Ed., 2007.
20. Behera SB, Das DP, Rout NK. Nonlinear feedback active noise control for broadband chaotic
noise. Appl Soft Comput. 2014;15:80–7. https://doi.org/10.1016/j.asoc.2013.10.025.
21. Chambers JM, Mallows CL, Stuck BW (1976) A method for simulating stable random variables.
J Am Stat Assoc 71(354):340–344. https://www.jstor.org/stable/2285309
Effect of Thermo-Mechanical Coupling
and Large Deformation on the Response
of SMA Structures

Animesh Kundu and Atanu Banerjee

Abstract This exposition presents a thermomechanical analysis of Shape Memory

Alloy structures by taking into account large strain and thermomechanical coupling.
The Total Lagrangian (TL) approach is utilized to implement the thermodynamically
consistent constitutive model of Lagoudas et al. (Int J Plast 16:1309–1343, [1]) in a
non-linear finite element (FE) framework. Using the Newton-Raphson (NR) iterative
approach, the mechanical and thermal equilibrium equations are solved concurrently
while taking into account the coupling factors, i.e. the latent heat of phase transforma-
tion and the thermoelastic heat. Coupled pseudoelastic analysis and thermal recovery
of an SMA plate with a hole are performed to explore the capability of the developed
finite element formulation. A delayed response occurs during transformation as a
result of the introduction of the heat equation with the thermomechanical coupling
factors; thus playing a significant contribution in determining the response of SMA
structures subjected to thermomechanical loading.

Keywords Material Non-linearity · SMA · Thermomechanical coupling ·

Hysteresis · Pseudoelasticity

1 Introduction

Shape Memory Alloys (SMAs) are a distinctive type of intelligent materials that
exhibit two fascinating properties, i.e. Shape Memory Effect (SME) and Pseudoe-
lasticity (PE). In shape memory effect, below martensite finish temperature, SMAs
retain large displacement at the cessation of loading and unloading process, that is
recovered by applying heat into it (shown in Fig. 1). In pseudoelasticity or superelas-
ticity phenomenon, they undergo a large deformation, resulting in hysteritic response

A. Kundu (B) · A. Banerjee

Department of Mechanical Engineering, IIT Guwahati, Assam, India
e-mail: [email protected]
A. Banerjee
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 181
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_12
182 A. Kundu and A. Banerjee

Fig. 1 Schematic
representation of
pseudoelasticity (PE) and
shape memory effect (SME)
in stress-strain-temperature
diagram

at higher temperature (above austenite finish temperature). These phenomena are

explored in actuator applications and in developing vibration damping or shock-
absorbing devices, respectively. Various applications in the field of biomedical,
aerospace and civil structures can be found in Machado and Savi [2], Hartl and
Lagoudas [3] and Song et al. [4], respectively.
An ample amount of research work had been carried out to determine the ther-
momechanical behaviour of shape memory alloy since 1980. Predominantly, these
models can be categorized into two types: (i) Micromechanical models and (ii) Phe-
nomenological model. The former [5, 6] makes use of the microstructure pattern to
predict the material’s macroscopic behaviour, but a significant computational work
is needed to emulate the material response, owing to the model’s complexity. On the
other side, phenomenological models [1, 7–10] consist of constitutive law, that gives
the relationship between stress, strain and temperature, and are related using the inter-
nal variables like martensite volume fraction and transformation strain. This employs
an energy potential considering various material parameters and applies a classical
plasticity-based approach to capture the bulk behaviour. An in-depth thorough study
of the existing constitutive models of SMA can be found in [11–15].
In most of the previous phenomenological models, temperature is considered as
input parameter; but during thermomechanical loading, the temperature evolves as
thermal energy is applied to the system. Moreover, this also depends on the boundary
conditions and internal material parameters. Also, during transformation phase, latent
heat and thermoelastic heat absorb or release as martensite volume fraction changes.
This in turn affects the temperature distribution in the domain. Furthermore, SMAs
undergo large deformation during phase transition from austenite to martensite or
vice versa. So, in order to predict the response of any SMA-based structure effectively,
a coupled thermomechanical analysis considering large deformation is required. To
Effect of Thermo-Mechanical Coupling and Large Deformation … 183

the most of the knowledge of the author, limited research have been reported utilizing
the commercial finite element software [16].
Tabesh et al. [17] developed a 1D analytical model for predicting SMA behaviour
considering latent heat and heat convection effect. The same formulation is also
implemented in 3D FE framework to find the response of SMA strip and antago-
nistic rod systems. In 2007, Thiebaud et al. [18] found the response of SMA plate
under biaxial tension considering partial loops. Yang and Xu [19] considered heat
conservation equation and stress equilibrium equation to simulate the response of
SMA beam by incorporating the constitutive model of Seelecke and Muller [20].
Alipour et al. [21] reported the thermomechanical response of an SMA wire actu-
ator considering Brinson’s 1D constitutive model, under different loading condi-
tions. However, the effect of latent heat was excluded in this study. Solomou et al.
[22] reported the coupled thermomechanical analysis of SMA structures consider-
ing Euler-Bernoulli beam theory and assumed polynomial temperature distribution
along the thickness [23, 24]. Recently, Kundu and Banerjee [25] proposed a FE
tool for finding coupled response of SMA-based components considering non-linear
strain-displacement relations.
This exposition reports a finite element (FE) analysis formulation of an SMA-
based structure by taking into account the effect of finite strain and thermomechani-
cal coupling. Utilizing the Total Lagrangian framework [26], a non-linear finite ele-
ment (FE) approach is used to implement the thermodynamical constitutive model of
Lagoudas et al. [1]. Using the Newton-Raphson (NR) iterative approach, the mechan-
ical and thermal equilibrium equations are solved concurrently while taking coupling
variables into consideration. To introduce a time scale in the response, the inclusion
of the heat equation with coupling effect is inevitable. The developed formulation to
emulate the response under several thermomechanical loading is implemented in a
C++ based finite element library, deal II [27, 28].
The arrangement of the article is the following. In Sect. 2, the constitutive mod-
elling of shape memory alloy is elucidated, subsequently, the FE formulation is
discussed in Sect. 3. Section 4 explicates the implementation of the developed cou-
pled FE model in case of pseudoelastic SMA wire and validated with the analytical
solution. Then, the response of a 2D plane stress problem of an SMA plate with a
hole is explored to understand the coupling phenomenon during transformation.

2 Constitutive Modelling of Shape Memory Alloy

In this study, the constitutive model of Lagoudas et al. [1, 10] is considered. The
thermodynamic potential is represented by Gibbs free energy (.G), expressed by the
second Piola-Kirchhoff stress (.S), temperature (.T ) and internal material variables,
[ ( )]
1 1 T 1
.G(S, T, ξ, ε t ) = − S : S : S − S : [α(T − T0 ) + εt ] + c (T − T0 ) − T ln − s0 T + u 0 + f (ξ ). (1)
2ρ ρ T0 ρ
184 A. Kundu and A. Banerjee

where .ε accounts for the Green-Lagrange strain. .S, .α, .c, .s0 , .u 0 , . f (ξ ) indicate effec-
tive compliance tensor, effective coefficient of thermal expansion tensor and effective
specific heat, effective entropy, effective specific internal energy and hardening func-
tion, respectively. .S is related with .ε, by,
[ ]
S = S −1 : ε − α(T − T0 ) − ε t .
. (2)

Transformation strain evolves with the generation martensite volume fraction using

.ε̇ t = Ʌξ̇ , (3)

where .Ʌ represents the transformation tensor. At any thermomechanical loading

path, the transformation is accounted by the transformation function, .φ; which is
.π − Y for forward transformation and .−π − Y for reverse transformation. .π and
.Y account for thermodynamic force conjugate and internal dissipation energy. For
negative value of .φ, elastic response is observed; otherwise one has to ensure .φ = 0,
by predictor-corrector return mapping algorithm [29].

3 Coupled Finite Element Formulation

In this section, stress equilibrium and thermal energy balance equations are con-
sidered simultaneously, taking into account the effect of coupling terms and large
deformation. From the virtual work principle [30], one can write
∫ ∫ ∫
.
n+1
0S : δ n+1
0 ε dV =
n+1 b
0 F δu dV + n+1 t
0 F δu d┌, (4)
V0
V0 ┌0

where .ε represents Green-Lagrange strain and right-hand side elucidates the work
done caused by the external body force (.n+1 b n+1 t
0 F ) and traction (. 0 F ) at time .tn+1 ,
due to the virtual displacement .δu. Using incremental decomposition of stress and
expressing the incremental stress with the help of strain increment and temperature
increment yields
∫ ∫ ∫ ∫
n n Θ : δ e ΔT d V + n S : δ η d V = n+1 R − n S : δ e dV . (5)
. 0 e : 0 L : δ0 e d V + 0 0 0 0 0 0
V0 V0 V0 V0

By expressing Eq. (5) with regard to nodal DOF increments, one obtains

. K UU ΔU + K U T ΔT = ΔF U . (6)

Here, . K UU and . K U T relate residue in mechanical load vector (.ΔF U ) with the
displacement increment (.ΔU) and temperature increment (.ΔT ), respectively. Con-
sidering the latent heat (. Q L ) and thermoelastic heat (. Q T ), the Heat energy balance
Effect of Thermo-Mechanical Coupling and Large Deformation … 185

equation can be expressed as,

ρC Ṫ = −∇.q + Q s + Q L + Q T .
. (7)

In Eq. (7), .T stands for the temperature at any point in the reference configura-
tion at time, .t. .∇.q illustrates the gradient of heat transfer due to conduction and
. Q s represents thermal energy input per unit volume. .ρ accounts for density and .C

explicates specific heat. . Q L and . Q T can be written as

( )
∂π
.Q L = π− T ξ̇ ; Q T = −α : ṠT . (8)
∂T

Utilizing conventional temporal discretization approach [31], Eq. (7) can be expressed
as ( )
Tn+1 − Tn
. ρC = θ χn+1 + (1 − θ )χn , (9)
Δtn

where .χ = [−∇.q + Q s + Q L + Q T ]. The time step size is denoted by .Δtn =

tn+1 − tn , and .(∗)n accounts for the measured quantity at any time instant .tn . By
integrating in the volume (.V0 ) at the reference configuration after multiplying (Eq. 9)
by the associated weight function (.w), one can get
∫ [ ∫ ∫ ∫ ] ∫
.− ρC(Tn+1 − Tn )w dV0 + θ Δtn − ∇.qw dV0 + Q s w dV0 + (Q L + Q T )w dV0 + (1 − θ )Δtn χn w dV0
n+1
V0 V0 V0 V0 V0
= 0. .
(10)

Further simplification of Eq. (10) and expressing increment in strain and temperature
in terms of displacement increment (.ΔU) and temperature increment (.ΔT ) DOFs
renders
. K T U ΔU + K T T ΔT = ΔF T , (11)

where . K T U and . K T T correlate the residual thermal energy vector (.ΔF T ) with incre-
ment in displacement and temperature, respectively. Finally, accumulating Eqs. (6)
and (11), the system of equation is expressed as
[ ]{ } { }
K UU K U T ΔU ΔF U
. = . (12)
K TU K TT ΔT ΔF T

At each time instant, Eq. (12) has to be solved by employing Newton-Raphson

iterative scheme, until the norm of the RHS becomes insignificant.
186 A. Kundu and A. Banerjee

Fig. 2 a Schematic representation of SMA wire subjected to uniaxial tensile load, b loading history
in pseudoelastic loading above austenite finish temperature

4 Results and Discussion

To demonstrate the potential of the proposed coupled formulation, two sample prob-
lems are simulated. Initially, basic pseudoelasticity test of an SMA wire is performed
by taking into consideration the effect of latent heat and thermoelastic heat. In the
discussion section, if these terms are considered, it is named as coupled response, oth-
erwise termed as uncoupled response. In the subsequent section, the shape memory
effect of a plate with hole is simulated.

4.1 Pseudoelasticity of 1D SMA Wire

In this case, The superelastic response of SMA wire is emulated and compared to
the analytical findings by Tabesh et al. [17]. A uniaxial tensile force is applied to the
SMA wire with a length of 0.2 m and a diameter of 3.8 mm, as shown in Fig. 2a, while
maintaining a constant temperature of 293 K. As shown in Fig. 2b, the applied force
is linearly raised from 0 to 800 MPa before being reduced. The material parameters
used for the study are taken from Tabesh et al. [17]. Adiabatic thermal boundary
condition is taken in to consideration for the analysis. Calculating the strain against
stress variation is the intent.
For this analysis, ten 2-node bar elements are considered. The stress generated at
the right end is plotted against strain for coupled and uncoupled model, as shown in
Fig. 3a. Simulated response is in accordance with the result as described in Tabesh et
al. [17]. The exothermic nature of forward transformation causes temperature to rise
during loading, obstructing austenite from undergoing martensite transformation and
leading to an increase in stiffness. This, in turn, leads to dilatory response during the
phase transformation, as depicted in Fig. 3b. Corresponding temperature variation is
plotted with stress in Fig. 3c. This reveals the accuracy of the developed model.
Effect of Thermo-Mechanical Coupling and Large Deformation … 187

Fig. 3 Comparison between the coupled and uncoupled responses in a stress-strain, b right end
displacement-time and c stress-temperature plot during pseudoelastic loading

Fig. 4 a Schematic diagram of SMA plate with hole subjected to horizontal traction force, b
Applied thermomechanical load

4.2 Thermal Recovery of SMA Plate with Hole

In this section, the proposed formulation is used to simulate the thermomechanical

response of a SMA plate with a hole. It has a dimension of .0.2 m .× .0.2 m, shown
in Fig. 4a. Initially, the material state is complete austenite phase at a temperature
below . As . Adiabatic thermal boundary condition is taken into consideration for the
simulation. Material properties are enlisted in Tabesh et al. [17]. The applied thermo-
mechanical load is shown in Fig. 4b. Firstly, increasing traction of maximum intensity
188 A. Kundu and A. Banerjee

Fig. 5 a Displacement response of point P in x-direction from coupled and uncoupled model, b
Distribution of temperatures as determined by the coupled model at the completion of mechanical
loading

Fig. 6 Martensite volume fraction distribution at the cessation of loading form a uncoupled, b
coupled analysis

100 MPa is applied for 25 sec, and gradually removed afterward. Then, an uniform
heat of 1.8 Wmm.−2 is applied for the next 50 sec, to restore it to its original shape.
By taking in to account the effect symmetry, only a quarter portion of the plate is
simulated. In the simulation, plane stress analysis is considered, and temperature and
displacement are represented by 2D four-node quadrilateral isoparametric elements.
Figure 5a depicts the displacement of point P (shown in Fig. 4a) in the x-direction,
as obtained from the coupled and uncoupled analysis. There is a significant distinc-
tion between the coupled and uncoupled responses. As the forward transformation
begins, temperature increases in the high-stress zone for the coupled model, shown
in Fig. 5b. Higher temperature makes austenite more stable, resulting in stiffening of
the structure and thus produces less displacement as compared to that of the uncou-
pled model. The distribution of the martensite volume fraction following loading
demonstrates this is shown in Fig. 6.
Effect of Thermo-Mechanical Coupling and Large Deformation … 189

5 Summary and Conclusions

This study develops a coupled thermomechanical analysis for shape memory alloys
in a non-linear finite element framework. The impact of latent heat and thermoelastic
heat are explored in this formulation. It is found that the coupled analysis leads to
a slower response, caused by the exothermic or endothermic nature of transforma-
tion. In addition, the effect of large deformation is also investigated by considering
proper stress and strain measures. This formulation can be extended for other prac-
tical applications of SMAs, e.g. stents, origami structures, to envisage the effect of
coupling terms and large deformation in the response.

Acknowledgements The Department of Mechanical Engineering at IIT Guwahati, and project

sponsored by SERB, SERB/CRG/2020/003585, are acknowledged and thanked by the authors for
providing the essential resources.

References

1. Qidwai M, Lagoudas D. On thermomechanics and transformation surfaces of polycrystalline

NiTi shape memory alloy material. Int J Plast. 2000;16(10–11):1309–43.
2. Machado L, Savi M. Medical applications of shape memory alloys. Braz J Med Biol Res.
2003;36(6):683–91.
3. Hartl DJ, Lagoudas DC. Aerospace applications of shape memory alloys. Proc Instit Mech Eng
Part G J Aerosp Eng. 2007;221(4):535–52.
4. Song G, Ma N, Li HN. Applications of shape memory alloys in civil structures. Eng Struct.
2006;28(9):1266–74.
5. Hackl K, Heinen R, Schmahl WW, Hasan M. Experimental verification of a micromechan-
ical model for polycrystalline shape memory alloys in dependence of martensite orientation
distributions. Mater Sci Eng A. 2008;481:347–50.
6. Levitas VI, Ozsoy IB. Micromechanical modeling of stress-induced phase transformations:
part 1 thermodynamics and kinetics of coupled interface propagation and reorientation. Int J
Plast. 2009;25(2):239–280.
7. Tanaka K. A thermomechanical sketch of shape memory effect; one dimensional tensile behav-
ior. Int J Numer Meth Eng. 1986;18:251–63.
8. Liang C, Rogers CA. One-dimensional thermomechanical constitutive relations for shape mem-
ory materials. J Intell Mater Syst Struct. 1990;2:207–34.
9. Brinson LC. One-dimensional constitutive behavior of shape memory alloys: thermomechani-
cal derivation with non-constant material functions and redefined martensite internal variable.
J Intell Mater Syst Struct. 1993;4:229–42.
10. Boyd JG, Lagoudas DC. A thermodynamical constitutive model for shape memory materials.
Int J Plast. 1996;12:805–42.
11. Birman V. Review of mechanics of shape memory alloy structures. Appl Mech Rev.
1997;50(11):629–45.
12. Lagoudas DC, Entchev PB, Popov P, Patoor E, Brinson LC, Gao X. Shape memory alloys, part
ii: modeling of polycrystals. Mech Mater. 2006;38(5–6):430–62.
13. Paiva A, Savi MA. An overview of constitutive models for shape memory alloys. Math Probl
Eng. 2006;2006.
14. Patoor E, Lagoudas DC, Entchev PB, Brinson LC, Gao X. Shape memory alloys, part I: general
properties and modeling of single crystals. Mech Mater. 2006;38(5–6):391–429.
190 A. Kundu and A. Banerjee

15. Khandelwal A, Buravalla V. Models for shape memory alloy behavior: an overview of modeling
approaches. Int J Struct Changes Solids. 2009;1(1):111–48.
16. Abaqus I. ABAQUS theory manual. RI: Providence; 2003.
17. Tabesh M, Lester B, Hartl D, Lagoudas D. Influence of the latent heat of transformation
and thermomechanical coupling on the performance of shape memory alloy actuators. In:
Smart materials, adaptive structures and intelligent systems, vol. 45103. American Society of
Mechanical Engineers; 2012. p. 237–248.
18. Thiebaud F, Collet M, Foltete E, Lexcellent C. Implementation of a multi-axial pseudoe-
lastic model to predict the dynamic behavior of shape memory alloys. Smart Mater Struct.
2007;16(4):935.
19. Yang SB, Xu M. Finite element analysis of 2D SMA beam bending. Acta Mech Sin.
2011;27(5):738.
20. Seelecke S, Muller I. Shape memory alloy actuators in smart structures: modeling and simula-
tion. Appl Mech Rev. 2004;57(1):23–46.
21. Alipour A, Kadkhodaei M, Ghaei A. Finite element simulation of shape memory alloy wires
using a user material subroutine: parametric study on heating rate, conductivity, and heat
convection. J Intell Mater Syst Struct. 2015;26(5):554–72.
22. Solomou AG, Machairas TT, Saravanos DA. A coupled thermomechanical beam finite element
for the simulation of shape memory alloy actuators. J Intell Mater Syst Struct. 2014;25(7):890–
907.
23. Solomou AG, Machairas TT, Saravanos DA, Hartl DJ, Lagoudas DC. A coupled layered ther-
momechanical shape memory alloy beam element with enhanced higher order temperature
field approximations. J Intell Mater Syst Struct. 2016;27(17):2359–84.
24. Solomou AG, Machairas TT, Karakalas AA, Saravanos DA. Co-rotational thermo-
mechanically coupled multi-field framework and finite element for the large displacement anal-
ysis of multi-layered shape memory alloy beam-like structures. Smart Mater Struct. 2017;26(6):
065028.
25. Kundu A, Banerjee A. Coupled thermomechanical modelling of shape memory alloy structures
undergoing large deformation. Int J Mech Sci. 2022;220:107102.
26. Bathe KJ, Ramm E, Wilson EL. Finite element formulations for large deformation dynamic
analysis. Int J Numer Meth Eng. 1975;9(2):353–86.
27. Bangerth W, Hartmann R, Kanschat G. Deal II—a general purpose object oriented finite element
library. ACM Trans Math Softw. 2007;33(4):24/1–24/27.
28. Arndt D, Bangerth W, Clevenger TC, Davydov D, Fehling M, Garcia-Sanchez D et al. The
deal II library, version 9.1. J Numer Math. 2019;27(4):203–213.
29. Qidwai MA, Lagoudas DC. Numerical implementation of a shape memory alloy thermo-
mechanical constitutive model using return mapping algorithms. Int J Numer Meth Eng.
2000;47:1123–68.
30. Bathe KJ. Finite element procedures. Klaus-Jurgen Bathe; 2006.
31. Crank J, Nicolson P. A practical method for numerical evaluation of solutions of partial differ-
ential equations of the heat-conduction type. In: Mathematical proceedings of the Cambridge
philosophical society, vol. 43. Cambridge University Press; 1947. p. 50–67.
Performance Enhancement of Bistable
Energy Harvester with Asymmetric
Potential Function Using an Elastic
Magnifier

K. Devarajan and B. Santhosh

Abstract Bistable piezoelectric energy harvesters are used as an alternative to

batteries in low-power electronic devices, wireless sensors, and wearable devices.
However, their energy harvesting capabilities are limited. Performance enhancement
methods for bistable energy harvesters (BEHs) with symmetric potential well have
been investigated in detail in the literature. Due to the eccentricity of the buckled
beam and repulsive forces due to magnetic interaction, the performance enhance-
ment of BEHs with asymmetric potential function remains an open challenge. In
this paper, the performance enhancement of a BEH with asymmetric potential func-
tion with an elastic magnifier (EM) is proposed to enhance the harvesting capability
under low amplitude external excitation. The effect of stiffness and mass ratio on the
dynamics and energy harvesting performance of the proposed system is investigated
numerically and analytically. The proposed system is compared to asymmetric BEH
and symmetric BEH with an EM, and it is revealed that the proposed system has
a large interwell motion, which leads to greater energy harvesting. To identify the
regions of co-existing solutions, basins of attraction are obtained for the proposed
system.

Keywords Bistable energy harvester · Asymmetric potential function · Elastic

magnifier · Basins of attraction · Harmonic balance method

K. Devarajan (B) · B. Santhosh

Department of Mechanical Engineering, Amrita of School of Engineering, Amrita Vishwa
Vidyapeetham, Coimbatore, India
e-mail: [email protected]
B. Santhosh
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 191
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_13
192 K. Devarajan and B. Santhosh

1 Introduction

Energy harvesters are electro-mechanical devices that harvest and convert ambient
energy into electrical energy. They have undergone rapid developments in recent
years and drawn the attention of many researchers as an alternative to batteries
in low-power electronic devices [1–3]. Ali and Adhikari [4] proposed a vibration
absorber combined with a piezoelectric transduction mechanism for both vibration
confinement and energy harvesting, and the fixed-point theory was used for opti-
mization. The dynamics, vibration absorption, and energy harvesting capabilities of
combined nonlinear vibration absorber with energy harvester were also investigated
[5–7]. The major concern of linear energy harvesters is that they perform well only
at frequencies close to resonance when subjected to harmonic excitation. Due to
broadband snap-through responses, BEHs have been widely used to improve the
energy harvesting performance over linear energy harvester. Erturk and Inman [8]
provided theoretical and experimental investigations of high energy orbits in the non-
resonant bi-stable piezo-magneto-elastic energy harvester over a range of excitation
frequencies. Number of review papers are available which discuss the importance
of nonlinear techniques in performance enhancement of vibration energy harvesters
(VEH) with symmetric potential wells [9–12]. Hu et al. [13] proposed a two degree
of freedom piezoelectric energy harvester with stoppers to improve the performance
of power output. Zhou et al. [14] investigated the dynamic behavior and nonlinear
analysis of a piezoelectric energy harvester with different stoppers in order to deter-
mine optimal impact energy harvesting configurations. Cao et al. [15] proposed an
impact driven piezoelectric VEH for lower resonance frequency and a higher output
power based on a layer separated piezoelectric laminated cantilever beam. Duan et al.
[16] proposed a system composed of piezoelectric beams and mechanical amplifier
to improve the energy conversion efficiency of a VEH.
Wang and Liao [17] investigated the dynamics and the performance enhancement
of BEH with symmetric potential function using an EM. Wang et al. [18] used a con-
figuration called EM to amplify base excitation and provide enough kinetic energy
to harvester, thus leading to large amplitude interwell motion. Galbier and Karami
[19] developed cantilevered-based energy harvesters with a linear EM to increase
power output. Liu et al. [20] focused on improving the capability of the cantilever
beam magnetostrictive harvester through structural configuration, pre-magnetization
condition, and EM experimentally. Zamani et al. [21] obtained larger frequency band
in the energy harvester system of two degree of freedom by adding EM to system.
Liu et al. [22] introduced the bistable structure, and the displacement amplifica-
tion mechanism to amplify the base excitation displacement in the magnetostrictive
bistable vibration harvester. The parameters of the bistable VEH and the displace-
ment amplification mechanism (DAM) are optimized for performance improvement
in a later work by the same authors [23]. It is difficult or almost impossible in prac-
tice to obtain a nonlinear system that exhibits perfectly symmetric potential well due
to various factors. Giri et al. [24] reviewed the dynamics of symmetric and asym-
metric potential well models based on piezoelectric energy harvesters. Kumar et
Performance Enhancement of Bistable Energy Harvester … 193

al. [25] explored the benefits of asymmetric monostable potential function over the
conventional symmetric bistable potential functions in broadband vibration energy
harvesting. Zhou and Zou [26] analyzed nonlinear dynamic analysis of asymmet-
ric tristable energy harvesters for enhanced vibration energy harvesting. Cao et al.
[27] studied time-varying potential energy function of BEH with respect to different
swing angles.
Wang et al. [28] investigated various solutions for BEH with asymmetric poten-
tial function under harmonic excitation, and obtained the basins of attraction map
under certain excitation to conclude that the harvester with asymmetric well has a
higher probability to oscillate in the deeper potential well. Wang et al. [29] studied
the influence of asymmetric potential well on the output and response probability
of BEH under Gaussian white noise excitation. Wang et al. [30] obtained the output
voltage responses of an asymmetric BEH excited by harmonic signals under different
bias angles. Wang et al. [31] explored the response characteristics of the asymmetric
monostable piezoelectric energy harvester with quadratic potential function under
random vibrations. In spite of these works, there have been limited research on
bistable energy harvester with asymmetric potential function, especially the method
for improving the performance of the harvester. Wang et al. [32] investigated the
effect of asymmetric potential well on the dynamics of BEHs and proposed a per-
formance improvement method using bias angle techniques. Santhosh et al. [33]
investigated the nonlinear dynamics and the possibility of enhancing the perfor-
mance of an asymmetric BEH using elastic constraints. Wang et al. [34] studied
the performance enhancement of BEH with asymmetric potential functions under
various excitation using bias angle approach.
Constructing an asymmetric bistable potential well of desired size and shape is
practically difficult and no experimental studies on the asymmetric BEH are reported
in the literature. The performance enhancement of asymmetric BEH with EM is
another method to harvest more energy and this work has not yet been reported in
the literature. This paper attempts to extend the work of [17, 18] to the asymmetric
potential function. This approach is easy to implement and more effective in terms
of energy harvesting than bias angle approach [32]. The energy harvesting perfor-
mance of asymmetric BEH with EM is evaluated numerically and analytically using
harmonic balance method in this work. The results show that by the addition of EM
to the asymmetric BEH, its performance is improved. The rest of the paper is orga-
nized as follows: The mathematical model and equations of motion of asymmetric
BEH with EM are given in Sect. 2. Potential energy analysis is presented in Sect. 3.
In Sects. 4 and 5, the effect of mass and stiffness ratios on the response of the sys-
tem are studied. Performance comparison between asymmetric BEH with EM and
without EM is presented in Sect. 6. In Sect. 7, the comparison between asymmetric
and symmetric BEH with EM is given. Finally, the important conclusion is given in
Sect. 8.
194 K. Devarajan and B. Santhosh

2 Mathematical Model and Equation of Motion

of Asymmetric BEH with EM

The asymmetric BEH with an EM configuration considered in this study is shown in

Fig. 1a. The asymmetric BEH comprises cantilever beam covered with two identical
thin piezoelectric layers (PZT) on both of its surfaces and two permanent magnets,
one of which is attached to the tip of the cantilever beam, and the other one is fixed at
the right wall of the U-shaped block, and its magnetic field orientation is of opposite
polarity to the field of a fixed magnet. Schematic diagram of asymmetric BEH with an
EM is shown in Fig. 1a can be modeled as a two degree of freedom spring mass system
as shown in Fig. 1b, where . Meq , .Ceq , and . K eq are the equivalent mass, damping, and
stiffness of the piezoelectric cantilever beam respectively. The EM consists of a mass,
a damper, and a spring element kept between the asymmetric BEH and the base. The
EM is intended to provide an enhancement for the system acting like a vibration
absorber and . Mb , . K b , and .Cb denote the equivalent mass, stiffness, and damping
of the EM. The .α and .C p represent the electro-mechanical coupling term and the
electrical capacitance of piezoelectric material. .V is the voltage harvested across the
load resistance . R, and . FN is the vertical component of the repulsive force induced
by the two magnets. The whole system is subjected to a given acceleration .ü b (t) on
the base and the output displacement of the mass . Meq and . Mb are written as .x(t) and
. yb (t), respectively. The equation of motion of the proposed system is given by

. Meq ẍ(t) + Ceq (ẋ(t) − ẏb (t)) + K eq (x(t) − yb (t)) − FN (t) − αV (t) = 0 (1)

. Mb ÿb (t) + (C b + Ceq ) ẏb (t) + (K b + K eq )yb (t) − Cb u̇ b (t) − Ceq ẋ(t) − K eq x(t) − K b u b (t) + αV (t) = 0 (2)

V (t)
α(ẋ(t) − ẏb (t)) + C p V̇ (t) +
. =0 (3)
RL

where. FN (t) = K 1 (x(t) − yb (t)) − K 2 (x(t) − yb (t))2 − K 3 (x(t) − yb (t))3 ,. K 1 ,. K 2

and . K 3 are the linear, quadratic and cubic spring constants. Using relative motion,

y (t) = x(t) − yb (t)

. eq (4)

Equations (1)–(3) can be written as follows:

. Meq ÿeq (t) + C eq ẏeq (t) +

2 (t) + K y 3 (t) − αV (t) = −M ÿ (t)
K eq (1 − r )yeq (t) + K 2 yeq (5)
3 eq b b

. Mb ÿb (t) + C b ẏb (t) + K b yb (t) − Ceq ẏeq (t) − K eq yeq (t) + αV (t) = −Cb u̇ b (t) + K b u b (t) (6)

V (t)
α ẏeq (t)) + C p V̇ (t) +
. =0 (7)
RL
Performance Enhancement of Bistable Energy Harvester … 195

The equations of motion are non-dimensional form by considering the following

non-dimensional parameters and variables.
/ /
K1 K eq K eq
. r = , τ = ωeq t, ωeq = , l= , yeq (t) = lYeq (τ ), yb (t) = lYb (τ ), u b (t) = lUb (τ )
K eq Meq K3

K eq lν(τ ) 1 α2
. V (t) = , θ= , κ2 =
α C p R L ωeq C p K eq

Time derivatives are then denoted as follows:

dyeq (t) dYeq (τ ) dyb (t) du b (t) V (t) αωeq ν̇(τ )
. = lωeq ˙ (τ ),
= lωeq Yeq = lωeq Y˙b (τ ), = lωeq U˙b (τ ), =
dt dτ dt dt dt Cp

The non-dimensional equations of motion from Eqs. (5)–(7) using non-dimensional

parameters are given below

Ÿ (τ ) + μẎeq (τ ) + (1 − r )Yeq (τ ) + Yeq

. eq
2
(τ ) + Yeq
3
(τ ) − v(τ ) = −Ÿb (τ ) (8)

rc rk 1 1 1 rc rk
.Ÿb (τ ) + μẎb (τ ) + Yb (τ ) − μẎeq (τ ) − Yeq (τ ) − v(τ ) = μU̇b (τ ) + Ub (τ ) (9)
rm rm rm rm rm rm rm

κ 2 Ẏeq (τ ) + ν̇(τ ) + θ ν(τ ) = 0

. (10)

Ceq
Where .μ = Meq ωeq
, .rm = Mb
Meq
is the mass ratio, .rk = Kb
K eq
is the stiffness ratio and
r =
. c
Cb
is the damping ratio. The state space form of Eqs.(8) to (10) can be expressed
Ceq
as follows.
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ ẋ1 ⎪ ⎪
⎪ x2 ⎪
⎪ ⎪ 0 ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ rk ⎪
⎪ ⎪
⎪ rk ⎪
⎪
⎪
⎨ẋ2 ⎪⎬ ⎪ ⎨−(1 − rm )x2 − (1 − r + rm )x1 − x1 − x1 + (1 + rm )x5 + rm x3 ⎪
1 1 2 3 1
⎬ ⎪ ⎨− rm Ub τ ⎪
⎬
. ẋ 3 = x4 + 0
⎪
⎪ ⎪
⎪ ⎪ ⎪ r ⎪ ⎪
⎪ ⎪ rk ⎪
⎪
⎪
⎪ẋ4 ⎪⎪ ⎪ ⎪ − rmk x3 + r1m μx2 + r1m x1 − r1m x5 ⎪
⎪ ⎪ r Ub τ ⎪
⎪ ⎪
⎩ ⎭ ⎪
⎩ ⎪
⎭ ⎩ m ⎭
ẋ5 2
−κ x2 − θ x5 0
(11)

where the state space variables are .x1 = .Yeq , .x2 = .Ẏeq , .x3 = .Yb , .x4 = .Ẏb , .x5 = .v. The
equations of motion of asymmetric BEH without EM are given by

ẍ (τ ) + μẋ1 (τ ) + (1 − r )x1 (τ ) + x12 (τ ) + x13 (τ ) − v(τ ) = −U¨b (τ )

. 1 (12)

κ 2 ẋ1 (τ ) + ν̇(τ ) + θ ν(τ ) = 0

. (13)

The state space form of Eqs.(12) and (13) is written as follows

196 K. Devarajan and B. Santhosh

Fig. 1 Schematics of a the general structure of proposed system and b two degree of freedom
nonlinear spring mass system of the proposed system

⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨ẋ1 ⎬ ⎨ x2 ⎬ ⎨ 0 ⎬
. ẋ2 = −μx2 − (1 − r )x1 − x12 − x13 + x3 + −Üb (τ ) (14)
⎩ ⎭ ⎩ ⎭ ⎩ ⎭
ẋ3 −κ 2 x2 − θ x3 0

where the state variables are .x1 = .Yeq , .x2 = .Ẏeq , .x3 = .v. The governing equations of
symmetric BEH are given in [17]. The base is excited by .Üb (τ ) = . f cos(Ωτ ), where
.Ω and . f are the amplitude and frequency of base excitation.

3 Potential Energy Analysis

The effective potential and forcing function of the proposed system can be written
as follows:
. F(x) = (1 − r )x 1 + x 1 + x 1
2 3
(15)

1 1 1
. V (x) = (1 − r )x12 + x13 + x14 (16)
2 3 4
The forcing and potential function of proposed system is shown in Fig. 2 for different
values of .r . It shows that when .r > 1, the system undergoes bistable motion and
behaves as monostable motion for .r < 1 when .rm = .rk = 25.
Performance Enhancement of Bistable Energy Harvester … 197

15 10
r = 1.0 r = 1.0
r = 2.5 r = 2.5
10

Potential Energy Fucntion

r = 5.0 r = 5.0
Forcing Fucntion

5
5

0
0
-5

-10
-2 -1 0 1 2 -5
-4 -2 0 2 4
Displacement Displacement
(a) (b)

Fig. 2 Forcing and potential functions of proposed system for .r = 1,2, 3 when .rm = .rk = 25

Fig. 3 Poincare points of

voltage against mass and
stiffness ratio .rm = .rk for the
following parameters . f =
0.05, .Ω = 0.80, .κ 2 = 0.0589,
.θ = 4.1250, .µ = 0.04, and .r =
1.1746

4 Effects of Mass and Stiffness Ratios in Time Domain

In this section, the bifurcation diagram for the voltage of the system with .rm and
r as parameters is obtained. The Poincare points of voltage is plotted along the . y
. k
axis and is shown in Fig. 3. Solving Eq. (11) with the system parameters .µ = 0.04,
.κ = 0.0589, .θ = 4.1250, and .r = 1.1746 and .r m = .r k = [5,17,25], the time domain
2

electro-mechanical response is obtained with different mass and stiffness ratio. The
time response of voltage and the phase portrait of the proposed system configuration
is shown in Figs. 4 and 6 for . f = 0.05 and .Ω = 0.80. The proposed system exhibits
periodic intrawell motion when .rm = .rk = 5 as shown in Fig. 4. When .rm = .rk = 17,
system exhibits chaotic interwell motion which is depicted in Fig. 5. When .rm and
.r k is increased to 25, the oscillator exhibits a large amplitude interwell motion as
shown in Fig. 6.
198 K. Devarajan and B. Santhosh

Non-Dimensional Voltage 0.015 0.3

0.15
0.006

Velocity
0 0
-3
10
-0.006 5
0 -0.15
-5
3740 3750 3760 3770
-0.015
2000 3000 4000 5000 6000 -0.3
-1.5 -1.2 -0.9 -0.6
Time Displacement
(a) (b)

Fig. 4 a Time history b Phase plot with Poincare points for .rm = .rk = 5

0.015 1.5
Non-Dimensional Voltage

0.75
0.006
Velocity

0 0

-0.006
-0.75

-0.015 -1.5
2000 3000 4000 5000 6000 -2.5 -1.5 -0.5 0.5 1.5
Time Displacement
(a) (b)

Fig. 5 a Time history b Phase plot with Poincare points for .rm = .rk = 17

0.015 1.5
0.01
Non-Dimensional Voltage

0
0.75
0.006
-0.01
Velocity

3490 3500 3510 3520 3530

0 0

-0.006
-0.75

-0.015 -1.5
2000 3000 4000 5000 6000 -2 -1 0 1 2
Time Displacement
(a) (b)

Fig. 6 a Time history b Phase plot with Poincare points for .rm = .rk = 25
Performance Enhancement of Bistable Energy Harvester … 199

5 Effects of Mass and Stiffness Ratios in Frequency Domain

Harmonic balance method is used to get solution of the proposed system in frequency
domain. From Eqs. (8)–(9), the reduced displacement .Yb is written as a function of
.Yeq and the voltage .ν as following expression

rm rm + 1 (1 − r )rm + 1 rm 2 rm 3 rm + 1
.Yb (τ ) = Ÿeq (τ ) + μ Ẏeq (τ ) + Ẏeq (τ ) + Y (τ ) + Y (τ ) − ν(τ ) + Ub (τ )
rk rk rk rk eq rk eq rk
(17)
Inserting Eqs. (17) into (9), one can obtain fourth-order nonlinear differential equa-
tion

rm .... rm + 1 ... (1 − r )rm + 1 rm rm

x (τ ) + μ x (τ ) + ẍ + (2 ẋ 2 + 2x ẍ) + (6rm x ẋ 2 + 3rm x 2 ẍ)
rk rk rk rk rk
. (18)
rm + 1
− ν̈ + Üb + ẍ + μẋ + (1 − r )x + x 2 + x 3 − ν = 0
rk

Periodic solution of Eq. (18) is solved by harmonic balance method analytically.

For the most basic response prediction, it consists of writing the first-order periodic
solution. We write the following time signals

Üb (τ ) = f cos(Ωt)
. (19)

. x(τ ) = a0 (τ ) + a1 cos(Ωt) + a2 sin(Ωt) (20)

. ν(τ ) = b0 (τ ) + b1 cos(Ωt) + b2 sin(Ωt) (21)

After substituting Eqs. (19)–(21) into (18), we equalize the constant, .sin(Ωt) and
cos(Ωt) terms to obtain a sixth order in .a nonlinear polynomial expression
.

( )
Ω4 κ 2 r m Ω4 κ 2 2 2 2
4rm Ω4 − 4Ω2 rk − 4Ω2 + 4rm Ω2 r − 4rm Ω2 − 4 −4 + 4rk − 4rk r − 3Ω2 rm2 a + 3r a + 4 rk Ω κ a2
2 2 2 2 k 2 2
Ω +θ Ω +θ Ω +θ
. ( )2 ( )2
(rm + 1) Ω2 − rk μΩ2 + θκ 2 + μθ 2 Ω2 a 2
+16 ( )2 = 16rk2 f 2
Ω2 + θ 2
(22)
2
(Ω2 +θ 2
Substitute.a =. V Ω 2κ4 into Eq. (22) in order to get the frequency response function of
the steady-state voltage of the proposed system. The amplitude and voltage frequency
of the proposed system for different .rm and .rk are shown in Fig. 7 for the following
parameters . f = 0.05, .r = 1.1746, .θ = 4.1250, .κ 2 = 0.0589 and .µ = 0.04. There exist
two resonant peaks in each harmonic response curve since proposed system is two
degree of freedom, and the two resonant peaks not only couple with each other, also
bend to right due to the nonlinear restoring force generated by the system. This leads
to the appearance of the broadband region as shown in Fig. 7. With increasing of
.r m and .r k , the response amplitudes and voltage of proposed system configuration

frequency band of the two peaks increases as shown, whereas symmetric BEH with
EM the frequency band of the two peaks becomes slender [17].
200 K. Devarajan and B. Santhosh

2.5 0.012
rm =k = 5 rm = r k = 5
2
Non-Dimensional Amplitude

rm =k = 10 rm = rk = 10

Non-Dimensional Voltage
2
rm =k = 25 0.009 r = r = 25
1 m k

1.5 r = = 38 rm = rk = 38
m k
0
0 0.5 0.4 0.006
1
0.2
0.003
0.5 0
1 1.5 2

0 0
0 0.5 1 1.5 2 0 1 2 3

(a) (b)

Fig. 7 Response amplitude and voltage as function of excitation frequency at constant amplitude
f = 0.05 with different value of .rm and .rk for the following parameters .r = 1.1746, .θ = 4.1250, .κ 2
.
= 0.0589 and .µ = 0.04

6 Performance Comparison

In this section, the performance comparison of proposed system and asymmetric

BEH without EM is investigated and phase plane is plotted for various excitation
force. At low excitation amplitude . f = 0.05, the proposed system configuration
can easily attain higher amplitude periodic interwell motion, and the asymmetric
BEH without EM configuration exhibits low amplitude periodic intrawell motion
as shown in Fig. 8a and b. The simulation results show that when EM is added to
the asymmetric BEH, the amplitude of the proposed system increases as shown in
Fig. 8c–f. Initial condition maps provide a global understanding of the dynamics by
mapping the regions of initial conditions leading to co-existing solutions of different
nature in nonlinear systems. Out of these, the interwell motions are of large amplitude
and can provide better energy harvesting compared to intrawell motions. For initial
condition selection, the basins of attraction are investigated for the asymmetric BEH
with and without EM in this part for the following parameters . f = 0.05, .µ = 0.04, .r =
1.1746, .θ = 4.1250, .κ 2 = 0.0589, and.rm =.rk = 25. Setting initial voltage .v(0) as 0 and
both initial displacement .x(0) and initial velocity .ẋ(0) were varied from .− 2 to 2, the
basins of attraction for proposed system is obtained by numerical simulation and each
numerical simulation is conducted for 500 cycles in order to identify the final state
of the system. Figure 9 shows the basins of attraction maps of the proposed system
for .Ω = [0.7, 0.8, 0.9, 1.0], respectively. In this study, the black color represents the
oscillation with a final state in the interwell periodic motion, while red color for those
chaotic interwell motion and magenta color represents intrawell periodic motion.
Figure 9 shows that the proposed system has a much greater occurring probabilities to
exhibit periodic interwell oscillation than the asymmetric BEH without EM exhibits
only periodic intrawell motion for . f = 0.05 and .Ω = 0.7, 0.8, 0.9 and 1.0.
Performance Enhancement of Bistable Energy Harvester … 201

Fig. 8 Comparisons of the a voltage time history and b phase plane of proposed system (Blue
color), and asymmetric BEH without EM (Red color) configurations under different excitation
when .Ω = 0.80, .κ 2 = 0.0589, .θ = 4.1250, .µ = 0.04, .rm = .rk = 25 and .r = 1.1746
202 K. Devarajan and B. Santhosh

2 2

1 1
Velocity

Velocity
0 0

-1 -1

-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
Displacement Displacement
(a) (b)

2 2

1 1
Velocity

Velocity

0 0

-1 -1

-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
Displacement Displacement
(c) (d)

Fig. 9 Initial Condition maps of the proposed system for the following parameters . f = 0.05, .µ
= 0.04, .r = 1.1746, .θ = 4.1250, .κ 2 = 0.0589, and .rm = .rk = 25. The black and red scatter points
represent periodic interwell and chaotic interwell responses. a .Ω = 0.70 b .Ω = 0.80 c .Ω = 0.90, d
.Ω = 1.00

7 Comparisons with Symmetric BEH with EM

In this section, the proposed system is compared with symmetric BEH with EM
to show the effectiveness of the proposed system. The parameters selected for this
study are . f = 0.05, .µ = 0.04, .r = 1.1746, .θ = 4.1250, .κ 2 = 0.0589, and .rm = .rk =
25. The phase plane and time series plot for both the systems are shown in Fig. 10.
It is observed that the proposed system outperforms symmetric BEH with EM under
low excitation amplitude . f = 0.05. The proposed system exhibits larger periodic
interwell motion than symmetric BEH with EM under different excitation frequency
are shown in Fig. 10a–f. The numerical simulation of the dimensionless electro-
mechanical equation shows that adding an EM to asymmetric BEHs has a positive
impact on the output response when compared to asymmetric BEHs without EM and
symmetric BEHs with EM.
Performance Enhancement of Bistable Energy Harvester … 203

Fig. 10 Comparisons of the proposed system (Blue color) and symmetric BEH with EM (Magenta
color) configurations when . f = 0.05, .κ 2 = 0.0589, .θ = 4.1250, .µ = 0.04, .rm = .rk = 25, .Ω = 0.70
and .r = 1.1746
204 K. Devarajan and B. Santhosh

8 Conclusion

In this work, the performance enhancement of bistable energy harvester with asym-
metric potential function using an EM is investigated using both analytical and
numerical techniques. A quadratic non-linearity is introduced to the restoring force
to describe the asymmetric BEH and the harmonic balance method is used to find
the steady-state solution of proposed system. The effect of stiffness and mass ratio
on the response of the system is studied in frequency domain. The results reveal
that proposed system has wider frequency bandwidth than symmetric BEH with EM
configuration. From the results, the following conclusions are made: Adding an EM
to a asymmetric BEH configuration is an efficient approach to maintain the oscilla-
tion in higher amplitude and enhance energy harvesting performance of asymmetric
BEH. Basins of attractions maps show that, for the same excitation and initial condi-
tions, the proposed system configuration can be attracted by the high energy periodic
interwell motion whereas asymmetric BEH is attracted by the low energy periodic
intrawell motion. Under low excitation amplitude, the proposed system outperforms
symmetric BEH with EM configurations in terms of energy harvesting capabilities.
This work provides a way to attain optimum performance for BEH with asymmet-
ric potential function, which are frequently presented due to imperfections in the
harvesters.

References

1. Elvin N, Erturk A. Advances in energy harvesting methods. New York: Springer Science; 2013.
2. Erturk A, Inman DJ. Piezoelectric Energy Harvesting. Wiley: John Wiley and Sons Ltd; 2011.
3. Priya S, Inman DJ. Energy harvesting technologies. In: Priya S, Inman DJ, editors. Springer;
2009.
4. Ali SF, Adhikari S. Energy harvesting dynamic vibration absorbers. J Appl Mech.
2013;80(4):4–9.
5. Santhosh B, Das AS. Energy harvesting from nonlinear vibration absorbers. Proc Eng.
2016;144:653–659.
6. Rasil R, Santhosh B. Parametric study and optimization of linear and nonlinear vibration
absorbers combined with piezoelectric energy harvester. Int J Mech Sci. 2019;152:268–79.
7. Rasil R, Santhosh B. A comparative study on the primary system response and energy harvesting
from linear and nonlinear tuned vibration absorbers. Advances in rotor dynamics, control, and
structural health monitoring. Singapore: Springer;2020. p. 429–440.
8. Erturk A, Inman DJ. Broadband piezoelectric power generation on high-energy orbits of the
bistable Duffing oscillator with electro mechanical coupling. J Sound Vib. 2011;330(10):2339–
53.
9. Yildrim T, Ghayesh MH, Li W, Alici G. A review on performance enhancement techniques for
ambient vibration energy harvesters. Renew Sustain Energy Rev. 2017;71:435–49.
10. Daqaq MF, Mansa R, Erturk A, Quinn DD. On the role of nonlinearities in vibratory energy
harvesting: a critical review and discussion. Appl Mech Rev. 2014;66(4):040801.
11. Zou HX, Zhao LC, hua Gao Q, Zuo L, Liu FR, Tan T, et al. Mechanical modulations for enhanc-
ing energy harvesting: principles, methods and applications. Appl Energy. 2019;225:113871.
12. Tran N, Ghyesh MH, Arjomandi M. Ambient vibration energy harvesters: a review on nonlinear
techniques for performance enhancement. Int J Eng Sci. 2018;127:162–85.
Performance Enhancement of Bistable Energy Harvester … 205

13. Hu G, Tang L, Das R, Marzocca P. A two-degree-of-freedom piezoelectric energy harvester

with stoppers for achieving enhanced performance. Int J Mech Sci. 2018;149:500–7.
14. Zhou K, Dai HL, Abdelkefi A, Ni Q. Theoretical modeling and nonlinear analysis of piezo-
electric energy harvesters with different stoppers. Int J Mech Sci. 2020;166:105233.
15. Cao D, Xia W, Hu W. Low-frequency and broadband vibration energy harvester driven
by mechanical impact based on layer-separated piezoelectric beam. Appl Math Mech.
2019;40(1):1777–90.
16. Duan X, Cao D, Li X, Shen Y. Design and dynamic analysis of integrated architecture for
vibration energy harvesting including piezoelectric frame and mechanical amplifier. Appl Math
Mech. 2021;42(6):1–16.
17. Wang GQ, Liao WH. A bistable piezoelectric oscillator with an elastic magnifier for energy
harvesting enhancement. J Intell Mater Syst Struct. 2016;38(3):392–407.
18. Wang G, Liao WH, Yang B, Wang X, Xu W, Li X. Dynamic and energetic characteristics of
a bistable piezoelectric vibration energy harvester with an elastic magnifier. Mech Syst Signal
Process. 2018;105:427–46.
19. Galbier AC, Karmi MA. Using an elastic magnifier to increase power output and performance
of heart-beat harvesters. Smart Mater Struct. 2017;26(9):1–31.
20. Liu H, Li W, Sun X, Cong C, Cao C, Zhao Q. Enhanced the capability of magnetostrictive
ambient vibration harvester through structural configuration, pre-magnetization condition and
elastic magnifier. J Sound Vib. 2021;492:1–22.
21. Zamani MM, Abbasi M, Forouhandeh F. Investigation of output voltage, vibrations and
dynamic characteristic of 2DOF nonlinear functionally graded piezoelectric energy harvester.
Eur Phys J Plus. 2020;135(298):1–18.
22. Liu H, Zhao L, Chang Y, Cong C. Design and characteristic analysis of magnetostrictive
bistable vibration harvester with displacement amplification mechanism. Energy Convers Man-
age. 2021;243:114361.
23. Liu H, Zhao L, Chang Y, Shan G, Gao Y. Parameter optimization of magnetostrictive bistable
vibration harvester with displacement amplifier. Int J Mech Sci. 2022;223:107291.
24. Giri AM, Ali SF, Arockiarajan A. Dynamics of symmetric and asymmetric potential well-based
piezoelectric harvesters: a comprehensive review. J Intell Mater Syst Struct. 2020:1–67.
25. Kumar A, Ali SF, Arockiarajan A. Exploring the benefits of an asymmetric monostable potential
function in broadband vibration energy harvesting. Appl Phys Lett. 2018;112(233901):1–6.
26. Zhou S, Zuo L. Nonlinear dynamic analysis of asymmetric tristable energy harvesters for
enhanced energy harvesting. Commun Nonlinear Sci Numer Simul. 2018;61:271–84.
27. Cao J, Wang W, Zhou S, Inman DJ, Lin J. Nonlinear time-varying potential bistable energy
harvesting from human motion. Appl Phys Lett. 2015;107(14):143904.
28. Wang W, Cao J, Bowen CR, Litak G. Multiple solutions of asymmetric potential bistable energy
harvesters: numerical simulation and experimental validation. Eur Phys J B. 2018;91(254):1–9.
29. Wang W, Cao J, Bowen CR, Zhang Y, Lin J. Probability and output analysis of asymmetric
bistable energy harvesters subjected to Gaussian white noise. Eur Phys J B. 2018;134(558):3–
16.
30. Wang W, Cao J, Bowen CR, Zhang Y, Litak G. Nonlinear response identification of an asym-
metric bistable harvester excited at different bias angles by multiscale entropy and recurrence
plot. J Comput Nonlinear Dyn. 2020;15(091004):1–7.
31. Wang W, Cao J, Wei ZH, Litak G. Stochastic analysis of asymmetric monostable har-
vesters driven by Gaussian white noise with moment differential equations. Eur Phys J Plus.
2021;136(104):1–16.
32. Wang W, Cao J, Bowen CR, Zhang Y, Lin J. Nonlinear dynamics and performance enhancement
of asymmetric potential bistable energy harvesters. Nonlinear Dyn. 2018;94:1183–94.
33. Santhosh B, Vijeykumar S, Vignesh K, Sumanth R. Nonlinear dynamics of asymmetric bistable
energy harvesting systems. In: AIP conference proceedings, vol. 2134(1). AIP Publishing;2019.
p. 1–9.
34. Wang W, Cao J, Bowen CR, Zhang Y, Lin J. Performance enhancement of nonlinear asymmetric
bistable energy harvesting from harmonic, random and human motion excitations. Appl Phys
Lett. 2018;112(1):1–6.
Vibration Reduction in Ambulance Using
Modified Stretcher with Vibration
Absorber

Laxmikant S. Dhamande and Vishal B. Jadhav

Abstract This paper aims to address the effect of vibrations on patient, when patien
is on a stretcher in a moving ambulance. Also, it aims to develop a modified stretcher
which will reduce the harmful vibrations which are transmitted to the stretcher from
suspension system of the current ambulance. Vibration data is collected experimen-
tally from the stretcher and chassis of a moving ambulance for different road condi-
tions. This data is then used for analyzing its effect from the point of view of safety
of the patient. A modified stretcher with a vibration absorber is proposed to reduce
the most harmful vibrations being transferred to the patient. The results show that the
modified stretcher experiences lower amplitude of vibration compared to the original
stretcher, which will avoid any minor injury to the patient and encourage a parametric
for providing emergency treatment, i.e., enable doctors to provide pre-treatment to
patients. The main contribution of the work has been the development of a novel
stretcher along with vibration absorber which provides more comfort to the patients
while escorting for emergencies.

Keywords Vibration absorber · Vibration measurement · Ambulance · Vibration

analysis · Patient comfort

Nomenclature

A Put your nomenclature here

E Modulus of elasticity of the absorber material. (MPa)
I Area moment of inertia of cantilever beam. (mm4)
ka Stiffness of absorber beam. (N/mm)
L Distance of absorber mass from the fixed end. (mm)
ma Mass attached to absorber beam. (Kg)

L. S. Dhamande (B) · V. B. Jadhav

Department of Mechanical Engineering, Sanjivani, College of Engineering, Kopargaon,
Maharashtra, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 207
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_14
208 L. S. Dhamande and V. B. Jadhav

1 Introduction

In the current situation of the COVID-19 pandemic, it is observed that quick and
comfortable ambulance service is important. An ambulance service is used to carry
the patient from the site of accident to the hospital as early as possible. It is not an easy
task to carry a patient comfortably in an ambulance. The patient is having an injury
and in this situation, it is not easy to provide any kind of treatment to the patient in a
moving ambulance hence, it is important to improve the currently available medical
emergency services. Treatment and movement of patient are influenced by time and
distance; while time is influenced by the severity of injury and possible death of
the patient [1]. As the ambulance moves over the uneven road surface, it results in
hazardous shocks and vibrations transmitted through it, which may increase the injury
and may keep the life of patient at risk. Many times the vibration level experienced
by a patient is very high when it is carried in the ambulance. The patient may get
secondary injury due to the vibrations induced in moving ambulance. This makes
it difficult for parametric to provide emergency treatment. It is the most common
practice that to avoid any injury to patient, he is properly tied to the stretcher and
carried to the ambulance. But the vibrations transmitted to the stretcher will not be
reduced by this practice [2, 3].
Alternative ways are tried to avoid the exposure of heavy vibrations to patient
by driving the ambulance slowly and using better roads which may be a long route
or may have minimum potholes, bumps, speed breakers, and other obstacles. These
obstacles will produce severe radially vertical vibrations. The vibration sensitivity
of human body is in the range of 0.1–80 Hz frequency and the ambulance stretcher
is subjected to the same range of vibrations.
Snook [1] presented the problems related to the movement of the injured people
by considering several different steps such as attending patient, escorting patient to
ambulance, carry the patient in the moving ambulance, and moving from ambulance
to hospital room. It is found that research work is going on the response of normal
person to vibrations and response of patient during transportation in ambulance. So to
know the effect of road transmitted vibration, Junhong Zhang et al. [4] have prepared
a mathematical model for describing the dynamics of road transport vehicles. It is
found that optimum reduction of cab vibration can be achieved when the cab–chassis
coupling is high and the carriage–chassis coupling is low. An effort is made to absorb
the vibration transmitted by road to the patients using dynamic vibration absorber.
Ghodge et al. [5] have determined how effectively the vibrations of primary system
at resonance are reduced by using a cantilever type dynamic vibration absorber. The
testing of the designed absorber is carried out for cantilever and simply supported
boundary conditions of beam creating harmonic excitation in motor/ rotor assembly.
The modal and harmonic response analysis of the setup is carried out using ANSYS
and validation of the same is done by preparing a prototype. The vibration analyzer
is used for conducting experimentation. When the vibration absorber was tuned to
operating frequency, the vibrations of the proposed system were reduced signifi-
cantly. Lavnis et al. [2] reviewed the important features and techniques used for
Vibration Reduction in Ambulance Using Modified Stretcher … 209

reducing vibration in ambulance by the use of Mechatronics along with electromag-

nets and conventional suspension system. When two same poles of different magnets
come together then they try to oppose each other, which is called as repulsion, and
by controlling the rate of repulsion the shocks and vibrations can be minimized.
Alexander Voitko et al. [3] have done the maximum acceleration analysis of ambu-
lance using three different shock absorbers for three different bump road conditions
at various speeds. It is found that a suspension with one-tube hydro-pneumatic shock
absorbers provides the most efficient damping of high-acceleration impulses.
A lot of information is available about the vibrations induced due to uneven
road surfaces and the suspension systems used in ambulances. A detailed analysis is
required to know the effect of vibrations on patient health. It is observed that very
less alternatives are available in market which can improve the condition of patients
while transporting them to hospitals in emergency conditions. Whatever products are
available in market, we can fit them into ambulance, but it will create economic issues.
Few patents and research papers are also available to solve the vibration problem in
ambulance. Solutions such as use of suspension system, use of gyroscopic stabilizer,
and use of magneto rheological fluid damper are available. Still a product that will
minimize the effect of vibrations is needed in addition to the available theoretical
solutions given by researchers. According to the available research, it is possible to
develop the mechanism which can be cost effective and provide good performance.
In this paper, an effort is made to develop a system which will isolate the patient
from shocks and vibrations. It aims to study the vibrations affecting patient health
when it is transferred on stretcher in a moving ambulance and to develop a new
system that can reduce the harmful vibrations being transferred to the patient by the
current ambulance suspension system. In this work, a novel thing is that a low cost
solution is developed to absorb the low frequency vibration amplitude. As most of the
ambulances used in India are commercially converted vans, hence such a simplified
absorber will be a better option for such ambulance. This will be achieved by the use
of tuned mass vibration absorber along with multi-reed frequency meter or Frahm’s
tachometer.

2 Methodology

The methodology followed in this work consists of ideas generation based on the
requirement. To understand the need of developing a device or system for current
situation of transportation of patients for hospitalization, a qualitative research is
carried out. The existing solutions of vibration isolation are analyzed using research
papers. Various methods used for vibration control are studied. The existing value of
shocks/vibration on the stretcher of ambulance is measured using FFT analyzer. The
tuned mass vibration absorber along with multi-reed frequency meter or Frahm’s
tachometer is designed. The setup is manufactured as per design. Measurement is
done to know actual readings for various road conditions & then the results are
compared with existing and actual.
210 L. S. Dhamande and V. B. Jadhav

2.1 Measurement of Vibration

To know the level of vibration in the ambulance, vibration measurement is done using
FFT analyzer. Figure 1a shows the ambulance in which measurement of vibration
is done in the existing condition, while Fig. 1b shows the position of accelerom-
eter mounted on the stretcher to measure the amplitude of vibration transmitted to
it. Figure 1c shows the position of accelerometer mounted on floor of ambulance
to measure the amplitude of vibration transmitted to the floor. When the ambu-
lance travels on rough road surface, we have kept accelerometer alternatively on
stretcher and floor of the ambulance. Then vibration displacement, velocity, and
acceleration are measured with the help of FFT analyzer. Above measurements are
repeated for smooth road conditions. This vibration measurement data obtained is
used for the design of vibration isolation system. Figure 2 shows the rough and
smooth road conditions used while collecting vibration measurement data. Table 1
indicates the measurement data collected from ambulance in running condition and
by keeping accelerometer at different locations as specified in column II. Using this
data, the transmissibility ratio i.e., vibrations transferred from floor to the stretcher
is calculated and as given in Table 2.
It is observed that the amplitude of vibration on floor is same as amplitude of
vibration on stretcher for rough road condition, while transmissibility ratio is more
than one for smooth road condition. It shows that it is necessary to isolate the stretcher.

2.2 Design of Vibration Absorber for Stretcher

From the vibration measurement carried out in the ambulance, it is observed that the
following amplitudes and the corresponding excitation frequencies are noted as given
in Table 3. Figure 3a, b shows the velocity spectrum indicating excitation frequencies
and the displacement spectrum indicating excitation frequencies respectively.
To absorb the vibrations at above mentioned excitation frequencies, it is decided to
make use of the principle used in multi-read frequency meter or Frahm’s tachometer
shown in Fig. 4 which consists of a number of plates or strips/reeds with cantilever
boundary condition along with small mass attached at its free ends. Every cantilever
strip has its own natural frequency which is marked on its surface. This instrument
needs to be mounted on the machine surface whose vibrations are to be measured or
natural frequency to be determined. The reed whose natural frequency will match with
the excitation frequency of machine structure, it will start vibrating with maximum
amplitude corresponding to the resonance condition. This concept is utilized for
avoiding resonance at above mentioned (Table 3) excitation frequencies using a
number of cantilever beams.
Vibration Reduction in Ambulance Using Modified Stretcher … 211

Fig. 1 Road excitation measurement of the ambulance: a Ambulance, b Accelerometer mounted

on stretcher, c Accelerometer mounted on floor

Fig. 2 Types of road conditions: a Rough road, b Smooth road

212 L. S. Dhamande and V. B. Jadhav

Table 1 Vibration amplitude in existing ambulance

Sr. no Position of Displacement Velocity Acceleration Avg. value of
accelerometer on Micron mm/s m/s2 displacement
1 Floor- for smooth 29.49 4.7895 1.3598 50.08
road 70.67 6.0156 1.3704
2 Stretcher- for 60.67 5.6066 1.0252 86.84
smooth road 113.04 9.7942 1.5038
3 Floor-for rough 101.97 13.147 2.7852 101.97
road
4 Stretcher-for rough 118.47 9.3936 1.4903 102.29
road 86.11 7.5723 1.8655

Table 2 Transmissibility ratio

Sr. No Condition Transmissibility ratio
1 Smooth road condition 1.734098
2 Rough road condition 1.003152

Table 3 Amplitude and excitation frequencies

Parameter/Condition Smooth road Rough road
Maximum amplitude on floor 30 μ 110 μ
Excitation frequencies 7, 8, 9, 13, 14, 15 7.5, 8, 9.5, 11.5, 13.5, 14.5, 15.5

3 Experimental Details

For the purpose of experimentation, a simple stretcher having a rectangular structure

is chosen as the primary system, as shown in Fig. 5. The absorber is chosen as a
leaf spring or a cantilever beam with dimensions 300 × 25.4 × 1 mm and a slot
provided with a mass which is movable from its position. The absorber can be tuned
to the required operating frequency by changing the stiffness of the absorber. This
will make it possible to match the natural frequency of absorber with the excitation
frequency to be measured. Following steps are followed to achieve the same. (i) the
natural frequency is given by Eq. (1)

√
ka/ma
f = (1)
2π
where, ma is mass of absorber to be considered on random basis to fulfill the require-
ment. Generally, ma is considered to be between 0.05 to 0.25 times the mass of the
primary system.
Vibration Reduction in Ambulance Using Modified Stretcher … 213

Fig. 3 a Velocity spectrum showing excitation frequencies on rough road, b Displacement spectrum
showing excitation frequencies on rough road

(ii) For a cantilever type absorber, the stiffness is given by Eq. (2).

3EI
ka = (2)
L3
(iii) Here E is modulus of elasticity which depends on the material of the absorber
beam and I is area moment of inertia, which depends on its cross section. L is the
length of cantilever beam which needs to be calculated. After calculating L we need
to decide the location of mass corresponding to this length.
First three natural frequencies of absorber are given by Eqs. (3–5), in rad/sec. [5]

3.52a
wn1 = (3)
L2
214 L. S. Dhamande and V. B. Jadhav

Fig. 4 Multi-read frequency

meter

Fig. 5 Stretcher Model for

experimentation

22a
wn2 = (4)
L2
61.7a
wn3 = (5)
L2
√
where a is constant, a = E I /ρ A
Considering mild steel strip having the following properties,
Modulus of Elasticity (E) = 2.1 × 1011 N/m2
Strip dimensions = 250 × 25.4 × 1 mm.
Moment of inertia (I) = (0.0254 × 0.0013 )/12 = 2.1167*10–12 m4 .
Density (ρ)/= 7800 kg/m3
Now, a = (
2.1×1011 ×2.1167×10−12 )
(7800×0.0254×0.001)
= 1.4978
Vibration Reduction in Ambulance Using Modified Stretcher … 215

First natural frequency, wn1 = 3.52×1.4978

0.02542
= 84.35 rad
sec

wn1 = 13.42 Hz

Second natural frequency, wn2 = 22×1.4978

0.2542
= 527.24 rad
sec

wn2 = 83.91H z

Third natural frequency, wn3 = 61.7∗1.4978

0.02542
= 1478.68 rad
sec

Table 4 Dimensions of various strips of absorber

Parameter/ 1 2 3 4 5 6 7
strips No
Width 25.4 25.4 25.4 25.4 25.4 25.4 25.4
(mm)
Thickness 1 1 1 1 1 1 1
(mm)
Length 300 290 280 270 260 250 240
(mm)
C/S Area 2.54E-05 2.54E-05 2.54E-05 2.54E-05 2.54E-05 2.54E-05 2.54E-05
(m2 )
Density 7800 7800 7800 7800 7800 7800 7800
(Kg/m3 )
Youngs 2.10E + 2.10E + 2.10E + 2.10E + 2.10E + 2.10E + 2.10E +
Modulus 11 11 11 11 11 11 11
(N/m2 )
Moment of 2.12E-12 2.12E-12 2.12E-12 2.12E-12 2.12E-12 2.12E-12 2.12E-12
Inertia (m4)
Constant 1.49 1.49 1.49 1.49 1.49 1.49 1.49
(a)
1st Natural 58.58 62.69 67.25 72.32 77.99 84.35 91.53
Frequency
(rad/s)
Hz 9.32 9.97 10.70 11.51 12.41 13.42 14.56
2nd Natural 366.14 391.83 420.31 452.02 487.46 527.24 572.09
Frequency
(rad/s)
Hz 58.27 62.36 66.89 71.94 77.58 83.91 91.05
3rd Natural 1026.86 1098.90 1178.80 1267.73 1367.13 1478.68 1604.48
Frequency
(rad/s)
Hz 163.43 174.89 187.61 201.76 217.58 235.34 255.36
216 L. S. Dhamande and V. B. Jadhav

wn3 = 235.34H z

In our case, we are going to use multi-read frequency meter. The frequency of
external excitation is 14 Hz and the acceptable limit of vibration is 1 to 8 Hz. Hence
we have to design the strips having natural frequencies of 9, 10, 11, 12, 13, and
14 Hz. The dimensions of various strips are shown in Table 4.
After doing the above analysis for the absorber, the length of cantilever beam is
decided to be 250 mm for the weight. This distance is measured from fixed end of
cantilever beam. From Table 4 it is observed that the vibration absorber proposed
above can significantly reduce the vibrations at operating frequency of the primary
system [4].

4 Results and Discussion

The proposed model of modified stretcher is as shown in Fig. 6. This setup is tested
with and without absorber. First, the natural frequencies of primary system are found
using FFT analyzer.
As the ambulance moves over the road surface the vibrations induced due to
uneven road conditions are transmitted to the leg of the stretcher from the wheels
through the chassis. All the vibrations induced in the ambulance cannot be controlled
only by the basic suspension system of the vehicle. When a structure that is externally
excited has undesirable vibrations, these vibrations can be eliminated by coupling a
properly designed auxiliary strip-mass system to the main system, which is known
as a tuned mass vibration absorber. After coupling the absorber to the structure,
the structure, which was excited before attaching the absorber, can now have zero
amplitude of vibration, and the absorber, which is coupled to it, is now vibrating
freely [5].

Fig. 6 Proposed model of

modified stretcher with
vibration absorber
Vibration Reduction in Ambulance Using Modified Stretcher … 217

Table 5 Natural frequencies of stretcher structure (primary system)

Condition Frequency of Displacement Velocity (mm/s) Acceleration (m/
external (microns) s2 )
excitation (Hz)
Without absorber 10 20.6 2.24 1.24
With tuned 8.24 0.89 0.5
vibration absorber

Testing of the experimental model was carried out on a vertical vibration shaker.
The mass corresponding to a typical patient was lumped on the stretcher for the
experiments. Table 5 shows the natural frequencies of the stretcher structure (primary
system) with and without a vibration absorber as found using an FFT analyzer. It was
observed that with absorber attached to the stretcher, there is a reduction in amplitude
of vibration by 150%.

5 Conclusion

The modified stretcher is developed and tested for vibrations it experiences over
it. It is observed that it experiences lower amplitude of vibration compared to the
original stretcher, which will avoid any minor injury to the patient and encourage a
parametric for providing emergency treatment, i.e., enable doctors to provide pre-
treatment to patients. The main contribution of the work has been the development
of novel stretcher along with vibration absorber which provides more comfort to the
patients while escorting for emergencies. It is possible to easily implement it in any
version of an ambulance and reduce the vibrations experienced by the patient in the
ambulance. This will enable doctors to provide pre-treatment and will prove a boon
by saving valuable time.

Acknowledgements This work is carried out in the PG lab and in testing of materials lab of SRES’s
Sanjivani College of Engineering, Kopargaon affiliated to Savitribai Phule Pune University, India.

References

1. Snook R. Transport of the injured patient: past, present and future. Br J Anaesth. 1977;49:651.
2. Lavnis AK, Parmar V and Nadaf M (2019) Innovative vibration dampers in ambulance. J Mech
Robot. 2019;22–25.
3. Alexander V, Victor D, Nikolay P, Aleksandr M. Improving safety of using ambulance vehicles
in large cities. Transp Res Procedia. 2020;50:716–26.
4. Junhong Z, Peng G, Jiewei L, Kainan W. A mathematical model for coupled vibration system
of road vehicle and coupling effect analysis. Appl Math Model. 2016;40:1199–217.
5. Ghodge V, Bhattu AP and Patil SB. Design and validation of a mass tuned dynamic vibration
absorber. Int J Adv Res Ideas Innov Technol. 2018;688–693.
Simulation Studies of Low-Velocity
Impact Damage in FRPS

L. Lokesh Kumar, B. S. Raviteja, S. V. Ashish, A. Arun, A. R. Anilchandra,

and C. M. Manjunatha

Abstract Fibre reinforced plastic (FRP) composites are one of the extensively used
materials in aerospace applications due to their high specific strength and stiffness.
Composites are vulnerable to low-velocity impact (LVI) damages, and they can lead
to premature failure in structural components. In this study, the behaviour of the
composite under LVI loading was modelled and simulated using the finite element
(FE) method. Modelling was carried out as inter- and intra-layer cracking, where
Hashin and Puck criteria were used to determine damage initiation in composites,
and cohesive zone modelling (CZM) was used to predict the delamination/interlayer
damages. A simulation study of the LVI test was carried out on an AS4/8552 graphite
epoxy flat plate composite using ABAQUS® software with a user-defined subrou-
tine VUMAT. Cohesive elements were inserted between the plies to predict the
delamination. The compression-after-impact (CAI) tests were simulated using shell
elements available in ABAQUS. The matrix damage profiles, cohesive damage initi-
ation profiles, force–time, and force–displacement graphs were obtained. The results
obtained were compared with available experimental results for similar material in
the literature, and a good agreement was found.

Keywords Composites · Low-velocity impact · Delamination ·

Compression-after-impact

L. Lokesh Kumar · B. S. Raviteja · S. V. Ashish · A. Arun · A. R. Anilchandra (B)

Mechanical Engineering Department, B.M.S.College of Engineering, Bengaluru, Karnataka, India
e-mail: [email protected]
C. M. Manjunatha
Structural Integrity Division, CSIR- NAL (National Aerospace Laboratories),
Bengaluru 560017, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 219
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_15
220 L. Lokesh Kumar et al.

1 Introduction

Fibre-reinforced polymers (FRPs) are being widely used in various sectors of Engi-
neering, especially in Aerospace as they offer tailor-made properties such as high
specific strength and stiffness, good corrosion resistance, and so on. The recent
Boeing 787 is 80% composite by volume [1] which gives superior fuel efficiency
compared to the conventional materials. Such widespread use of composites requires
a thorough understanding of its failure mechanism to design for safety. Failure mech-
anisms in such advanced materials are difficult to understand and predict while
designing for safety [1]. One of the prominent forms of damage encountered by
composites is the barely visible impact damage (BVID) which is caused due to the
low-velocity impact (LVI) from the ‘small’ objects during maintenance or service of
the aircraft [2].
LVI can be defined as an impact event which occurs in the velocity range of
1–10 m s−1 . An LVI event can occur during service or under maintenance and is
considered as one of the most dangerous types of loads on composite laminates
[2]. During service, an aircraft generally experiences LVI events such as tool-drop
or debris from runways. This results in internal damage (BVID), undetected from
visible observation. Variable service loads, particularly compressive loads, can lead
to continuous growth of the damaged area resulting in complete structural collapse.
Therefore, BVID from LVI is an important design and analysis parameter in lami-
nated FRPs. LVI can cause intra-layer damage such as matrix cracking and fibre
breakage, or interlayer damage (called delamination that occurs mainly between plies
of different orientations), and these failure mechanisms interact in a complicated way
[3].
Dramatic progress in computational technologies over the last few years has made
it possible to design and analyse numerically such complicated failures in composite
structures. The use of computational tools makes analysis of damages such as BVID
economical and optimizes experimentation. According to Shi et al. [4], the impact
can be described accurately using numerical methods which can be developed using
the finite element methods. It also helps in predicting, in a short period of time, the
internal damage mechanisms, when compared to physical testing. A user-defined
damage model in 3D was developed using solid elements and implemented into
the FEM code (Abaqus/Explicit) for predicting the extent and type of damage. The
numerical and experimental results were in good agreement with each other. Perillo
et al. [3] used a finite element (FE) investigation on laminated composites involving
impact using advanced modelling techniques. The analysis and modelling of the
events were carried out using FE commercial software, Abaqus/Explicit and Abaqus/
CAE. A user-defined VUMAT was written in Fortran language. Zhang et al. [5] used
a surface based-cohesive contact model in the ABAQUS software with the objective
of reducing the computation time. Riccio et al. [6] used two modelling strategies: the
first strategy used a refined mesh in the impacted area and coarser mesh in the rest of
the model. In the second strategy, the cohesive zone model (CZM) was adopted to
study the inter-lamina damage using the quadratic nominal stress (QUADS) criterion.
Simulation Studies of Low-Velocity Impact Damage in FRPS 221

Soroush et al. [7] introduced a low-cost, reliable, and simple method of modelling
the dynamic simulation impact on composites based on progressive damage model
(PDM) and CZM techniques, available in ABAQUS without using any subroutines.
They found that the experimental results were more accurate and acceptable than
the numerical work, especially in the case of higher impactor velocity. The literature
review indicates the scope to improve the prediction of BVID from LVI. The use of
user-defined subroutine VUMAT and subsequent modelling using cohesive elements
to predict the delamination is limited in the literature, especially when the BVID
material is to be tested for compression-after-impact (CAI) strength [8] and is the
aim of the present work.

2 Material and Methods

The materials were modelled differently for the dynamic (LVI simulation) and the
quasi-static (CAI simulation) conditions as follows.

2.1 Dynamic Simulation

The material parts modelled in the study are composite lamina, cohesive element, and
impactor (TUP), as shown in Fig. 1. The composite lamina was modelled using an
ABAQUS user-subroutine VUMAT, since ABAQUS CAE does not have the provi-
sion to model 3D laminates. The cohesive elements and TUP were modelled in
ABAQUS CAE.

Fig. 1 Assembled model with the boundary conditions

222 L. Lokesh Kumar et al.

2.1.1 Composite

The composite plies were modelled as rectangular solids of dimension 125 mm ×

75 mm, corresponding to the cutout dimensions as specified in the ASTM D7136
standard. The composite specimen was modelled by assembling unidirectional (UD)
composite plies/lamina of different orientations. Each ply of the composite was
modelled using 3D stress elements as a single layer. The UD fibre plies were modelled
using an orthotropic damage elasticity model, as used by [9]. The stress–strain
relationship for the composite is shown in Eq. (1).
The intralayer damage model was implemented using the Hashin criteria for fibre
and Puck criteria for matrix damage, Eqs. (4–7); a similar approach is adopted by
researchers in [3, 6, 10]. The damage variables dft , dfc , dmt , and dmc , correspond to
the fibre damage due to tension and compression and matrix damage due to tension
and compression, respectively. They are obtained from Eqs. (4–7):
⎛ ⎞ ⎛ ⎞⎛ ⎞
σ11 C11 C12 C13 000 ε11
⎜ σ22 ⎟ ⎜C C C ⎟⎜ ε22 ⎟
⎜ ⎟ ⎜ 12 22 23 000 ⎟⎜ ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ σ33 ⎟ ⎜ C13 C23 C33 000 ⎟⎜ ε33 ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟ (1)
⎜ σ12 ⎟ ⎜ 000 2G 12 0 0 ⎟⎜ ε12 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎝ σ23 ⎠ ⎝ 000 0 2G 23 0 ⎠⎝ ε23 ⎠
σ31 000 0 0 2G 31 ε31
( )( )
df = 1 − 1 − dft 1 − dfC (2)

dm = 1 − (1 − dmt )(1 − dmc ) (3)

( Ʌ )2 ( Ʌ )2
σ 11 τ 12
F tf = +α (4)
XT SL
( Ʌ )2
σ 11
F cf = (5)
Xc
( Ʌ )2 ( Ʌ )2
σ 22 τ 12
Fmt = + (6)
YT SL
( Ʌ )2 [( )2 ] Ʌ ( Ʌ )2
σ 22 YC σ 22 τ 12
Fmc = + −1 + (7)
2S T 2S T Yc SL

XT , XC —the longitudinal tensile and compressive strength;

YT , YC —transverse tensile and compressive strength;
SL , ST —longitudinal and transverse shear strength;
α—a coefficient used to determine the contribution of shear stress to the fibre
tension; and
Simulation Studies of Low-Velocity Impact Damage in FRPS 223

Table 1 Hexply AS4/8552 properties from [8]

Properties Value
Density [kg/m3 ] 1590
Elastic properties [GPa] E1 = 128.0; E2 = 7.63; E3 * = 7.63; G12 = G13 * = G23 * = 4.36;
Poisson’s ratio ν12 = 0.35; ν13 * = 0.35; ν23 = 0.45
Strength [MPa] XT = 2300; XC = 1531; YT = ZT = 26; YC = ZC * = 199.8; S12 =
S13 * = S23 * = 78.4
* Assumed as in [4]

Ʌ Ʌ Ʌ

σ 11 , σ 22 , τ 12 —the effective stress tensor.

Ʌ

σ is used to evaluate the initiation criteria and is computed from

Ʌ

σ = Mσ (8)

σ is the true stress, and M is the damage operator given by

⎡ 1
⎤
0 0
⎢ (1−d f ) ⎥
M=⎣ 0 1
(1−dm )
0 ⎦ (9)
1
0 0 (1−ds )

The internal damage variables df , dm , and ds are used to characterize fibre, matrix,
and shear damage, and are derived from variables, dtf , dcf , dtm , and dcm , corresponding
to the four modes previously discussed, as follows:
{ Ʌ

d tf i f σ 11 ≥ 0,
df = Ʌ

d cf i f σ 11 < 0,
{ Ʌ

dmt i f σ 22 ≥ 0,
dm = Ʌ

dmc i f σ 22 < 0,
( )( )( )( )
ds = 1 − 1 − d tf 1 − d cf 1 − dmt 1 − dmc (10)

The carbon-epoxy (AS4/8552) UD prepreg was used as the material of study. Its
properties are listed in Table 1.

2.1.2 Cohesive Element

In order to capture the delamination pattern and area, cohesive layers of very fine
thickness (h = 0.01 mm) were placed between the plies. These cohesive elements
were positioned between layers having different orientations since the delamination
is observed to occur at the interface of plies. Modelling interlayer damage using the
224 L. Lokesh Kumar et al.

Table 2 Cohesive element properties from [8]

Properties Value
Density [kg/m3 ] 1590
Elastic properties [GPa] in three directions Knn = 7.63; Kss* = Ktt* = 4.36
Strength [MPa] in three directions tn = 26.0; ts * = tt * = 78.4
SERR [kJ/m2 ] in three modes GIC = 0.28; GIIC = GIIIC = 0.79
* Assumed as in [8]

cohesive zone model (CZM) has been carried out in [3, 5, 8]. CZM is based on the
existence of a cohesive damage zone in the vicinity of the crack front: connecting
the displacement jumps at the interface where the crack might occur to tractions.
The interfacial strength of the material is connected to the damage initiation, and the
critical strain energy release rates (SERR) are related to the damage evolution. Both
the crack initiation and crack propagation can be accommodated in the same model
which renders this theory quite popular. Further, the complexity of the numerical
modelling is reduced by CZM. The main disadvantage with CZM is the requirement
of knowing the crack path in advance. This is made possible using the cohesive law
which introduces discrete interface elements in the FE mesh. Further in the case of
composite materials, since delamination is mostly between the differently oriented
plies, the crack path definition is not a problem.
ABAQUS has several damage initiation criteria for cohesive elements. In this
study, QUADS defined by a traction separation law [3, 5, 8] was used. After the
damage initiation criteria are satisfied, the material response changes according to
the selected damage evolution law. Damage is assumed to initiate when Eq. 11 is
satisfied [11]. The properties of the cohesive elements adopted from [8] are presented
in Table 2:

{ }2 { }2 { }2
⟨tn ⟩ ts tt
+ 0 + 0 =1 (11)
tn0 ts tt

where to n , to s , and to t are the peak nominal stress values.

2.1.3 Impactor

The impactor was modelled using a planar shell formulation in a hemispherical shape
of diameter d = 16 mm, considering it as a rigid element in ABAQUS: such elements
help in reducing computation time [3]. The impactor deformation is relatively small
compared to the composite specimen. The impactor was assigned a mass of 5 kg (M)
and loaded with selected velocities to achieve the required impact energies.
Simulation Studies of Low-Velocity Impact Damage in FRPS 225

2.1.4 Contact Formulation

The following contact formulation was carried out during the modelling exercise:
(i) Between the impactor or TUP and the first layer of the composite and,
(ii) the successive layers after the complete failure of the cohesive element.
A general contact definition in ABAQUS is used to formulate these contacts. The
interaction between the impactor and the first layer of the composite has been consid-
ered to have a friction coefficient of 0.3 [8]. The layer interactions are considered
to be frictionless. The cohesive elements are bound to the composite plies by using
tie constraints. The tie constraint fuses the cohesive surface and the composite ply
surface, and also constrains the relative motion between the surfaces [8].

2.1.5 Assembling the Model and Boundary Conditions

The individual plies are stacked upon one another with a cohesive element in between
every two plies. The obtained composite specimen is pinned on all the edges (Ux =
Uy = Uz = 0) (see Fig. 1).
The impactor is constrained only to move normally to the plate (here, the Z axis).
The impactor is assigned a predefined velocity, corresponding to the impactor mass M
= 5 kg, to achieve the various impact energies. The velocities and the corresponding
impact energies are summarized in Table 3.
Meshing
The meshing strategy used in this study was a finer mesh towards the point of impact
and a relatively coarser mesh elsewhere. For this purpose, each ply and cohesive
element layer were partitioned; for further details, please refer to [3–6] (see Fig. 2).
Composite
The composite plies were meshed using 3D hexahedral solid elements with 8 nodes
and reduced integration (C3D8R). The reduced integration reduces the number of

Table 3 Simulation
Impactor velocity Impact energy Corresponding
performed for various
[ms−1 ] [J] drop height [mm]
velocities of impact under
constant mass of impactor (M 4 40 815.5
= 5 kg) 3.46 30 611.6
3.163 25 509.7
2.83 20 407.7
2.45 15 305.8
2 10 203.9
1.414 5 101.9
226 L. Lokesh Kumar et al.

Fig. 2 Meshed assembly

integration points, thereby reducing the computational effort and storage. Also, the
stress and strains in the case of elements with reduced integration are calculated at
points having optimal accuracy.
Cohesive Elements
The cohesive elements were meshed using 3D hexahedral cohesive elements
(COH3D8). The meshing size and strategy were the same as those used for composite
elements. By meshing both the elements with the same size, we can have coincident
meshes, which are stable and take less computation time.
Impactor
The impactor was meshed using 4-node 3D bilinear rigid quadrilateral elements
(R3D4).

2.1.6 Simulation of Impact Event

The simulation of impact tests was modelled and analysed in ABAQUS CAE and
ABAQUS Explicit, respectively. ABAQUS Explicit uses solvers to analyse dynamic
problems, in contrast to ABAQUS Standard, which is used for static problems. Equa-
tions were solved for each time step in ABAQUS Standard while in Explicit, the
solution for a particular time step is solved based on the previous step. The analysis
was solved for a total time period of 10 ms, which is the average time that an impact
event lasts. As presented before, the simulation was analysed for different impact
energies, and the results were compiled.
Simulation Studies of Low-Velocity Impact Damage in FRPS 227

2.2 Static Simulation—Material and Modelling

For conducting the CAI simulation, the test was modelled using shell elements to
simplify the problem. A laminate part of the same dimensions and properties as in
Sect. 2.2 were designed.

2.2.1 Composite

The composite laminate was defined using the composite layup manager in ABAQUS
CAE. The layup consists of 8 layers with a dimension of 125 mm × 75 mm.
The Hashin damage model for the plane stress case was used to determine the
damage profiles. The initiation criteria had the following general forms: Fibre tension
Ʌ Ʌ Ʌ

(σ 11 ≥ 0) in Eq. (4), fibre compression (σ 11 < 0) in Eq. (5), matrix tension (σ 22 ≥

Ʌ

0) in Eq. (6), and matrix compression (σ 22 < 0) in Eq. (7).

Ʌ
Prior to the damage initiation and evolution, M is equal to the identity matrix
(σ =σ). Once damage initiates for at least one mode, the damage operator becomes Ʌ

significant in the criteria for damage initiation of other modes. The effective stress, σ ,
represents the stress acting over the damaged area that effectively resists the internal
forces [11].

2.2.2 Assembling the Model and the Boundary Conditions

In the CAI test, the assembly consists of only a single part, the composite sheet. The
composite sheet is fixed on one of the shorter edges (Ux = Uy = Uz = 0), and the
opposite shorter edge is given a displacement boundary condition, i.e., simulating
compression. The face of the composite sheet is restrained from moving out of plane,
that is, Uz = 0, to prevent buckling (see Fig. 3). The composite sheet is also given a
predefined boundary field pertaining to the damage that occurred during the impact
test by simulating the impact in a different study and importing the damage results
to a new CAI simulation.
Meshing
A similar meshing strategy as used in dynamic modelling is used—fine mesh towards
the centre and coarser mesh elsewhere.

a b

Fig. 3 a Boundary conditions and b Meshing of the composite using the shell element
228 L. Lokesh Kumar et al.

Composite
The composite sheet was meshed using 4-node shell elements (S4R: A 4-node doubly
curved thin or thick shell) with reduced integration. The element size in the finer mesh
was 1.5 mm and 4 mm in the coarser mesh region. (See Fig. 3.)

2.2.3 Simulation of Compression After Impact (CAI) Test

CAI tests are conducted to assess the residual compressive strength of the composite
sheet after the impact. Prior to the CAI simulation, the modelled composite was
subjected to an impact test and the damaged boundary condition was applied.

3 Results and Discussion

The numerical results obtained are presented and discussed in this section. To estab-
lish confidence in our design and simulation procedure, the results from this study are
compared with standard literature [3, 12]. The recorded similarities are mentioned
in the forthcoming sections.

3.1 Results of Low-Velocity Impact (LVI) Test

A force–time graph is a typical output of any impact test. Figure 4 shows the
representative force versus time graphs for a laminated composite plate.
There are two different stages during the impact event (see Fig. 4a, b). Initially,
when the threshold value Fd of the impact force is reached, an unstable crack prop-
agation takes place which leads to an instantaneous large delaminated area. The
impact force often drops due to this, which indicates a sudden loss of stiffness. Small

Fig. 4 a Representative force of impact versus time history (ASTM D7136) [8]; b Representative
experimental impact force versus time history showing initial damage load and peak load [9]
Simulation Studies of Low-Velocity Impact Damage in FRPS 229

delaminations and sub-critical micro cracks may initiate as lower forces. Therefore,
F d does not physically represent the initiation of damage. Rather, it indicates the
initial value at which damage can be detected using non-destructive inspection tech-
niques, and a change in the stiffness properties of the laminate could occur. F d and
the peak force, F max , are indicated in Fig. 4 and depend on the impact velocity.
The damage profiles obtained by simulation studies from the present work tally
well with the experimental results of Gonzalez [12], as seen in Fig. 5. However,
the total impact time is longer than the experimental results. Further, the damage
thresholds F d are underpredicted, hence the predictions are conservative. Moreover,
it can be seen that the values of F d change with changing impact velocity. The
underpredicted values can be attributed to (i) simplification of the model from the real-
life conditions, (ii) discrepancy in material properties, and (iii) excessive stiffness of
the contact constraints defined between the surfaces of the composite plies [12]. The
pressure produced by the impactor can make these surfaces remain in contact without
losing the modelled elements, yielding a stiffer response. Future simulations with
surface-based cohesive interactions need to be performed to check the F d values.
The displacement versus time and velocity versus time graphs are shown in Fig. 6a,
b. As the impactor moves down, the displacement increases in the negative axis.
The impactor moves down until almost all kinetic energy (KE) is absorbed by the
composite specimen. After all the energy from the impact is absorbed, the impactor
rebounds as the energy stored in the plate (Elastic energy) is transferred back to
the impactor. This causes the displacement to increase in the positive direction as
seen in Fig. 6a. In the velocity versus time graph, the velocity of the impactor keeps
decreasing as the KE is absorbed by the specimen. At the point where the velocity
is minimum, the displacement of the impactor is maximum. The rebounding of the
impactor imparts velocity to it, as seen in Fig. 6b. The results from this work tally well
with overall behaviour from experimental results from the literature [3], as shown in
Fig. 6c, d.

3.2 Intralayer Damage—Matrix Cracking

The intralayer damage in the composite specimen occurs predominantly due to matrix
cracking. Figure 7 highlights the envelopes of matrix cracking for different energies
of impact at the end of the simulation, with black areas representing elements that are
damaged and grey (or) white areas representing pristine elements. The matrix damage
is initiated from the bottom layer (opposite to the impact surface), and propagates
to the other layers during the event of impact. In each layer, the matrix damage is
oriented towards the direction of the fibre.
Matrix cracking is the predominant damage mechanism in composite laminates.
Matrix cracking in itself does not generally result in the failure of a laminate. Nonethe-
less, the concomitant effect of matrix cracking is moisture absorption, reduction
in matrix-dominated stiffness, and crack propagation into the interface between
adjacent layers, initiating delamination [13].
230 L. Lokesh Kumar et al.

Fig. 5 Force versus time history at different impact energies of the present work (a, c, e) compared
with Gonzalez [12] (b, d, f)

3.3 Interlayer Damage—Delamination

Delamination is the separation between the lamina and is frequently caused by an

external impact from an object on the laminate. The damage in the cohesive layer
between the lamina (or delamination) results in the classical peanut shape, as shown in
Fig. 8. Figure 9 reports the delamination pattern and area, from the present work. The
value of the term quadratic nominal stress damage initiation criterion (QUADSCRT)
in the figures determines the probability of failure of those elements as determined
Simulation Studies of Low-Velocity Impact Damage in FRPS 231

Fig. 6 a Displacement versus time graph and b KE versus time graph from the current study;
comparison with c displacement versus time and d KE versus time graphs from [3]

(a): 40 Joules (b): 30 Joules

(c): 25 Joules (d): 20 Joules

Fig. 7 Matrix cracking profiles for different impact energies, in each layer and damages overlaid
upon each other
232 L. Lokesh Kumar et al.

by the QUADS criterion. The model captures the classical peanut shape and the
orientation of the delamination in the fibre direction of the lower layer (depicted in
Fig. 8).
Delamination reduces the strength of the laminate, under compression. Upon
further loading, the size of the delamination may increase. Once a critical size
is reached, the delamination growth becomes unstable leading to a drastic loss
of compressive strength [3]. Figure 9 shows the obtained damage profiles for the
cohesive layers for different impact energies. The figures reiterate the peanut shape
damage profile and also match with Figs. 10 and 11 from [12]. The delamination
profile increases with the increase in impact energy and decreases with the layers of
the composite sheet.

Fig. 8 Delamination between two plies, favouring lower ply orientation (−45°) [9]

(a): 40 Joules (b): 30 Joules

(c): 25 Joules (d): 20 Joules

Fig. 9 Damage to the cohesive elements placed between adjacent layers, and the overlaid damage
along with scale for various energies of impact
Simulation Studies of Low-Velocity Impact Damage in FRPS 233

Fig. 10 Typical delamination area at the end of impact event from Gonzalez [12]

Fig. 11 Projected delamination areas for different impact energies (experimental: black line;
numerical: dashed grey line), Gonzalez [12]

3.4 Results of Compression After Impact (CAI) Test

The result obtained from the CAI simulation test is shown in Fig. 12a. The plot shows
the change in the reaction force with an increase in displacement of the test specimen
(synonymous with loading during experimentation). The force versus displacement
plot was simulated for specimens subjected to various energies of impact as shown
in Table 3, with induced damages as shown in Fig. 9. Generally, BVID lowers the
residual compressive strength of the composite depending on the energy of impact
and layup sequence [12]. The model in this study did not account for any manufac-
turing defect or residual deflections. Figure 12b shows the CAI test results from [14].
The difference in the force observed in the current study (Fig. 12a) and the compar-
ison work (Fig. 12b) is due to the difference in materials and layup sequence used.
However, the trend is similar. The reaction force increases linearly till it reaches a
threshold after which the force suddenly drops indicating that the laminate has failed,
as shown in Fig. 12a. The inset in Fig. 12a shows a magnified view of the failure
region where ‘1’ represents a small damage growth for the amount of displacement
applied. An increase in damage growth is predicted from ‘2’ to ‘3’ where it reaches
the peak reaction load, occurring at the larger existent delamination from the LVI and
234 L. Lokesh Kumar et al.

the plies close to the impacted side. ‘4’ represents unstable damage growth leading to
specimen collapse. The predicted damage sequence was similar for different impact
energies.
As the impact energy increases, the BVID would be larger in size, and hence
the residual compressive strength would be lower. The estimated residual strength,
measured in terms of peak reaction load (from Fig. 12a), is plotted for various impact
energy simulations and is shown in Fig. 13. The plot shows two distinct regions: (i)
Initial region where the decrease in reaction force with increase in impact energy
is very low and transits to (ii) region where the slope is drastic. Further work is
necessary to validate the simulation results and establish a relation between layup
sequence–impact energy–reaction force in the specimen.

Fig. 12 a Force versus displacement plot for different impact energies from current study;
b Comparison force versus displacement graph (experimental and numerical) from [14] for 20 J for
TeXtreme weave laminate

Fig. 13 Reduction in Reaction Force with increase in impact energy

Simulation Studies of Low-Velocity Impact Damage in FRPS 235

4 Conclusions

In this work, a numerical study of low-velocity impact (LVI) events on AS4-8552

was conducted using commercially available finite element software, ABAQUS. The
material behaviour for the composite was modelled using a VUMAT subroutine (user
variable material code). Both interlayer and intralayer barely visible impact damages
(BVID) were predicted using Cohesive Zone Modelling (CZM) and Hashin and Puck
Criteria, respectively. The following are the conclusions from the work:
(1) The simulation results were found to be in good agreement with the literature
(e.g., classical ‘peanut shape’ of the delamination profiles, force–time graph
profiles).
(2) However, the results from force versus time graphs were found to be conserva-
tive.
(3) The peanut shape of delamination was oriented towards the direction of fibre in
the adjacent lower ply.
(4) The matrix cracking profiles obtained in the current study are consistent with
the experimental literature results.
(5) The intralayer damage study showed that the main cause for it was the matrix
cracking rather than fibre breakage, which needs to be further investigated with
experimental results.
(6) Simulation results of compression strength after impact (CAI) showed no
damage propagation until final failure, a characteristic of brittle behaviour.
The force–time profiles from the CAI tests were in good agreement with the
experimental literature results.
(7) It was also observed that the reaction force (measure of CAI strength) decreased
with an increase in LVI exhibiting two distinct slopes indicating a threshold
impact energy until the strength degradation is marginal.
Overall, the present work demonstrates the successful use of numerical modelling
techniques on complex low-velocity impact problems on composites. Further
numerical work is planned on the residual fatigue strength of such composites.

Acknowledgements The authors thank the Principal, Head of the Department, Mechanical Engi-
neering, B.M.S. College of Engineering (BMSCE) for the support rendered in carrying out the
work. One of the authors, Dr. CMM, profusely thanks the Director, CSIR-NAL. Useful discussions
held with Jayaram Pothnis (Post-doc.), Abhilash (Research scholar), Aerospace Department, I.I.Sc.,
Bengaluru, and Vishwas Patil (Post-graduate student), BMSCE, are also acknowledged.

References

1. LABIDI Aymen, “Boeing 787 Dreamliner Represents Composites Revolution”, Research gate.
2. Safri SNA, Sultan MTH, Yidris N, Mustapha F. Low velocity and high velocity impact test on
composite materials—a review. Int J Eng Sci. 2014.
236 L. Lokesh Kumar et al.

3. Perillo G, Vedvik NP, Echtermeyer AT. Numerical analyses of low velocity impacts on
composite. Adv Modell Techn. Simulia Conference. 2016.
4. Shi Y, Swait T, Soutis C. Modelling damage evolution in composite laminates subjected to low
velocity impact. Compos Struct. 2012.
5. Zhang J, Zhang X. Simulating low-velocity impact induced delamination in composites by a
quasi-static load model with surface-based cohesive contact. Compos Struct. 2015.
6. Riccio A, De Luca A, Di Felice G, Caputo F. Modelling the simulation of impact induced
damage onset and evolution in composites. Compos Struct. 2014.
7. Soroush M, Malekzadeh Fard K, Shahravi M. Finite element simulation of interlaminar and
intralaminar damage in laminated composite plates subjected to impact. Latin Amer J Solids
Struct. 2018.
8. Gonzalez EV, Maimi P, Camanho PP, Turon A, Mayugo JA. Simulation of drop-weight impact
and compression after impact tests on composite laminates. Compos Struct. 2012.
9. Pederson J. Finite element analysis of carbon fiber composite ripping using ABAQUS. Master
Thesis.
10. Liu H, Falzon B, Tan W. Experimental and numerical studies on the impact response of damage-
tolerant hybrid unidirectional/woven carbon-fibre reinforced composite laminates. Compos
Part B Eng. 2018.
11. ABAQUS User Manual and Documentation
12. Gonzalez EV. Simulation of interlaminar and intralaminar damage in polymer-based compos-
ites for aeronautical applications under impact loading. PhD Thesis. 2011.
13. Mechanics of Composite Structures, Laszlo P. Kollar, Cambridge University Press. 2003.
14. Soto A, Gonzalez EV, Maimi P, Martin de la Escalera F, Sainz de Aja JR, Alvarez E. Low
velocity impact and compression after impact simulation of thin ply laminates. Compos Part
A. 2018.
Experimental Identification of Unbalance
and Crack Parameters in an Internally
Damped Rotor System Integrated
with an Active Magnetic Bearing

Nilakshi Sarmah and Rajiv Tiwari

Abstract Imbalance and crack in rotors are the extremely prevalent faults in rotating
machinery. Fault detection and identification techniques thus play a vital role in
highly reliable operations without intervening in the machine’s rundown time. In the
present work, experimental identification of dynamic characteristic parameters of
active magnetic bearing (AMB) together with other fault parameters, such as the addi-
tive stiffness of fatigue crack, the internal and external damping, and residual unbal-
ance are evaluated using a model-based methodology. A laboratory test rig consists
of an electromagnetic actuator, which is mounted on the cracked-shaft supported
on deep groove ball bearings. A closed loop feedback control law is implemented
with the help of PID controller gains through dSPACE control desk for generating
control current in the magnetic actuator for producing electromagnetic forces that
will control the additional excitation forces arising in the rotor system due to the
presence of faults. The AMB control currents are not only suppressing the vibration
generated due to crack and unbalance forces, but also utilized along with vibra-
tional displacement responses in the identification procedure. To analyze the forced
responses of the test rig, three sets of vibrational displacement and AMB control
current signals are acquired for two speed ranges of 15–20 Hz and 20–26 Hz with
a step of 1 Hz. These time series responses are converted into multi-harmonic full
spectrum responses. The amplitudes and phases of displacement and control current
of full spectrum after removal of residual shaft bow from 1X harmonic are used in the
identification methodology to estimate multiple faults present in the laboratory test
bench. To check the reliability and efficacy of the identification methodology, stan-
dard deviations of the estimated multiple fault parameters are evaluated and found
well concordant with each other irrespective of the different data sets considered.

N. Sarmah (B)
Research Scholar, Department of Mechanical Engineering, Indian Institute of Technology
Guwahati, Guwahati, Assam, India
e-mail: [email protected]
R. Tiwari
Department of Mechanical Engineering, Indian Institute of Technology Guwahati, Guwahati,
Assam, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 237
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_16
238 N. Sarmah and R. Tiwari

Keywords Fatigue crack · Internal damping · Unbalance · Active magnetic

bearing (AMB) · Full spectrum

Nomenclature

[A1 ] Regression matrix

{B1 } Regressor
cV External damping
cH Internal damping
[C] Damping matrix
d Shaft diameter
eu Disc eccentricity
{f} Force vector
i Number
√ of harmonics
j −1
ko Intact shaft stiffness
l Shaft length
m Disc mass
p Participation factor
s Complex response vector
s Amplitude of response vector
[T] Transpose matrix
t Time instant
{X 1 } Unknown fault vector

Greek letters

Δk Additive crack stiffness

φu Unbalance phase

Subscripts

amb AMB
cr Crack
unb Unbalance

Abbreviation

AMB Active magnetic bearing

PID Proportional-integral-derivative
Experimental Identification of Unbalance and Crack Parameters … 239

1 Introduction

In the recent development in the modern industry the online health monitoring gets
wide range of popularity. For this, the smart condition monitoring concept in rotating
machinery through AMBs play a vital role. This is because of the contact free support
and controlling actuation through AMB, which avoids lubrication [1]. The presence
of faults, such as the unbalance and crack pose the greatest danger by producing
high amplitude of vibration in the machine. AMBs can also be used as a controller to
attenuate the vibration [2]. Moreover, the internal damping in the rotor also possesses
a significant role in the dynamics of the system. It is found in the literature that small
cracks do not affect much on the dynamics of the system and are difficult to detect,
but with the increase of time the crack grows leading towards the complex dynamic
behavior and even catastrophic failures.
Model based technique is capable to identify magnitude and location of faults.
Many research works have been successfully done for the fault detection and diag-
nosis of machinery using the model-based method. Lees et al. [3] overviewed the
identification and quantification of faults in rotating machines using a model based
approach. The identification of faults, like the rotor cracks, unbalances, rubs, rotor
bends, misalignment, and fluid induced instability with the analysis of vibration
signatures is based on qualitative measured data. These data are practically not
feasible to significant numbers of plant testing for which model-based identifica-
tion quantification of these faults is essential. Vania and Pennacchi [4] gave insight
to different techniques of fault identification using the model-based approach. These
models showed their usefulness in detecting faults in spite of the inaccuracy in the
fault model in a fully assembled machine. Singh and Tiwari [5] discussed the cracked
rotor-AMB system’s dynamic behavior and numerically estimated the additive crack
stiffness, unbalance and AMB parameter using the regression technique.
Self-excited vibrations induced in machine are mainly due to internal (rotating)
damping, when the machine operates in supercritical speed ranges. This type of
damping increases the whirling amplitude in the supercritical speeds and leads
to instability beyond the threshold speed. The rotor bearing system encompassing
internal viscous and hysteretic damping stability is analyzed using Timoshenko beam
theory including gyroscopic effect by Nelson and Zorzi [6]. Ku [7] discussed the
collective effect of shear deformation and internal damping on the forward and back-
ward spinning of the flexible rotor system with its instability threshold. Montagnier
and Hochard [8] illustrated the stability differences between two models of super-
critical driveshaft mounted on a viscoelastic support, such as the internal viscous
and hysteretic damping using the finite element method. Das et al. [9] studied the
performance of active vibration control technique for vibrational amplitude control
and how the instability of a rotor system due to the presence of internal viscous
damping is postponed to some extent.
Till now, researchers worked very less in the experimental fault detection of
rotating machine using AMB. Quinn et al. [10] experimentally verified that the
time-varying stiffness arises due to the breathing mode of crack on shaft applying
240 N. Sarmah and R. Tiwari

combination resonance. The magnetic bearing actuator in the mid shaft location
supplied the time-dependent forcing, which induces combination resonance in the
system. Al-Shudeifat and Butcher [11] proposed a different breathing function of the
crack for a rotor having transverse crack. They verified the theoretical results with
the experimental one and observed unique vibration signatures for various ranges
of crack depths in the neighborhood of the subcritical speed region of the rotor.
Guo et al. [12] carried out the experimental study of the crack detection proce-
dure in a Jeffcott rotor having a transverse crack by implementing the empirical
mode decomposition (EMD) method. The crack detection and identification were
performed of the reduction of stiffness quantitatively using the forward and back-
ward whirl responses in the frequency domain by Shravankumar and Tiwari [13]. Roy
and Tiwari [14] performed an experiment to analyze the influence of the external and
internal damping on a cracked rotor system without the gyroscopic effect and then
identified the fault parameters using a full spectrum. Friswell et al. [15] discussed
detection and diagnosis of shaft cracks by giving additional excitation using AMB
on to the rotor. The AMB excites the rotor with sinusoidal force for which the rotor’s
ultimate response is a combination of AMB excitation frequency and the spin speed
of the rotor. Sarmah and Tiwari [16] did a theoretical illustration of multiple fault and
system parameters identification of an internally damped cracked rotor system inte-
grated with AMB. Sarmah and Tiwari [17, 18] proposed identification methodologies
of cracked rotor- AMB system to estimate the fault parameters without residual bow
and one with residual bow and phase. They illustrated the effect of residual bow on
the cracked shaft of the rotor and the system dynamics.
The research areas are mostly focused on the stability of the rotating system due to
an internal damping effect. There is a lack of research in the estimation of the internal
damping parameter, which needs special attention along with other rotor faults in
condition monitoring of the machine. The present methodology is an extension of
[16] for validation with the experimental test results acquired from the laboratory
cracked rotor-AMB test rig.
The originality of the proposed work is as per following:
(a) Based on the previous theoretical work provided in Ref. [16], the detection
of multiple fault and system parameters in the rotor-AMB system is done
experimentally.
(b) The multiple harmonics of the directional spectra due to the presence of fatigue
crack on the shaft from the measured responses of both displacement and control
current are implemented into the methodology of the identification algorithm
for estimation.
(c) Experimental study is carried out in a laboratory developed cracked rotor-
AMB test rig at multiple speed ranges. The characteristic parameters of a tuned
AMB are estimated along with the additive crack stiffness, external and internal
damping and unbalance parameters. To check the efficacy of the identification
methodology, standard deviations of the estimated parameters are evaluated.
Experimental Identification of Unbalance and Crack Parameters … 241

2 Analytical Approcah of Cracked Internally Damped

Rotor-Amb System

Mathematical modeling and related assumptions for the cracked internally damped
rotor-AMB system are shown in Fig. 1. The rotor bearing system comprises of a
flexible shaft with a rigid disc supported on two conventional bearings and integrated
with a magnetic bearing. The rotor is connected with a flexible coupling to the motor.
The assumptions considered in the modelling are (i) shaft to be massless and bearings
are to be rigid and (ii) weight dominance effect that exhibits static displacement due
to dead weight of the rotor is greater than the dynamic displacement of the rotor.
The thin rigid disc has mass ‘m’ that inherited a residual unbalance, f un and it is
mounted at the central location of the shaft span. Thus, the wobbling of disc would
not take place that is emphasized and gyroscopic effect can be ignored. Due to this,
in the present analytical modeling, the gyroscopic effect is neglected. The intact
shaft stiffness is modeled with stiffness coefficient ‘ko ’, the reduction in stiffness
due to the presence of crack on it is defined as additive crack stiffness, ‘Δk’, and the
external damping due to the bearing is with a bearing damping coefficient, ‘cV ’. The
shaft’s cracked faces due to rubbing aggravates internal damping that is modeled by
the internal damping coefficient ‘c H ’. The rotor is assumed to have two degrees of
freedom (DOFs) with u x and u y translational vibrational displacements in vertical
and horizontal directions, respectively. The static displacement of the rotor due to its
dead weight in the vertical direction is denoted as u xo .
The equation of motions (EOMs) of the proposed rotor bearing system equipped
with an AMB will be briefly discussed in the following subsection.

2.1 Equations of Motion

The EOMs for the internally damped rotor system, bearing, disc, and AMB having
a transverse crack on it are well derived in Refs. [2, 16]. The overall EOM of the
proposed rotor-AMB system is represented as

Sensor Disc
Rotor core Shaft
Crack

Electromagnetic actuator
Bearing

Fig.1 Schematic representation of cracked rotor bearing equipped with an AMB

242 N. Sarmah and R. Tiwari

[ ]
[M]{s̈(t)} + [C]{ṡ(t)} + K + K C {s(t)} = { f un } + { f c } − { f amb } − ω[C H ]{so } (1)

where {s(t)} represents the response vector, {s0 (t)} defines the static displacement
vector due to dead weight of the rotor and {f un }, {f c } and {f amb } are the unbal-
ance, crack and magnetic force vectors, respectively. A linear PID control algorithm
is implemented to generate the control current (ic ) about the electromagnets of the
AMB having displacement-stiffness and current- stiffness coefficients as k s and k i for
vibration suppression. Matrices [M], [C], [C H ], [K] and [K C ] are the mass, external
damping, internal damping, stiffness and circulatory stiffness matrices, respectively.
Detailed descriptions of these matrices are discussed in Ref. [16] and added in
Appendix A.
To make Eq. (1) in a concise manner, all the forces are denoted as {F} and may
write as

[M]{s̈(t)} + [C]{ṡ(t)} + [K + K C ]{s(t)} = {F} − ω[C H ]{so } (2)

The time domain EOM of Eq. (1) is converted into frequency domain by assuming
a multi-harmonic (‘i’ denotes the number of harmonics) solution as {s(t)} = {s i }ejiωt
for displacement and for the control current of AMB as {i c } = {i ci }ejiωt for all
the force vectors {F} = {F}ejiωt . Upon inserting the assumed solution, Eq. (1) is
written as,
( )
[K + K C ] + ji ω[C] − (iω)2 [M] s i = [F] − ω[C H ]s o (3)

The multi-harmonic frequency domain Eq. (3) is responsible for collection of

amplitude, phase information of the rotor-AMB system, and is further employed in
the identification algorithm development to estimate the fault parameters associated
with the system.

2.2 Identification Algorithm

The multi-harmonic frequency domain equation Eq. (3) is utilized for the develop-
ment of the identification equation and segregating the unknown parameters and the
known ones in the right hand side we may get as

∑
n
(−ji ωcV + j(i − 1)ωc H + jωc H u x0 )s i − Δku x0 pi
i=−n
( )
− meu ω2 ejφu − ks s i + ki i ci = −(i ω)2 m + k0 s i
(4)

To identify the multiple faults, a linear regression method is practiced as follows

Experimental Identification of Unbalance and Crack Parameters … 243

[A1 (ω)]14×7 {X 1 }7×1 = {B1 (ω)}14×1 (5)

with {X 1 }7×1 = { cV c H Δk eu φu ks ki }T
where the disc eccentricity and its phase are defined as eu and φu respectively.
Equation (5) has been segregated such that the unknowns, i.e. the fault parameters
are arranged in vector {X 1 } and the known elements are in vector {B1 (ω)}. The
coefficients of parameters to be estimated, such as external and internal damping,
crack additive stiffness, disc unbalance and its orientation and force-stiffness and
force-current factor of AMB are stacked in the regression matrix [A1 (ω)]. Now,
splitting of Eq. (5) in the real and imaginary terms of responses and unbalance, crack
and magnetic force is represented as

[A1 (ω)]28×8 {X 1 }8×1 = {B1 (ω)}28×1 (6)

where subscript in matrices and vectors represent its size.

The identification equation can be arranged in the matrix form as
⎡ ⎤
−ωs̄ 1,Im 0 −u x0 p1 −mω2 0 −s̄ 1,Re ī c1 ,Re
⎢ 0 ωs̄ 0,Im −u x0 p0 0 0 −s̄ 0,Re ī c0 ,Re ⎥
⎢ ⎥
⎢ −2ωs̄ 2,Im −ωs̄ 2,Im −u x0 p2 −s̄ 2,Re ⎥
⎢ 0 0 ī c2 ,Re ⎥
⎢ −3ωs̄ 3,Im −2ωs̄ 3,Im −u x0 p3 −s̄ 3,Re ⎥
⎢ 0 0 ī c3 ,Re ⎥
⎢ ⎥
⎢ −4ωs̄ 4,Im −3ωs̄ 4,Im 0 0 0 −s̄ 4,Re ī c4 ,Re ⎥
⎢ ⎥
⎢ −5ωs̄ 5,Im −4ωs̄ 5,Im −u x0 p5 0 0 −s̄ 5,Re ī c5 ,Re ⎥
⎢ ⎥
⎢ −6ωs̄ 6,Im −5ωs̄ 6,Im 0 0 0 −s̄ 6,Re ī c6 ,Re ⎥
⎢ ⎥
⎢ .. .. .. .. .. .. .. ⎥
⎢ . . . . . . . ⎥
⎢ ⎥
⎢ ωs̄ −1,Im 2ωs̄ −1,Im −u x0 p−1 0 0 −s̄ −1,Re ī c−1 ,Re ⎥ ⎧ ⎫
⎢ ⎥⎪ cV ⎪
⎢ 2ωs̄ −s̄ −2,Re ⎥
ī c−2 ,Re ⎥⎪⎪
⎪ ⎪
⎪
⎢ −2,Im 3ωs̄ −2,Im 0 0 0
cH ⎪
⎢ ⎥⎪⎪ ⎪
⎪
⎢ 3ωs̄ −3,Im 4ωs̄ −3,Im −u x0 p−3 0 0 −s̄ −3,Re ī c−3 ,Re ⎥⎪⎪
⎪
⎪
⎪
⎢ ⎥⎨ Δk ⎪ ⎬
⎢ 4ωs̄ −4,Im 5ωs̄ −4,Im 0 0 0 −s̄ −4,Re ī c−4 ,Re ⎥
⎢ .. .. .. .. .. .. .. ⎥⎪⎥ e
⎢ u,Re
⎪
⎢ . . . . . . . ⎥⎪ ⎪ eu,Im ⎪
⎪
⎢ ⎥⎪⎪
⎪
⎪
⎪
⎪
⎢ ωs̄ 1,Re 0 0 0 −mω2 −s̄ 1,Im ī c1 ,Im ⎥ ⎪
⎪ s ⎪
k ⎪
⎢ ⎥⎪⎩ ⎪
⎭
⎢ 0 ω(s̄ 0,Re + u x0 ) 0 0 0 −s̄ 0,Im ī c0 ,Im ⎥ ki
⎢ ⎥
⎢ 2ωs̄ ωs̄ 2,Re −s̄ 2,Im ī c2 ,Im ⎥
⎢ 2,Re 0 0 0 ⎥
⎢ 3ωs̄ −s̄ 3,Im ī c3 ,Im ⎥
⎢ 3,Re 2ωs̄ 3,Re 0 0 0 ⎥
⎢ ⎥
⎢ 4ωs̄ 4,Re 3ωs̄ 4,Re 0 0 0 −s̄ 4,Im ī c4 ,Im ⎥
⎢ ⎥
⎢ 5ωs̄ 5,Re 4ωs̄ 5,Re 0 0 0 −s̄ 5,Im ī c5 ,Im ⎥
⎢ ⎥
⎢ .. .. .. .. .. .. .. ⎥
⎢ . . . . . . . ⎥
⎢ ⎥
⎢ −ωs̄ −1,Re −2ωs̄ −1,Re 0 0 0 −s̄ −1,Im ī c−1 ,Im ⎥
⎢ ⎥
⎢ −2ωs̄ −3ωs̄ −2,Re 0 0 0 −s̄ −2,Im ī c−2 ,Im ⎥
⎣ −2,Re ⎦
.. .. .. .. .. .. ..
. . . . . . .
244 N. Sarmah and R. Tiwari
⎧ ⎫
⎪
⎪ (ω2 m − k0 )s̄ 1,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −k0 s̄ 0,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (4ω m − k0 )s̄ 2,Re
2 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (9ω2 m − k0 )s̄ 3,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (16ω2 m − k0 )s̄ 4,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (25ω2 m − k0 )s̄ 5,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (36ω2 m − k0 )s̄ 6,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ .. ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ . ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (ω2 m − k0 )s̄ −1,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪(4ω2 m − k0 )s̄ −2,Re ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪(9ω2 m − k0 )s̄ −3,Re ⎪
⎪
⎪
⎨ ⎪
⎬
(16ω2 m − k0 )s̄ −4,Re
= (7)
⎪
⎪ .. ⎪
⎪
⎪
⎪ . ⎪
⎪
⎪
⎪ (ω2 m − k )s̄ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
0 1,Im ⎪
⎪
⎪
⎪ −k s̄ ⎪
⎪
⎪
⎪ 0 0,Im ⎪
⎪
⎪
⎪ ⎪
(4ω − 0 2,Im ⎪
)s̄
2
⎪
⎪
m k ⎪
⎪
⎪
⎪ (9ω 2
− )s̄ ⎪
⎪
⎪
⎪ m k 0 3,Im ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (16ω 2
m − k )s̄ ⎪
⎪
⎪
⎪
0 4,Im
⎪
⎪
⎪
⎪ (25ω 2
m − k )s̄ ⎪
⎪
⎪
⎪
0 5,Im
⎪
⎪
⎪
⎪ .
.. ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (ω 2
m − k )s̄ ⎪
⎪
⎪
⎪ 0 −1,Im ⎪
⎪
⎪
⎪ ⎪
(4ω − 0 −2,Im ⎪
)s̄
2
⎪
⎪
m k ⎪
⎪
⎪
⎩ .. ⎪
⎭
.

For solving Eq. (6), on implementing the pseudo-inverse technique, it is written as

( )−1
{X 1 }8×1 = {A1T }8×28 {A1 }28×8 {A1T }8×28 {B1 }28×1 (8)

To get the more accurate estimation, large data sets are required for different speed
ranges. Therefore, identification of fault parameters in a combined speed case may
be written as (it is assumed here that parameters are independent of speed)
⎧ ⎫ ⎧ ⎫
⎪
⎪ [A1 (ω1 )] ⎪
⎪ ⎪
⎪ {B1 (ω1 )} ⎪
⎪
⎪
⎨ [A1 (ω2 )] ⎪
⎬ ⎨ {B1 (ω2 )} ⎪
⎪ ⎬
. {X 1 }8×1 = . (9)
⎪
⎪ .. ⎪
⎪ ⎪
⎪ .. ⎪
⎪
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
[A1 (ωn )] 28×8 {B1 (ωn )} 28×1

Equation (9) would be used to estimate the multiple fault and system parameters
as proposed in Fig. 1 the rotor-AMB system. In the subsequent sections, detailed
experimental investigation to evaluate the fault parameters in a laboratory test bench
using the developed identification algorithm will be discussed.
Experimental Identification of Unbalance and Crack Parameters … 245

3 Experimental Analysis

In order to carry out the experimental tests, the vibration signatures from the labo-
ratory test rig are acquired in two orthogonal translational directions. The pictorial
view of the rotor test bench shown in Fig. 2 is developed in the Advanced Dynamics
and Vibration laboratory of IIT Guwahati. The test rig comprises of a shaft with a
fatigue crack, where the rigid disc is mounted on it. The shaft is supported on two
deep-groove ball bearings out of which one bearing is mounted on the driven end of
the CNC spindle motor connected with a flexible coupling. The artificial fatigue crack
generation on the shaft is described in detail [19]. An eight-pole hetero-polar AMB is
mounted near on the shaft, which supplies the control current leading vibration atten-
uation. Two proximity sensors in orthogonal directions are mounted near the disc
location to acquire the test data from the rig. In order to implement the control algo-
rithm, a PID (proportional-integral-derivative) controller is used through dSPACE
platform linked with MATLAB® . The control currents supplied to the magnetic
actuator are amplified through PWM servo amplifiers in the current control mode
to generate adequate amount of magnetic force. A 2 A DC bias current is supplied
through the electromagnets of the actuator. The sampling rate taken for the analysis
is 10 kHz with sampling time 0.0001 s.
The experimental tests are performed with the impact test for finding out the
natural frequency of the system and are summarised in the next section. The forced
response test data is collected for the slow run condition, i.e. at 5 Hz that is used to
remove the effect of residual bow from the other responses at 1X harmonic of the
signals during various spin speeds of operation [2, 14]. The bow effect is eliminated
from the complex displacement for the high spin speed at 1X harmonic of the full
spectrum by subtracting that from the slow roll displacement and it can be expressed
as

Zoomed view of
Magnetic bearing actuator fatigue crack on shaft

Disc
DC power source
Flexible coupling
Motor

Power amplifier
Bearing block Eddy current sensor dSPACE MicroLabBox
with bearing

AC/DC probe

(a) (b)

Fig. 2 Pictorial view of a the developed cracked rotor-AMB test bench in the laboratory b the
controller instrumentation
246 N. Sarmah and R. Tiwari
⎛ ⎞
⎜ ⎟
S(t) =⎝ S1 − S 1 ⎠e j wt (10)
Response after bow removal High speed Slow roll

The multi-harmonic full spectrum, except 1X frequency component, remains

the same and the amplitude and corrected phases are applied to the identification
algorithm to estimate the fault parameters. The phase correction can be done by
subtracting the phase of the rotor displacement and control current response to that
of the reference signal to avoid the leakage error as discussed elaborately in Refs.
[2, 5, 14]. Subsequent sections will be containing the results and discussion of the
experimental analysis.

4 Results and Discussions

The static and dynamic analysis of the extracted data from the test bench along with
the estimation of fault parameters is illustrated in the following subsections.

4.1 Analysis of Time Domain and Frequency Domain Signal

from the Extracted Test Data

An impact force is applied on the rotor in static condition without and with AMB case.
The time domain signal after impact is acquired and the frequency domain conversion
is done through Fast Fourier Transform (FFT) for evaluating the natural frequency
of the test system. The details of the cracked rotor-AMB test rig are presented in
Table 1. Figure 3 shows the first fundamental frequency of the cracked rotor system
without and with AMB and found for two conditions, such as

(i) if the cracked face in the shaft is fully open (cracked face is in downward
direction), natural frequency of the system with no AMB is 58.2 Hz and using
AMB is 60.5 Hz as shown in Fig. 3a.
(ii) if the cracked face of the shaft is in fully closed condition (cracked face of
the shaft is in perpendicular to the bearing axis, i.e. in the horizontal direction),
with no AMB case, natural frequency is 61.02 Hz and using AMB as 65.37 Hz.

A time domain response is obtained near the disc location at slow roll condition,
i.e. 5 Hz as shown in Fig. 4. To remove the effect of residual shaft bow from whole
analysis, slow roll speed is required. The reference signal is shown in Fig. 4a consists
of the high and low peaks due to the presence of two slots 180° apart in the motor
spindle from where this signal is extracted. Figures 4b, c depict the vibrational
displacements in the transverse directions, respectively. The required control current
to attenuate the displacement in two orthogonal directions is shown in Fig. 4d, e. A
Experimental Identification of Unbalance and Crack Parameters … 247

Table 1 Rotordynamic system parameters and actuator properties with PID controller gains
Parameters Data set 1 Actuator PID controller gains
properties
Shaft length (l) 0.47 m Pole 8 Proportional 4000 Am−1
gain (K P )
Shaft diameter 0.16 m No. of turns 100 Integral gain 1500 A (ms)−1
(d) (N) (K I )
Density (ρ) 7860 kgm−3 Copper wire 20 AWG Derivative 3 As m−1
diameter gain (K D )
Disc mass (m) 1.87 kg Outer 0.09 m
diameter
SKF 61,802 Inner 0.0328 m
diameter
Motor power 1.5 kW Air gap 0.0008 m

Fig. 3 Natural frequency of the cracked rotor exclusively and with AMB implementing impulsive
force a cracked face is fully open b cracked face in fully closed condition

similar approach is followed to acquire the time domain responses at various spin
speeds and time responses, at the one of the high speeds of 15 Hz, are shown in Fig. 5
for the reference. The orbit plots of the rotor system spinning at 15 Hz speed is shown
in Fig. 6. A comparison is done to show the effect of residual bow onto the cracked
rotor and found that with bow, the displacement is 3.63 × 10−4 m and without the
bow effect it is 1.25 × 10−4 m as observed in Fig. 6a. Similarly, the control current
needed for attenuating vibration without the bow case is lower, i.e. 0.6 A, whereas
for the bowed cracked rotor it is 1.94 A.
The full spectrum is a highly effective tool that displays the directional nature of a
signal (such as for the rotor with crack on it) both forward and backward precessional
whirls. Time responses extracted from the test data are converted into a full spectrum
through the FFT algorithm. The full spectrum displacement and control current
amplitude and phases of the multiple harmonic signals of the proposed rotor system
are shown in Figs. 7 and 8 at the spin speed of 5 Hz and 15 Hz, respectively. The
1X harmonic of slow roll full spectrum is utilized to remove the bow effect from 1X
harmonic of the high speed full spectrum using Eq. (10). Figures 7a, d depict the
248 N. Sarmah and R. Tiwari

Fig. 4 Time waveform of the cracked rotor-AMB system spinning at 5 Hz a Reference signal
b–c Vertical and horizontal displacements near the disc position d–e Vertical and horizontal control
currents of AMB

Fig. 5 Time waveform of the cracked rotor-AMB system spinning at 4 Hz a Reference signal
b Vertical c Horizontal displacements near the disc position d Vertical and e Horizontal control
currents of AMB
Experimental Identification of Unbalance and Crack Parameters … 249

Fig. 6 Orbit comparison of cracked rotor-AMB system with and exclusive of residual bow at the
spin speed of 18 Hz a displacement signal b control current signal of AMB

reference signal amplitude and phase at the spinning of 5 Hz, Figs. 7b, e show the
amplitude and phase of displacement signal, and the amplitude and phase of control
current is shown in Fig. 7c, f. Full spectrum amplitude and phase of reference,
displacement and control current signal at the rotor spinning of 15 Hz are shown in
Fig. 8a–f.
The multiple harmonic full spectrum amplitude and corrected phases are further
utilized in the estimation procedure that will be discussed in the next subsection.

Fig. 7 Multi-harmonic full spectrum of cracked rotor spinning at 5 Hz a, d reference signal ampli-
tude and phase b, e displacement signal amplitude and phase c, f control current signal amplitude
and phase
250 N. Sarmah and R. Tiwari

Fig. 8 Multi-harmonic full spectrum of cracked rotor spinning at 15 Hz a, d reference signal

amplitude and phase b, e displacement signal amplitude and phase c, f control current signal
amplitude and phase

4.2 Estimation of Fault and System Parameters

The displacement and control current amplitudes and phase corrections of full spec-
trum responses after removal of the residual bow effect, as discussed in the previous
section, are implemented into the developed identification methodology to estimate
fault and system parameters in the proposed system. Table 2 represents the esti-
mated fault parameters associated with the cracked rotor-AMB system. Responses
are measured for the two speed range cases, such as 15–20 Hz and 20–26 Hz with
the steps of 1 Hz and to check the repeatability of the estimated parameters, three
data sets of responses are acquired at each spin speed.
The estimation done for the multiple speed conditions to get a better accuracy,
mean and standard deviation is also evaluated to check the efficacy of the developed
methodology. It has been interpreted from Table 2 that the estimated parameters are
in good agreement irrespective of different data sets for the above mentioned two
speed ranges. In order to combine the results of different sets of test data, mean or
expected value, and standard deviations of each of the fault parameters for three test
sets in the speed range of 15–20 Hz and 20–26 Hz are presented in Table 2.
Mean (μ) and standard deviations (σ ) of the parameters are obtained as in Ref.
[20] respectively as
┌
| N
1 ∑N |∑ (xi − μ)2
μ= xi σ = √ (11)
N i=1 i=1
N

where xi are values of estimated parameters and μ is the mean value of these
estimations, while N is the total number of the estimated parameters.
Table 2 Experimentally estimated multiple fault parameters
Estimated parameters Speed (15–20 Hz) Mean (μ) Speed (20–26 Hz) Mean (μ)
Data set 1 Data set 2 Data set 3 and Data set 1 Data set 2 Data set 3 &
standard deviation Standard deviation
(σ ) (σ )
cV Nsm−1 213.3 215.4 210.9 213.2 261.2 263.6 266.2 263.6
1.8 2.1
c H Nsm−1 182.6 185.5 180.8 182.97 210.7 214.5 212.8 212.6
1.9 1.5
Δk Nm−1 9.65 × 104 9.06 × 104 1.49 × 105 1.12 × 105 1.75 × 105 1.67 × 105 1.48 × 105 1.63 × 105
2.6 0.11
eu m 2.29 × 10−4 1.27 × 10−4 2.51 × 10−4 2.02 × 10−4 2.82 × 10−4 3.06 × 10−4 2.65 × 10−4 2.84 × 10−4
0.5 0.16
φu deg 18.2 15.6 17.5 17.1 19.9 19.5 19.9 19.8
Experimental Identification of Unbalance and Crack Parameters …

1.09 0.18
ks N m−1 10,620.5 10,615.2 10,623.1 10,619.6 10,553.3 10,560.6 10,563.2 10,559.0
3.3 4.1
ki NA−1 5.6 5.1 5.9 7.2 6.9 7.7 6.9 7.2
0.4 0.38
251
252 N. Sarmah and R. Tiwari

It has been observed that the standard deviations of each estimated parameters
are below 5 and the maximum standard deviations are found to be 3.3 and 4.1 for
the displacement-stiffness factor of AMB for both the speed ranges. As the lower
standard deviation intricate closer estimates to the mean or expected value, thus,
it can be concluded that the estimates are quite close to each other regardless of
different data sets in both the speed ranges.

5 Conclusions

The proposed work aims to model a faulty rotor bearing system integrated with
AMB to mimic the faults present in a practical rotating machinery. A methodology
for estimating multiple fault and system parameters for a cracked internally damped
rotor-AMB system is presented here. The faults considered for the modeling are
the crack, residual unbalance and its orientation, the external and internal damping,
and AMB force factors. A simple rotor model with a central disc having a trans-
verse crack is considered to obtain the system’s dynamic equations. The develop-
ment of the identification algorithm using the frequency domain EOMs to evaluate
the fault parameters is discussed. The experimental investigation of the developed
cracked rotor-AMB test rig in the laboratory is done to obtain the forced responses
of displacement and current at various spin speeds. A discussion of removal of the
residual bow effect from the high speed test data with respect to the slow roll condi-
tion is carried out. The displacement and control current amplitudes and corrected
phases of full spectrum after removing the bow are utilized in the identification
algorithm for estimating the external and internal damping, additive crack stiffness,
unbalance and its orientation or phase, displacement-stiffness and current-stiffness
factors of AMB. The parameters are estimated for three sets of response data at two
different multiple speed ranges, such as 15–20 Hz and 20–26 Hz. It is observed that
estimated parameters are well concordant with each other for all data sets. To validate
the accuracy and efficacy of the identification methodology, standard deviations of
the estimated parameters are obtained. For the two spin speed ranges, the maximum
standard deviations are found to be 3.3 and 4.1 for the displacement-stiffness (ks )
parameter. As a whole, a good agreement of the estimated multi fault parameters are
obtained with the test data. The proposed identification approach has the flexibility
to incorporate multiple discs, cracks and AMBs. Thus, the future scope will focus
on the experimental investigation of the multi-discs cracked rotor-AMB system to
relate with the real turbogenerator sets.

Acknowledgements The authors would like to acknowledge the facility and infrastructure of the
Advanced Dynamics and Vibration Laboratory, IIT Guwahati provided to carry out experiments.
Experimental Identification of Unbalance and Crack Parameters … 253

Appendix A: Mathematical Formulation of Cracked

Internally Damped Rotor—Amb System

The EOM of the cracked rotor system integrated with AMB as

[ ]
[M]{s̈(t)} + [C]{ṡ(t)} + K + K C {s(t)} = { f un } + { f c } − { f amb } − ω[C H ]{so } (A1)

where
[ ] [ ]
m 0 (cV + c H ) 0
[M] = ; [C] = ;
0 m 0 (cV + c H )
[ ] [ ]
k0 0 0 ωc H
[K ] = ; [K C ] = (A2)
0 k0 −ωc H 0

Unbalance { force, crack and }magnetic force is defined

{ as follows }
meu cos(θ + φu ) u (1 + cos 2θ )
{ f un } = ; { f cr } = 21 τ (t)Δk x0 with τ repre-
meu sin(θ + φu ) sin 2θ u x0
sents the switching crack excitation function resembling the shaft’s cracked faces of
the opening and closure.

1 2 2 2 2 2
τ (t) = + cos θ − cos 3θ + cos 5θ − cos 7θ + cos 9θ · · · (A3)
2 π 3π 5π 7π 9π
{ }
−ks u x + ki i cx
{ f amb } = − (A4)
−ks u y + ki i cy

The control current regulated by PID controller is represented as

∫ ∫
i cx = K P u x + K I u x dt + K D u̇ x ;i cy = K P u y + K I u y dt + K D u̇ y (A5)

The vibrational displacement vector {s} and static displacement vector {so } of
the rotor are }
{ written as. { }
ux u x0
{s} = and {so } = .
uy 0

References

1. Schweitzer G, Maslen EH. Magnetic bearings: theory, design, and application to rotating
machinery. New York: Springer; 2009.
2. Tiwari R. Rotor systems: analysis and identification: Boca Raton: CRC Press;2017.
3. Lees A, Sinha J, Friswell M.L. Model-based identification of rotating machines. Mech Syst
Signal Process. 2009;23:1884–1893.
254 N. Sarmah and R. Tiwari

4. Vania A, Pennacchi P. Experimental and theoretical application of fault identification measures

of accuracy in rotating machine diagnostics. Mech Syst Signal Process. 2004;18:329–52.
5. Singh S, Tiwari R. Model-based fatigue crack identification in rotors integrated with active
magnetic bearings. J Vib Control. 2017;23:980–1000.
6. Zorzi E, Nelson H. Finite element simulation of rotor-bearing systems with internal damping.
J Eng Power. 1977;99:71–6.
7. Ku DM. Finite element analysis of whirl speeds for rotor-bearing systems with internal
damping. Mech Syst Signal Process. 1998;12:599–610.
8. Montagnier O, Hochard C. Dynamic instability of supercritical driveshafts mounted on dissi-
pative supports effects of viscous and hysteretic internal damping. J Sound Vib. 2007;305:378–
400.
9. Das A, Nighil M, Dutt J, Irretier H. Vibration control and stability analysis of rotor-shaft system
with electromagnetic exciters. Mech Mach Theory 2008; 43:1295–1316.
10. Quinn DD, Mani G, Kasarda ME, Bash T, et al. Damage detection of a rotating cracked shaft
using an active magnetic bearing as a force actuator-analysis and experimental verification.
IEEE/ASME Trans Mechatron. 2005;10:640–7.
11. Al-Shudeifat MA, Butcher EA. New breathing functions for the transverse breathing crack of
the cracked rotor system: approach for critical and subcritical harmonic analysis. J Sound Vib.
2011;330:526–44.
12. Guo C, Yan J, Yang W. Crack detection for a Jeffcott rotor with a transverse crack: an
experimental investigation. Mech Syst Signal Process. 2017;83:260–71.
13. Shravankumar C, Tiwari R. Experimental identification of cracked rotor system parameters
from the forward and backward whirl responses. Arch Mech Eng. 2019;329–353.
14. Roy DK, Tiwari R. Estimation of the internal and external damping from the forward and
backward spectrum of a rotor with a fatigue crack. Propuls Power Res. 2020;9(1):62–74.
15. Friswell M, Litak G, Sawicki J. Crack identification in rotating machines with active bearings.
In Proceedings of ISMA. 2010. pp. 2843–2855.
16. Sarmah N, Tiwari R. Identification of crack and internal damping parameters using full spectrum
responses from a jeffcott rotor incorporated with an active magnetic bearing. In: Proceedings
of the 10th International Conference on Rotor Dynamics—IFToMM. Cham: Springer;2018.
pp. 34–48.
17. Sarmah N, Tiwari R. Dynamic analysis and identification of multiple fault parameters in a
cracked rotor system equipped with active magnetic bearings: a physical model based approach.
Inverse Probl Sci Eng. 2020;28(8):1103–34.
18. Sarmah N, Tiwari R. Analysis and identification of the additive and multiplicative fault param-
eters in a cracked-bowed-unbalanced rotor system integrated with an auxiliary active magnetic
bearing. Mech Mach Theory. 2020;146: 103744.
19. Sarmah N, Tiwari R. Experimental Investigation of Active Control of Cracked Rotor-Bearing
System Equipped With Magnetic Bearing. In: Proceedings of the ASME 2019 Gas Turbine
India Conference: Compressors, Fans, and Pumps; Turbines; Heat Transfer; Structures and
Dynamics, vol. 1. Chennai, Tamil Nadu, India; 2019.
20. Donner A, Zou G. Closed-form confidence intervals for functions of the normal mean and
standard deviation. Stat Methods Med Res. 2012;21(4):347–59.
Vibration Response Prediction in Rotor
Systems with External Damping by Deep
Learning Using Geometrical Features

R. Siva Srinivas, Rajiv Tiwari, and Amar Singh

Abstract This paper attempts to predict the response of a rotor-bearing system with
external damping by training a neural network model using the frequency domain
historical data obtained from multiple simulations. In general prediction of response
for a rotor dynamic system is performed by detailed finite element modeling and/
or testing on the test bench. This involves the modeling of rotors, bearings, and
housings. Accurate finite element modeling involves considering various important
aspects such as material damping, structural damping, and bearing non-linearities.
Design is an iterative process and any change in the design of bearings or housings
or dampers necessitates the repetition of the FE modeling and analysis. In this work,
a method to overcome this repetitive process is presented. A neural network model is
built by mapping the geometric features of bearing housings and dampers, unbalance
amplitude and phase, oil viscosity, and rotational speed with the frequency domain
response at the front and rear bearing locations. The model is trained using a set of
input features and target variables by tuning the hyperparameters of the model. The
performance of the model is then tested on unseen data. The predicted results closely
match the true vibration responses. This neural network regression model can be
extended to real data from the rotor dynamic test rig as well.

Keywords Neural network · Regression · Training · Testing · Damper · Bearing

R. S. Srinivas (B) · A. Singh

Aero Engine Research and Design Centre, Hindustan Aeronautics Limited, Bengaluru, India
e-mail: [email protected]
A. Singh
e-mail: [email protected]
R. Tiwari
Department of Mechanical Engineering,
Indian Institute of Technology Guwahati, Guwahati, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 255
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_17
256 R. S. Srinivas et al.

Nomenclature

c Clearance
d Depth
e Eccentricity
k Stiffness
l Length
m Mass
n Number of webs
u Complex displacement
w Width
U 1, U 2 Unbalance amplitude

Greek Symbols

ε Eccentricity ratio
ω Angular velocity
μ Oil viscosity
β 1, β 2 Phase angles
δ Pocket depth

Abbreviations

ANN Artificial neural network

SFD Squeeze film damper
rpm Revolutions per minute
rad Radial

1 Introduction

Machine learning algorithms have applications in sciences, engineering, medicine,

and finance. There are broadly classified as linear algorithms, non-linear algorithms,
and ensemble algorithms [1]. Based on the task they perform they are classified as
classification and regression algorithms. Based on how they learn they are classi-
fied as supervised and unsupervised algorithms. In the current literature survey, the
focus is limited to condition monitoring using applied machine learning. Rapur and
Tiwari [2] classified the various hydraulic and mechanical faults in a centrifugal pump
using support vector machines (SVM) by inputting features obtained from wavelet
Vibration Response Prediction in Rotor Systems with External Damping … 257

transforms. Tiwari et al. [3] performed the detection of blockage and cavitation in
centrifugal pumps using artificial neural networks (ANN) by obtaining input features
from dynamic pressure signals. Kumar et al. [4] identified the level of inlet pipe
blockage by using pressure, vibration, and current signals obtained at various speeds
and feeding the corresponding time domain features such as mean, standard devia-
tion, skewness, and kurtosis as input features to ANN model. Gangsar and Tiwari
[5] performed condition monitoring of mechanical and electrical faults in induc-
tion motors using multi-class SVM. Choudary et al. [6] performed fault diagnosis
in bearings with convolutional neural networks (CNN) that uses thermal images as
input features. The novelty of the method is that it uses non-contact infrared thermog-
raphy measurements as against the conventional methods such as using accelerometer
signals.
The present paper develops a new method in which bearing housing’s geometric
features, oil properties, unbalance and rotor speed are used to predict the vibra-
tion response of a rotor system at multiple speeds. The rotor system considered is
supported by rolling element bearings at the front and the rear. Viscous dampers are
used at bearing locations for the purpose of vibration attenuation. A squirrel cage is
used as flexible support at the front and an elastic ring gives flexibility at the rear
support. In the response of the rotor system to various levels of unbalance for various
configurations of bearing housings, two different values of oil viscosity are obtained
from the finite element model. These vibration responses are split into training and
test data. The ANN model receives the training data as input and maps the input
features to the output variables. The performance of the trained model is tested using
test data. The predicted response matches closely with the output of the test data
validating the usefulness of the algorithm. In future, this model shall be used on
experimental data.

2 System Configuration

2.1 Rotor Bearing System

The rotor system considered for this study is shown in Fig. 1. It consists of a rotor
with two discs mounted on a ball bearing at the front and a roller bearing at the rear in
the simply supported arrangement. The bubbles shown represent the discs mounted
on the shafts.
The front bearings are mounted in a soft mount or flexible mount (squirrel cage)
with a squeeze film damper in parallel. The schematic of the squirrel cage and squeeze
film damper is shown in Figs. 2 and 3. In the rear bearing plane, the flexibility is
obtained by using an elastic ring that is slid on the outer race of the bearing. The
schematic of the elastic ring is shown in Fig. 4. Oil with a certain viscosity is supplied
to both dampers which generate damping under radial load.
258 R. S. Srinivas et al.

Fig. 1 Finite element model of rotor-bearing system

web Cut out

Fig. 2 Squirrel cage flexible support

Vibration Response Prediction in Rotor Systems with External Damping … 259

Damper housing

Oil film

Bearing outer race

O-ring seal

Fig. 3 Squeeze film damper

Fig. 4 Elastic ring damper (source: [7])

2.2 Input Features Selection

2.2.1 Front Bearing Plane

For a given rotor geometry the vibration response along the length of the rotor depends
on the stiffness and damping in bearing planes, speed, and amount of unbalance

(xfront , yfront , xrear , yrear ) = f (kfront , cfront , krear , crear , Udisc1 , Udisc2 , ω) (1)

In the present case front-bearing plane stiffness depends on the stiffness of the
bearing and that of the flexible housing. In the present case, squirrel cage acts as the
flexible housing.

kfront = f (kfrontbearing , ksquirrelcage ) (2)

260 R. S. Srinivas et al.

If the linear stiffness of the bearing is assumed to be constant then

kfront = f (ksquirrel cage ) (3)

The expression for the stiffness of the squirrel cage is given in [7]. It essentially
depends on its geometric features such as a number of webs, length, width, and depth
of the webs.

ksquirrel cage = f (l, w, d, n) (4)

The damping in the front-bearing plane depends on the damping of the bearing
and that of the squeeze film damper (SFD). The damping of rolling element bearings
is small compared to that of SFD.

cfront = f (csfd ) (5)

The expressions for stiffness and damping coefficients in a π -film SFD and 2π -
film SFD are given in [8]. In the present case, a full film is assumed. Therefore, the
stiffness contribution from SFD is zero and its damping capacity is a function of its
features such as radial clearance (c), eccentricity ratio (ε), damper radius (d), damper
length (l), oil viscosity (μ), speed (ω), i.e.,

Csfd = f (c, l, μ, ω, d, ε) (6)

Of the above parameters damper radius (d), damper length (l) are considered
constant in this work. Eccentricity ratio ε depends on the unbalance Udisc1 , Udisc2
and angular velocity ω. If unbalance values are considered as independent input
features, then

Csfd = f (c, μ, ω) (7)

2.2.2 Rear Bearing Plane

The expression for the stiffness of the elastic ring damper is given in [7]. There are
no empirical formulae in the literature to estimate the damping coefficient of ring
damper. The oil film coefficients of the ring damper have been estimated in [8, 9]
using the finite element method. However, both stiffness and damping of elastic ring
damper are functions of its design features such as projection width, height, and no.
of projections.

(K , C)elasticring_damper = f (δ1 , δ2 , n) (8)

Vibration Response Prediction in Rotor Systems with External Damping … 261

amp phase amp phase

If Udisc1 , Udisc2 , Udisc1 , Udisc2 are considered along with the features given in
Eqs. (4), (7), and (8) the total number of input features required to model the neural
network are 14.

3 Response Generation from Numerical Simulation

In this section the elemental matrices of shaft, discs, bearings, and squeeze film
damper are presented. The global equations of motion (EOM) of the Rotor-Bearing-
SFD system are then derived by assembling the sub-system matrices [10, 11].

3.1 Shaft

In this work, Timoshenko beam finite elements have been used to discretize the
rotor system. The elemental matrices of size 4 × 4 given in [10] that correspond to a
complex nodal displacement vector have been used for modeling each finite element.
 T  T
u e = u x z + ju yz = x1 ϕ y1 x2 ϕ y2 + j y1 ϕx1 y2 ϕx2 (9)

The mass, stiffness, and gyroscopic matrices of the shaft and discs are given in
the next subsections.

3.1.1 Translational Mass Matrix

⎡ ⎤
m1 m2 m3 m4
ρ Ar l e ⎢ m2 m5 −m 4 m6 ⎥
MT = ⎢ ⎥ (10)
420(1 + )2 ⎣ m 3 −m 4 m1 −m 2 ⎦
m4 m6 −m 2 m5

The coefficients of the translational mass matrix are given by

m 1 = 156 + 294 + 140 2

m 2 = le 22 + 38.5 + 17.5 2

m 3 = 54 + 126 + 70 2
m 4 = le −13 − 31.5 − 17.5 2

m 5 = le2 4 + 7 + 3.5 2
m 6 = le2 −3 − 7 − 3.5 2

12E I
=
k AG L 2
262 R. S. Srinivas et al.

3.1.2 Rotational Mass Matrix

⎡ ⎤
n1 n2 −n 1 n2
ρ Ie ⎢ n2 n3 −n 2 n4 ⎥
MR = ⎢ ⎥ (11)
30le (1 + )2 ⎣ −n 1 −n 2 n1 −n 2 ⎦
n2 n4 −n 2 n3

The coefficients of the rotational mass matrix are given by

n1 = 36 n2 = le (3 − 15 )

n 3 = le2 4 + 5 + 10 2
n 4 = le2 −1 − 5 +5 2

3.1.3 Stiffness Matrix

⎡ ⎤
k1 k2 −k1 k2
E Ie ⎢ k2 k3 −k2 k4 ⎥
Ks = ⎢ ⎥ (12)
+
le3 (1 ) ⎣ −k1 −k2 k1 −k2 ⎦
k2 k4 −k2 k3

The coefficients of the stiffness matrix are given by

k1 =12 k2 =6le k3 = (4 + )le2 k4 = (2 − )le2

3.1.4 Gyroscopic Matrix

⎡ ⎤
n1 n2 −n 1 n2
ρ Ie ⎢ n2 n3 −n 2 n4 ⎥
Gs = ⎢ ⎥ (13)
15le (1 + )2 ⎣ −n 1 −n 2 n1 −n 2 ⎦
n2 n4 −n 2 n3

The coefficients of the gyroscopic matrix are given by

n1 =36 n2 = le (2−15 ) n 3 = le2 4 + 5 + 10 2
n 4 = le2 −1 − 5 +5 2

l e is the shaft length, E is Young’s modulus and I e is the polar inertia of the shaft.

3.2 Disc

The mass and gyroscopic matrices of rigid discs are given by

Vibration Response Prediction in Rotor Systems with External Damping … 263

3.2.1 Mass Matrix

m d1 0
Md = (14)
0 Id1

where m d1 and Id1 are the mass and diametrical mass moment of inertia of the disc.

3.2.2 Gyroscopic Matrix

0 0
Gd = (15)
0 −I p1

where I p is the polar mass moment of inertia of the disc.

3.3 Bearings

3.3.1 Force Due to Bearing Stiffness

Rolling element bearings are isotropic with no cross-coupled stiffness coefficients.

If krad is the radial stiffness of the bearing then the force vector due to stiffness is
 
0.5 krad (u bi + u bi ) + krad (u bi − u bi )
f bki = (16)
0

3.3.2 Force Due to Bearing Damping

Generally, the equivalent viscous damping of rolling element bearings is much lower
than that of journal bearings. If crad is the radial stiffness of the bearing then the force
vector due to damping is given by
 
0.5 crad (u̇ bi + u̇ bi ) + crad (u̇ bi − u̇ bi )
f bci = (17)
0

3.4 Squeeze Film Damper

SFD is commonly used in rotor systems in conjunction with a rolling element bearing
for the purpose of attenuating rotor vibration. The stiffness and damping of SFD are
264 R. S. Srinivas et al.

given in [10] and are repeated here.

2μR L 3 εω
K = (18)
c3 (1 − ε2 )2

μR L 3 π
C= (19)
2c3 (1
− ε2 )3/2

where μ is the viscosity of oil, c is the SFD clearance, ε is the eccentricity ratio (e/
c; e is the journal eccentricity), ω is the angular speed of rotor, R is SFD radius, and
L is SFD land length.

4 Neural Networks

Figure 4 shows the model of a neuron which is the building block for neural networks.
Each neuron receives weighted inputs and produces outputs. Activation functions
or transfer functions are used to perform computation inside each neuron. The
commonly used non-linear activation functions are Tanh, ReLu, and Sigmoid. For
classification problems, Softmax activation function is used while for regression
problems no activation function (linear) is used.
The architecture of the neurons in the network is called the network topology
[12]. It consists of input layer, hidden layer(s), and output layers. A typical neural
network is shown in Fig. 5. A network that has many hidden layers is called a deep
network and a network with many neurons in hidden layers is called wide network.
A classification problem may have two or more neurons in the output layer indi-
cating multiple classes, whereas a regression problem may have one or more neurons
in the output layer indicating the output variables.

Fig. 5 Model of simple

neuron
Output

Activation

Weights

Input
Vibration Response Prediction in Rotor Systems with External Damping … 265

The commonly used training algorithm for neural networks is stochastic gradient
descent (SGD). To begin with, small random weights are initialized in the first
forward pass, where in input values travel through the network activating neurons
in each layer and finally producing output(s) in the last layer. An error is calcu-
lated by comparing the output of the network to the expected output. This error is
then broadcasted back through the network, layer after layer, and weight updating
happens directly proportional to the amount the weights contributed to the error. This
process is called the Back Propagation. This forward–backward process is repeated
for all of the examples in the training data and the network is said to be learning. One
complete round of updating the weights in the network for the entire training dataset
is called an epoch. The number of epochs varies from problem to problem. After a
neural network has been trained it can be used to make predictions. An unseen set
of input features is given and the forward pass makes predictions on this new data.

5 Results and Discussion

The rotor system considered in the present paper is shown in Fig. 6. The geometric
properties of the rotor, bearings, bearing housings, and dampers are shown in Table 1.
The rotor system is made of steel with a mass of 72 kg and a bearing span of 790 mm.
It is discretized into 35 Timoshenko beam elements,
Table 2 lists the input features used to train the neural network. The length and
width of the squirrel cage web shown in Fig. 1 are kept constant and depth is varied.
The length and diameter of the damper shown in Fig. 2 are kept constant while
changing the radial clearance and viscosity of the supply oil. The projection width of
the elastic ring shown in Fig. 3 is kept constant and its height is changed, effectively
modifying the pocket depth δ. Four different unbalance values are considered for
disc-1 and disc-2. The phase angles are chosen such that they excite the in-phase
cylindrical mode and out-of-phase conical mode.

Fig. 6 A typical neural

network Output layer

Hidden layer

Input layer
266 R. S. Srinivas et al.

Table 1 Properties of rotor and housings

Rotor Squirrel cage Elastic ring damper
m sh , kg 30 l, mm 20 D1 , mm 101.6
m d1 , m d2 , kg 21.3 d, mm 2, 2.5, 3 D2 , mm 104
I p1 , I p2 kg-m^2 0.2755 w, mm 3.5 b1 , mm 5
U1 , cm.gm 0.08, 0.16, 0.24, 0.5 n 25 δ, mm 0.2, 0.25
U2 , cm.gm 0.08, 0.2, 0.3, 0.6 Squeeze film damper
c, mm 0.125, 0.15, 0.175
β1 , degrees 0 μ, CentiPoise 2.66, 5
β2 , degrees 0, 180 L, mm 25
Total mass, kg 72.3 R, mm 160
Bearings
Front, N/mm 105,000 Rear, N/mm 500,000

Table 2 Input features considered in training neural network

Input features
d, mm c, mm μ, δ, mm U1 cm.gm β1 degrees U2 cm.gm β2 degrees rpm
centistokes

A total of 72 different combinations of rotor-bearing-damper systems are obtained

by varying the input features shown in Table 2. Keras framework which wraps around
the functionalities of TensorFlow is part of Anaconda distribution and is used to model
the neural network. The response is obtained in steps of 100 rpm up to 20,000 rpm
from unbalance response analysis performed on the finite element model. A total of
14,400 data points or rows are obtained from the simulation. Each row consists of 9
input features and 1 output variable which is front-bearing vibration in the present
case. Out of all the data points the maximum response corresponding to critical
speed due to highest unbalance is 191 µm, while the smallest value of vibration
corresponding to low unbalance at lower speeds is 1 µm.
This large variance in output variables makes it hard to train the ANN model. So
a range of vibration values less than 150 µm and greater than 50 µm has been used
to train the model. An 80: 20 split has been chosen to partition this filtered data set
containing 492 data points into training and test data. A neural network has been
trained using the hyperparameters shown in Table 3. The model is then tested against
the input features of the stand-out data (99 points) to predict the output. Figure 7
shows the predicted versus true response of the front-bearing vibration of test data.
It can be seen that some vibration data points are overpredicted and the rest are
underpredicted. Figure 8 shows the error percentage in the prediction of data points.
Vibration Response Prediction in Rotor Systems with External Damping … 267

Out of the 99 test data points only 8 data points showed more than 10% error. This
shows the effectiveness of the model in predicting the vibration response. The ANN
model can be further improved by (a) adapting dropout and batch normalization, (b)
performing Grid search cross validation, and (c) using scalers such as minmax scaler
and standard scaler. The novelty of the current work lies in the use of geometric
features of bearing housings to train the ANN model. Since geometric features such
as squirrel cage webs, SFD clearance, and damper ring pockets translate into stiffness
and damping capacity of the housings it makes perfect sense to use them as input
features. The future work consists of improving the ANN model and testing the
model using experimental data.

Table 3 Tuned hyperparameters

Number of layers Number of hidden Learning rate Momentum Mini-batch size Epoch
nodes
4 256, 256, 256 0.001 0.9, 0.999 50 1500

Fig. 7 Predicted versus actual value of front-bearing vibration

268 R. S. Srinivas et al.

Fig. 8 Error percentage in the prediction of test data points

Acknowledgements The authors acknowledge T. Ramesh, General Manager (AERDC) for giving
permission to present this work at the conference.

References

1. Brownlee J. Master machine learning algorithms. 2017. p. 162.

2. Rapur JS, Tiwari R. Experimental fault diagnosis for known and unseen operating conditions of
centrifugal pumps using MSVM and WPT based analyses. Measurement. 2019; 147:106809.
3. Tiwari R, Bordoloi DJ, Dewangan A, Blockage and cavitation detection in centrifugal pumps
from dynamic pressure signal using deep learning algorithm. Measurement. 2020;173:108676.
4. Kumar D, Dewangan A, Tiwari R, Bordoloi DJ. Identification of Inlet pipe blockage level in
centrifugal pump over a range of speeds by deep learning algorithm using multi source data.
Measurement. 2021;186: 110146.
5. Gangsar P, Tiwari R. Comparative investigation of vibration and current monitoring for predic-
tion of mechanical and electrical faults in induction motor based on multi class-support vector
machine algorithms. Mech. Syst. Signal Process. 2017; 464–481.
6. Choudary A, Mian T, Fatima S. Convolutional neural network based bearing fault diagnosis of
rotating machine using thermal images. Measurement. 2021;176: 109196.
7. Leontiev M. Damper supports. Aircraft engine department. Moscow Aviation Institute, pp. 1–
49.
Vibration Response Prediction in Rotor Systems with External Damping … 269

8. Han Z, Sing Q, Zhang W. Dynamical analysis of an elastic ring squeeze film damper-rotor
system. Mech Mach Theory. 2019;131:406–19.
9. Han Z, Ma Z, Zhang W, Han B, Dong Q. Dynamic analysis of an elastic ring squeeze film
damper supported rotor using a semi-analytic method. Eng. Appl. Comput. Fluid Mech.. 2020;
1263–1278.
10. Chen WJ, Gunter EJ. Dynamics of rotor-bearing systems. Virginia: Trafford publishing; 2007.
p. 482.
11. Tiwari R. Rotor systems: analysis and identification, 1st ed. CRC press. p. 1092.
12. Brownlee J. Deep learning with python. 2016. p. 266.
Effect of Bluff Body on the Performance
of Flutter Based Energy Harvester

Ankit Agarwal and Ashish Purohit

Abstract In the present work, an experimental study has been carried out to inves-
tigate the effect of bluff body on the performance of flutter based energy harvester.
A cantilever flexible plate is used for the wind tunnel experiment, and PZT types
piezoelectric material is employed for energy harvesting purposes. A high-speed
camera is used to visualize the vibration of the flexible plate, and displacement is
sensed using a piezoelectric transducer. It has been observed that the presence of
bluff body reduces the flutter velocity and improves the performance of the harvester
significantly. The presence of bluff body creates wake disturbance over the plane
structure and causes a drop in critical velocity. The power obtained from the case
of a single sheet in flow and sheet with 20 mm square bluff body is compared and
found that the response corresponding to 10.6 m/s (case I: without bluff body) and
6.6 m/s (case II: with bluff body) flow velocities show an equivalent power output
of 6.67 µW.

Keywords Flutter · Flow-induced vibration · Wind tunnel experiment ·

Piezoelectric energy harvesting · Bluff Body

1 Introduction

To power remotely located, micro-powered electronic devices, where traditional

supply through cables is unfeasible and it is also difficult to service batteries once
the device is installed, wind induced vibration associated with the surrounding envi-
ronment has been identified as a suitable tool. Gauging the potential of flow-induced
vibration (FIV) in solids in terms of energy harvesting, significant research has been

A. Agarwal (B) · A. Purohit

Departmentt of Mechanical Engineering, T.I.E.T, Patiala, Punjab, India
e-mail: [email protected]
A. Purohit
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 271
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_18
272 A. Agarwal and A. Purohit

carried out in the recent past. Based on the encouraging results from initial investiga-
tion, energy harvesting using FIV has gained significant popularity in the last decade
[1, 2]. A flexible structure subjected to flow loses stability exhibits high amplitude
oscillation [3] and generates electric charge when piezoelectric material is associ-
ated. The output of a piezoelectric structure is highly dependent on strain distribution,
frequency of oscillation, circuit resistivity, etc. The paradox in materializing these
means to improve output is that it requires higher flow velocity, which in turn restrains
the applicability of the harvester at normal outdoor wind conditions. Therefore, in
order to truly enhance the power output of flow-energy harvester, serious efforts to be
made to understand the complexity of coupled system and, hence, suitable changes
can be made to improve power output at ambient flow velocity.
In the present scenario, different excitation mechanisms i.e., wake galloping,
transverse galloping, flutter, etc. are utilized to harvest energy; nevertheless, each
mechanism shows relative advantages and limitations when implemented at real
ambient conditions. The wake galloping and vortex-induced vibration differs based
on the location of the oscillator with respect to the upstream body. In the wake
galloping, trailing structure is excited due to shed vortices, wherein in VIV, structure
is subjected to periodic lift force by the action of its own vortex shedding [4, 5]. Flutter
instability is caused by the flow elastic coupling of the attached flow and occurs over
a certain flow velocity. It can be seen in a variety of conditions, e.g., a gentle breeze
over a flexible flag creates instability in the flag, a gust creates flutter in airplane
wings, etc. Typically, flutter is characterized by the coupled mode vibration e.g.,
coupling of bending and torsion modes [6]. A lot of research work has been reported
in surveys on reckoning of flutter in flexible plates, papers, flags, membranes, etc.,
but few studies on energy harvesters have been reported [7, 8].
From the existing work, it is observed that majority of work is done on unstable
flutter vibration of plane structure. Most of these works is emphasized on investi-
gation of flow dynamics of coupled systems and suppression of flutter. However,
limited research is reported in the context of energy harvesting. It is also noted that
the presence of an upstream bluff body results in reduction of critical velocity of
flutter, but there is a lack of its detailed exploration. This reduction can be further
employed to improve the output power of piezoelectric harvester. Hence, by noting
the present demand of utilizing non-conventional sources of energy for useful work, a
detailed investigation on the effect of various parameters i.e., the role of front obstacle
on the vibration characteristics of a flexible structure in flow is warranted. In this
proposed work, an experimental investigation of vibration of a plate like structure in
the presence of an upstream obstacle is presented.

2 Methodology and Experimental Details

All experiments are performed on a plane laminated sheet that is made by layering a
paper sheet between two polythene films. The sheet used has dimensions of 195, 295,
and 0.28 mm as length, width and thickness, respectively. The sheet is mounted in
Effect of Bluff Body on the Performance of Flutter Based Energy Harvester 273

Fig. 1 Schematic top view of the test section of experimental setup

a conventional flag type orientation; the fixed end is towards the upstream side. The
cantilever structure is created by fixing the leading edge between two metal strips
of a thickness of 1.5 mm. Figure 1 depicts a schematic top view of the experimental
setup. The top and bottom wall of the test section test model is rigidly clamped.
Figure 2a–c shows wind tunnel experimental setup, flexible plate without bluff body,
and cantilever plate with 20 mm square bluff body. To minimize the impact of self-
weight, the plane of the sheet is vertically aligned. Before, with tunnel testing, the free
vibration frequency of the cantilevered sheet is measured by solving finite element
model based on the Euler–Bernoulli beam equation [9] Eq. 1, and found first three
bending modes corresponding to three natural frequencies as 2.1, 11.1, and 36.4 Hz.

E I ϕ '''' + ρ Aϕ̈ = 0 (1)

All of the experiments were carried out in a low speed wind tunnel with a section
size of 0.4 m × 0.4 m at the fluid machinery laboratory at TIET, Patiala. To ensure
laminar flow at inlet and uniform flow condition, a honeycomb structure is used
before the test section. For measuring flow velocity, hotwire anemometers (ESF-
35–2, Greystone energy system) are installed at inlet and Pitot tubes areplaced at the
outlet of the test section. The flow velocity is controlled by a variable frequency drive
(Atsel VFD 3HP) with a minimum change of 0.2 m/s.

2.1 Case 1: Flutter Prediction of Cantilever Flexible Plate

Without Bluff Body

The vibration response of the cantilever flexible sheet is noted by incremental change
in the flow velocity varying from 0 to 12 m/s. It is observed the cantilever flexible sheet
does not appear to be disturbed until the flow velocity hits 10.2 m/s. Additionally,
there is some movement at a flow velocity of 10.4 m/s, and abrupt high amplitude
oscillations are observed at 10.6 m/s flow velocity. Figure 3a shows the time history
274 A. Agarwal and A. Purohit

Fig. 2 a Wind tunnel experimental setup, b Flexible plate without bluff body, and c Flexible plate
with 2 cm square bluff body

and Fig. 3b frequency spectrum of the captured signal from the PZT element at a
rate of 10,000 samples/s. Higher harmonics of the frequency spectrum can be seen,
and 17.3 Hz is the dominant frequency. The harmonics are expected because the
piezo-element is not attached to the sheet surface, and also multiple harmonicas
in the frequency spectrum exist due to periodic motion. However, the frequency
spectrum suffices the objective of knowing oscillation frequency which has been
further verified by the sequence image processing of recorded video. The sequential
images of the sheet vibration taken at various times are displayed in Fig. 3c. The figure
illustrates that the vibration mode is mostly a combination of first and second bending
modes of cantilevered structure, wherein the second mode frequency dominant the
vibration. It is also observed that frequency increases slightly with increase in the
flow velocity. Additionally, the visual observation supports the presence of minor
torsional mode.
The observed critical velocity is verified with the standard numerical result [3]
reported in the literature and a good matching (less than 3% variation) is noted.
Effect of Bluff Body on the Performance of Flutter Based Energy Harvester 275

Fig. 3 a Voltage variation with time, b frequency spectrum, and c images of vibration at different
time instances

2.2 Case 2: Flutter Prediction of Cantilever Flexible Plate

with 2 cm Square Bluff Body

The experiments are now carried out again by placing an upstream square section
body in the leading direction. When the flow velocity reaches 6.4 m/s, it is noted that
there is no appreciable sheet movement. However a small level oscillation is noticed
at about 3 m/s velocity, and it again disappears completely as flow velocity increases.
From the visual observation, an equivalent initial bending mode oscillation of the
sheet is demonstrated, which may have resulted from vortex-induced oscillation. A
high amplitude vibration is seen when the flow is further increased. Figure 4a shows
the voltage variation and Fig. 4b spectrum of the captured voltage signal from the
PZT. The frequency spectrum displays a dominant frequency component of 14 Hz,
along with its higher tones, which shows the periodic nature of the signal. Figure 4c
shows superimposed image of the sheet motion. It can be noticed that there is a
reduction of critical flutter speed in the presence of bluff body, as compared to the
flutter observed for the case without bluff body. The probable reason of change in
the critical velocity due to the presence of wake flow from the upstream body is an
276 A. Agarwal and A. Purohit

Fig. 4 a Voltage variation with time, b frequency spectrum, and c images of vibration at different
time instances with 2 cm square bluff body

additional pressure gradient on the surface of the trailing sheet. The alternate pressure
field on both top and bottom sides causes early flutter than observed in the normal
condition.

3 Energy Harvesting

In this section, energy harvesting aspect from the flow governed vibration of the sheet
is investigated and a comparison is made for the two reference cases of vibration of
the sheet in plane flutter instability and flutter in the presence of bluff body. Six PZT
elements are used in series connection for this purpose and placed at root area of
the sheet, as illustrated in Fig. 5. The partially sandwiched piezoelectric material
is anticipated to be stretched during vibration and will generate charge. The output
voltage has been recoded using digital oscilloscopes (Siglent SDS1102CML up to
100 MHz frequency). The peak to peak of voltage was measured over a period of 1 s
Effect of Bluff Body on the Performance of Flutter Based Energy Harvester 277

Fig. 5 Flexible plate along

with six piezoelectric
elements

through the X10 probe. The power is measured across a calibrated resistor that is
found by first comparing output across a number of resistors ranging from 20 to 100
kΩ.
For the maximum power output, when the internal resistance is equal to the load
resistance, harvester produces maximum power, and we find that 39 KΩ is the optimal
load resistance at which a 6.67 µW maximum power output. Figure 6 (a-b) depicts
the effect of external load resistance (R L ) on the flutter based output power and
comparison of output power with and without bluff body. In experiments, every test
is performed at least five times to reduce measurement error, and the average value
is used for post data analysis. The summary of the FEH at 39 KΩ load resistance is
shown in Table 1.
The average output power P was calculated using the following relationship.

V pp
Ve f f = √ (1)
2 2
Ve f f 2
P= (2)
RL

Figure 6b shows the power generated. The results obtained show that the presence
of 20 mm bluff body (square shape) produces an equivalent level of power at 37.7%
lower velocity (6.6 m/s in case II) when compared with case I (without bluff body)
at 10.6 m/s flow speed. It is established that the introduction of additional wake
disturbance from an upstream object lowers the critical flutter. From the point of
278 A. Agarwal and A. Purohit

Fig. 6 a Load resistance vs power output, and b power output at different flow velocity with and
without bluff body

Table 1 Summary of the

Bluff body Frequency Critical velocity Power
FEH at optimum load
(Hz) (m/s) (µW)
resistance
Without bluff body 17.32 10.6 6.52
2 cm 14 6.6 6.67

view of performance, the input energy reduces significantly at lower velocity and
enables the harvester to operate at normal ambient conditions.

4 Conclusions

In this paper, an investigation of energy harnessing from flow-induced instability is

performed. Flutter is a potential means to convert mechanical energy into electrical
energy, however, it is less explored as high flow velocity is required. This work
proposes a novel solution to improve the performance of an FEH by reducing the
threshold velocity of flutter at the approachable range of flow at ambient conditions.
In this study, the effect of bluff bodies on cut off velocity of flutter of a cantilever
laminated plate is first studied and then a comparison of energy harnessing between
the traditional flutter and flutter in the presence of a bluff body is carried out. Wind
tunnel testing is performed on a test model of a laminated cantilever sheet. The
sheet is allowed to oscillate in wind tunnel at different levels of flow velocity. It has
been found that cut off velocity of flutter reduces up to 37.7% by using an upstream
bluff body (square shape). When the flutter arrives, the cantilever structure shows
Effect of Bluff Body on the Performance of Flutter Based Energy Harvester 279

vigorous oscillation in combination of first and higher bending mode. The high-
speed image processing indicates that in both the cases, the level of tip displacement
is comparable however, in the case of bluff body, both first and second modes are
evident in the oscillation, however, in the case without the obstacle, the second mode
is dominated in vibration. From energy harnessing aspect, a noticeable increase in
the power efficiency, while compared in terms of input flow velocity, is observed.

References

1. Sanchis A, Sælevik G, Grue J. Study of vortex induced vibrations in two degrees of freedom on
a spring-mounted circular cylinder with PIV . J Fluids Struct. 2004;Viv:1–4.
2. Robbins WP, Morris D, Marusic I, Novak TO. Wind-generated electrical energy using flexible
piezoelectric materials. Am Soc Mech Eng Aerosp Div. AD 2006;1–10. https://doi.org/10.1115/
imece2006-14050.
3. Manela A, Howe MS. The forced motion of a flag. J Fluid Mech. 2009;635:439–54. https://doi.
org/10.1017/S0022112009007770.
4. AkaydIn HD, Elvin N, Andreopoulos Y. Wake of a cylinder: a paradigm for energy harvesting
with piezoelectric materials. Exp Fluids. 2010;49(1):291–304. https://doi.org/10.1007/s00348-
010-0871-7.
5. Jung H, Lee S. The experimental validation of a new energy harvesting system based on the
wake galloping phenomenon. 2011. https://doi.org/10.1088/0964-1726/20/5/055022.
6. Huang L. Flutter of cantilevered plates in axial flow. J Fluids Struct. 1995;9(2):127–47. https://
doi.org/10.1006/jfls.1995.1007.
7. Jiang J, Tang L, Paı MP. Cantilevered flexible plates in axial flow : energy transfer and the
concept of flutter-mill. 2009;326:263–276. https://doi.org/10.1016/j.jsv.2009.04.041.
8. Tang DM, Yamamoto H, Dowell EH. Flutter and limit cycle oscillations of two-dimensional
panels in three-dimensional axial flow. J Fluids Struct. 2003;17(2):225–42. https://doi.org/10.
1016/S0889-9746(02)00121-4.
9. Rao SS. Mechanial vibration, vol. 33, 5th ed., no. 6. 1981.
A Systematic Review of Rotor Unbalance
Diagnosis in Rotating Machinery Based
on Machine Learning Algorithms

Sudhar Rajagopalan, Ashish Purohit, and Jaskaran Singh

Abstract Rotor unbalance initiates high vibration, which gives rise to other faults
that lead to reduction in equipment life, loss of energy, and productivity. To date,
various machine learning algorithms have been proposed in the literature as they
have proven to be fruitful in detecting rotor unbalance and this research topic is fast-
growing. Therefore, this present paper aims at a comprehensive review of the develop-
ment and advancement of machine learning algorithms in the field of vibration-based
diagnostics of rotor unbalance in the last decade. Advantages and disadvantages of
the reviewed machine learning algorithms have been discussed for academic as well
as industrial operations. Finally, possible challenges in the implication of machine
learning algorithms for rotor balancing have been examined in detail as results and
hence, inferences have been drawn.

Keywords Machine learning · Artificial intelligence · Rotor unbalance · Fault

diagnosis

Nomenclature

ANN Artificial neural networks

BP Backpropagation
CAE Computer-aided engineering
CNN Convolutional neural network
DBN Deep neural network
DFT Discrete Fourier Transform

S. Rajagopalan (B) · A. Purohit · J. Singh

Department of Mechanical Engineering TIET, Patiala, Punjab, India
e-mail: [email protected]
A. Purohit
e-mail: [email protected]
J. Singh
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 281
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_19
282 S. Rajagopalan et al.

FCM Fuzzy c-means clustering

FEM Finite element analysis
GPC Gaussian processes classification
GW Gigawatt
IVHM Integrated vehicle health management
KDA Kernel discriminant analysis
KLT Karhunen–Loeve transform
KNN K-nearest neighbors
LDA Linear discriminant analysis
LSTM Long short-term memory
LSTMAM LSTM with attention mechanism
MFCCs Mel-frequency cepstral components
MLP Multilayer perceptron
MW Megawatt
NN Neural network
PCA Principal component analysis
PCC Pearson correlation coefficient
PNN Probabilistic neural network
PSD Power spectral density
PSMG Permanent magnetic synchronous generator
RBF Radial basis function
RNN Recurrent neural network
SBM Similarity-based model
SVDD Support vector data description
SVM Support vector machine
UAV Unmanned aerial vehicle
WPT Wavelet packet decomposition

1 Introduction

We are living in a modern era where high-speed rotating machines are playing a vital
role to keep our daily life comfortable. High-Speed rotating machines are used in
numerous industries [1] such as aero engines, thermal power plants, nuclear power
plants, hydropower plants, manufacturing industries, petrochemical industries, oil
and gas industries, etc., and their failure causes a huge impact on the asset, brand
image as well as on the environment and the human life. The rotor system under-
goes failure due to various reasons such as misalignment, mass imbalance, bearing
failures, seal/bearing rub, housing and foundation bolt looseness, rotor instability,
fatigue fracture, etc. Out of the many reasons, the mass imbalance is one of the
common problems noticed frequently. Mass imbalance causes excessive vibration
which affects the connected rotor bearing components, which may lead to perfor-
mance limitation or in certain cases to catastrophic failure of the machines [1, 2].
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 283

Even at normal operation conditions, the heat generation, wear, erosion, etc., [3]
cause loss of mass from the rotor system that leads to the static, couple, quasi-static
as well as dynamic imbalance [4–6], which further led to other problems such as
subharmonic resonance, variation in critical speed, rotor rub, elevated mass imbal-
ance, and imbalance-induced rotor instability [7, 8]. In summary, research on rotor
dynamics has gained attention due to the industrial needs of a high-speed rotor system
for high performance [9–11] and to be in compliance with environmental, health,
and safety norms [12, 13].
In the past, many researchers have focused on the identification of imbalance influ-
ence on cracks, imbalance identification in combination with other problems, and
case studies which influenced imbalance and caused catastrophic failures [14–23]. In
a comprehensive review, Laws has reported many case studies in which rotor instabil-
ities do get influenced by the residual mass imbalance even a small amount is left gets
aggravated and causes catastrophic failures [24]. Recently, different approaches have
been employed to identify the mass imbalance of actual industrial rotating machinery
to avoid catastrophic failure in different applications, viz., 125 MW turbo-gas gener-
ator in combined cycle plants in Italy [25], 320 MW steam turbogenerator unit in
Italy [26], and 1.3 GW (Gigawatt) steam turbo-generator in nuclear power plant
[27]. The dynamically balanced rotor not only overcomes the possibilities of many
potential failures but also ensures reliable, safe, and productive operation. The diag-
nosis and prognosis for the unbalance is the fundamental concern in the design and
operation of the high-speed rotor bearing system. Many high-speed rotor bearing
systems are equipped with vibration monitoring system for rotor shaft vibration and
casing vibration. Researchers developed automatic fault diagnostic and prognostic
systems using statistical, model-based, combined modeling and statistical, artificial
intelligence, and machine learning approaches [28].
Fault detection is broadly classified into vibration-based (signal-based and model-
based), modal testing, and non-traditional methods [29]. Vibration-based and modal-
based fault detection requires lots of effort to collect the data, domain knowl-
edge, and expertise in the diagnosis of fault with the involvement of human inter-
vention physically for data collection and analysis. The scarcity of skilled engi-
neers with diagnosis capability and the continuous online fault detection and diag-
nosis requirement opened the pathway for non-traditional fault detection methods.
Researchers have focused on the development of non-traditional fault diagnosis and
prognosis methods using artificial intelligence, advanced signal processing tech-
niques, borescope inspection, etc. The speed of data collection and processing,
storage, and computational accuracy have reached peak in the current decade. More-
over, with the advancement in the sensor technology and manufacturing process, the
industrial critical and semi-critical machines are now equipped with various sensors
to collect data like pressure, flow, temperature, vibration, current, voltage, etc. As
Industry 4.0 brought the change of data-driven maintenance strategies, the industries
realized the need to achieve fast, accurate, and cost-effective diagnosis and prognosis
solution [30] based on AI, machine learning, and deep learning approaches.
284 S. Rajagopalan et al.

This review is primarily focused to explore mass imbalance diagnosis and prog-
nosis for high-speed rotor using machine learning algorithms for the period of 2014–
2021. The reader firstly will get the comprehensive understanding about machine
learning algorithms, types of algorithms, and their application in rotor mass imbal-
ance diagnosis and prognosis along with advantages and disadvantages. Secondly, the
various techniques and methods used for the development of fault detection system
are discussed with their performance and efficiency. Section 2 explains in detail
about non-traditional machine learning approach. The various algorithms and their
relevant studies are reviewed with their data collection techniques, method of diag-
nosis, and prognosis used for mass imbalance with their performance and efficiency.
Section 3 focuses on the advantages and disadvantages of the algorithms. Section 4
concludes with the remarks observed by comparing various articles reviewed for
mass imbalance.

2 Non-traditional Method in Mass Imbalance Diagnosis

and Prognosis: Machine Learning

In the traditional method, the vibration-based fault diagnosis is done using signal-
based and/or model-based methods. But in non-traditional methods, such as artifi-
cial intelligence, machine learning, and deep learning, gained attention due to their
automatic fault diagnosis capabilities. The advancement in the field of electronics,
instrumentation and computer engineering, and Industry 4.0 brought the need usage
of data-driven maintenance with the help of AI. Machine learning is the subset of
artificial intelligence that involves the design and development of algorithms that
make computers eligible to learn based on data, for example, data from the sensors
or databases. Hence, machines can automatically learn to understand complicated
patterns and make smart choices depending on the data. Machine learning mainly
comprises of fields like statistics, data mining, pattern recognition, and artificial
intelligence [31].
Machine learning is subdivided into various types such as supervised learning,
unsupervised learning, semi-supervised learning, neural network, deep learning,
instance-based learning, etc., as shown in Fig. 1 [32]. The machine learning algo-
rithms are trained by using the input parameters for the known output condition. Then
they are tested to find their effectiveness in predicting the output condition for the
unknown input conditions. The input variables are collected from the rotor bearing
systems, such as vibration signals, current, power, voltage, temperature, flow, etc.
After the training and testing stages of machine learning algorithms, they are used
to predict output conditions using known input parameters.
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 285

Machine
learning

Semi- Neural Instance-

Supervised Unsupervised Supervised
Network Deep Learning Based
learning learning learning Learning

Fig. 1 Types of machine learning [32]

2.1 Supervised Learning

Supervised learning algorithms need external assistance during the learning process.
Training and testing are the major requirements of this kind of learning to predict
the outcome of the model, as per classification. The training data is labeled data.
Testing is the prediction based on the training. Algorithms such as decision tree,
naïve Bayes, support vector machine, KLT algorithm, and CART are examples of
supervised learning [32]. The important research works about rotor mass imbalance
are discussed in the below examples.
Deng et al. [33] proposed a modified classification and regression Tree (D-CART)
algorithm to predict the presence of imbalance and location of imbalance and quan-
tify the imbalance using dichotomy with high speed and accuracy. The author also
compared with other improved CART algorithms, as shown in Fig. 2, and found
that computation is 23.92% faster than the Adaboost-CART algorithm with 96.77%
accuracy. Test data collected from the rig has the antifriction bearing with proximity,
speed, rotational, and acceleration sensors. In the pre-processing, the amplitude and
acceleration data, i.e., timewave form data used as input to the machine learning
algorithm because the D-CART machine learning algorithm can handle continuous
and intermittent data without discretization.

Fig. 2 Comparison of the CART and D-CART algorithms[33]

286 S. Rajagopalan et al.

de Moura et al. [34] utilized supervised (Karhunen–Loeve transform (KLT), NN,

and Gaussian classifier) and unsupervised (K-means) machine learning algorithm in
the combination of detrended fluctuation analysis to predict wind turbine 7 imbal-
ance conditions for a rotational frequency using vibration signals. The experimental
setup was a scaled wind turbine with acceleration sensors used for signal collection,
and features from this acceleration data were extracted as the input to the machine
learning algorithm. The supervised algorithm could perform well and predict its
mass imbalance as output, specifically KLT, with 99.8% accuracy compared to the
supervised algorithms. However, other algorithms have shown average accuracy of
more than 97% in the prediction of mass imbalance.
Hübner et al. [35] proposed wind turbine imbalance identification using the rotor’s
speed fluctuation. Figures 3 and 4 explain the proposed methodology. The speed of
the wind turbine is estimated using PSMG’s current and voltage data. Then the wind
turbine speed fluctuations were trained for different mass imbalances using SVM-
based machine learning algorithms. This trained data of mass imbalance position and
type of mass imbalance is stored in the Azimuth database. The SVM receives pre-
processed current and voltage data as input and outputs the mass imbalance severity
and mass imbalance location by correlating with the Azimuth database.
Meng et al. [36] established the benefits of orbital analysis in machine learning
as it is a classic fault identification tool that can be utilized in machine learning pre-
processing of feature extraction. The orbital-based extracted features as input to the
machine learning algorithm provide the best prediction for mass imbalance identifica-
tion. The SVM algorithm was used to train and test the orbital features. Furthermore,

Fig. 3 The schematic diagram from training and testing the algorithm for detecting the magnitude
and position of the mass imbalance through computational simulation [35]

Fig. 4 The schematic diagram for detecting the mass imbalance position in real wind turbine. The
last block is the same used in the computational simulation [35]
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 287

the SVM algorithm compared orbital features extracted accuracy with the accuracy
extracted from the frequency domain using WTP and PCA. The comparison proved
that the SVM algorithm, which is orbital-based, outperformed in identifying the mass
imbalance with an accuracy of 92.5%–100%, depending upon training and testing
sample size.
Pinheiro et al. [37] explored the SVM machine learning algorithm for mass imbal-
ance diagnosis using an FFT-based extracted feature as an input. The acceleration
data is fed into pre-processing, and then 60 features were used as input to the SVM to
predict 7 types of mass imbalance conditions. The SVM could identify the imbalance
with 90% accuracy.

2.2 Unsupervised Learning

Unsupervised learning algorithms could learn a few features initially and then use
the previously learned features while new data is introduced to the algorithm it
understands to classify them. These algorithms are mainly used for feature reduction
and clustering of the data [32]. Algorithms such as K-means clustering, principal
component analysis, and fuzzy C-means clustering (FCM) are widely used machine
learning algorithms; some important research works based on these algorithms for
rotor mass imbalance are discussed in the below examples.
Żabiński et al. [38] tested the seven algorithms to predict the mass imbalance
for accuracy, sensitivity, and specificity. The time domain features and FFT-based
frequency domain features were extracted. From 14 features using only 3 features
as the input to the machine learning algorithms, mass imbalance with 4 classes was
predicted. All seven machine learning algorithm’s performances were found equally
well and high, but k-mean scored at all classes of imbalance as 1. Thus, this research
suggests that real-time implementation to detect mass imbalance is possible on CNC
machines by selecting the algorithm requiring low computational power.

2.3 Semi-supervised Learning

Semi-supervised learning takes the advantage of supervised and unsupervised algo-

rithms. Thus, semi-supervised machine learning algorithms are effectively utilized
if unlabeled data is available at the beginning of the learning process and receiving
labeled data later, which are scarce, hence used in applications such as machine
learning and data mining [32]. Some of the algorithms in this type are boosting and
bagging algorithms.
288 S. Rajagopalan et al.

2.4 Ensemble Learning

Ensemble learning is a collection of individual learning algorithms to form only

one algorithm to learn and make decisions. As all algorithms are best in their way,
combining them as one learning algorithm enhances prediction capabilities [32]. The
relevant field research works are presented below.
Duan et al. [39] introduced a new method to predict mass imbalance by combining
the support vector data description (SVDD) with the binary tree, as shown in Fig. 5.
The Mahalanobis distance used in SVDD and parameters optimized using particle
swarm optimization algorithm. For classification, SVDD was used in the binary
tree structure from top to bottom. The unbalance data from proximity sensors on
sleeve bearings were pre-processed using multifractal detrended fluctuation and
extracted eigenvalues were sent to machine learning algorithms as input. The devel-
oped machine learning algorithm compared with other traditional algorithms such
as BP neural network, SVM, and FMC. The author concludes that the proposed
algorithm outperforms also could achieve 95% classification accuracy.
Martins et al. [40] proposed two different classification techniques to predict rotor
mass imbalance. The SBM and KDA were combined with a random forest machine
learning algorithm, as shown in Fig. 6. The time and frequency domain features were
used as input to the machine learning algorithm for three imbalance conditions and
considered angular velocity to quantify the mass imbalance. The experimental setup
was used with acceleration sensors and a tachometer. The classification accuracy of
the proposed method was found to be approximately ±0.42% lesser compared to the
random forest method. However, feature space reduction using KDA comes down
from 31 to 6, giving advantages such as faster computation with less computational
load for mass imbalance prediction.

Fig. 5 Framework of the proposed method [39]

A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 289

Fig. 6 Random forest algorithm diagram [40]

2.5 Instance-Based Learning

In this learning algorithm, once the particular pattern is learned, the algorithm applies
the same pattern to new data. These algorithms hold training data until get the testing
data to work together; one way it’s a lazy learner. The algorithm’s complexity depends
upon the data volume [32]. The k-nearest neighbor (KNN) algorithm is the best
example of instance-based learning. In instance-based learning, The mass imbalance
identification using KNN gives better clarity on its fault identification capacity.
Jablon et al. [41] introduced features extracted from orbit plots as input to the
machine learning algorithm to identify the rotor mass imbalance on the antifriction
bearing. The features were extracted from the orbit constructed from the vibration
sensors. The antifriction bearing system was used to extract the orbital informa-
tion due to its vast usage in the industry, as shown in Fig. 7. The author addressed
the difficulties of extracting useful information from the antifriction bearing. The
author utilized the information multiplicity present in the orbits from the acceleration
and proximity sensor with PCA for feature extraction. The extracted features were
given as input to KNN and SVM algorithms, and their performance was compared.
The effectiveness of the algorithms was tested against the signal-to-noise ratio from
the test rig data with three different levels of mass imbalance under normal oper-
ating conditions. The author proved that KNN performance is better than SVM and
concluded that KNN is suitable for industrial environments. The KNN could predict
up to 85% accuracy for different SNRs, while with less than -25 dB algorithm could
provide 60% accurate classification.
290 S. Rajagopalan et al.

Fig. 7 Flowchart [41]

Gohari and Eydi [42] carried out studies to predict mass imbalance value, location
of the disc, and location of the disc from the axis center using KNN and decision
tree. Acceleration sensors from multidisc rotor used for data collection with IR phase
sensor. Time and frequency domain features were extracted from the collected data,
and these features were given as input to the machine learning algorithm. The KNN
was compared with the decision tree and found KNN outperforming with the accuracy
of 87% for identifying the location of the disc and 88% for distance from center to
imbalance mass.
Yan et al. [43] explored the efficiency of SVM and KNN for the imbalance predic-
tion of the rotor using the frequency domain features for three different imbalance
conditions. The acceleration sensors were used to measure the signals. For the 29
features extracted and given as input to the machine learning algorithms, the author
observed prediction accuracy for SVM as 95.87% and KNN as 77.51%. The author
concludes that SVM has better prediction accuracy, but KNN is found much faster.
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 291

2.6 Neural Network

The neural network concept is developed from the biological concept of neuron’s
functional mechanism. There are many neurons interconnected to arrive at final
decisions based on many inputs, and their weights and nodes are connected to a
complex form of hidden layers [32, 44], as shown in Fig. 8. Algorithms such as
artificial neural network (ANN), multilayer perceptron (MLP), radial basis function
(RBF), and probabilistic neural network (PNN) are examples of neural networks.
Kornaev et al. [45], to predict rotor mass imbalance, extended a similar data
collection system and the ANN architecture which was used in thermal wedge and
bubble formation in sleeve bearing fault detection. The author used the combined
measurement data like displacement, pressure, temperature, resistance ratio, flow
rate, and torque. The pre-processing of the signal for its smoothness was carried
out using the third-order Savitzky–Golay filter. And, this was given as input to the
ANN algorithm. The ANN algorithm predicts three different imbalance conditions
with 97.3% accuracy. It is also observed that the data volume and sensor numbers
influence the algorithm’s accuracy.
Malik and Mishra [46] used a backpropagation neural network (BPNN) for wind
turbine imbalance fault prediction. The input current signal to the algorithm was
simulated and fed into pre-processing for feature selection and noise elimination
through the RapidMiner-based PCA method. The BPNN algorithm could perform
better than power spectral density with 99.12% of accuracy during the testing phase.
Dwi, Alfaradin, and Darojah [47] developed an algorithm based on ANN and
tested the data collected from multidisc rotor with four mass imbalance conditions.
The algorithm either provides early warning of the fault or localizes the discs’ imbal-
ance. The ANN uses the time and frequency domain features extracted by statistical
and DFT methods to predict mass imbalance. The ANN can provide a classification
accuracy of 90.12% for online and offline mass imbalance prediction.

Fig. 8 Schematic architecture of an ANN [44]

292 S. Rajagopalan et al.

Pavlenko et al. [48] proposed an identification method for mass imbalance predic-
tion for liquid rocket engine’s turbopump unit. The author integrated numerical and
analytical parameters of the rotor system dependencies with ANN. The deflection
and imbalance vectors are obtained from the FEM calculations for different oper-
ating speeds and given as input to ANN for training and testing, and then prediction
accuracy is calculated. The experimental imbalance data was then compared with
the ANN. This unique system eliminates the initial balancing trial required for the
critical systems, and the authors suggest system can be utilized for virtual balancing
of the rotor system.
Iannace et al. [49] carried out a noise-based prediction of rotor blade mass imbal-
ance of unmanned aerial vehicle (UAV) using the machine learning algorithm of
ANN. The experiment was carried out in the laboratory to collect data using an
acoustic sensor during take-off. The data was collected for three different mass imbal-
ances and normal operating conditions. The frequency domain features are extracted
and given as input to the ANN. The ANN found that the accuracy of imbalance
detection was 97.63%.
Tajik et al. [50] authors studied the prediction of mass imbalance of a 162.1 MW
gas turbine using PNN, MLP, and RBF neural network algorithms. The PCA and
LDA classifiers are used to reduce the dimensionality of the input data and then
reduced dimensional data is given as input to the machine learning algorithm. The
false and miser alarm rate is used to evaluate the performance of the PNN algorithm
with PCA, and MLP & RBF algorithms with LDA are found to perform well.
Gohari et al. [51] studied mass imbalance prediction through ANN. The input
data to the ANN was from the modeled and obtained from the rotor rig with four
discs. The acceleration values from the time domain were fed as input data to the
ANN. The ANN algorithm could predict accurately for mass imbalance location up
to 96% and imbalance value up to 94%.
Walker et al. [52] studied the prediction of different mass imbalance localization
scenarios using ANN for multiplane rotor rig considering system nonlinearities. The
study was carried out on a single and multiplane rotor rig with four different masse
imbalance conditions. The data from the acceleration sensor was used to extract the
frequency domain data and given as input to the ANN. The goal of this research is
to implement the study techniques in integrated vehicle health management (IVHM)
systems.
Carbajal-Hernández et al. [53] proposed orbit-based mass imbalance prediction
system for induction motor using ANN. The acceleration data were extracted from
different induction motors to get orbit patterns. The collected time domain data were
pre-processed using the Butterworth passband filter and given as input data to the
ANN. The noise is also added to the signal to generate a similar kind of signal, which
is usually available in harsh industrial environments. The ANN could generate the
accuracy for mass imbalance fault detection without noise up to 99.53% and with
noise up to 92.55%.
Junior et al. [54] proposed machine learning algorithms for wind turbine mass
imbalance identification and prediction. The ANN, KLT, and Gaussian discriminator
algorithms with a combination of detrended fluctuation analysis. The acceleration
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 293

vibration data was collected from the scaled wind turbine and pre-processed, and
then this data was fed to the machine learning algorithms. It is observed that all the
algorithms show more than 95% prediction efficiency.
Pavlenko et al. [55] integrated computer-aided engineering software capabilities
with machine learning to explore the feasibility of virtual balancing. This integration
opens a pathway for virtual balancing by considering rotor bearing system nonlinear-
ities, stiffness dependencies, the relationship between deflection direction and mass
location, and the critical speed of the rotor. Thus identifying imbalance mass and
its value as output is predicted from ANN through a combination of CAE-based
nonlinear dependencies relation along with the experimental deflection.

2.7 Deep Learning

Deep learning is a subset of machine learning. Deep learning also utilizes the neural
network concept of the human brain, but it is deeper than the neural network. When
the hidden layers are more than three layers inclusive of input and output [56], as
shown in Fig. 9, it is called a deep neural network. The algorithms such as LSTM,
CNN, and DBN are discussed with the relevant experimental work in this section.

Fig. 9 Deep neural network [56]

294 S. Rajagopalan et al.

Fig. 10 The overall structure of the imbalance fault detection [57]

Chen et al. [57] proposed the LSTM combined with attention mechanism
(LSTMAM), as shown in Fig. 10, for the wind turbine blade mass imbalance predic-
tion. The mass imbalance happens due to ICE formation in harsh environments. Inter-
estingly the input parameters to the machine learning algorithms used are not vibra-
tion but the power, voltage, current, wind speed, and torque of the hub. The authors
used simulated wind turbine time series parameters as an input to the algorithm. Three
different mass imbalance conditions on wind turbine blades were compared with
different machine learning algorithms such as RNN, SVM, GPC against LSTMAM,
and LSTMAM predicted best and up to 99.8% accuracy.
Mey et al. [58] authors studied different algorithms, such as CNN, random forest
classifications, etc., for their performance to predict the mass imbalance with and
without pre-processing of data. Also, the authors published data in the public domain
and the Python codes used in the research. The four different mass imbalance condi-
tions were tested. The CNN with FFT transformed frequency domain data as an
input to the algorithm provides better performance, up to 98.6%, compared with
other algorithms and CNN with the raw signal. The process is shown in Fig. 11.
Yan et al. [43] explored the multi-DBN, as shown in Fig. 12 which can take
multiple sensors from the machine, such as acceleration and displacement sensors,
to form heterogeneous information. The algorithm compared with vibration sensor
based on frequency domain feature which was used as input to machine learning
algorithms with DBN, orbital feature as input DBN, and then both the vibration and
orbital feature to the multi-DBN. The comparison proved that multi-DBN performs
better. So, the authors conclude that multi-DBN can predict mass imbalance up to
100% accuracy with a testing and training samples ratio of 1:9, and even with only
two testing data, it can predict up to 86.46%.
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 295

Fig. 11 Sketch of the used neural network architecture for the classification of the raw vibration
samples. Nconv describes the number of hidden convolutional and pooling layers used [58]

Fig. 12 Diagnosis procedure based on the multi-DBN model with multi-source heterogeneous
information [43]

3 The Advantages and Disadvantages of Machine Learning

Algorithms

In general, machine learning algorithms require pre-processing and feature extraction

on the input vibrational signal, and then these features are effectively used in the
diagnosis and prognosis purpose through testing and training. However, the subset
of machine learning deep learning differs from this aspect; it has the ability to extract
296 S. Rajagopalan et al.

the features required for diagnosis and prognosis purposes. The various machine
learning algorithms used in rotor mass imbalance diagnosis and prognosis have been
studied and summarized in Table 1.

4 Conclusion

The various algorithms were studied and reviewed for mass imbalance diagnosis and
prognosis using machine learning. The machine learning algorithms are showing
better mass imbalance diagnosis and prognosis accuracy for the application methods.
The review on mass imbalance indicates the potential scope to utilize machine
learning algorithms in the real world’s defect diagnosis and prognosis requirements.
It is also observed that deep learning reduces the need for feature extraction require-
ments more than other machine learning algorithms. Even though all the algorithms
follow different feature extraction methods, they show high accuracy; thus, the other
factors such as noisy environment, the behavior of the algorithms, speed of the algo-
rithm, computational cost, volume of the data, consistency of the data, loss of some
input data, trustworthy of the input data, etc., can be considered for future analysis.

Table 1 Advantages and disadvantages of machine learning

Technique Advantages Disadvantages
SVM SVM is more adaptable to a small sample Classification of faults becomes
of data and is responsive to constant and challenging in SVM if the range and
adaptive online learning [35]. The complexity of the data increase [56]
performance of SVM is considerably
better in fault detection and classification
because of its exceptional capacity in
classification problems [42]
RNN RNN is proven to work better in managing The performance of RNN decreases
time series data [56] with the growth of data [56]
ANN It is beneficial for nonlinear, complex, It requires more data for training
multidimensional data modeling, and purposes to achieve higher accuracy,
multivariate analysis can be processed which is time-consuming. Slow
faster with ANN [43] convergence nonlinear approximation
capability is constrained, leading to
poor performance in complex
classification problems [43]
PNN PNN has the ability to manage large PNN requires more memory space to
amount of data because it needs less store the model
iteration process for learning. So, it is
faster than ANN. This PNN feature results
from the Bayesian technique’s behavior
[49]
(continued)
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 297

Table 1 (continued)
Technique Advantages Disadvantages
RBF Training time requires less due to global It requires more memory and time
maxima convergence than classification for fault diagnosis [58]
perceptron-based. algorithms with less
training error [58]
MLP It may require more than one-time training It requires less memory and is quickly
due to local minima convergence [58] trained for fault diagnosis [58]
BPNN The backpropagation nature allows for The initial weight value should not be
optimizing the weights, thus valuable for unreasonable to avoid local
nonlinear problems [59] convergence [59]
Random RF can be used for regression as well as As it uses the probabilistic model, the
forest classification that produces an ensemble of extracted feature can’t classify
decision trees during the training phase. effectively if it has mixed problems
RF has been proven an excellent tool for which are not known [59]
defect detection and prediction [39]
Decision Interpretation of data through data Statistical feature selection is made
tree visualization is possible.as data can be for one feature not used by other nods
visualized, spurious data can be [41]
eliminated. As simultaneous feature
selection and classification happen more
effective [43]
KNN Multi-feature selection increase prediction Training and testing time is much
accuracy for imbalance mass identification faster than other algorithms [42]
[41]
CNN Effectively works with image and time Weight parameters initialization takes
series data. Combining the different time [59] network is complicated
convolution and pooling layers yields
better performance [59]. Multidimensional
data can be analyzed
LSTM LSTM is an enhanced form of RNN and Feature extraction becomes difficult
helps in extracting long-term dependency with the increase in the length of data
features [56] [56]
It has high efficiency in processing time
series data [56]
DBN It can work comfortably with Learning is possible from raw data and
multi-feature combinations, but training does not require any feature extraction
time is more with overfitting [59] before training and testing [59]

Furthermore, the algorithm’s effectiveness in identifying other faults will also be

explored.
298 S. Rajagopalan et al.

References

1. Bouzidane A, Thomas M. An electrorheological hydrostatic journal bearing for controlling

rotor vibration. Comput Struct. 2008;86(3–5):463–72.
2. Ding W. Rotor Whirl. In: Self-excited vibration. Springer; 2010. p. 191–242.
3. Sudhakar G, Sekhar A. Identification of unbalance in a rotor bearing system. J Sound Vib.
2011;330(10):2299–313.
4. Sharp R. Flexible rotor balancing: a review of principles and practices. Tribol Int.
1980;13(5):211–7.
5. Parkinson AG. Balancing of rotating machinery. Proc Inst Mech Eng Part C-J Mech Eng Sci.
1991;205(1):53–66.
6. Darlow MS. Balancing of high-speed machinery.Springer Science and Business Media;2012.
7. Jacquet-Richardet G et al. Rotor to stator contacts in turbomachines. Review and application.
Mech Syst Signal Process. 2013;40(2):401–420.
8. Adams ML. Large unbalance vibration analysis of steam turbine generators. The Institute;1984.
9. Boyce MP. Gas turbine engineering handbook. Elsevier;2011.
10. Lewis RI. Turbomachinery performance analysis. Butterworth-Heinemann;1996.
11. Wright T. Fluid machinery: performance, analysis, and design. CRC press;1999.
12. Krut R, Gleckman H. ISO 14001: A missed opportunity for sustainable global industrial
development. Routledge;2013.
13. Darabont DC, Antonov AE, Bejinariu C. Key elements on implementing an occupational health
and safety management system using ISO 45001 standard. In: Matec web of conferences. EDP
Sciences;2017
14. Cheng L, et al. The influence of crack breathing and imbalance orientation angle on the
characteristics of the critical speed of a cracked rotor. J Sound Vib. 2011;330(9):2031–48.
15. Gomez-Mancilla J, et al. The influence of crack-imbalance orientation and orbital evolution
for an extended cracked Jeffcott rotor. CR Mec. 2004;332(12):955–62.
16. Yang Y, et al. Response evaluation of imbalance-rub-pedestal looseness coupling fault on a
geometrically non-linear rotor system. Mech Syst Signal Process. 2019;118:423–42.
17. Gasch R, Markert R, Pfützner H. Acceleration of unbalanced flexible rotors through the critical
speeds. J Sound Vib. 1979;63(3):393–409.
18. He Q, et al. The effects of unbalance orientation angle on the stability of the lateral torsion
coupling vibration of an accelerated rotor with a transverse breathing crack. Mech Syst Signal
Process. 2016;75:330–44.
19. Sekhar AS. Identification of unbalance and crack acting simultaneously in a rotor system: modal
expansion versus reduced basis dynamic expansion. J Vib Control. 2016;11(9):1125–45.
20. Jalan AK, Mohanty AR. Model based fault diagnosis of a rotor–bearing system for misalign-
ment and unbalance under steady-state condition. J Sound Vib. 2009;327(3–5):604–22.
21. Behzad M, Alvandi M. Unbalance-induced rub between rotor and compliant-segmented stator.
J Sound Vib. 2018;429:96–129.
22. Ishida Y. Cracked rotors: industrial machine case histories and non-linear effects shown by
simple Jeffcott rotor. Mech Syst Signal Process. 2008;22(4):805–17.
23. Wauer J. On the dynamics of cracked rotors: a literature survey. Appl Mech Rev. 1990;43(1):13–
7.
24. Laws C. Turbine instabilities-case histories. Instab Rotating Mach. 1985;2409:65.
25. Bachschmid N, et al. Case studies of fault identification in power plant large rotating machinery.
In: IFToMM 6. International conference on rotor dynamics. 2002.
26. Bachschmid N, et al. Identification of rub and unbalance in 320 MW turbogenerators. Int J
Rotating Mach. 2003;9(2):97–112.
27. Pennacchi P, et al. Unbalance identification in large steam turbo-generator unit using a model-
based method. In: International design engineering technical conferences and computers and
information in engineering conference. American Society of Mechanical Engineers;2013.
28. Jardine AK, Lin D, Banjevic D. A review on machinery diagnostics and prognostics
implementing condition-based maintenance. Mech Syst Signal Process. 2006;20(7):1483–510.
A Systematic Review of Rotor Unbalance Diagnosis in Rotating … 299

29. Sabnavis G, et al. Cracked shaft detection and diagnostics: a literature review. Shock Vib Dig.
2004;36(4):287.
30. Lee YO, Jo J, Hwang J. Application of deep neural network and generative adversarial network
to industrial maintenance: a case study of induction motor fault detection. In 2017 IEEE
international conference on big data (big data). IEEE;2017.
31. Keijsers NLW. Neural networks. In: Encyclopedia of movement disorders. Elsevier; 2010.
p. 257–9.
32. Dey A. Machine learning algorithms: a review. Int J Comput Sci Inf Technol. 2016;7(3):1174–9.
33. Deng H, et al. A high-speed D-CART online fault diagnosis algorithm for rotor systems. Appl
Intell. 2020;50(1):29–41.
34. de Moura EP, et al. Classification of imbalance levels in a scaled wind turbine through detrended
fluctuation analysis of vibration signals. Renew Energy. 2016;96:993–1002.
35. Hübner G, et al. Detection of mass imbalance in the rotor of wind turbines using support vector
machine. Renew Energy. 2021;170:49–59.
36. Meng Y, Lu L, Yan J. Shaft orbit feature based rotator early unbalance fault identification.
Procedia CIRP. 2016;56:512–5.
37. Pinheiro AA, Brandao IM, Da Costa C. Vibration analysis in turbomachines using machine
learning techniques. Eur J Eng Technol Res. 2019;4(2):12–6.
38. Żabiński T, et al. CNC milling tool head imbalance prediction using computational intelligence
methods. In: International conference on artificial intelligence and soft computing. Springer;
2015.
39. Duan L, et al. A new support vector data description method for machinery fault diagnosis with
unbalanced datasets. Expert Syst Appl. 2016;64:239–46.
40. de SS Martins DH, et al. Application of machine learning to evaluate unbalance severity in
rotating machines. In: International conference on rotor dynamics. Springer; 2018.
41. Jablon LS, et al. Diagnosis of rotating machine unbalance using machine learning algorithms
on vibration orbital features. J Vib Control 2020: 1077546320929830.
42. Gohari M, Eydi AM. Modelling of shaft unbalance: modelling a multi discs rotor using K-
nearest neighbor and decision tree algorithms. Measurement. 2020;151: 107253.
43. Yan J, Hu Y, Guo C. Rotor unbalance fault diagnosis using DBN based on multi-source
heterogeneous information fusion. Procedia Manufs. 2019;35:1184–9.
44. Singh J, et al. A systematic review of machine learning algorithms for prognostics and health
management of rolling element bearings: fundamentals, concepts and applications. Meas Sci
Technol. 2020;32(1): 012001.
45. Kornaev N, Kornaeva E, Savin L. Application of artificial neural networks to fault diagnostics
of rotor-bearing systems. In: IOP conference series: materials science and engineering. IOP
Publishing;2020
46. Malik H, Mishra S. Artificial neural network and empirical mode decomposition based imbal-
ance fault diagnosis of wind turbine using TurbSim, FAST and Simulink. IET Renew Power
Gener. 2017;11(6):889–902.
47. Dwi H, Alfaradin F, Darojah Z. Artificial Neural Network based identification system for
abnormal vibration of motor rotating disc system. In: 2015 International electronics symposium
(IES). IEEE;2015.
48. Pavlenko I, et al. Ensuring vibration reliability of turbopump units using artificial neural
networks. In: International scientific-technical conference manufacturing. Springer; 2019.
49. Iannace G, Ciaburro G, Trematerra A. Fault diagnosis for UAV blades using artificial neural
network. Robotics. 2019;8(3):59.
50. Tajik M, et al. Gas turbine shaft unbalance fault detection by using vibration data and neural
networks. In: 2015 3rd RSI international conference on robotics and mechatronics (ICROM).
IEEE;2015.
51. Gohari M, Kord A. Unbalance rotor parameters detection based on artificial neural network.
Int J Acoust Vib. 2019;24(1):113–8.
52. Walker R, et al. Unbalance localization through machine nonlinearities using an artificial neural
network approach. Mech Mach Theory. 2014;75:54–66.
300 S. Rajagopalan et al.

53. Carbajal-Hernández JJ, et al. Rotor unbalance detection in electrical induction motors using
orbital analysis. In: Mexican conference on pattern recognition. Springer;2014.
54. Junior FEdAM, et al. Unbalance evaluation of a scaled wind turbine under different rota-
tional regimes via detrended fluctuation analysis of vibration signals combined with pattern
recognition techniques. Energy 2019;171:556–565.
55. Pavlenko I, et al. Using Computer Modeling and Artificial Neural Networks for Ensuring the
Vibration Reliability of Rotors. In: CMIS. 2019.
56. Kavlakoglu E. AI vs. machine learning vs. deep learning vs. neural networks: what’s the
difference? IBM cloud. 2020.
57. Chen J, et al. An imbalance fault detection algorithm for variable-speed wind turbines: a deep
learning approach. Energies. 2019;12(14):2764.
58. Mey O, et al. Machine learning-based unbalance detection of a rotating shaft using vibra-
tion data. In: 2020 25th IEEE international conference on emerging technologies and factory
automation (ETFA). IEEE;2020.
59. Cabrera D, et al. Fault diagnosis of spur gearbox based on random forest and wavelet packet
decomposition. Front Mech Eng. 2015;10(3):277–86.
Tuning of PID Parameters for Misaligned
Rotor Bearing System

Kuppa Sampath Kumar and Mohit Lal

Abstract In the online vibration monitoring technique, the selection of controlling

parameters is the real challenge. Wrongly tuned and selected controlling parame-
ters lead to an increase in settling time that is directly related to the exposure time
to the exaggerated vibration. In the present work, the characteristic parameters of
the Proportional–Integral–Derivative (PID) type controller are tuned for the stable
operation of the active magnetic bearing (AMB) integrated misaligned rotor bearing
system. The finite element method (FEM) is used to formulate equations of motion
of the system. Accordingly, a SIMULINK™ model is developed to acquire displace-
ment and current responses. The PID controller parameters are tuned to obtain a
steady-state response. The effect of misalignment in terms of misalignment forces
and moments is calculated. From analysis, it is evident that misalignment forces
and moments are very high at or near critical speeds of the system and that reveals
the misalignment forces and moments depend on the spin speed. An appreciable
reduction in misalignment force and moment magnitude is observed by varying
controller parameters which is an indication of stable operation of the system at
higher operating speeds. The identification algorithm is developed using the least
squares method to estimate the characteristic fault parameters. The effectiveness and
robustness of the proposed methodology are tested against measurement error by
incorporating different levels of noise percentage. The estimated parameters exhibit
good agreement against measurement error.

Keywords Active magnetic bearing (AMB) · Proportional–integral–derivative

(PID) controller · Misalignment · Finite element method (FEM) · Online vibration
monitoring

K. S. Kumar · M. Lal (B)

Department of Industrial Design, NIT Rourkela, Rourkela, Odisha, India
e-mail: [email protected]
K. S. Kumar
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 301
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_20
302 K. S. Kumar and M. Lal

1 Introduction

Nowadays, active magnetic bearings’ (AMBs) usage in rotating machinery gains

increasing acceptance by various high-speed machinery applications. The frequency
adaptability, less friction losses and zero lubrication are some of the features that
make AMB useful for high-speed machinery. Its use in online condition monitoring,
the balancing of the rotor for a wide range of frequencies and the selection of PID
controller parameters for AMB are the real challenges [1, 2].
The authors of [3, 4] developed a novel trail misalignment approach to estimate
the rotor unbalance and AMB misalignment in the misaligned rotor system balanced
by AMB. This approach is similar to the trail unbalance method in rotor balancing.
Article [5] developed a method to quantify the misalignment effect and additive
coupling stiffness. Here, the angular coupling stiffness is developed as the sum of
direct and time-varying coupling stiffness. A PID controller of fractional order type is
used to enhance the dynamic and stability of the system, and the gain parameters are
tuned and optimized using the particle swarm optimization method [6]. To overcome
the disturbances and unbalanced vibration in the rotor, AMB system is controlled
using a self-tuning PID controller [7].
In the present study, the Euler–Bernoulli beam theory is used to model the shaft
for the finite element (FE) formulations. According to the FE method, the equations
of the motion (EOMs) of the system are derived. The EOM of the system is modelled
using SIMULINK™ and solved using the Runge–Kutta method of fourth order to
generate time domain and current signals. These signals are converted into frequency
signals using a fast Fourier transform (FFT). The system stability and performance
at different running conditions can be validated by tuning the proportional–inte-
gral–derivative (PID) controller. The characteristic response is stable with the lower
steady-state error obtained by the rise time (T r ) and settling time (T s ). The PID tuning
parameters are varied at different running speeds to verify the rise time and settling
time. Based on PID tuning parameters, the misalignment forces and moments are
checked at different speeds. The condition monitoring technique is used to identify
the fault parameters of the misaligned rotor bearing system with AMB. The simula-
tions have been performed to check the effectiveness of the identification algorithm
by varying the PID controller gain parameters. The estimation technique is based on
the least squares fit method, and the parameters are estimated against measurement
noise.

2 Modelling of Misaligned Rotor-Bearing System

In this section, the assumptions and modelling of the misaligned rotor bearing system
with AMB have been stated.
Tuning of PID Parameters for Misaligned Rotor Bearing System 303

2.1 Assumptions and Description of the Model

Figure 1 shows a misaligned rotor bearing system with AMB. The system consists
of two flexible shafts connected with flexible coupling, and each shaft is supported
by two flexible bearings. Here, each shaft has a rigid disc and AMB at the centre.
The FE model of the shaft with the elemental and nodal numbers is shown in Fig. 2.
The individual substructure equations of motion (EOMs), viz. shaft, disc, bearing,
coupling, AMB and unbalance force, are derived in [1, 8–11]. Individual EOMs are
combined together to obtain the assembled EOM for the misaligned rotor bearing
system with AMB as

M ϑ̈ + (C − jωG)ϑ̇ + K ϑ = f unb + f AMB (1)

Here, M, C, G and K are the global mass, damping, gyroscopic and stiffness matrices
and f unb , f AMB and ϑ are the unbalance force, AMB controlling force and nodal
displacement vector. Here, two AMBs are isotropic in nature and characterized by
two linear displacement stiffness k s , and current stiffness k i , AMB force may be
derived as

f AMB = −ks ϑ + ki i (2)

The AMB controlling current i may be expressed as

Fig. 1 Misaligned rotor bearing system with AMB

Fig. 2 Elemental representation of the misaligned rotor bearing system with AMB
304 K. S. Kumar and M. Lal

i = k p ϑ + kd ϑ̇ + ki ϑ

Substitute the force, nodal and current displacement in the form f unb (t) =
F unb (t)ejωt , ϑ(t) = V (t)ejωt and i(t) = I(t) ejωt in Eq. (1). The obtained system EOMs
in the frequency domain may be expressed as
 2  
−ω M + jω(C − jωG) + K V = Funb + FAMB (3)

3 Simulated System Responses Based on Rise and Settling

Times

Equation (1) can be modified according to the SIMULINK™ model as

1
ϑ̈(t) = { f unb (t)} + { f AMB (t)} − (C − jωG)ϑ̇(t) − K ϑ(t) (4)
M
From Eq. (3), a SIMULINK™ model is developed to generate the time domain
displacement and current responses of the system. The system EOM is solved using
the Runge–Kutta method of fourth order to generate the responses at various oper-
ating speed conditions. Using the fast Fourier transformation (FFT) technique, the
time domain responses are converted into a frequency domain. The specifications
and properties of the misaligned rotor bearing system with AMB for the simulation
are listed in Table 1 [8, 12].
Figure 3 represents the characteristic linear displacement and phase response
against the spin speed at the bearing 1 location. The first ten natural frequencies
observed from Fig. 3 are ωnf 1 = 66.08 rad/s, ωnf 2 = 200.00 rad/s, ωnf 3 = 347.48 rad/s,
ωnf 4 = 517.37 rad/s, ωnf 5 = 795.16 rad/s, ωnf 6 = 1200.06 rad/s, ωnf 7 = 1728.30 rad/
s, ωnf 8 = 2067.40 rad/s, ωnf 9 = 2630.10 rad/s and ωnf 10 = 3584.90 rad/s. The
effectiveness of the algorithm is checked for a range of operating speeds and is
considered to generate the responses at selected operating speeds for estimating the
parameters.
The system stability condition is based on the Routh–Horwitz criteria [13]. The
optimal characteristic parameters of the system are estimated by tuning the PID
control parameters for lower steady-state error and faster stabilization. The PID
controller parameters are tuned using the Ziegler–Nichols tuning method [14–17] for
stable performance at different running conditions. The steady-state error and stable
characteristic response of the system are obtained with the help of rise time (T r ) and
settling time (T s ). The rise time and settling time should be small for faster responses
of the system. The settling time and rise time are varied throughout the simulations
by tuning the PID parameters and system frequencies. The rise time and settling
time play a major role in faster responses and system parameter identification. For
Tuning of PID Parameters for Misaligned Rotor Bearing System 305

Table 1 Properties of misaligned rotor bearing system with AMB

Rotor
Parameters Values (Units) Parameters Values (Units)
Disc mass m1 and m2 2 kg Shaft length l 2.5 m
Polar moment of inertia 45 × 10–4 kg-m2 Disc eccentricity e1 and 24 µm
Ip e2
Diametrical moment of 0.0024 kg-m2 Shaft diameter d 0.016 m
inertia I d
Shaft density ρ 7850 kg-m3 Phase change β 1 and β 2 300 and 360
deg
Rayleigh’s coefficients for the damping a0 = 0.154 a1 = 0.00001
proportional:
Actuator forces
Force-current factor, k i 48 N/A Force-disp. Factor, k s 105,210 N/m
Bearing
Stiffness parameters (N/m) Damping parameters (Ns/m)
K b1 x 2.21 × 105 K b1 y 2.35 × 105 C b1 x 280 C b1 y 350
K b2 x 2.35 × 105 K b2 y 2.42 × 105 C b2 x 287 C b2 y 350
K b3 x 2.13 × 105 K b3 y 2.24 × 105 C b3 x 225 C b3 y 275
K b4 x 2.01 × 105 K b4 y 2.20 × 105 C b4 x 315 C b4 y 290
Coupling
Direct and Cross stiffness parameters (N/m) Angular stiffness parameters (Nm/rad)
KC xx 2.12 × 105 KC xy 1.99 × 105 K C ϕyϕy 2.50 × 105
KC yy 1.11 × 105 KC yx 1.31 × 105 KC ϕxϕx 2.20 × 105
Damping parameters (Ns/m)
C C xx 225 C C yx 295
CC xy 20 C C yy 50

testing the PID parameters, two cases are studied: (a) constant spin speed by varying
the K i and K d parameters, and (b) different spin speeds by varying the K i and K d
parameters.
In case (a), for testing K p = 4200, 800 ≥ K i ≥ 4200 and 1 ≥ K d ≥ 18 are
considered for simulations. The rise time and settling time variations at a constant
speed (35 Hz) by varying the K i and K d parameters are shown in Fig. 4. From
Fig. 4a, b, it may be observed that the settling time is constant by varying the integral
parameter, and time shows some variation by varying the differential parameter. The
rise time is varied by varying the integral parameter, and it is constant by varying
the differential parameter shown in Fig. 4c, d. In case (b), the rise time and settling
time variations for different spin speed range from 0 to 100 Hz by varying the K i
and K d parameters are shown in Fig. 5. From Fig. 5a, b, it is observed that the rise
time is varied with the differential parameters, and it varied up to a certain speed and
is constant for the remaining speed by varying the integral parameters. The settling
306 K. S. Kumar and M. Lal

5
10
Amplitude (m)

0
10

-5
10
4

2
Phase (rads)

-2

-4
0 500 1000 1500 2000 2500 3000 3500
Spin speed (rad/s)

Fig. 3 Amplitude and phase of linear displacement with spin speed

time shows variations with the increase of speed and integral parameters, and it is
varied up to a certain speed and the remaining speeds remain constant by varying
the differential parameters as shown in Fig. 5c, d.

10-5 10-5
(a) 8.06 (b) 9
Kp =4200
8
Rise time Tr (sec)

8.04 Kd =3
Rise time Tr (sec)

7
8.02
6 Kp =4200
8
5 Ki =4200
7.98 4

7.96 3
1000 1500 2000 2500 3000 3500 4000 0 5 10 15
K K
i d
(c) 2 (d)
Kp =4200 0.9998
Settling time T (sec)

Settling time Ts (sec)

Kd =3 0.9996
1.5
Kp =4200
s

0.9994
Ki =4200
1 0.9992
0.999
0.5
0.9988

0 0.9986
1000 1500 2000 2500 3000 3500 4000 0 5 10 15
K
Ki d

Fig. 4 Variation of rise time and settling time with constant speed by varying K i and K d parameters
Tuning of PID Parameters for Misaligned Rotor Bearing System 307

(a) 10 -5 (b) 10 -5
Rise time T (sec) 8.5 8.1

Rise time T (sec)

8.05
r

7.5

r
Kp =4200 8
7 Kd =3
Kp =4200
7.95
6.5 Ki =4200

6 7.9
20 40 60 80 100 20 40 60 80 100
Frequency (Hz) (varying K ) Frequency (Hz) (varying Kd)
(c) i (d)
1 1
Settling time Ts (sec)

Settling time T (sec)

0.999
0.999 Kp =4200

s
0.998 Kp =4200 Ki =4200
0.997 Kd =3
0.998
0.996

0.995 0.997

20 40 60 80 100 20 40 60 80 100
Frequency (Hz) (varying Ki) Frequency (Hz) (varying Kd )

Fig. 5 Variation of rise time and settling time with different speeds by varying K i and K d parameters

4 Calculation of Misalignment Forces and Moments

For checking the effectiveness of the characteristic parameters, misalignment forces

and moments are calculated. The misalignment forces and moments expressions are
presented in [8]. The frequency range considered for calculating the misalignment
forces and moments is 0–100 Hz. To check the effect of misalignment forces and
moments of the system at higher operating speeds for different PID controller tuning
parameters, different case studies are developed. These case studies are as follows
tuning parameters, (b) K p and K d are constant and varying the K i parameters, and
(c) K p and K i are constant and varying the K d parameters.
In case (a), misalignment forces and moments for the rotor system with and
without AMB are calculated for the constant tuning parameters (i.e., K p = 4200,
K d = 1, K i = 1200) as shown in Fig. 6. From Fig. 6, it may be observed that
the misalignment forces and moments are varying and show minor variation in the
vibration amplitude by applying the constant tuning parameters. In case (b), the
misalignment forces and moments for the rotor system with and without AMB are
calculated for the tuning parameters (i.e., K p = 4200, K d = 3, 800 ≥ K i ≥ 4200) as
shown in Fig. 7. From Fig. 7, it may be observed that the appreciable reduction in
the forces and moments by varying the integral parameters. The increase of integral
parameters with speeds to regulate and stabilize the system at higher operating speeds.
In case (c), the rotor system with and without AMB misalignment forces and moments
is calculated for the tuning parameters (i.e., K p = 4200, K i = 4200, 1 ≥ K d ≥ 18)
as shown in Fig. 8. From Fig. 8, it may be observed that the misalignment forces
and moments are varying with the increase of differential parameters. From all these
308 K. S. Kumar and M. Lal

(a) (b)
2 1.5
mis mis
Misalignment Force (N)

Misalignment Force (N)

F x with AMB F y with AMB
mis mis
1.5 F without
x
AMB F
y
without AMB
1
1
0.5
0.5

0 0
20 40 60 80 100 20 40 60 80 100
(c)
Frequency (Hz) Frequency (Hz)
(d)
Misalignment Moment (N-m)

Misalignment Moment (N-m)

8 4
Mmiswith AMB Mmiswith AMB
x-z y-z
mis mis
6 Mx-z without AMB 3 My-z without AMB

4 2

2 1

0 0
20 40 60 80 100 20 40 60 80 100
Frequency (Hz) Frequency (Hz)

Fig. 6 Misalignment forces and moments (with and without AMB) with respect to running
frequencies for constant PID controller parameters

cases, it may be observed that the increase in integral parameter (K i ) may lead to
stabilization of the system and reduction in vibration amplitude at higher operating
speeds.

5 Formulation of Identification Algorithm

For developing the identification algorithm to estimate the characteristic parameter,

Eq. (3) can be rearranged to obtain the characteristic parameters (i.e., unbalances,
coupling stiffness, coupling damping and AMB force stiffness parameters) to the left-
hand side and the system parameters that are known (i.e., global mass, rotor stiffness
and damping, bearing stiffness and damping, and gyroscopic) to the right-hand side.
These rearrangements result in the regression equation as
   
− {Funb } − {FAM B } + K C + jω C C {V }
     
= ω2 [M] − K B − j ω C B − ω2 [G] {V } (5)

Equation (5) can be expressed in linear regression equation form as

[A(ω)]{X } = [B(ω)] (6)

with
Tuning of PID Parameters for Misaligned Rotor Bearing System 309

(a) (b)
2 1.5
mis mis

Misalignment Force (N)

Misalignment Force (N)
F x with AMB F y with AMB
mis mis
1.5 F without AMB F y without AMB
1
1
0.5
0.5

0 0
20 40 60 80 100 20 40 60 80 100
(c)
Frequency (Hz) (d) Frequency (Hz)
Misalignment Moment (N-m)

Misalignment Moment (N-m)

8 2.5
mis mis
M with AMB M with AMB
x-z y-z
2
6 Mmiswithout AMB Mmiswithout AMB
x-z y-z
1.5
4
1
2
0.5

0 0
20 40 60 80 100 20 40 60 80 100
Frequency (Hz) Frequency (Hz)

Fig. 7 Misalignment forces and moments (with and without AMB) with respect to running
frequencies for constant K p , K d and varying K i controller parameters

(a) (b)
2 1.5
mis mis
Misalignment Force (N)

Misalignment Force (N)

F x with AMB F y with AMB

mis mis
1.5 F without
x
AMB F
y
without AMB
1
1
0.5
0.5

0 0
20 40 60 80 100 20 40 60 80 100
Frequency (Hz) Frequency (Hz)
(c) (d)
Misalignment Moment (N-m)

Misalignment Moment (N-m)

8 4
Mmiswith AMB Mmiswith AMB
x-z y-z
mis mis
6 Mx-z without AMB 3 My-z without AMB

4 2

2 1

0 0
20 40 60 80 100 20 40 60 80 100
Frequency (Hz) Frequency (Hz)

Fig. 8 Misalignment forces and moments (with and without AMB) with respect to running
frequencies for constant K p , K i and varying K d controller parameters
310 K. S. Kumar and M. Lal

Fig. 9 Effect of measurement noise on estimated parameters

 
{X } = e1 e2 K sx1 K i x1 K sx2 K i x2 K sy1 K i y1 K sy2 K i y2 K xCx K xCy K yx
C KC KC C C C C C
yy ϕx ϕx K ϕ y ϕ y C x x C x y C yx C yy

The regression matrix [A] and vector [B] are in complex form, and the above
equation requires independent sets of displacement and current response information
to estimate the characteristic parameters in vector {X}.

6 Fault Parameter Estimation

The identification algorithm is developed using the least squares fit method to esti-
mate characteristic fault parameters of the system. The estimated parameters are
coupling, unbalance and AMB characteristic parameters. The effectiveness of the
algorithm for a range of operating speeds (31–51 Hz) for a suitable PID controller
(K p = 4200, K i = 2400, K d = 3) is estimated and shown in Fig. 9. From Fig. 9,
it is observed that the AMB parameters exhibit good agreement with the addition
of measurement noise up to 5%. The unbalance parameters show a small devia-
tion with the increase of measurement noise. The maximum error obtained for the
unbalance parameter is around (e1 = 4.89%) at a 5% measurement noise condition.
The coupling parameters exhibit good agreement with the addition of measurement
noise. The effectiveness of the algorithm is found to be robust with suitable PID
tuning parameters.

7 Conclusions

In the present work, an online vibration monitoring technique is proposed for the
estimation of unbalance, coupling and dynamic AMB parameters of a misaligned
rotor bearing system with AMB. In this article, an identification algorithm based on
Tuning of PID Parameters for Misaligned Rotor Bearing System 311

the least squares fit technique is developed to estimate various fault parameters of
the system. The PID controller is used to stabilize the system at different operating
speeds. The stable characteristic response of the system with lower steady-state error
is obtained with the help of rise time and settling time. The effective changes in the
rise time and settling time for various speed conditions are tested against different
PID controlling parameters for the misalignment force calculation and characteristic
parameter estimation. From case studies, it is found that the increase of integral
parameter of the PID increases stabilization and reduces the vibration amplitude
in the system. The fault parameters estimated are coupling dynamic and inherent
unbalance parameters along with AMB’s characteristic parameters. The effectiveness
and robustness of the proposed methodology are tested against measurement error by
incorporating different levels of noise percentage. The unbalance parameters show a
small deviation with the increase of measurement noise. The maximum error obtained
for the unbalance parameter is around (e1 = 4.89%) at a 5% measurement noise
condition. The estimated parameters exhibit good agreement against measurement
error.

References

1. Rao JS. Rotor dynamics. New Age International. New Delhi;1996.

2. Bleuler H, Cole M, Keogh P, Larsonneur R, Maslen E, Okada Y, Schweitzer G, Traxler A.
Magnetic bearings: theory, design, and application to rotating machinery. Springer Science
and Business Media;2009
3. Kumar P, Tiwari R. Finite element modelling, analysis and identification using novel trial
misalignment approach in an unbalanced and misaligned flexible rotor system levitated by
active magnetic bearings. Mech Syst Signal Process. 2021;152: 107454. https://doi.org/10.
1016/j.ymssp.2020.107454.
4. Kumar P, Tiwari R. Dynamic analysis and identification of unbalance and misalignment in a
rigid rotor with two offset discs levitated by active magnetic bearings: a novel trial misalignment
approach. Propuls Power Res. 2021;10(1):58–82. https://doi.org/10.1016/j.jppr.2020.06.003.
5. Srinivas RS, Tiwari R, Kannababu C. Model based analysis and identification of multiple fault
parameters in coupled rotor systems with offset discs in the presence of angular misalignment
and integrated with an active magnetic bearing. J Sound Vib. 2019;450:109–40. https://doi.org/
10.1016/j.jsv.2019.03.007.
6. Shata AMAH, Hamdy RA, Abdelkhalik AS, El-Arabawy I. A fractional order PID control
strategy in active magnetic bearing systems. Alex Eng J. 2018;57(4):3985–93. https://doi.org/
10.1016/j.aej.2018.01.020.
7. Chen KY, Tung PC, Tsai MT, Fan YH. A self-tuning fuzzy PID-type controller design for
unbalance compensation in an active magnetic bearing. Expert Syst Appl. 2009;36(4):8560–70.
https://doi.org/10.1016/j.eswa.2008.10.055.
8. Lal M, Tiwari R. Multi-fault identification in simple rotor-bearing-coupling systems based on
forced response measurements. Mech Mach Theory. 2012;51:87–109. https://doi.org/10.1016/
j.mechmachtheory.2012.01.001.
9. Lal M, Tiwari R. Quantification of multiple fault parameters in flexible turbo-generator systems
with incomplete rundown data. Mech Syst Signal Process. 2013;41:546–63. https://doi.org/10.
1016/j.ymssp.2013.06.025.
312 K. S. Kumar and M. Lal

10. Kuppa SK, Lal M. Dual flexible rotor system with active magnetic bearings for unbalance and
coupling misalignment faults analysis. Sādhanā. 2019;44(8):1–16. https://doi.org/10.1007/s12
046-019-1168-z.
11. Kuppa SK, Lal M. Characteristic parameters estimation of active magnetic bearings in a coupled
rotor system. J Verif Valid Uncertain Quantif. 2019;4(3). https://doi.org/10.1115/1.4045295
12. Tiwari R, Chougale A. Identification of bearing dynamic parameters and unbalance states in a
flexible rotor system fully levitated on active magnetic bearings. Mechatronics. 2014;24:274–
86. https://doi.org/10.1016/j.mechatronics.2014.02.010.
13. Tiwari R. Rotor systems: analysis and identification. Boca Raton: CRC Press; 2017.
14. Ziegler JG, Nichols NB. Optimum settings for automatic controllers. Trans. ASME.
1942;64(11).
15. Levine WS. The control systems handbook: control system advanced methods. CRC
press;2010.
16. Anantachaisilp P, Lin Z. An experimental study on PID tuning methods for active magnetic
bearing systems. Int J Adv Mechatron Syst. 2013;5(2):146–54. https://doi.org/10.1504/IJA
MECHS.2013.055991.
17. Borase RP, Maghade DK, Sondkar SY, Pawar SN. A review of PID control, tuning methods
and applications. Int J Dyn Control. 2020;1–10. https://doi.org/10.1007/s40435-020-00665-4.
Machine Learning-Based Fault
Prediction of Electromechanical System
with Current and Vibration Signals

Purushottam Gangsar, Vikas Singh, Manoj Chouksey, and Anand Parey

Abstract This paper describes the development of intelligent fault diagnosis for
electromechanical systems (EMS) and examines various combinations of mechanical
and electrical faults in such systems. Specifically, the study focuses on a three-phase
asynchronous motor (IM) with an outer rotor bearing system. The faults investigated
in this study include a healthy system (F1), healthy motor with outer bearing faults
(F2), healthy motor with unbalance in outer rotor (F3), bearing fault in the motor
with healthy outer rotor (F4), inner motor bearing and outer rotor bearing fault (F5),
motor bearing fault with outer unbalanced rotor (F6), motor stator fault with outer
healthy rotor (F7), motor stator fault with outer unbalanced rotor (F8), motor stator
fault with outer bearing fault (F9), and motor bearing fault with outer bearing fault
and unbalanced rotor (F10). In order to detect combined faults in the motor-rotor-
bearing assembly, the paper proposes using wavelet characteristics extracted from
the current and vibration signals, which are used to develop a support vector machine
(SVM)-based defect detection system. Finally, the paper concludes with a discussion
of the research results.

Keywords Support vector machine · Electromechanical system · Rotor · Bearing

P. Gangsar (B) · V. Singh · M. Chouksey

Mechanical Engineering Department, Shri G S Institute of Technology and Science, Indore, India
e-mail: [email protected]
V. Singh
e-mail: [email protected]
M. Chouksey
e-mail: [email protected]
A. Parey
School of Mechanical Engineering, IIT Indore, Indore, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 313
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_21
314 P. Gangsar et al.

1 Introduction

EMS comprises electrical and mechanical components and is extensively used in

light as well as heavy industry. EMS is exposed to various kinds of stresses that
cannot be avoided at work, resulting in various types of electrical and mechanical
breakdown of the body. A minor fault of these, if not detected and rectified in a timely
manner, can cause serious damage to the EMS, resulting in downtime of the entire
production process, loss of production, and sometimes serious injury. Therefore,
early diagnosis of EMS failure is important to avoid its serious consequences [1, 2].
In the last two decades, multiple diagnostic techniques have been developed to
prevent any kind of malfunction in EMS, including FFT, high-resolution spectral
analysis, wavelet transform, Hilbert transform, Park vector method, and Hilbert-
Hong transform [3, 4]. These tests are based on signal analysis and utilize a range of
signals such as vibration, stator current, induced voltage, air gap torque, acoustics,
and others [5]. Note that current and vibration techniques are most prevalent and
favored due to its higher accuracy and simplicity of measurement [6]. It is important
to remember that traditional methods are not always consistent because they are
influenced by many factors such as defect levels, motor running conditions, noise,
and others [4]. Another fault diagnosis method based on mathematical modeling of
the machine has also been developed. However, many assumptions need to be made to
construct reliable mathematical models to deal with nonlinear and stochastic systems,
but this is still weak in the presence of distortions and noise [7].
In order to diagnose defects in machines using signals and computational models,
one needs to have adequate knowledge and experience gained through practice.
However, with the constant rotation of machines for all kinds of industries, it is
challenging to fulfill the need for skilled professionals. Consequently, in the last
decade, new diagnostic methods utilizing machine learning (ML) technology have
been developed and adopted in the machinery field. These methods aim to automate
traditional diagnostic techniques, increase reliability and accuracy, and lower costs
[8]. ML techniques are data-driven techniques that do not involve knowledge of
models. Several AI techniques, including ANN, fuzzy logic, hidden Markov model,
genetic algorithm (GA), and SVM, have been applied in the identification of system
defects [9, 10]. In ML, the support vector machine has attractive features such as
classification efficiency, short duration, and strong flexibility compared to other ML
techniques [11].
To use ML-based diagnostics, different processing methods have been developed
for example time-domain, frequency-domain, and multi-time frequency methods
such as wavelets. Time domain and frequency domain theory assumes signal stability
and system linearity; However, differences and inconsistencies and/or changes in
electronic components may occur during normal operation of the EMS. Due to
many faults, EMS operates in unstable conditions. Time–frequency or wavelet-based
feature extraction methods are also used to solve complex and non-local problems
[12]. For defect detection, three types of wavelets, namely, WPT, CWT, and DWT
are employed [13]. In all the above studies, fault detection is determined at constant
Machine Learning-Based Fault Prediction of Electromechanical System … 315

speed of the motor and less diagnostic for ramp speed problem. In addition, fault iden-
tification of EMS is limited in the literature. In addition, fault detection of electrical
and mechanical defects is rare in AI-based EMS [14–16].
Therefore, in the current study, defect detection of EMS is attempted based
on wavelet and SVM. Ten different combinations of EMS defects are considered.
Wavelets (continuous wavelets) are considered for diagnosis due to their special
properties of simultaneously processing time and frequency information. SVMs are
used with RBF kernel. Since the accuracy of the SVM is influenced by the kernel &
SVM parameters, the optimal values of parameters are selected by grid search &
cross-validation algorithm. Also, performance evaluation also depends on the input,
so here three important wavelet properties are measured for research. Lastly, identi-
fication was performed to confirm the effectiveness of the method at various motor
operating conditions.

2 SVM Introduction

SVM is a statistical learning method that leverages examples to assign labels to data
points [17]. It was proposed by Vapnik in 1995 and is characterized by four main
points: the hyperplane with the maximum margin, the separating hyperplane, the
flexible margin, and the kernel function. In high-dimensional space, a separating
hyperplane is a boundary that divides two object classes. The objective of SVM is
aimed to determine the hyperplane that provides the maximum or best separation
between the classes. The margin, as depicted in Fig. 1, is defined as the gap between
the separator hyperplane and the data point which is closer to it from each class.
The optimal hyperplane is achieved using the subsequent optimization problem;

Fig. 1 The X2
+

-ξ /|W| Margin
+ +

maximum-margin
-
+ + +
+
+ +

hyperplane

+
-- --
SV
+

W SV

Hyperplane
- - -
(W)
SV -
- -
- - - --
-
X1
316 P. Gangsar et al.

1  m
Minimum [τ (w)] = w2 + C ξj (1)
2 j=1

Subjected to, y j (w.x j + b) ≥ 1 − ξ j , ξ j ≥ 0, j = 1, 2, ..., m (2)

 m
where training datasets is defined by (x j , y j ) j=1 x j ∈ R l , y j ∈ {−1, 1}, input
vectors is defined by x j , label of x j is defined by y j , w represents normal direction
of a hyperplane, scaler is defined by b, positive slack variables is defined by ξ j , and
C represents the generalization parameter. Linear discriminate function for SVM
training can be written as


m
∗ ∗
f (x) = w · x + b = {yi αi∗ xi , x} + b∗ (3)
i=1

The optimization problems mentioned above are designed for linearly separable
data only. However, real-world data may not always be linearly separable and may
exhibit complex patterns in the input space. In such cases, Support Vector Machine
(SVM) algorithms can generate hyperplanes that allow for larger margin of sepa-
ration even in nonlinearly separable data. One of the key advantages of support
vector machines (SVMs) is their ability to transform data from an input space with a
lower number of dimensions to a feature space with a higher number of dimensions,
using kernel functions. This enables more flexible decision boundaries and facilitates
the capture of complex relationships between data points. Various kernel functions
k(x, xi ) = φ(x)φ(xi ), including Gaussian RBF, polynomials, and sigmoid functions,
can be utilized to map data into higher-dimensional spaces. This transformation can
aid in reducing computational load by retaining the effects of Multidimensional trans-
formations. In the present study, the RBF (Radial Basis Function) kernel, which is a
popular and widely used option, is employed.

  1
k(x, xi ) = exp −γ x − xi 2 ; γ = >0 (4)
2σ 2
The parameter σ in the RBF kernel represents the width of the kernel. While
basic SVMs are designed for binary classification tasks, in reality, there may be
scenarios where multiple classes need to be classified. Several methods have been
proposed to tackle this issue, including the OAO, OAA, and DAGS algorithms [18].
These approaches break down multinomial classification problem into a sequence of
bivariate classification problems. The OAA method involves training k-SVM models,
where k represents the number of classes, while the OVO method requires training
k(k − 1)/2 SVM models, with each model differentiating between two classes. Hsu
and Lin [18] compared the hybrid method with three binary classification-based
methods (OVO, OAA, and DAGS) and concluded from their experiments that OAO
and DAGS are effective in practice.
Machine Learning-Based Fault Prediction of Electromechanical System … 317

3 Experimental Setup

Tests were conducted using a Mechanical Failure Simulator (MFS) as illustrated

in Fig. 2. The MFS consists of a 0.5 hp, 50 Hz three-phase asynchronous motor
induction motor connected to a rotor bearing system through an elastic coupling. A
pulley connects the motor shaft to the gearbox, and an electromagnetic brake clutch
is connected to the gearbox to deploy external load to the IM. A speed controller is
connected to the IM for changing the speed. Three AC current probes and a triaxial
accelerometer are used to record the current and vibration signals, respectively. Data
acquisition was carried out using a National Instruments data acquisition system
(DAQ). A tachometer with constant DC power was used to measure motor speed.
The acquired data were analyzed using NI-LabVIEW data acquisition software.
The tests were conducted to simulate various fault conditions, including healthy
system (F1), healthy motor with outer bearing faults (F2), healthy motor with unbal-
ance in outer rotor (F3), bearing fault in the motor with healthy outer rotor (F4),
inner motor bearing and outer rotor bearing fault (F5), motor bearing fault with outer
unbalanced rotor (F6), motor stator fault with outer healthy rotor (F7), motor stator
fault with outer unbalanced rotor (F8), motor stator fault with outer bearing fault
(F9), and motor bearing fault with outer bearing fault and unbalanced rotor (F10).
Raw data was collected at a sampling rate of 20,480 Hz in the time domain, resulting
in a total of 80 raw data sets (80 × 20,480 sampling points) for all the defect condi-
tions. Data was recorded for different motor running conditions, including ramp-up
speeds up to 10, 20, 30, and 40 Hz, under two different torque conditions: without
load and high load (0% of rated load; 0.55% of rated load).

Fig. 2 Experimental test-rig

318 P. Gangsar et al.

4 Feature Calculation Based on CWT

In order to achieve accurate results, this article employs the inference method based
on Continuous Wavelet Transform. CWT is a significant approach for analyzing
time–frequency information as it transforms time-domain information into time–
frequency information. The fundamental concepts of CWT and its application for
defect identification have been extensively discussed in numerous articles [19, 20].
CWT is utilized by combining the signal with the wavelet family, and it can be
described as follows:

+∞
ω(s, τ ) = x(t)ψ ∗ (t)dt (5)
−∞

with
1 t −τ
ψ ∗ (t) = √ ψ( ) (6)
s s

Among these, the master wavelet, also known as the window function, is denoted
by ψ(t). The parameter “s” that controls the scale is related to the frequency charac-
teristics of the compressed or expanded signal, with higher scales indicating lower
frequencies and vice versa. The parameter “τ” that controls the translation is used
to determine the position of the window as it moves through the signal, analogous
to the direction of the red light. The wavelet transform utilizes the master wavelet
function to break down the signal into a weighted set of scaled wavelet energies.
While wavelets and Fourier transforms share similarities, the wavelet family differs
in that it employs an infinite number of fundamental functions to transform sine and
cosine functions.
Time–frequency information can be extracted from time-series data [19]. In this
study, HAAR wavelets are utilized to obtain time–frequency information. The time-
series data is initially decomposed into 2^7 sub-signals, yielding wavelet coefficients
that are associated with each parameter. Each individual data point, for instance,
containing 10,000 data points, encompasses the wavelet coefficients for each param-
eter, resulting in a matrix of coefficients for the dataset. It is important to note that
calculating wavelet coefficients for each parameter requires significant storage space
and time due to the large amount of data involved. Furthermore, choosing the appro-
priate wavelet scale can be challenging as an indicator chosen from a negative (or
positive) scale may not fully capture the wave, while other scales might be over-
looked. Hence, in this study, relative wavelet power (RWE) was employed as a
criterion for selecting a suitable scale. Based on the RWE criteria, the wavelet scale
with the highest energy is considered as the appropriate scale. RWE is a time–
frequency measure that can describe specific events in the time–frequency domain.
Mathematically, RWE is defined as the energy allocation given by
Machine Learning-Based Fault Prediction of Electromechanical System … 319

E(m)
pm = (7)
E overall

where pm = 1, & overall energy is
m

 2 
E overall = Cm, j = E(m) (8)
m i n

where Cn, j represents “j” denotes the wavelet coefficient corresponding to the “n”th
scale, “m” represents the quantity of wavelet coefficients, where m = 1, …, n,. Total
energy of detail signal can be represented by


n
2
E(m) = Cm, j (9)
j=1

Power is now calculated for each scale. Then, the RWE of each parameter was
calculated by Eq. (7). Select the ratio with the largest RWE as the best ratio. Now the
wavelet coefficients of all data corresponding to the best measurement are obtained.
Wavelet features may be extracted from the wavelet coefficients. This function uses
three parameters such as standard deviation (σ), skewness (χ), and kurtosis (k).

5 Results and Discussion

In this study, one-on-one SVM is utilized to complement the diagnosis of multiple

faults in EMS. A total of 80 datasets containing statistical data are split into training
and testing subsets, with 80% utilized for training and 20% for testing. SVM training
is conducted for each speed and load condition, using the training data for each
fault. It should be noted that the RBF kernel is employed for SVM training, which
involves two initial hyperparameters: kernel parameter γ and Lagrangian multiplier
C. These parameters need to be optimized for error detection. In this study, grid
search and cross-validation methods are used to fine-tune these two measures. Various
combinations of (C, γ) are attempted, and the one that produces the greatest level
of accuracy or education is chosen. This optimization process is performed for each
wavelet feature one at a time. Figure 3 illustrates the optimization of (C, γ) during
training at 10 Hz ramp-up and high torque condition, showing that the SVM training
accuracy is 82.5%. It should be noted that the final estimate may be influenced by
the training of the distributor. Once the best (C, γ) pair is selected, it is used for the
final training. The SVM model that has been trained is then utilized to categorize or
detect ten faults. The SVM prediction function is represented by the percentage of
the predictive value, which is the number of successful test data among all the test
data.
320 P. Gangsar et al.

5 80 -14
0 60 82
log3Y

log3Y
-5 -14.5 81.5
40
-10 81
-15 20 -15
5 10 15 13 13.5 14
log3C
log3c
(a) (b)

Fig. 3 Cross-validation accuracy for 40 Hz and T2 load

In the next step, the SVM model is evaluated at the same velocity and load condi-
tions as during training. The selected wavelet features are fed one by one, and diag-
nosis is performed. The solution is evaluated by checking the performance of various
functions of the IM, including four ramp-up and two torque conditions. Three main
features, namely, standard deviation (σ), kurtosis (k), and skewness (χ) are utilized
in this work. Fault detection is initially carried out in the zero load environment,
and the outcomes are presented in Fig. 4a. The lowest accuracy achieved is 70.7%,
while the highest accuracy is 99.3%, both of which are obtained at 40 and 10 Hz. All
faults, except SWF_HR and SWF_UR, were successfully classified for all speeds
with more than 80% accuracy. The average classification accuracy based on the
no-load condition is 87.3%.
The analysis was also performed for high load conditions, and the results are
presented in Fig. 4b. The lowest and highest classification accuracies achieved were
94% and 66%, which were obtained at 20 and 40 Hz ramp-up conditions, respectively.
All faults, except MBF_UR, MBF_HR, and SWF_HR, were successfully classified
for all speeds with more than 80% accuracy. The overall mean performance at high
torque conditions was 85.5%. This indicates that the dispersion of misclassifications
is slightly higher, by around 2%, in high torque conditions compared to no-torque
conditions. In general, the identification of faults using the three main features of
Haar wavelets was successful in predicting the faults even at different load levels.

6 Conclusions

Here, Haar wavelet features are calculated to evaluate combined defects of EMS using
OAO SVM technique. Three characteristic standard deviations, skewness, skewness,
and signal current are used for EMS diagnostics. Finally, diagnostics were performed
for various speed tests at mechanical loads under different engine conditions, to
determine the diagnostics of the EMS in an emergency. The average performance
is 85.5 and 87.5% for high and zero load environment, individually. This study
presents that the integration of wavelet properties & SVM can detect EMS defects
at all loads, even at ramp-up speeds. Here, single Haar wavelet was utilized in EMS
Machine Learning-Based Fault Prediction of Electromechanical System … 321

(a) Zero load case

100.0

80.0
Performance, %

60.0

40.0

20.0

0.0

10 Hz 20 Hz 30 Hz 40 Hz Average

(b) High Load Case

100.0

80.0
Performance, %

60.0

40.0

20.0

0.0

10 Hz 20 Hz 30 Hz 40 Hz Average

Fig. 4 Result for present diagnosis

defect analysis, but Shannon, Gaussian, and other wavelet functions may be further
utilized.

Acknowledgements We are thankful to NPIU (a project of Education ministry of India & the
World Bank) for funding research through the Collaborative Research Program. The ID number is
1-5770792503.
322 P. Gangsar et al.

References

1. Bazzi AM, Krein PT. Review of methods for real-time loss minimization in induction machines.
IEEE Trans Ind Appl. 2010;46(6):2319–28.
2. Henao H, Capolino GA, Fernandez-Cabanas M, Filippetti F, Bruzzese C, Strangas E, Hedayati-
Kia S. Trends in fault diagnosis for electrical machines: a review of diagnostic techniques. Ind
Electron Mag IEEE. 2014;8(2):31–42.
3. Nandi S, Toliyat HA, Li X. Condition monitoring and fault diagnosis of electrical motors-a
review. IEEE Trans Energy Convers. 2005;20(4):719–29.
4. Lee S, Bryant MD, Karlapalem L. Model-and information theory-based diagnostic method for
induction motors. J Dyn Syst Meas Contr. 2006;128(3):584–91.
5. Alsaedi MA. Fault diagnosis of three-phase induction motor: a review. Optics. Special Issue:
Appl Opt Signal Process. 2015;4(1–1):1–8.
6. Thomson WT, Orpin P. Current and vibration monitoring for fault diagnosis and root cause anal-
ysis of induction motor drives. In: Proceedings of the thirty-first turbomachinery symposium.
2002. p. 61–7.
7. Bilski P. Application of support vector machines to the induction motor parameters identifica-
tion. Measurement. 2014;51:377–86.
8. Siddique A, Yadava GS, Singh B.. Applications of artificial intelligence techniques for induc-
tion machine stator fault diagnostics: review. In: Diagnostics for electric machines, power
electronics and drives, 2003. SDEMPED 2003. In: 4th IEEE international symposium on.
IEEE; 2003. p. 29–34.
9. Tran VT, Yang BS, Oh MS, Tan ACC. Fault diagnosis of induction motor based on decision
trees and adaptive neuro-fuzzy inference. Expert Syst Appl. 2009;36(2):1840–9.
10. Li L, Mechefske CK. Induction motor fault detection and diagnosis using artificial neural
networks. Int J COMADEM. 2006;9(3):15.
11. Salem SB, Bacha K, Chaari A. Support vector machine based decision for mechanical fault
condition monitoring in induction motor using an advanced Hilbert-Park transform. ISA Trans.
2012;51(5):566–72.
12. Gangsar P, Tiwari R. Diagnostics of mechanical and electrical faults in induction motors using
wavelet-based features of vibration and current through support vector machine algorithms for
various operating conditions. J Braz Soc Mech Sci Eng. 2019;41(2):71.
13. Ali Z, Gangsar P, Chouksey M, Parey A. Intelligent and robust fault diagnostics for an elec-
tromechanical system using vibration and current signals. In: International conference on future
technology (ICOFT)-2020. NIT; Puducherry:2020.
14. Yu M, Xiao C, Jiang W, Yang S, Wang H. Fault diagnosis for electromechanical system via
extended analytical redundancy relations. IEEE Trans Industr Inf. 2018;14(12):5233–44.
15. Gangsar P, Tiwari R. Signal based condition monitoring techniques for fault detection and
diagnosis of induction motors: a state-of-the-art review. Mech Syst Signal Process. 2020;1(144):
106908.
16. Chouhan A, Gangsar P, Porwal R, Mechefske CK. Artificial neural network–based fault diag-
nosis for induction motors under similar, interpolated and extrapolated operating conditions.
Noise Vib Worldw. 2021;6:09574565211030709.
17. Vapnik VN. An overview of statistical learning theory. IEEE Trans Neural Netw.
1999;10(5):988–99.
18. Hsu CW, Lin CJ. A comparison of methods for multiclass support vector machines. IEEE Trans
Neural Netw. 2002;13(2):415–25.
19. Rafiee J, Rafiee MA, Prause N, Tse PW. Application of Daubechies 44 in machine fault diag-
nostics. In: 2nd international conference on computer, control and communication, 2009. IC4
2009. IEEE;2009. p. 1–6.
20. Peng ZK, Chu FL. Application of the wavelet transform in machine condition monitoring and
fault diagnostics: a review with bibliography. Mech Syst Signal Process. 2004;18(2):199–221.
Identifying Condition Indicators
for Artificially Intelligent Fault
Classification in Rolling Element
Bearings

Mohd Atif Jamil and Sidra Khanam

Abstract Bearing condition monitoring is significant in industries due to increased

machine reliability and decreased production loss due to machinery breakdown. With
the advancement of Artificial Intelligence (AI), Machine Learning (ML) techniques
are reasonably useful to build condition monitoring systems for real-world applica-
tions. ML algorithms help distinguish faulty bearings from healthy ones and clas-
sify the related fault types using the extracted time-domain and frequency-domain
features. This study recognizes distinctive features or condition indicators that effec-
tively separate different fault groups and are worthy of training an ML model. Box
plot and scatter plot of fault features are used to identify these condition indicators.
Vibration datasets representing various faults are taken from the open-source Case
Western Reserve University (CWRU) bearing database. A number of time-domain
features are extracted from the ensemble data of bearing fault classes, consisting of
healthy bearing, inner race fault, ball fault, and outer race fault. Our investigation
indicates that more than one condition indicator is better for separating the fault cate-
gories. Six different ML models are trained using the condition indicators and the
best-performing model is found through the classification accuracy, training time,
and prediction speed of the classifier.

Keywords Condition monitoring · Bearing faults · Vibration features · Condition

indicators · Machine learning · Fault classification

Nomenclature

CWRU Case Western Reserve University

CF Crest Factor

M. A. Jamil (B) · S. Khanam

Department of Mechanical Engineering, Aligarh Muslim University, Aligarh, India
e-mail: [email protected]
S. Khanam
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 323
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_22
324 M. A. Jamil and S. Khanam

IF Impulse Factor
Ku Kurtosis
REB Rolling Element Bearing
Sk Skewness
X Mean
σ Standard Deviation

1 Introduction

Modern machinery has a high starting cost, and its efficient operation depends on
minimal operating and maintenance expenses. Rotating machinery must have rolling
element bearings (REBs). As Rolling Element Bearings (REBs) are essential compo-
nent of rotating machinery, their health monitoring is essential to reduce unintentional
machinery shutdowns, minimize downtime for maintenance, and to enhance relia-
bility and safety. Condition monitoring and fault diagnostics of REBs have emerged
as key characteristics to meet these requirements. Condition monitoring is a proce-
dure of knowing machinery health by capturing the operational information and
examining it to put a figure/label on the state of equipment. It helps to detect and
diagnose potential problems early in their development and fixing them by suitable
recovery activities before they become hazardous enough to cause machine failure
and other severe consequences. Subsequently, there is a requirement for a large
amount of data for analysis. But the relationship between the bearing health and the
condition monitoring data produced is not always well understood [1]. So, it is a dare
to extract meaningful information from the data for condition monitoring.
For machine monitoring and diagnosis, a number of reliable techniques are
well-established. These techniques include visual examination, stator current anal-
ysis, temperature monitoring, and vibration-based monitoring. The most often used
parameter for identifying damage and monitoring machine condition is the vibration
signature [2, 3]. Condition monitoring of bearings is also carried out using acoustic
emission [4], non-contact infrared thermography [5, 6], and laser Doppler vibrometer
[7]. Recent advances in sensing technology and the internet of things have introduced
an intelligent framework to monitor bearing healthiness [8]. The bearing fault diag-
nosis is more problematic than detection because different faults can have analogous
characteristics and different faults can happen at the same time. The two stages of
the fault diagnosis procedure are vibration feature extraction and the classification
of defects. Typically, features are retrieved using the Fast Fourier Transform (FFT),
Hilbert Transform (HT), Short-Time Fourier Transform (STFT), Wavelet Transform
(WT), both in continuous and discrete form, and the envelope analysis [9]. In order to
identify the most important features, effective dimensionality reduction and feature
selection methods have been used, namely, linear discriminant analysis, Principal
Components Analysis (PCA), Sequential Floating Forward Selection (SFFS) [10],
and Genetic Algorithm (GA), etc. Using ML techniques, fuzzy logic, and Deep
Identifying Condition Indicators for Artificially Intelligent Fault … 325

Learning (DL) approaches, fault classification deals with detecting bearing fault
category. For condition monitoring of REBs, many AI techniques have been used,
such as Artificial Neural Network (ANN) [11], Support Vector Machine [12, 13],
K-Nearest Neighbors (KNN) [14, 15], and Hidden Markov Model (HMM) [16].
The Deep Neural Network (DNN) has also been documented in recent literature to
determine the Remaining Useful Life (RUL) of bearings [17].
The identification of condition indicators, or features in the bearing data whose
behavior change predictably as the bearing deteriorates due to presence of defects,
is a crucial stage in the creation of condition monitoring algorithms. It may be
recognized whether a bearing is healthy or defective using condition indicators. For
fault classification and RUL estimation, they can be extracted from preprocessed data.
The objective of feature extraction is to identify the most condensed and informative
set of features (distinct patterns) to improve the effectiveness of the ML classifier,
thereby achieving accurate classification.
To diagnose faulty and healthy rolling element bearing states, statistical char-
acteristics of time signals can be used as condition indicators. The mean value of
a particular signal or the standard deviation of the signal, for instance, may shift
by a significant amount as the health of the bearing deteriorates. The worsening in
the bearing healthiness may also be visible in some higher-order moments, such as
kurtosis and skewness of the signal. The threshold values that differentiate healthy
operation from defective one can be defined with such features, or changes in the
bearing state can also be revealed by sudden or abrupt changes in the corresponding
values. In the present analysis, the time-domain features namely, mean, median,
kurtosis, skewness, crest factor, and impulse factor, are extracted to check their indi-
vidual and combined effectiveness in classifying different fault categories using box
plot and scatter plot approach, respectively.

2 Experimental Data and Machine Learning

Implementation

Figure 1 illustrates the setup used to collect vibration data for healthy and defective
ball bearings available at the bearing data centre of Case Western Reserve University
(CWRU). The arrangement at its left has a 2 hp induction, a torque transducer in
the center, and a dynamometer attached on the right. Single point faults of 0.007'' ,
0.014'' , 0.021'' , and 0.028'' in diameter were formed artificially on the inner race,
rolling balls, and outer race of test bearings by electro-discharge machining (EDM).
At sampling frequencies of 12,000 and 48,000 Hz, vibration data was obtained
for motor speeds of 1797 to 1720 rpm using two accelerometers that are positioned
on the fan and drive ends of the housing of motor. The CWRU bearing data center
documented the data and made it accessible to the public [18]. It can be used as a
benchmark dataset to assess how well ML algorithms perform.
326 M. A. Jamil and S. Khanam

Fig. 1 Setup for CWRU bearing data collection [18]

2.1 Data Subset

The subset of CWRU data considered for the present analysis consists of 10 samples
of 6000 data points/observations, each from four fault categories; healthy bearing,
inner race fault, ball fault, and outer race fault located at 6 o’clock angular position.
The fault codes 1, 2, 3, and 4 are assigned to these fault categories. Samples are
collected at 12,000 samples per second at an average motor speed of 1772 rpm, and
the fault size is 0.007''.

2.2 Fault Features

The features used have a significant impact on the effectiveness of ML-based bearing
defect detection techniques. The time-domain features described in Table 1 have been
extracted from the raw vibration data and are assessed to be used to train the ML
models for bearing fault classification.
∑n data for i = 1, 2, …, n; n is the total no. of
xi corresponds to some time-series
observations in a sample, x = n1 i=1 |xi | represents the absolute mean, σ denotes
the standard deviation, and xmax = max|xi |.
Identifying Condition Indicators for Artificially Intelligent Fault … 327

Table 1 Description of
S. No Feature Feature description
time-domain features ∑n
extracted from bearing 1 Mean X = n1 i=1 xi
vibration data (n+1)th
2 Median Med = 2 ter m; n is even
n th
ter m+( n2 +1)th ter m
Med = 2
2 ; n is odd
∑n
i=1 (x i −x)
4
3 Kurtosis Ku = (n−1)σ 4
∑n
i=1 (x i −x)
3
4 Skewness Sk = (n−1)σ 3
x max
5 Crest factor CF = xr ms
x max
6 Impulse factor I F = x

2.3 Cross-Validation Approach

The k-fold cross-validation procedure is implemented to assess machine learning

models by splitting the original sample into k equal-sized subsamples. A single
subsample is retained as the validation data, and the remaining k − 1 subsamples
are used as training data. The cross-validation process is then repeated k times, with
each of the k subsamples used exactly once as the validation data. The k outcomes
from the folds are then combined to produce a single estimation. A 8-fold cross-
validation is used on 40 samples, 10 each from the four fault categories considered
in the present work.

3 Results and Discussion

Ensembled raw vibration data and power spectrum of different fault categories are
presented. Boxplots are used to understand how each of the time-domain features;
mean, median, kurtosis, skewness, crest factor, and impulse factor perform indi-
vidually in differentiating between types of faults. Scatter plots of the combination
of features are obtained to investigate how well a particular combination separates
different kinds of faults and can be used as condition indicators to train machine
learning models. Finally, the best-performing ML classifier that has been trained
with the extracted features is shown by its confusion matrix.

3.1 Raw Data

Figure 2 shows the ensemble vibration signal of about 50 ms and the power spectrum
containing measurements of healthy bearings corresponding to fault code 1 and faulty
328 M. A. Jamil and S. Khanam

bearings with three fault types: inner race fault, ball fault, and outer race fault located
at the 6 o’clock position, each with a size of 0.007'' . These fault classes are assigned
with fault codes 2, 3, and 4.
The time-domain ensemble vibration plot includes data of all the healthy and
faulty conditions. The trend in the raw vibration signals of different fault categories
gives an idea of their relative amplitude. As shown in the power spectrum plot, the
resonance due to impact caused by interaction of the respective defect with the mating
element excites the resonant frequencies of the bearing. The resonant frequency band

Fig. 2 a Ensemble vibration signal and b Power spectrum for different fault categories designated
as 1, 2, 3, and 4, shown by blue, red, yellow, and purple colors, respectively
Identifying Condition Indicators for Artificially Intelligent Fault … 329

is almost the same for the inner race fault, ball fault, and outer race fault, in contrast
to the healthy bearing.

3.2 Box Plot

A box plot, commonly referred to as a box and whisker plot, is a form of a chart
that is used in explanatory data analysis in descriptive statistics. Box plots use the
quartiles (or percentiles) of the data and averages to visually depict the distribution of
numerical data. Box plots are used to display numeric data distributions, particularly
when comparing values between various groups. In the present work, the box plots
of individual features are compared with respect to fault classes of the bearing to
check whether a feature can be considered as a condition indicator to train a ML
classifier.
The box plots of time-domain features, mean, median, kurtosis, skewness, crest
factor, and impulse factor, are shown in Fig. 3. In the case of (a) Mean and (d)
Skewness, it can be observed that box plots for fault class 3 and 4 overlap, which
implies that the distribution of mean and skewness values for these two fault classes
are in the same respective ranges and therefore they may not distinguish between
fault classes 3 and 4. Likewise, the (b) Median and (e) Crest Factor for the fault
classes 2 and 4 are not very effective in classifying them because they lie in the same
range owing to their corresponding overlaying box plots. For (c) Kurtosis, it can be
seen that the box plots for fault classes 1 and 3 somewhat overlap, suggesting that
the distribution of kurtosis values for these two fault classes is in the same range. As
a result, they may not be able to differentiate between these classes.
As the boxes don’t overlap for any of the four fault classes in the plot of (f)
Impulse Factor only, there is a difference between the associated data groups, so
Impulse Factor can be used to distinguish the fault categories under consideration.
The other five features may not be considered promising for classifying the faults
when used individually.

3.3 Scatter Plot

To indicate how much one variable affects another, scatter plots are used to exhibit
data points on a horizontal and a vertical axis. A marker is used to represent each
row in the data table, and the position of the marker depends on the values of the
columns that are set up on the X and Y axes.
The scatter plots of the time-domain features in combination with each other are
presented in Fig. 4. The scatter plot of (a) Skewness versus Mean shows a significant
overlapping of data samples of fault classes 3 and 4, which indicates that these two
classes are not clearly distinguished by Skewness and Mean. Similar is the case
of (b) Kurtosis versus Impulse Factor where the fault classes 1 and 3 are not well
330 M. A. Jamil and S. Khanam

(a) (b) 10

(c) (d)

(e) (f)

Fig. 3 Box plot of time-domain features for fault classification—a Mean b Median c Kurtosis
d Skewness e Crest factor and f Impulse factor

differentiated. The plot of (c) Crest Factor versus Mean shows a somewhat better
separation of the fault classes but with a minor closeness of the samples of fault classes
3 and 4. On the other hand, the scatter plots of (d) Impulse Factor versus Mean, (e)
Kurtosis versus Skewness, and (f) Kurtosis versus Mean show good parting of all
the fault classes with maximum separation in the case of (f) Kurtosis versus Mean.
Identifying Condition Indicators for Artificially Intelligent Fault … 331

Thus, these are good condition indicators and may be used to train the ML models
for fault classification.

(a) (b)

(c) (d)

(e) (f)

Fig. 4 Scatter plots for different pairs of time-domain features—a Skewness versus Mean
b Kurtosis versus Impulse Factor c Crest Factor versus Mean d Impulse Factor versus Mean
e Kurtosis versus Skewness f Kurtosis versus Mean
332 M. A. Jamil and S. Khanam

3.4 Confusion Matrix

The confusion matrix of a ML classifier is a summary of classification problem

prediction outcomes. Count values are used to describe the number of accurate and
inaccurate predictions for each class. It gives an insight into the errors being made by
the classifier. The confusion matrices for (a) K-Nearest Neighbor, (b) Decision Tree,
(c) Ensemble Classifier, (d) Discriminant Analysis, (e) Support Vector Machine, and
(f) Gaussian Naïve Bayes models are shown in Fig. 5. The associated results of fault
classification accuracy, time to train the classifier, and the corresponding prediction
speed of these classification models are reported in Table 2.
It is worth noting that the classification accuracy obtained by SVM and Gaussian
Naïve Bayes is the same, but the latter outperforms in terms of training time and
prediction speed. The same is applicable when comparing the Ensemble Classifier
and Discriminant Analysis. Overall, the Gaussian Naïve Bayes model performs best
in the current analysis.

4 Conclusions

Investigating different fault types using a box plot shows that a single time-domain
feature may not be sufficient to classify the faulty behavior, especially in multi-class
fault classification problems. One cannot distinguish between all the fault types
as the box plots of some of the time-domain features overlap, due to which these
features are not enough to set fault types apart. The scatter plot of a combination
of features reveals that two condition indicators are better than one for separating
different faults. Different combinations of features may be tried to see which ones
are better at classifying the defects.
In the present analysis, all the time-domain features except the impulse factor are
insufficient to classify the fault categories when used independently. However some
of their combination in different pairs may successfully distinguish between the fault
classes, which can be easily understood from the scatter plots. Further, the combina-
tion of mean and kurtosis results in the best prediction of different fault categories.
Therefore, these time-domain features and the other extracted features showing a
bit lesser prediction are good candidates to train ML models. The condition indi-
cators thus identified are used to train the ML models. Finally, the best-performing
model may be selected by checking its accuracy using a confusion matrix. In case
the accuracy of two or more classifiers match, the training time and prediction speed
parameters may be used to recognize the best classifier. In this case, out of 6 ML
classifiers, namely, k-Nearest Neighbors (k-NN), Support Vector Machines (SVM),
Naïve Bayes classifier, Discriminant Analysis, Ensemble Classifier, and Decision
Trees. The Gaussian Naïve Bayes is discovered to be the most effective.
There is no pre-determined number regarding how many features are enough to
train a machine learning model. So, as a future direction, the feature extraction step
Identifying Condition Indicators for Artificially Intelligent Fault … 333

(a) (b)

(c) (d)

(e) (f)

Fig. 5 Confusion matrices showing fault classification results—a K-nearest neighbor b Decision
tree c Ensemble classifier d Discriminant analysis e Support vector machine f Gaussian Naïve
Bayes

may be reconsidered, and machine learning models be trained with different sets of
features to check for the possible improvement in the accuracy of ML models. It is
also important to remember that ML models can benefit from a high-dimensional set
of distinguishing features and can effectively differentiate the fault types.
334 M. A. Jamil and S. Khanam

Table 2 Performance results of different ML models

ML classifier Accuracy Training time Prediction speed
(%) (S) (Obs/S)
K-nearest neighbor 90 0.69843 640
(KNN)
Decision tree 92.5 9.6881 1200
Ensemble classifier 95 1.8984 350
Discriminant analysis 95 0.72548 1000
Support vector machine (SVM) 97.5 0.79327 730
Gaussian Naïve Bayes 97.5 0.64558 980

References

1. McArthur SDJ, Strachan SM, Jahn G. The design of a multi-agent transformer condition moni-
toring system. IEEE Trans Power Syst. 2004;19(4):1845–52. https://doi.org/10.1109/TPWRS.
2004.835667.
2. Goyal D, Pabla BS. The vibration monitoring methods and signal processing techniques for
structural health monitoring: a review. Arch Comput Methods Eng. 2016;23(4):585–94. https://
doi.org/10.1007/s11831-015-9145-0.
3. Sehgal S, Kumar H. Damage detection using derringer’s function based weighted model
updating method. Conf Proc Soc Exp Mech Ser. 2014;5:241–53. https://doi.org/10.1007/978-
3-319-04570-2_27.
4. Glowacz A. Fault diagnosis of single-phase induction motor based on acoustic signals. Mech
Syst Signal Process. 2019;117:65–80. https://doi.org/10.1016/j.ymssp.2018.07.044.
5. Younus AMD, Yang BS. Intelligent fault diagnosis of rotating machinery using infrared thermal
image. Expert Syst Appl. 2012;39(2):2082–91. https://doi.org/10.1016/j.eswa.2011.08.004.
6. Lim GM, Bae DM, Kim JH. Fault diagnosis of rotating machine by thermography method on
support vector machine. J Mech Sci Technol. 2014;28(8):2947–52. https://doi.org/10.1007/s12
206-014-0701-6.
7. Vass J, Šmíd R, Randall RB, Sovka P, Cristalli C, Torcianti B. Avoidance of Speckle noise in
laser vibrometry by the use of kurtosis ratio: application to mechanical fault diagnostics. Mech
Syst Signal Process. 2008;22(3):647–71. https://doi.org/10.1016/j.ymssp.2007.08.008.
8. Tokognon Jr. CA, Gao B, Tian GY, Yan Y. Structural health monitoring framework based on
internet of things: a survey. IEEE Internet Things J. 2017;4(3):619–635. https://doi.org/10.
1109/SYSCO.2016.7831328.
9. Shi DF, Wang WJ, Qu LS. Defect detection for bearings using envelope spectra of wavelet
transform. J Vib Acoust Trans ASME. 2004;126(4):567–73. https://doi.org/10.1115/1.180
4995.
10. Pudil P, Novovičová J, Kittler J. Floating search methods in feature selection. Pattern Recognit
Lett. 1994;15(11):1119–25. https://doi.org/10.1016/0167-8655(94)90127-9.
11. Haj Mohamad T, Samadani M, Nataraj C. Rolling element bearing diagnostics using extended
phase space topology. J Vib Acoust Trans ASME. 2018;140(6):1–9. https://doi.org/10.1115/1.
4040041.
12. Widodo A, Yang BS. Support vector machine in machine condition monitoring and fault diag-
nosis. Mech Syst Signal Process. 2007;21(6):2560–74. https://doi.org/10.1016/j.ymssp.2006.
12.007.
13. Jamil MA, Khan MAA, Khanam S. Feature-based performance of SVM and KNN classifiers
for diagnosis of rolling element bearing faults. Vibroeng Procedia. 2021;36–42. https://doi.org/
10.21595/vp.2021.22307.
Identifying Condition Indicators for Artificially Intelligent Fault … 335

14. Pandya DH, Upadhyay SH, Harsha SP. Fault diagnosis of rolling element bearing with
intrinsic mode function of acoustic emission data using APF-KNN. Expert Syst Appl.
2013;40(10):4137–45. https://doi.org/10.1016/j.eswa.2013.01.033.
15. Jamil MA, Khanam S. Fault classification of rolling element bearing in machine learning
domain. Int J Acoust Vib. 2022;27(2):77–90. https://doi.org/10.20855/ijav.2022.27.21829.
16. Zhou H, Chen J, Dong G, Wang H, Yuan H. Bearing fault recognition method based on neigh-
bourhood component analysis and coupled hidden markov model. Mech Syst Signal Process.
2016;66–67:568–81. https://doi.org/10.1016/j.ymssp.2015.04.037.
17. Xia M, Li T, Shu T, Wan J, De Silva CW, Wang Z. A two-stage approach for the remaining
useful life prediction of bearings using deep neural networks. IEEE Trans Ind Inform.
2019;15(6):3703–11. https://doi.org/10.1109/TII.2018.2868687.
18. Case Western Reserve University (CWRU) Bearing Data Center [Online]. https://csegroups.
case.edu/bearingdatacenter/pages/welcome-case-western-reserve-university-bearing-data-cen
ter-website.
Influence of Local Surface Cracks
on Dynamic Parameters of Multi-span
Beam

Mitesh J. Mungla and Dharmendra S. Sharma

Abstract The cracks present in a mechanical/structural member alter flexibility

and thereby change modal parameters. The one-dimensional open cracks in the
multi-span Euler–Bernoulli beam are modeled as rotational spring. The supports
are modeled as a combination of rotational and linear springs. A computer code
is developed to obtain numerical results from the formulation. The effect of crack
locations, crack depths, several supports, and cracks on various modal parameters
are presented for isotropic multi-span cracked beams.

Keywords Multi-span beam · Crack · Modal parameter · Natural frequency ·

Rotational spring

Nomenclature

A First constant of beam expression

Ar Cross-sectional Area
Cd Crack Depth
B Second constant of beam expression
C Third constant of beam expression
D Fourth constant of beam expression
E Young’s Modulus
I Moment of Inertia
Bt Beam Thickness
i Span Numbering
j Support Numbering
l Length of Beam

M. J. Mungla (B)
Indus University, Ahmedabad, India
e-mail: [email protected]
D. S. Sharma
The Maharaja Sayajirao University of Baroda, Vadodara, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 337
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_23
338 M. J. Mungla and D. S. Sharma

k Number of intermediate supports

m Number of cracks
n Number of spans
R Intermediate support location
x Positive linear distance from the nearby support
y Lateral Displacement of beam
α Normalized Location
ω Natural Frequency
ρ Density
ζ Crack Location

1 Introduction

The beam kind of components are omnipresent in most of the systems. With time,
a crack either develops or propagates in such components. The modal parameters
get influenced by the presence of crack/s. Hence, the modal parameters of the beam/
structure contain details about the presence of crack/s, number of cracks, position, and
intensity of crack/s. Modeling of crack is a key parameter to establish the correlations
between modal parameters and crack parameters.
In the bar-like structural members, the crack has been modeled as translation
spring [1, 2] to investigate longitudinal vibration. Also, to study lateral vibrations of
the beam, the crack can be theoretically modeled as rotational spring [3–11]. Also,
the correlation of spring constant and crack parameters have provided theoretical
quantitative information about the variation of crack/s on modal characteristics.
Initially, many researchers [12] have presented studies on the influence of crack
parameters on the modal parameter like fundamental frequencies for the clamped-free
beam. Tsai and Wang [13] developed a theory for the stepped shaft to investigate
the significance of cracks on various parameters. The influence of double cracks
[14, 15] on modal parameters of beam-like structures also opens many paradigms.
The perturbation method as well as transfer matrix method (TMM) can also be
employed to understand the effect of minor cracks on frequencies. Khiem and Lien
[3, 16] used the TMM to develop a corelation between crack parameters with modal
parameters. Finite element model (FEM) can also be used to analyze the effect
of coupling mechanics of mutually perpendicular bending vibrations due to cracks
on mode shapes. Moreover, crack effects, crack detection mechanisms, and health
monitoring strategies in various engineering components such as beam, bar, shaft,
disk, plates, and blades have been widely compiled [17–19].
Above mentioned discussion was limited to the existence and detection of
crack/s in single-span structural components. In many engineering/industrial applica-
tions, structural components are long and slender, which need intermediate supports
to control deflection and to ensure sufficient rigidity. However, such intermediate
supports significantly influence the modal parameters. The effect of intermediate
Influence of Local Surface Cracks on Dynamic Parameters … 339

supports along with the number of cracks on modal parameters attracted many
researchers [20–22] to correlate various structural, crack, and modal parameters.
These methods and studies to correlate crack parameters and modal parame-
ters have been applied to various applications such as to identify damage locations
and intensities [7, 23–27] in beam-like structures or pipes filled with pressurized
fluid [28].
This paper is focused on the effect of various crack parameters like location, inten-
sity, numbers cracks, and intermediate supports, as well as its locations on eigen-
values and modes for multi-span beams. To obtain a generalized formulation of the
beam having multiple intermediate support using thin beam theory, the end support
conditions are generalized using combination of linear and/or torsional springs. The
numerical illustrations reveal many important observations out of the study.

2 Methodology

For the “ith” span of the continuous Euler–Bernoulli beam (Fig. 1), the fundamental
equation can be written as [29] follows

∂ 4 yi 
+ λi4 yi |αi = 0 (1)
∂ x 4 αi
 2 
ω ρi Ari li4 1/4
where λi = E i Ii
. Moreover uniform cross-sectional beam is made of
isotropic and homogeneous material.
The generalized solution of Eq. 1,

yi = Ai cos(λi ∗ αi ) + Ci cosh(λi ∗ αi ) + Bi sin(λi ∗ αi ) + Di sinh(λi ∗ αi ) (2)

The generalized lateral displacement of the multiple span beam (Fig. 1) having
“n” spans and “n + 1” supports can be written as follows:

yi (α) = Ai cos(λi ∗ α) + Ci cosh(λi ∗ α) + Bi sin(λi ∗ α) + Di sinh(λi ∗ α),

0≤α≤1 (3)

Fig. 1 Uncrack multi-span beam supported on linear and torsional springs

340 M. J. Mungla and D. S. Sharma

To generalize the end/support conditions, the beam is supported on springs. The

supports are modeled using torsional and linear springs theoretically. The generalized
end support conditions can be expressed using torsional and linear spring stiffnesses
at the ends (K tl , K ll , K tr , and K lr ).
   
d 2 y1  dy1  d 2 y1  dy1 
EI = K tl = ktl (4a)
d x 2 α=0 d x α=0 d x 2 α=0 d x α=0
   
d 2 yn  dyn  d 2 yn  dyn 
EI = K = k (4b)
d x 2 α=1 d x α=1 d x 2 α=1 d x α=1
tr tr

 
d 3 y1  d 3 y1 
EI = K ll y1 |α=0 = kll y1 |α=0 (5a)
d x 3 α=0 dx3  α=0
 
d 3 yn  d 3 yn 
EI = K lr yn |α=1 = klr yn |α=1 (5b)
d x 3 α=1 d x 3 α=1

where ktl = K tl /E I , ktr = K tr /E I , kll = K ll /E I , and klr = K lr /E I .

The intermediate support locations are normalized and expressed as

1
Rk = lk (6)
L k

The intermediate supports can also be modeled by linear springs having stiffness
(K l(k) ). The generalized intermediate support conditions can be written as
   
dyk  dyk+1  d 2 yk  d 2 yk+1 
= E k Ik = E k+1 Ik+1 (7a–b)
d x α=Rk d x α=Rk d x 2 α=Rk d x 2 α=Rk
 
d 3 yk  d 3 yk 
E k Ik = K y | = kl(k) yk |α=Rk (8a)
d x 3 α=Rk d x 3 α=Rk
l(k) k α=Rk

 
d 3 yk+1  d 3 yk+1 
E k Ik = K l(k) yk+1 |α=Rk = kl(k) yk+1 |α=Rk (8b)
d x 3 α=Rk d x 3 α=Rk

where kl(k) = K l(k) /E k I k .

Applying the end conditions (Eqs. (4a)–(5b)) and intermediate support conditions
(Eqs. (7a–b)–(8b)) in the deflection equation (Eq. (3)), total “4n” simultaneous
equations are derived. They are represented in terms of matrices as follows.

Puc · Q uc = 0 (9)
 
Q uc = A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 − − −
T 1×4n
An−1 Bn−1 Cn−1 Dn−1 An Bn Cn Dn ,
Influence of Local Surface Cracks on Dynamic Parameters …

C(λ) = cos(λ) Ch(λ) = cosh(λ) S(λ) = sin(λ) Sh(λ) = sinh(λ)

Vk1 = (λ/L)3 · S(λRk ) − kl(k) · C(λRk ), Vk2 = −((λ/L)3 · C(λRk ) + kl(k) · S(λRk )), Vk3 = (λ/L)3 · Sh(λRk ) − kl(k) · Ch(λRk ), Vk4 = (λ/L)3 · Ch(λRk ) − kl(k) · Sh(λRk ),
M1 = −(λ/L) cos(λ) + ktr sin(λ) M2 = −(λ/L) sin(λ) − ktr cos(λ) M3 = (λ/L) cosh(λ) − ktr sinh(λ) M4 = (λ/L) sinh(λ) − ktr cosh(λ)
N1 = (λ/L)3 sin(λ) − klr cos(λ) N2 = −(λ/L)3 cos(λ) − klr sin(λ) N3 = (λ/L)3 sinh(λ) − klr cosh(λ) N4 = (λ/L)3 cosh(λ) − klr sinh(λ)
341
342 M. J. Mungla and D. S. Sharma

“m” number of cracks on a beam (Fig. 2) are assumed to have torsional stiffness
(K r ). It can be correlated with depth ratio (C d /Bt ) as shown below [30]:
K r = h f (CEdI/Bt ) kr = h f (C1d /Bt )
where kr = K r /E I and function f (Cd /Bt ) can be computed as

f (Cd /Bt ) = 355.83(Cd /Bt )10 − 769.7(Cd /Bt )9 + 919.5(Cd /Bt )8 − 678.4(Cd /Bt )7
+410.6(Cd /Bt )6 − 199(Cd /Bt )5 + 87.5(Cd /Bt )4 − 21.1(Cd /Bt )3 + 10(Cd /Bt )2
(11)

The beam having “n” spans (with identical material and geometrical properties)
and “m” cracks is shown in Fig. 2. A crack present in any span of the beam divides
it into two segments.
Hence, the lateral displacements of two segments (before and after) of “jth” crack
( j = 1, 2, 3,……, m) and “ith” span can be expressed as

yi j (α) = Ai j cos(λi ∗ α) + Ci j cosh(λi ∗ α) + Bi j sin(λi ∗ α) + Di j sinh(λi ∗ α),

(12a)

yi( j+1) (α) = Ai( j+1) cos(λi ∗ α) + Ci( j+1) cosh(λi ∗ α) + Bi ( j+1) sin(λi ∗ α)
+ Di( j+1) sinh(λi ∗ α) (12b)

The crack in the beam influences elasticity of the beam. At the location of crack,
the slope of deflection curve gets changed. At the “jth” crack location ζ j , the
following compatibility conditions are required to be met,
 
yi j ζ j = yi ( j+1) ζ j (13)
 
d 2 yi j  d 2 yi ( j+1) 
E i Ii = E i Ii (14)
d x 2 ζ j d x 2 ζ j
 
d 3 yi j  d 3 yi ( j+1) 
E i Ii = E I (15)
d x 3 ζ j d x 3 ζ j
i i

  
dyi j  λ d 2 yi j  dyi ( j+1) 
+ = (16)
d x ζ j kr (i) L d x 2 ζ j d x ζ j

Applying compatibility conditions at every intermediate support (Eqs. (7a–b)–

(8b)), at every crack location (Eqs. 13–16), and end support conditions (Eqs. 4a–5b)
in Eqs. 12a–12b, total “4(m + n)” equations are obtained as follows:

Pc · Q c = 0 (17)
 
Qc = A11 B11 C11 D11 A12 B12 C12 D12 A13 B13 C13 D13 A23

B23 C23 D23 − − − A(n−1)m B(n−1)m C(n−1)m D(n−1)m Anm Bnm Cnm
T 1×4(m+n)
Influence of Local Surface Cracks on Dynamic Parameters …

Dnm An(m+1) Bn(m+1) Cn(m+1) Dn(m+1) ,

       
Lkr ( j ) Lkr ( j ) Lkr ( j ) Lkr ( j )
T1 j =− λ sin(λζ j ) + cos(λζ j ) , T3 j = λ sinh(λζ j ) + cosh(λζ j ) , T5 j = λ sin(λζ j ) , T7 j = − λ sinh(λζ j ) ,
       
Lkr ( j ) Lkr ( j ) Lkr ( j ) Lkr ( j )
T2 j = λ cos(λζ j ) − sin(λζ j ) , T4 j = λ cosh(λζ j ) + sinh(λζ j ) , T6 j = − λ cos(λζ j ) , T8 j = − λ cosh(λζ j ) .
343
344 M. J. Mungla and D. S. Sharma

Fig. 2 Multi-span beam supported on linear and torsional springs having multiple cracks

3 Results and Discussion

A generalized solution to study eigenvalues and eigenvectors of a cracked/uncracked

beam is presented in the previous section. A code, capable of handling different end
fixity conditions, number of cracks at different locations, crack severities, and number
of intermediate supports, is prepared.
Eigenvalues (λ) and eigenvectors of pre and post-crack/s conditions can be calcu-
lated from Eqs. 9 and 17 respectively. For non-trivial solutions of the above two
equations, the following conditions must be met.


det Puc λ, K ll , K lr , K tl , K tr, K l(1) , K l(2) , ..., K l(n−1) , R1 , R2 , ..., Rn−1 4n×4n = 0
(18)
 
Pc (λ, K r (1) , K r (2) , ..., K r (m) , ζ1 , ζ2 , ...., ζm , K ll , K lr , K tl , K tr ,
det =0
K l(1) , K l(2) , ..., K l(n−1) , R1 , R2 , ..., Rn−1 )
4(m+n)×4(m+n)
(19)

By setting the values of linear and torsional spring stiffness (K ll , K lr , K tl & K tr )

and stiffness of intermediate linear springs (K l(1) , K l(2) ,…, K l(n − 1) ), different types
of the beam with cracks can be obtained.
The modal parameters of uncrack and (multiple) cracked single-span beam can
be obtained from Eqs. 18 and 19 respectively by replacing the magnitudes of spring
stiffness as well as intermediate dimensionless support locations (R1 , R2 ,…, Rn − 1 ).
For the sake of comparison, eigenvalues for the virgin beam with known boundary
conditions are presented in Table 1.
Also, frequencies of a single-span beam having two cracks (Table 2) are obtained
and compared with available literature (Hu and Liang 1993).
In the case of the two-span (pinned–pinned-pinned (PPP)) beam, eigenvalues of
the uncracked beam are shown in Table 3 and also compared with available reference.
The PPP beam arrangement is obtained by replacing the magnitudes of linear spring
stiffness (K ll & K lr ) equal to zero, torsional spring stiffness (K tl & K tr ) equal to
infinity, intermediate dimensionless support locations (R1 = 0.5 & R2 = R3 = … =
Table 1 Eigenvalues of pinned–pinned (PP) beam (uncracked) with extreme conditions of spring stiffness
Sr. No Torsional spring Linear spring Torsional spring Linear spring End conditions Eigenvalues
(K tl ) (K ll ) (K tr ) (K lr ) First Second Third Fourth Fifth
1 ∞ ∞ 0 0 Fix-Free 1.87515 4.69410 7.85480 10.99555 14.13720
2 0 ∞ 0 ∞ Pin-Pin 3.14160 6.28320 9.42480 12.56640 15.70800
3 ∞ ∞ ∞ ∞ Fix-Fix 4.73005 7.85325 10.99565 14.13720 17.27880
4 0 Fix-Pin 3.92665 7.06860 10.21020 13.35180 16.49340
Influence of Local Surface Cracks on Dynamic Parameters …

∞ ∞ ∞
345
346 M. J. Mungla and D. S. Sharma

Table 2 Comparison of frequencies (eigenvalues) of the single-span beam having two cracks
Crack Crack depth Frequencies (eigenvalues) End
locations ratio First Second Third condition

Proposed 0.25 and 0.08 and 0.1 58.8818 235.0582 529.9584 Pin-Pin
method 0.45 (3.13825) (6.27025) (9.41495)
[14] 0.25 and 0.08 and 0.1 58.625 235.142 528.096
0.45 (3.1314) (6.27137) (9.39839)

Table 3 Comparison of eigenvalues of uncrack two-span beam (PPP)

Natural frequency
First Second Third Fourth Fifth Sixth
Proposed approach 3.142 3.927 6.283 7.069 9.425 10.210
[20] 3.14 3.926602 6.28 7.068582 9.42 10.21017

Rn − 1 = 0), and intermediate linear spring stiffness (K l(1) , K l(2),…, K l(n − 1) ) equal to
infinity in Eq. 18. Similarly, any other end conditions can be obtained.
The eigenvalues of multi-span beam largely depend upon the end fixity conditions
and intermediate support conditions. Depending upon the end conditions the spring
stiffness can be other than zero and infinity, correspondingly different problems of
the beam on elastic support can be solved.

3.1 Number of Spans

Long and slender beam-like structures always need intermediate supports. These
intermediate supports affect modal parameters (eigenvalue and eigenvector) of the
system. The eigenvalues of the uncracked multiple supported beam are arrived at
from Eq. 18.
For a simply supported beam, Eq. 18 is reduced to sin(λ) = 0 and corresponding
eigenvalues are obtained (λ = 3.142, 6.283, 9.4248, etc.). Similarly, for the two-span
beam with all pinned supports, the above equation is reduced to sin(λ)(tan(λ) −
tanh(λ)) = 0 and correspondingly two sets of eigenvalues are obtained, first set
sin(λ) = 0(λ = 3.142, 6.283, 9.4248, etc.) and second set tan(λ) − tanh(λ) = 0 (λ =
3.9266, 7.0686, 10.2102, etc.). Similarly, sets of frequencies of different spans and
boundary conditions can also be obtained.
The eigenvalues of single, two, three, and four spans beam having simply
supported (P_P) and clamped (F_F) end conditions and all intermediate pinned (_P_)
supports (without crack/s) are presented in Table 4. The intermediate supports are
placed equidistant to each other.
Influence of Local Surface Cracks on Dynamic Parameters … 347

Table 4 Eigenvalues of multi-span beam (uncracked)

Support conditions Eigenvalues
First Second Third Fourth
PP 3.142 6.283 9.425 12.566
PPP 3.142 3.927 6.283 7.069
PPPP 3.142 3.558 4.304 6.283
PPPPP 3.142 3.394 3.927 4.473
FF 4.730 7.853 10.996 14.137
FPF 3.930 4.724 7.073 7.865
FPPF 3.559 4.306 4.743 6.708
FPPPF 3.394 3.932 4.475 6.541

It is important to observe (from Table 4) that every intermediate support changes

frequency subsets. The number of frequency subsets is proportional to the number
of spans.
In the case of two spans (PPP) beam having a crack, a three-dimensional graph
(Fig. 3) represents variation of crack locations and intensity on the first two frequen-
cies. The fundamental eigenvalue is significantly reduced at the mid location of each
span (0.25 and 0.75) but remains unaffected at extreme ends for any crack depth. On
the contrary, the second eigenvalue remains unaffected at mid locations of the span,
regardless of any crack depth, but significantly reduced at ends.
A similar graph of the FPF (fixed-pinned-fixed) supports for two-span beam with
a crack is presented in Fig. 4.
The eigenvalues of the beam having two, three, and four spans having a crack are
obtained (using Eq. 19) and presented in Tables 5, 6, and7 respectively (P_ _ _P).
Figures 5, 6, and 7 show the eigenvectors (mode shape) difference of uncracked

Fig. 3 Variation of crack parameters on first two normalized eigenvalues of two-span (PPP) beam
348 M. J. Mungla and D. S. Sharma

Fig. 4 Variation of crack parameters on first two normalized eigenvalues of two-span (FPF) beam

and cracked beams with two, three, and four spans respectively. The mode shape
difference for three spans cracked beam shows close confirmation with [22].
Similarly, eigenvalues and corresponding mode shape differences of the beam
having a crack with two and three intermediate pinned supports and fixed ends are
presented in Tables 8 and 9 and Figs. 8 and 9 respectively.

Table 5 Eigenvalues of the two-span beam (PPP) having a crack

Crack location Crack depth First Second Three Four
0.15 0.1 3.14112 3.92590 6.28185 7.06738
0.15 0.3 3.13732 3.92033 6.27145 7.05815
0.15 0.5 3.12643 3.90500 6.24255 7.03465
0.25 0.1 3.14085 3.92598 6.28320 7.06837
0.25 0.3 3.13510 3.92113 6.28320 7.06670
0.25 0.5 3.11880 3.90815 6.28320 7.06205

Table 6 Eigenvalues of the three-span beam (PPPP) having a crack

Crack location Crack depth First Second Three Four
0.15 0.1 3.14087 3.55537 4.29708 6.28299
0.15 0.3 3.13517 3.54743 4.29367 6.28197
0.15 0.5 3.11838 3.52677 4.28497 6.27910
0.25 0.1 3.14123 3.55632 4.29743 6.28170
0.25 0.3 3.13837 3.55555 4.2967 6.26997
0.25 0.5 3.13045 3.5535 4.29465 6.23552
Influence of Local Surface Cracks on Dynamic Parameters … 349

Table 7 Eigenvalues of the four-span beam (PPPPP) having a crack

Crack location Crack depth First Second Three Four
0.15 0.1 3.14093 3.39235 3.92629 4.45812
0.15 0.3 3.13560 3.38572 3.92414 4.44542
0.15 0.5 3.11975 3.36911 3.91866 4.42154

Fig. 5 The difference in Eigenvector of uncracked and cracked two spans (PPP) beam having a
crack (ζ = 0.15)

Fig. 6 Eigenvector difference of intact and cracked three spans (PPPP) beam having a crack (ζ =
0.2)
350 M. J. Mungla and D. S. Sharma

Fig. 7 Eigenvector difference of uncracked and cracked four spans (PPPPP) having a crack (ζ =
0.15)

Table 8 Eigenvalues of the three-span beam (FPPF) having a crack

Crack location Crack depth First Second Three Four
0.15 0.1 3.55620 4.29670 4.72928 6.69854
0.15 0.3 3.55448 4.29027 4.72353 6.68452
0.15 0.5 3.54967 4.27228 4.70900 6.68241
0.25 0.1 3.55583 4.29718 4.73003 6.69745
0.25 0.3 3.55138 4.29448 4.72987 6.69351
0.25 0.5 3.53908 4.28723 4.72942 6.68784

Table 9 Eigenvalues of the four-span beam (FPPPF) having a crack

Crack location Crack depth First Second Three Four
0.15 0.1 3.2939 3.92579 4.46541 6.51235
0.15 0.3 3.39031 3.91949 4.44512 6.50247
0.15 0.5 3.38286 3.90195 4.44121 6.50021

The existence of crack/s in the beam increases flexibility. Also, the mode shape
difference of crack and uncrack beam (Figs. 5, 6, 7, 8, and 9) shows due to the
presence of crack, the slope abruptly changes. As appears from the tables (Tables 5,
6, 7, 8, and 9), the increment in depth leads to a further reduction in flexibility.
Influence of Local Surface Cracks on Dynamic Parameters … 351

Fig. 8 Eigenvector difference of uncracked and cracked three spans (FPPF) beam having a crack
(ζ = 0.2)

Fig. 9 Eigenvector difference of uncracked and cracked four spans (FPPPF) having a crack (ζ =
0.15)

3.2 Multiple Cracks

The presence of multiple cracks in the multiple span beam shows significantly
different behavior compared to uncracked and single cracked beams. To under-
stand the effect of multiple cracks on eigenvalues and eigenvectors of the multiple
352 M. J. Mungla and D. S. Sharma

span beam, various four crack locations (0.15, 0.35, 0.65 & 0.85) and crack depths
(0.1, 0.2, 0.3 & 0.4) are taken for the study. Eigenvalues of the two-span beam (PPP
and FPF) having one, two, three, and four cracks are presented in Tables 10 and 11
respectively. Mode shape difference of crack and uncrack beam having two, three,
and four cracks in two, three, and four span beam are presented in Figs. 10, 11, and
12 respectively.

Table 10 Eigenvalues of the two-span beam (PPP) having one, two, three, and four cracks
ζ1 ζ2 ζ3 ζ4 (a/h) Eigenvalues (λ)
First Second Third Fourth
Uncracked beam – 3.14160 3.92663 6.28320 7.06860
0.15 – – – 0.1 3.14112 3.95590 6.28185 7.06738
0.15 0.35 – – 3.14063 3.92588 6.28050 7.06605
0.15 0.35 0.65 – 3.14015 3.92582 6.27915 7.06473
0.15 0.35 0.65 0.85 3.13965 3.92510 6.27780 7.06350
0.15 – – – 0.2 3.13975 3.92385 6.27808 7.06498
0.15 0.35 – – 3.13790 3.92378 6.27288 7.05908
0.15 0.35 0.65 – 3.13605 3.92368 6.26790 7.05398
0.15 0.35 0.65 0.85 3.13422 3.92093 6.26283 7.04930
0.15 – – – 0.3 3.13732 3.92033 6.27145 7.05815
0.15 0.35 – – 3.13312 3.92013 6.25932 7.04725
0.15 0.35 0.65 – 3.12887 3.91988 6.24820 7.03538
0.15 0.35 0.65 0.85 3.12470 3.91355 6.23668 7.02465
0.15 – – – 0.4 3.13330 3.91455 6.26063 7.04898
0.15 0.35 – – 3.12533 3.91423 6.23662 7.02905
0.15 0.35 0.65 – 3.11712 3.91375 6.21625 7.00555
0.15 0.35 0.65 0.85 3.10925 3.90158 6.19447 6.98498

Table 11 Eigenvalues of the two-span beam (FPF) having one, two, three, and four cracks
ζ1 ζ2 ζ3 ζ4 (a/h) Eigenvalues (λ)
First Second Third Fourth
Uncracked beam – 3.92663 4.73005 7.06860 7.85323
0.15 – – – 0.1 3.92660 4.72987 7.06728 7.85158
0.15 0.35 – – 3.92585 4.72973 7.06602 7.84997
0.15 0.35 0.65 – 3.92513 4.72955 7.06480 7.84833
0.15 0.35 0.65 0.85 3.92510 4.72940 7.06350 7.84670
(continued)
Influence of Local Surface Cracks on Dynamic Parameters … 353

Table 11 (continued)
ζ1 ζ2 ζ3 ζ4 (a/h) Eigenvalues (λ)
First Second Third Fourth
0.15 – – – 0.2 3.92652 4.72943 7.06354 7.84708
0.15 0.35 – – 3.92378 4.72880 7.05880 7.84105
0.15 0.35 0.65 – 3.92100 4.72818 7.05427 7.83478
0.15 0.35 0.65 0.85 3.92093 4.72755 7.04930 7.82860
0.15 – – – 0.3 3.92640 4.72860 7.05703 7.83935
0.15 0.35 – – 3.92007 4.72720 7.04587 7.82597
0.15 0.35 0.65 – 3.91378 4.72575 7.03585 7.81138
0.15 0.35 0.65 0.85 3.91355 4.72433 7.02465 7.79728
0.15 – – – 0.4 3.92617 4.72727 7.04620 7.82712
0.15 0.35 – – 3.91408 4.72465 7.02410 7.80283
0.15 0.35 0.65 – 3.90195 4.72182 7.00603 7.77412
0.15 0.35 0.65 0.85 3.90158 4.71915 6.98498 7.74720

Fig. 10 Eigenvector difference of cracked and uncracked two spans (PPP) beam having two cracks
(ζ = 0.25 and 0.75)

It is noticed (from Tables 10 and 11) that the availability of crack/s decreases the
rigidity of the multi-span beam and the presence of each additional crack further
reduces its rigidity. Eigenvalues are also affected by crack severity. The cracks with
significant depth decrease the eigenvalues of the beam.
354 M. J. Mungla and D. S. Sharma

Fig. 11 Mode shape difference of intact and cracked three-span (PPPP) beam having three cracks
(ζ = 0.1, 0.5 and 0.85)

Fig. 12 Mode shape difference of intact and cracked four-span beam (PPPPP) having four cracks
(ζ = 0.1, 0.35, 0.65, and 0.85)

3.3 Influence of Support Location

The behavior of eigenvalues concerning intermediate support location was observed

to be much noticeable. To understand the influence of intermediate support location
on frequency, two cases are discussed,
Influence of Local Surface Cracks on Dynamic Parameters … 355

(i) Intermediate support location is varied (0.1 ≤ R1 ≤ 0.9), and crack is located
at fixed place.
(ii) A crack is at the center of the first span (0.05 ≤ ξ1 ≤ 0.45) with intermediate
support location (0.1 ≤ R1 ≤ 0.9).
Figures 13a–b and 14a–b show the variation in frequency ratio with support loca-
tion for PPP (pinned–pinned-pinned) and FPF (fixed-pinned-fixed) arrangements for
first and second cases respectively.
In the first case, for all the crack depths under consideration (a/h = 0.1, 0.3, 0.5),
for both arrangements, negligible change in frequency ratio is observed for support
location 0.4 ≤ R1 ≤ 0.6. The intermediate support in the vicinity of the crack

Fig. 13 Effect of support location on normalized frequency for two spans arrangement having a
fixed location crack a PPP beam b FPF beam

Fig. 14 Effect of support location on normalized frequency for two spans arrangement having a
movable location crack a PPP beam b FPF beam
356 M. J. Mungla and D. S. Sharma

minimizes the reduction in the rigidity of the beam and thereby makes this crack less
severe than intermediate support away from the crack.
In the second case, along with the intermediate support (0.1 ≤ R1 ≤ 0.9), the
crack also moves (0.05 ≤ ξ1 ≤ 0.45) from left to right. For both the arrangement
(PPP and FPF), minor variation in frequency is observed for intermediate support
0.1 ≤ R1 ≤ 0.35, while a significant change in frequency is observed for intermediate
support location 0.35 ≤ R1 ≤ 0.57. Also, minimum rigidity is observed for interme-
diate support location at 0.57. The presence of intermediate support toward the left
or right support alters the rotational freedom there and leads toward some increment
ωc1 /ωu1 . This can be observed from Fig. 14a, b. The increment in ωc1 /ωu1 the range
0.57 ≤ R1 ≤ 1 is significantly higher compared to support location 0.1 ≤ R1 ≤ 0.35,
due to the availability of a crack toward the left side of the intermediate support. The
dip in the graph (in the range of intermediate support (0.35 ≤ R1 ≤ 0.57)) is due to
the reduced flexibility of the beam.

4 Concluding Remarks

A general solution to study the effect of the availability of crack on vibration parame-
ters in multi-span beams is presented. The presence of the crack/s alters the flexibility
of the beam and thereby reduces the modal frequency. The cracks are modeled as
torsional springs and support a combination of rotational and linear springs.
The modal parameters depend on crack locations, number of cracks, number
of intermediates supports, and crack depths of multi-span beam. The formula-
tion presented here is easy to apply and valid for time-invariant, isotropic, and
homogeneous materials.

References

1. Adams RD, Cawley P, Pye CJ, Stone BJ. A vibration technique for non-destructively assessing
the integrity of structures. J Mech Eng Sci. 1978;20(2):93–100.
2. Adams RD, Walton D, Flitcroft JE, Short D. Vibration testing as a non-destructive test tool for
composite materials. Am Soc Test Mater. 1975;580:159–75.
3. Khiem NT, Lien TV. A simplified method for natural frequency analysis of a multiple cracked
beam. J Sound Vib. 2001;245(4):737–51. https://doi.org/10.1006/jsvi.2001.3585.
4. Mungla MJ, Sharma DS, Makwana H, Bhavsar RR. Inverse dynamics of cracked cantilever
beam. Paper presented at: international conference on advances in design and manufacturing
(ICAD&M’14). NIT Tiruchirapalli, India;2014.
5. Mungla MJ, Sharma DS, Trivedi RR. Detection of crack using genetic algorithm.
Paper presented at: international conference on advances in design and manufacturing
(ICAD&M’14). NIT Tiruchirapalli, India;2014.
Influence of Local Surface Cracks on Dynamic Parameters … 357

6. Ostachowicz WM, Krawczuk M. Vibration analysis of a cracked beam. Comput Struct.

1990;36(2):245–50. https://doi.org/10.1016/0045-7949(90)90123-J.
7. Rizos PF, Aspragathos N, Dimarogonas AD. Identification of crack location and magnitude in
a cantilever beam from the vibration modes. J Sound Vib. 1990;138(3):381–8.
8. Mungla MJ, Sharma DS. Forward dynamics to the euler-bernoulli beam having multiple cracks.
Paper presented at: international conference on vibration problems (13th) ICOVP. Indian
Institute of Technology Guwahati, India;2017.
9. Mungla MJ, Sharma DS, Trivedi RR. Identification of a crack in clamped-clamped beam using
frequency-based method and genetic algorithm. Procedia Eng. 2016;144:1426–34. https://doi.
org/10.1016/j.proeng.2016.05.174.
10. Mungla MJ, Sharma DS, Trivedi RR. Inverse method to identify crack parameters in multi-span
beam using genetic algorithm. Nondestruct Test Eval. 2016;32(3):301–18. https://doi.org/10.
1080/10589759.2016.1226302.
11. Sharma DS, Mungla MJ, Barad KH. Vibration-based non-destructive technique to detect crack
in multi-span beam. Nondestruct Test Eval. 2015;30(4):291–311. https://doi.org/10.1080/105
89759.2015.1029475.
12. Chondros TG, Dimarogonas AD. Identification in welded joints of cracks of complex structures.
J Sound Vib. 1980;69(4):531–8.
13. Tsai TC, Wang YZ. Vibration analysis and diagnosis of a cracked shaft. J Sound Vib.
1996;192(3):607–20.
14. Hu J, Liang RY. An integrated approach to detection of cracks using vibration characteristics.
J Franklin Inst. 1993;330(5):841–53.
15. Shifrin EI, Ruotolo R. Natural frequencies of a beam with an arbitrary number of cracks. J
Sound Vib. 1999;222(3):409–23.
16. Khiem NT, Lien TV. The dynamic stiffness matrix method in forced vibration analysis of
multiple-cracked beam. J Sound Vib. 2002;254(3):541–55. https://doi.org/10.1006/jsvi.2001.
4109.
17. Dimarogonas AD. Vibration of cracked structures: a state of the art review. Eng Fract Mech.
1996;55(5):831–57. https://doi.org/10.1016/0013-7944(94)00175-8.
18. Salawu OS. Detection of structural damage through changes in frequency: a review. Eng Struct.
1997;19(9):718–23. https://doi.org/10.1016/S0141-0296(96)00149-6.
19. Sekhar AS. Multiple cracks effects and identification. Mech Syst Signal Process.
2008;22(4):845–78. https://doi.org/10.1016/j.ymssp.2007.11.008.
20. Ichikawa M, Miyakawa Y, Matsuda A. Vibration analysis of the continuous beam subjected to
a moving mass. J Sound Vib. 2000;230(3):493–506. https://doi.org/10.1006/jsvi.1999.2625.
21. Khiem NT, Khang PX, Hang PT. A Method for Calculation natural frequenies of multispan
beam with arbitrary number of cracks. Paper presented at: National conference on engineering
mechanics. Hanoi, Vietnam; (2014)
22. Lien TV, Hao TA. Determination of the shape function of a multiple cracked beam element
and its application for the free vibration analysis of a multiple cracked frame structure. Am J
Civ Eng Arch. 2014;2(1):12–25. https://doi.org/10.12691/ajcea-2-1-2.
23. Barad KH, Sharma DS, Vyas VN. Crack detection in cantilever beam by frequency based
method. Procedia Eng. 2013;51:770–5. https://doi.org/10.1016/j.proeng.2013.01.110.
24. Chinchalkar S. Determination of crack location in beams using natural frequencies. J Sound
Vib. 2001;247(3):417–29. https://doi.org/10.1006/jsvi.2001.3748.
25. Dado MH. A comprehensive crack identification algorithm for beams under different end
conditions. Appl Acoust. 1997;51(4):381–98. https://doi.org/10.1016/S0003-682X(97)000
05-4.
26. Patil DP, Maiti SK. Detection of multiple cracks using frequency measurements. Eng Fract
Mech. 2003;70(12):1553–72. https://doi.org/10.1016/s0013-7944(02)00121-2.
27. Ratcliffe CP. A frequency and curvature based experimental method for locating damage in
structures. J Vib Acoust. 2000;122(3):324–324. https://doi.org/10.1115/1.1303121.
358 M. J. Mungla and D. S. Sharma

28. Murigendrappa SM, Maiti SK, Srirangarajan HR. Frequency-based experimental and theo-
retical identification of multiple cracks in straight pipes filled with fluid. NDT and E Int.
2004;37(6):431–8. https://doi.org/10.1016/j.ndteint.2003.11.009.
29. Rao SS. Vibration of continuous systems. Wiley; 2007.
30. Dimarogonas AD, Paipetis SA. Analytical methods in rotor dynamics. London: Elsevier
Applied Science; 1983.
Crack Detection in a Shaft Using Wavelet
Packet Transform

Ramnivas Kumar and Sachin K. Singh

Abstract The existence of a crack in a shaft causes a local slope discontinuity in

the shaft elastic line. The crack in a shaft can be located by detecting this local slope
discontinuity in the shaft elastic line. Wavelet transform is a powerful technique for
extracting the local information from the signal. One drawback of discrete wavelet
transform (DWT) is that it analyzes the signal at the coarser level). Wavelet packet
transform (WPT) overcomes this difficulty by analyzing the signal at finer level. In
DWT, only the output low-pass filter is taken for further decomposition. However,
in WPT, the output of low passes, as well as high passes filters, are taken for further
level of decomposition. In this way, the WPT performs a complete decomposition
at each level and hence can reach a higher resolution in the high-frequency zone. In
the present work, WPT is used to locate the crack along the shaft length. Another
critical issue for wavelet-based crack detection is spatial measurement resolution. In
the present work, the photographic image-based technique is used to obtain the high
spatial measurement resolution of shaft deflection. The algorithm is also employed
for two simultaneous crack detection.

Keywords Crack detection · Wavelet packet transform · Shannon entropy · Image

edge extraction

Nomenclature

ψ(x) Mother wavelet

u Translation parameter
s Scale parameter
x Location along the shaft length

R. Kumar · S. K. Singh (B)

Department of Mechanical Engineering, Indian Institute of Technology (Indian School of Mines),
Dhanbad, Jharkhand 826004, India
e-mail: [email protected]
R. Kumar
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 359
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_24
360 R. Kumar and S. K. Singh

k Translation index
i Scale index
h(u) Discrete filter corresponding to the scaling function
g(u) Discrete filter corresponding to the mother wavelet function
m Modulation parameters
S Scale parameter
u Translation parameters
A Approximate coefficient
D Detailed transformation
J Level of decomposition

1 Introduction

Fatigue cracks present in a machine element make the machine vulnerable to catas-
trophic failure. It is necessary to detect all the potentially unsafe cracks present in a
beam before the occurrence of failure. Assessment of damage severity is also impor-
tant to predict the remaining useful life, schedule the maintenance, and prepare the
inventory.
The presence of a crack in a beam engenders a slope discontinuity (SD) in the
beam elastic line. SD in the beam elastic line occurs due to a change in local flexibility
at the vicinity of crack. Different models have been proposed to model the crack in
a shaft. Adams et al. [1] used the concept of a linear spring of small infinitesimal
length to model the effect of a crack in a beam. Dimarogonas and Massouros [2]
modeled a cracked rotor by using strain energy release rate. The details of different
methods to model the effect of a crack in structures were presented by Dimarogonas
[3].
When a crack appears in a structure, it can significantly affect the modal parame-
ters, including natural frequencies, modal damping, and mode shapes. This is because
the presence of a crack changes the stiffness and mass distribution of the structure,
leading to changes in its dynamic behavior [4, 5]. Among all the modal parameters
the change in natural frequency was widely used for the identification of crack [6,
7]. However, the natural frequency of the structures may alter due to wear and envi-
ronmental degradation [8]. Some researchers utilized the concept of mode shape to
locate the crack [9–11]. Further, the concept of modal curvature was utilized to locate
the crack along the beam length [12]. The operating deflection shape of the shaft is
obtained by exciting it at a particular frequency [13, 14]. The existence of cracks
creates local SD in the deflection shape of the beam. The location of a crack in the
spatial domain can be revealed by exposing the SD. The concept of crack probability
functions was proposed to identify the crack in a shaft [14].
In actual practice, the mode shapes are affected by measurement noise, making the
crack detection intricate. Wavelet transform (WT) is a powerful method to locate the
discontinuities in the presence of noise [15–17]. A discrete wavelet transform (DWT)
Crack Detection in a Shaft Using Wavelet Packet Transform 361

can be viewed as applying a multi-resolution filter bank on the signal [18]. In the
case of DWT, only the output of LPF is subjected to further levels of decomposition.
The output of HPF, called detailed information, is generally analyzed to extract
the signal information. Thus the multi-resolution analysis (MRA) is the result of
repeatedly projecting the approximation to a function of other subspaces to generate
coarser approximations [18]. The wavelet packet transform (WPT) decomposes both
the detailed and approximate coefficients [19]. In this way, the WPT is capable of
achieving a higher resolution in the high-frequency zone because it conducts a full
decomposition at each level [20].
On the other hand, the DWT suffers from a relatively low-frequency resolution in
the high-frequency zone [20]. This deficiency makes DWT inapt for differentiating
the high-frequency noise and SD caused by the crack. The WPT is the extension of
the WT that offers a complete level-by-level decomposition. This complete level-by-
level decomposition enables to extract the discontinuity information in the presence
of high noise [21].
In the present work, WPT is used to locate the cracks in beams. First, the simulated
responses of a cracked shaft are used to study the effect of measurement noise
on the crack detection capabilities of WPT. Thereafter, the beam spatial responses
are obtained from real single and doubled-cracked shafts. The high measurement
resolution is a requirement of localized crack detection. In this work, the photographic
method is used to get the shaft deflection at a relatively lower cost. Also, the high
spatial resolution shrinks the edge effect zone and enables crack detection close to
the ends of the shaft.

2 Methodology

2.1 Mathematical Modeling of Cracked Shaft

The Timoshenko beam theory is used for analyzing the response of cracked shafts.
In this approach, the shaft is discretized using FEM, as shown in Fig. 1. The effect of
the crack is modeled by the concept of increase in the local flexibility of the beam.
A shaft element with a crack is shown in Fig. 2.

Fig. 1 FE discretization of a cracked shaft

362 R. Kumar and S. K. Singh

Fig. 2 A shaft element having a crack

Let, P1 be the axial force, P2 and P3 the shearing forces, P4 and P5 the bending
moments, and P6 the torque, d the crack depth. The supplementary flexibility matrix
due to the presence of a crack is taken from reference [22]. It was assumed that
the axial and torsional forcing of the shaft are negligible and only the transverse
forcing is considered. Due to the existence of a crack, the supplementary flexibility
coefficients matrix can be expressed as [22],
⎡ ⎤
C22 0 0 0
⎢ 0 C33 0 0 ⎥
[CC ]e = ⎢
⎣ 0
⎥ (1)
0 C44 C45 ⎦
0 0 C54 C55

The total flexibility matrix of a cracked beam can be given as [23, 24]

[c] = [C0 ]e + [Cc ]e (2)

Here, [C0 ] and [Cc ] denotes the flexibility matrices of an element without a crack and
with crack, respectively. The cracked element stiffness matrix [S]e can be obtained
as [24]

[S]e = [T ][C]−1 [T ]T (3)

where [T ] is the transformation matrix, expressed as [24],

Crack Detection in a Shaft Using Wavelet Packet Transform 363

⎡ ⎤
−1 0 0 −l 1 0 0 0
⎢ 0 −1 l 0 0 1 0 0⎥
[T ]T = ⎢
⎣ 0
⎥ (4)
0 −1 0 0 0 1 0⎦
0 0 0 −1 0 0 0 1

To model the effects of damping, proportional damping is considered [23]. For

the intact and cracked elements, the elemental equations of motion can be expressed
as

¨ +[Dwc ]e {q(t)}
[M]e {q(t)}
e
˙ +[K wc ]e {q(t)}e = { f (t)}e
e
(5)

and

¨
[M]e {q(t)}
e
˙
+ [Dc ]e {q(t)}
e
+ [K c ]e {q(t)}e = { f (t)}e (6)

The elemental mass, damping, and stiffness matrices are denoted by [M]e , [D]e ,
and [K ]e respectively. The vectors {q(t)}e represent the elemental response, whereas
{ f (t)}e denotes the excitation force. The indexes “c” and “wc” indicate the cracked
and intact elements, respectively. All the elemental matrices are taken from reference
[25]. To form the global matrices, these elemental matrices are assembled. The simu-
lated shaft deflection is plotted in Fig. 3. In actual practice, the shaft deflection must
be polluted by measurement noise. To demonstrate the consequences of measure-
ment noise, white Gaussian noise is used to pollute the simulated shaft deflection
[15]. The shaft deflection with 80 dB of noise is plotted in Fig. 3.

Fig. 3 The shaft deflection, contaminated with noise

364 R. Kumar and S. K. Singh

2.2 Wavelet Packet Transform

The continuous WT of a signal, q(x) can be given as

+∞
1 x −u
Wu,s (x) = √ q(x)ψ( )d x (7)
s −∞ s

Here, u and s are the translation and scale indexes, respectively, and x denotes the
location in the spatial domain. For dyadic case (s = 2 j ) is mostly used in practice.
The DWT can be obtained from the dyadic orthonormal wavelets. [18]. The DWT
formation of a mother wavelet ψ(x) can be obtained as [18]:

ψi,k (x) = 2i/2 ψ(2i x − u) (8)

Here, ψ(x) is the mother wavelet, u, and i are the translational and scale indexes.
A discrete wavelet transform can be viewed as applying a multi-resolution filter
bank on the signal. In the case of DWT, only the low-pass filter (LPF) output is
subjected to further decomposition levels. The high-pass filter (HPF) output, called
detailed information, is generally analyzed to detect the signal information. The HPF
coefficients correspond to wavelet analyzing, and the LPF coefficients correspond
to wavelet scaling functions, respectively. The output of LPF is used for the next
level of decomposition. In the next level of decomposition, the wavelet analyzing
and scaling filter coefficients get coarser. Thus the DWT method is limited by its
lower resolution in the high-frequency region. This deficiency leads to difficulty in
separating the SD due to a crack from the high-frequency noise. The WPT, however,
decomposes both the detailed and approximate coefficients. In this way, the WPT is
able to achieve a higher resolution in the high-frequency zone because it conducts
a full decomposition at each level [19]. In the present work, WPT is used for the
localization of the cracks in a shaft.
A WPT is the extended form of DWT that offers decomposition of signals at each
level [26]. A wavelet packet (WP) ψs,u m
(x) is a function having additional modulation
index “m”. The ψs,u (x) can be written as [26]:
m

s
ψs,u
m
(x) = 2 2 ψ m (2s x − u) (9)

where m, s ϵ z + and uϵ z in the WP. The wavelet function ψ m (x) can be obtained as
[26]:

√ ∑
u=+∞
ψ 2s (x) = 2 h(u)ψ m (2x − u) (10)
u=−∞

√ ∑
u=+∞
ψ 2s+1 (x) = 2 g(u)ψ m (2x − u) (11)
u=−∞
Crack Detection in a Shaft Using Wavelet Packet Transform 365

where ψ 2s (x) and ψ 2s+1 (x) are the wavelet scaling and wavelet analyzing functions,
respectively. The discrete filters h(u) and g(u) correspond to ψ 2s (x) and ψ 2s+1 (x).
The recursive relationship between the s th and (s + 1)th levels is as follows:

qsm (x) = qs+1

2m−1
(x) + qs+1
2m
(x) (12)

2m−1
qs+1 (x) = Lqsm (x) (13)

2m
qs+1 (x) = H qsm (x) (14)

where L and H are the filtering operators corresponding to the h(u) and g(u)filters.
The corresponding relations between the filtering operators L and H and the filters
h(u) and g(u) are shown as

∑
u=+∞
L{.} = h(2x − u) (15)
u=−∞

∑
u=+∞
H {.} = g(2x − u) (16)
u=−∞

The spatial domain signal f (x), after the j level of decomposition, can be written
as

∑
s
2
q(x) = qsm (x) (17)
m=1

The WP component signal f sm (x) is obtained as follows [26]:

∑
u=+∞
qsm (x) = ϕs,u
m
(x)ψ m
s,u
(x) (18)
u=−∞

The WP coefficients ϕs,u

m
.(x) can be obtained as

+∞
ϕs,u
m
(x) = m
q(x)ψs,u (x)d x (19)
−∞

It is also known that the WP functions are orthogonal, i.e.,

α β
ψs,u (x).ψs,u (x) = 0, i f α /= β (20)

The signal decomposition using WPT is shown in Fig. 4. The “A” denotes approx-
imate coefficients, D depicts the detailed transformation, and “m” shows the level
366 R. Kumar and S. K. Singh

Fig. 4 Signal decomposition using; a Discrete wavelet transform and b wavelet packet transform

of decomposition. The symbol “H” indicates the high-pass filter and the “L” depicts
low-pass filter.

3 Results and Discussion

Crack detection using WPT is tested for two numerical simulations. In Simulation
I, a crack is located at a distance of 0.66 m from the left support of the simply
supported shaft. The physical properties and the simulation parameters of the shaft
are presented in Table 1.

Table 1 Simulation parameters for cracked shaft

Parameters Simulation I Simulation II
Radius of the shaft (R) 0.005 m 0.005 m
Crack depth (d) 0.0033 m 0.0033 m
Beam length (L) 1m 1m
Number of FE (N) 2000 2000
Modulus of elasticity (E) 2.06×1011 N/m2 2.06×1011 N/m2
Crack location from the left end 0.66 m 0.065 and 0.66 m
Excitation frequency (Ω) 45 rad/s 45 rad/s
Crack Detection in a Shaft Using Wavelet Packet Transform 367

Fig. 5 WP coefficients of shaft deflection for numerical simulation I

The rbio2.2, mother wavelet, is taken to locate the crack along the shaft length
[27]. Four noise levels are considered to study the ability of WPT to locate the crack
along the shaft length. The WP coefficients, DD, at the four different noise levels are
plotted in Fig. 5. From Fig. 5d, it can be clearly seen that the WPT can detect the
crack up to 60 SNR.
Now, to visualize the consequence of poor measurement resolution, the number
of finite elements is reduced to 100 over a length of 1 m. The simulated response of
the shaft having a crack at 0.66 m from the left end is taken as an input of WPT. The
WP coefficients are shown in Fig. 6. From Figs. 5a and 6, it can be concluded that
the WP gives localized crack information with high spatial resolution. Additionally,
the edge effect shrinks with high spatial resolution. Shrinkage of edge effect offers
more spans available for crack detection. The crack detection very close to the ends
is discussed in the subsequent section.

3.1 Multiple Crack Detection

Now, Simulation II is taken for multiple crack localization. The WP coefficients for
simulation II are plotted in Fig. 7. The appearance of two clear spikes denotes the
368 R. Kumar and S. K. Singh

Fig. 6 WP coefficients of shaft deflection with low spatial resolution

location of the crack along the shaft length. The abrupt change in the WP coefficients
at the two ends is due to the edge effect. The edge effect occurs due to the unavail-
ability of data at the ends of the signal [15]. The high spatial measurement resolution
of the shaft deflection shrinks the edge effect as more data points are available in a
shorter span. This shrinkage of edge effect enables the localization of cracks very
close to the shaft ends [15].

Fig. 7 WP coefficients of shaft deflection for numerical simulation II

Crack Detection in a Shaft Using Wavelet Packet Transform 369

4 Experimental Details

For experimental crack detection, a photographic-based experiment is performed

[26]. The experiment involves testing a cantilever beam having a rectangular cross-
section of 50 mm × 6 mm and a length of one meter. The setup for experimental is
shown in Fig. 8. The parameters for experimentations are given in Table 2.
To introduce an open crack on the upper surface of the beam, a shaper machine
is used to create a notch. An initial excitation is then applied at the beam’s free end
to obtain the beam deflection. The deflection shape of the cracked beam is taken by
a camera.
For obtaining the beam’s deflection shape, the captured images are kept as JPG
files in the RGB color space. The RGB images are then transformed into binary
space, and the Sobel method is employed to find the edges from the cracked binary
images [28]. The non-cracked edges of the beam are extracted from the edged image
to obtain the deflection shape of the beam. The bottom of the edges image is taken as a
reference for obtaining the beam deflection. The deflection of the beam is determined

Crack

Fig. 8 Experimental setup, a the beam with the camera, b length of crack
370 R. Kumar and S. K. Singh

Table 2 Parameters for experimentation

Parameters Case I Case II Case III
Poisson’s ratio (ν) 0.3 0.3 0.3
Crack’s depth (d) 3 mm 2 mm 3 and 3 mm
Crack location from the fixed end 0.66 m 0.66 m 0.065 and 0.66 m

by measuring the distances of the white pixel indexes with respect to the bottommost
of the edged image. These deflection data are used to generate the deflection shape
of the beam. The deflected shape of the beam is plotted in Fig. 9. The WPT is applied
on the obtained deflection shape. The resulting WP coefficients are plotted in Fig. 10.
It can be observed that a dominant spike is present near the location of the crack.

Fig. 9 Experimental beam deflection for Case I

Fig. 10 WP coefficients for Experiment I

Crack Detection in a Shaft Using Wavelet Packet Transform 371

Now, the experimentation is carried out for the detection of crack with lesser
crack depth. The experimental parameters are mentioned in Table 2. An isolated
and higher peak at the crack location is seen in the WP coefficients for case II,
as plotted in Fig. 11. The high spatial measurement resolution results in shrinkage
of edge distortion zone. Case III is considered for employing the multiple crack
detection with WPT. The WP coefficients for case III are shown in Fig. 12. Two
spikes corresponding to the crack locations are seen. The crack very close to the
clamped end is detected as the higher measurement resolution shrinks the WP edge
effects.

Fig. 11 WP coefficients for Experiment II

Fig. 12 WP coefficients for Experiment III

372 R. Kumar and S. K. Singh

5 Conclusions

A WP-based crack localization algorithm is employed for locating the crack along the
shaft length. The FE method is used to get the simulated shaft deflection. To demon-
strate the consequences of measurement noise, white Gaussian noise is used to pollute
the simulated shaft deflection. Crack detection with different noise levels is studied.
The WPT performs well even in the presence of noise. For experimental testing,
low-cost photographic-based experimentation is employed to get high spatial reso-
lution beam deflection. The high spatial resolution measurement of beam deflection
results in localized crack detection and a narrower edge distortion zone. The numer-
ical simulation and the experimentation show that the proposed WP-based algorithm
can detect the crack location correctly. The lower crack depth is considered to test
the algorithm for early crack detection. Furthermore, the WPT is also employed for
multiple crack localization.

References

1. Adams RD, Cawley P, Pye CJ, Stone BJ. A Vibration technique for non- distructively assessing
the integrity of structures. J Mech Eng Sci. 1978;20:93–100. https://doi.org/10.1243/JMES_J
OUR_1978_020_016_02.
2. Dimarogonas AD, Massouros G. Torsional vibration of a shfat with a circumferential crack.
Eng Fract Mech. 1981;15(3–4):439–44. https://doi.org/10.1016/0013-7944(81)90069-2.
3. Dimarogonas AD. Vibration of cracked structures: A state of the art review. Eng Fract Mech.
1996;55:831–57. https://doi.org/10.1016/0013-7944(94)00175-8.
4. Sekhar AS. Multiple cracks effects and identification. Mech Syst Signal Process.
2008;22(4):845–78. https://doi.org/10.1016/j.ymssp.2007.11.008.
5. Dimarogonas AD. Vibration of cracked structures: a state of the art review. Eng Fract Mech.
1996;55(5):831–57. https://doi.org/10.1016/0013-7944(94)00175-8.
6. Zhu HP, He B, Chen XQ. Detection of structural damage through changes in frequency. Wuhan
Univ J Nat Sci. 2005;10(6):1069–73. https://doi.org/10.1007/BF02832469.
7. Labib A, Kennedy D, Featherston CA. Crack localisation in frames using natural
frequency degradations. Comput Struct. 2015;157:51–9. https://doi.org/10.1016/j.compstruc.
2015.05.001.
8. Singh SK, Tiwari R. Detection and localisation of multiple cracks in a shaft system: An experi-
mental investigation. Meas J Int Meas Confed. 2014;53:182–93. https://doi.org/10.1016/j.mea
surement.2014.03.028.
9. Roy K, Ray-chaudhuri S. Fundamental mode shape and its derivatives in structural damage
localization. J Sound Vib. 2013;332:5584–93. https://doi.org/10.1016/j.jsv.2013.05.003.
10. Lee E, Eun H. Damage detection of damaged beam by constrained displacement curvature. J
mech sci technol. 2008;22:1111–20. https://doi.org/10.1007/s12206-008-0310-3.
11. Rrzos PF. Identification of crack location and magnitude in a cantilever beam from the vibration
mode. J Sound Vib. 1990;138:381–8. https://doi.org/10.1016/0022-460X(90)90593-O.
12. Pandey A, Biswas M, Samman M. Damage detection from mode changes in curvature. J Sound
Vib. 1991;145:321–32.
13. Farrar CR, James GH. System identification from ambient vibration measurements on a bridge.
J Sound Vib. 1997;205:1–18. https://doi.org/10.1006/jsvi.1997.0977.
14. Singh SK, Tiwari R. Identification of a multi-crack in a shaft system using transverse frequency
response functions. Mech Mach Theory. 2010;45:1813–27. https://doi.org/10.1016/j.mechma
chtheory.2010.08.007.
Crack Detection in a Shaft Using Wavelet Packet Transform 373

15. Kumar R, Singh SK. Crack detection near the ends of a beam using wavelet transform and high
resolution beam deflection measurement. Eur J Mech A/Solids. 2021;88: 104259. https://doi.
org/10.1016/j.euromechsol.2021.104259.
16. Nigam R, Singh SK.. Crack detection in a beam using wavelet transform and photographic
measurements. Structures 2020;25:436–447. (Elsevier).
17. Kim H, Melhem H. Damage detection of structures by wavelet analysis. Eng Struct.
2004;26:347–62. https://doi.org/10.1016/j.engstruct.2003.10.008.
18. Mallat S. A wavelet tour of signal processing. A Wavelet Tour Signal Process 2009. https://
doi.org/10.1016/B978-0-12-374370-1.X0001-8.
19. Peng XL, Hao H, Li ZX. Application of wavelet packet transform in subsea pipeline bedding
condition assessment. Eng Struct. 2012;39:50–65. https://doi.org/10.1016/j.engstruct.2012.
01.017.
20. Khushaba R. Feature extraction using multisignal wavelet packet (WP) decomposition [Tuto-
rial]. 1 Aug 2014 University of Higher Education. Cambridge, MA; 2003. See also http://www.
abc.edu.
21. Sun Z, Chang CC. Statistical wavelet-based method for structural health monitoring. J Struct
Eng. 2004;130(7):1055–62. https://doi.org/10.1061/~ASCE!0733-9445~2004!130:7~1055.
22. Papadopoulos CA, Dimarogonas AD. Coupled longitudinal and bending vibration of rotating
shaft with an open crack. J Sound Vib. 1987;117:81–93. https://doi.org/10.1016/0022-460
x(87)90437-8.
23. Clough RW, Penzien H. Dynamics of structures. McGraw-Hill computers and structures,
Inc.1995.
24. Darpe AK, Gupta K, Chawla A. Coupled bending, longitudinal and torsional vibrations
of a cracked rotor. J Sound Vib. 2004;269(1–2):33–60. https://doi.org/10.1016/S0022-460
X(03)00003-8.
25. Kumar R, Singh SK. A variance-based approach for the detection and localization of cracks in
a beam. Structures 2022;44:1261–1277. (Elsevier).
26. Wickerhauser MV. Adapted wavelet analysis: from theory to software. CRC Press; 1996.
27. Kumar R, Nigam R, Singh SK. Selection of suitable mother wavelet along with vanishing
moment for the effective detection of crack in a beam. Mech Syst Signal Process. 2022;163:
108136. https://doi.org/10.1016/j.ymssp.2021.108136.
28. Woods RE, Eddins SL, Gonzalez RC. Digital image processing using MATLAB®, Second.
Tata McGraw-Hill;Chennai:2010.
Design and Modeling of Piezo-resistive
MEMS Accelerometer for Crash Test
Application

N. Kusuma, J. Pradyumna, A. Tulasi, and Asha R. Upadhyay

Abstract Micro Electro Mechanical Systems (MEMS) are the latest technology
in which various microdevices have been developed globally. Many researchers are
working on MEMS Accelerometers (vibration sensors) and their focus is on miniatur-
ization, Integration, high reliability, and high resolution. MEMS accelerometers are
having more demand for various applications in the field of automotive industries,
Machine tool industries, Aerospace industries, military applications, and biomed-
ical applications. Vehicle manufacturers use the accelerometer as an impact sensor
for Automotive Crash test applications which are useful for validating computer
simulations of road traffic accidents. This paper focuses on designing, modeling,
and analyzing MEMS accelerometers for automotive crash test applications using
the piezoresistive transduction principle. It involves piezoresistive-based MEMS
accelerometer design with Wheatstone bridge configuration suitable for crash test
application, involves mathematical model development to form design procedure
required to find the influence of geometric attributes of a four cross-bridged MEMS
accelerometer and also involves suitable accelerometer fabrication process. Stress,
deformation, and sensitivity are determined by simulating and analyzing the solid
model using design tools such as ANSYS and Coventorware softwares and results
are validated.

Keywords MEMS · Accelerometer · Piezo-resistive · Crash test · Coventorware ·

ANSYS

N. Kusuma (B) · J. Pradyumna · A. Tulasi

Scientist-E, Centre Head (C-SVT), Centre for Sensors and Vision Technology, Bengaluru,
C.M.T.I Karnataka, India
e-mail: [email protected]
J. Pradyumna
e-mail: [email protected]
A. R. Upadhyay
Scientist-F, Centre Head (AEAMT), Bengaluru, C.M.T.I Karnataka, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 375
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_25
376 N. Kusuma et al.

1 Introduction

Micro Electro Mechanical Systems (MEMS) are presently the latest technology in
the global scenario and various microdevices have been developed using MEMS
Technology for many applications. Even today lot of research is going on in this
field especially on inertial MEMS accelerometers for applications such as automotive
industries, machine tools industries, aerospace industries, military applications, and
biomedical applications and dynamic measurement of vibration can be done using
MEMS Accelerometer sensors. Currently, there is more demand for Micro Electro
Mechanical System (MEMS) devices mainly due to their outstanding features such
as miniaturization, low power consumption, reduced cost, and ease of integration
with sub-systems. The sensor die size ranges from 1 to 100 μm and the packaged
sensor size ranges approximately from 20 μm to 10 mm.
MEMS Accelerometers are used for crash testing application in automobile indus-
tries especially for airbag deployment by inertial sensing mechanism. Here MEMS
Accelerometer continuously monitors the acceleration of the car, when the measured
value is more than the predetermined threshold, a microcontroller based control unit
will compute the integral of the acceleration (i.e., change in velocity) which can be
used to deploy or fire the airbag.
The Airbags are placed in front near the steering wheel of the car and side near
the car door, which need to be fired within few milliseconds based on the reading
from the MEMS Accelerometers, the main aim of this research work is to design and
develop such MEMS Accelerometer for crash testing applications.
Some of the Technological advancements in the design of MEMS accelerometers
are to increase the number of beams which leads to full bridge-length amplification
of the output voltage and also Sensitivity. Also, high natural frequency and resonant
are required for automotive and machine tool applications, hence these parameters
have to be considered during the design stage. Along with these parameters, before
the fabrication of the MEMS accelerometer, other design parameters such as stress,
displacement, input voltage, and output voltage have to be considered.
Design implementation: Design considerations, analytical design calculations,
design modeling, design simulation and analysis such as Mechanical (stress, displace-
ment, modal, harmonics), Electrical analysis such as PZR, Voltage, Current, capac-
itance matrix, charge measurement, thermal analysis and validation of design are
very important.
Literature review: Finite element method (FEM simulation) was used for proving
the mathematical model of the MEMS accelerometer, and the static and dynamic
response of a piezoresistive eight-beam, and the three-axis accelerometer was studied
by Song [1]. Mohanty [2] highlights the usage of the Wheatstone bridge configura-
tion in designing a piezoresistive-based MEMS accelerometer to reduce cross-axis
sensitivity. Zhang [3] emphasizes the fabrication process steps and stress distribution
of the MEMS accelerometer. Ghemari [4] explains on step and impulse response of
the PZR accelerometer with a bode plot by using damping rate and frequency range
and also the method of improving the stability and accuracy of the accelerometer.
Design and Modeling of Piezo-resistive MEMS Accelerometer … 377

Zhao et al. [5] focus on fabrication process steps wherein proof mass can be fabri-
cated by ICP-RIE etching at the bottom side of the wafer till the proof mass structure
is released, this also gives very fine edges for the proof mass structure. Liu and
Zhao [6] say optimization of geometrical dimensions is essential for improving the
performance of MEMS accelerometers.
In this research, a single-axis Piezo-resistive MEMS Accelerometer design is
chosen because the fabrication process and Signal conditioning of a Piezo-resistive
[1] accelerometer are very simple and have high linearity and sensitivity compared
to the capacitive type and piezoelectric type. The Piezo-resistive method is very
powerful as it changes its value when deformation occurs due to strain of the beam
and mass structure of the accelerometer. It also shows linear characteristics over a
wide range of frequencies.
The following technical device specifications are selected for designing of
single-axis piezo resistive-based MEMS Accelerometer for automotive crash testing
applications:
Type: Piezo-resistive range: ±500 g (1000 g)
Sensitivity: ±1 mV/g
Frequency response: 0 to 5 kHz
Resonant Frequency: 26 kHz
Excitation Voltage: 5VDC
Operating temperature: −18 to 66 °C
In this paper there are about 5 sections, in that, Section-I introduces the paper and
describes literature reviews on design principles such as piezoresistivity, piezore-
sistive effect, and static and dynamic response of MEMS Accelerometer. Section II
briefs on design considerations, design calculations, and mathematical modeling [1]
and also highlights on MEMS fabrication process flow. Section III discusses briefly
the modeling, and meshing, and Section IV discusses briefly simulations and analysis
results for the design selected. In Section V, the paper concludes with the results from
the Simulation and Analysis of the designed piezoresistive MEMS Accelerometer
for the automotive crash test application.

2 Design Principles and Considerations

2.1 Effects of Piezo-Resistive Sensors

The working principle of the Piezo-resistive Accelerometer can be as follows: when

the proof mass is applied with a force (in the form of acceleration) on the top, it
deforms or moves downward due to which the beams will deflect, and maximum
stress is induced on each beam at both frame and mass end. The piezo resistors are
placed on each beam where the stress is maximum [4]. When force is applied, stress
is induced which leads to a change in the resistance value of piezo resistors, it further
378 N. Kusuma et al.

leads to a change in voltage at the output. The Acceleration value from the sensor
can be inferred by measuring the output voltage from the sensor after fabrication,
characterization, and packaging.
The effect of change in resistance of PZR due to induced mechanical stresses
at beam end and Proof mass end when force is applied on MEMS Accelerometer
is known as piezoresistivity. In sensors such as Accelerometers, cantilever force
sensors, pressure sensors, and inertial sensors, PZR sensing elements are used for
transduction.
By using a 6 × 6 Piezo-resistive coefficient matrix and the stress tensor, change
in resistivity can be calculated.
Piezo-resistive coefficient matrix with 3 non-zero independent components (π11,
π12 , π44 ) are shown in following Eq. (1)
⎡ ⎤
π11 π 12 π 12 0 0 0
⎢ π12 π11 π 12 0 ⎥
⎢ 0 0 ⎥
⎢ ⎥
⎢ π12 π 12 π 11 0 0 0 ⎥
π =⎢ ⎥ (1)
⎢ 0 0 0 π 44 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 π 44 0 ⎦
0 0 0 0 0 π 44

The simplified expression for resistance change in a piezo-resistor is given by the

following Eq. (2):

∆R
= πlσ l + πtσ t (2)
R
πt and πl are the piezoresistive coefficients in transverse and longitudinal direction,
respectively. Similarly, σt and σl are the stresses induced on the surface of the piezo-
resistor in transverse and longitudinal direction, respectively.
A second-order spring-mass-damper system can be used to represent an
accelerometer as shown in Fig. 1.

Fig. 1 Spring mass damper

system
Design and Modeling of Piezo-resistive MEMS Accelerometer … 379

Force (F applied) is calculated by Eq. (3), with proof mass (m) and Acceleration
(a) as below

Fapplied = maapplied (3)

The force exerted in the system by springs and damping can be defined in Eqs. (4)
and (5) with a spring constant (k) and damping (b), the force (F applied) on the proof
mass:

Fspring = kx (4)

Fdamping = b ẋ (5)

Force applied is calculated by using Newton’s second law as in Eqs. (6) and (7):

Fapplied = m ẍ + Fspring + Fdamping (6)

m ẍ + b ẋ + kx = Fapplied = maapplied (7)

The transfer function H(s) is calculated as given by Eqs. (8–10):

ms 2 x(s) + bsx(s) + kx(s) = F(s) = ma(s) (8)

b k F(s)
s 2 x(s) + sx(s) + x(s) = = a(s) (9)
m m m
x(s) 1 1
H (s) = = 2 b = ω◦ (10)
a(s) s + ms + k
m
s2 + Q
s + ω◦2

In Eq. (11), ω0 is the resonance frequency given by

√k
ω0 = (11)
m
and in Eq. (12) Q is the quality factor given by
m
Q= ω0 (12)
b
As Accelerometers are inertial sensors, it works in low frequency domain (ω <<
ω0 ), and mechanical sensitivity is calculated by setting s = 0 in the transfer function
H(s) as in Eq. (13)
380 N. Kusuma et al.

x m 1
∼ = 2 (13)
a k ω◦

A high resonant frequency is required to get a higher bandwidth, it is possible

by proof mass size reduction and by increasing the stiffness of the beam (spring).
In our research to get the required resonant frequency ω0 = 26 kHz, the following
real value in microns are considered, and these dimensions are selected based on
suitability for actual fabrication at the foundry.

2.2 Static and Dynamic Response of MEMS Accelerometer

2.2.1 Static Response of the MEMS Accelerometer

A static structural analysis identifies the displacements; stresses, strains, and forces
induced in microstructures of the Accelerometer and it does not cause substantial
inertia or damping effects. Under steady-state conditions, all structures are supposed
to be stable with and without load. However, Accelerometer also experiences different
types of forces and varied moments which results in high deformation and more stress.
Hence, the design of the accelerometer should be analyzed statistically as well as
dynamically. The static analysis depends on the type of material, number of beams
involved, proof mass, and frame. Here the frame is fixed. The load is applied on proof
mass and the structure moves in an axial direction. Therefore directional deformation
in the Z-axis is considered for static analysis.

2.2.2 Dynamic Response of the MEMS Accelerometer

The dynamic response is the response of the sensor output to its change in input. The
dynamic response of the MEMS Accelerometer is often expressed as the frequency
response of the sensor at the resonant frequency and operating frequency. The
frequency response of the MEMS Accelerometer is a very important parameter to
be considered where dynamic changes in acceleration can be observed over small
intervals of time. The frequency response is governed by the sensing element (PZR)
that is placed on the beam structure with proof mass acts like a spring-mass system.
The natural frequency of this MEMS Accelerometer sensor is typically higher due
to its small size and high Young’s modulus of silicon value. This can be checked at
different modes of operation by using modal analysis.
Considering the above design aspects, a piezoresistive-based MEMS Accelerom-
eter was designed using Coventorware software with 2D schematic, 3D modeling,
mapped grid meshing, simulation, and analysis.
The final structure of the piezoresistive-based MEMS Accelerometer designed is
shown in Fig. 2. The dimensions and materials have been arrived at after detailed
analytical solution and simulation analysis studies.
Design and Modeling of Piezo-resistive MEMS Accelerometer … 381

Fig. 2 Microstructure of
MEMS accelerometer

The MEMS structure of the Piezo-resistive Accelerometer being designed consists

of a proof mass at the center, having four flexural silicon beams in a cross-bridged
structure, the two extremities of each flexure[2] are doped with Piezo-resistors of
p-type single crystal silicon <110> to sense the maximum stress and to measure
deflection where it is dominant. The design procedure involves the determination of
certain parameters such as proof mass side length, height, and dimensions of flexures
(l × b × h).

2.3 Mathematical Design Calculation

The mathematical design calculations were done as below using the governing
equations mentioned in Sect. 2 Eq. (1–12).
For silicon, Density (ρ) = 2330 kg/m3, E = Young’s Modulus = 160 GPa, the
following other parameters are calculated based on structural dimensions given in
Table 1 for the proposed Piezo-resistive MEMS Accelerometer design.
Mass m = 8.388 × 10–6 kg,
Spring constant k = 223.85 kN/m.
Resonant frequency ω0 = 26 kHz.
Bandwidth (frequency response) = 8.66 kHz.
Displacement (x) = 367.6 nm.

Table 1 Structural dimensions considered

Structure Geometrical Dimensions(μm)
(l × b × t)
Proof mass 3000 × 3000 × 400
Beam (Guided cantilever beam) 1200 × 200 × 150
Frame 5600 × 5600 × 420
382 N. Kusuma et al.

Fig. 3 Wheatstone bridge

configuration R1m R1f

R3m R3f

Vin Vout
R2f R2m

R4f R4m

2.4 Piezo-Resistor Positioning (Placement) and Output

The change in the resistivity of a Piezo-resistive material is directly proportional

to the stress, so it is very important to place these resistors at the maximum stress
points to get higher sensitivity of the Accelerometer. In this research Piezo-resistors
are arranged in a Wheatstone bridge configuration as shown in Fig. 3, under steady-
state (no-load) condition it is said to be in a balanced condition, and change in output
voltage is given by ∆V = ∆R × (V b /R), where V b is the wheat stone bridge supply
voltage, ∆R is the change of resistance of the Piezo-resistor and R is the initial
resistance.
The relationship between the input voltage and output voltage is
Vout
Vin
= ∆R R0
, R0 is initial resistance and Sensitivity (S) = Output voltage/
acceleration = V out /a.

2.5 Material Selection

Silicon is particularly associated with Microsystems design as it is strongly desirable

for inertial sensors with its greater flexibility in design and fabrication. Specifically,
silicon is mechanically stable and hence signals transduction elements or sensing
elements such as p-type or n-type Piezo-resistor can be integrated with silicon as a
substrate.
In the present work, both finite element analysis and analytical calculations
are done by considering single crystal silicon with <100> orientation and Yield
strength of 7GPa and it is approximated as an isotropic material. Its property Young’s
modulus(1.6E + 11) is also considered and silicon orientation of <110> is considered
for a proof mass of the accelerometer.
Other materials: Silicon dioxide as an insulating layer, photoresist for pattern
transfer at various stages during photolithography operation, and Pyrex 7740 glass
[3] as a protection layer at the bottom of the mass structure during the dynamic
condition of the accelerometer, vacuum is created between mass and Pyrex glass at
the bottom.
Design and Modeling of Piezo-resistive MEMS Accelerometer … 383

2.6 Fabrication Process Flow/Steps for Piezo-Resistive Based

MEMS Accelerometer

MEMS fabrication process steps for Piezo-resistive-based MEMS Accelerometer

have been created as shown in Fig. 4 below: this is purely based on the fabrication
facilities available at a particular foundry.
To start with a 100 mm diameter, the n-type 4-inch wafer of <100> orienta-
tion prime quality double side polished Si is taken and cleaned by general and
piranha cleaning process. In Step 1, possible Organic or Inorganic contaminations
are removed from the wafer surface by a wet chemical cleaning treatment. In Step-2,
a silicon dioxide layer of 50 nm is deposited on the silicon layer as an insulating
layer using thermal oxidation; In Step-3, a photolithography process is done to etch
the SiO2 by using wet etching to etch windows of the Piezo-resistors. In Step 4, the
first time Ion implantation process is done to form the p- region by doping Boron,
and the second time Ion implantation process is done into the etched window to form
the p+ region by heavy boron doping.
The remaining SiO2 layer of 50 nm is etched by using wet etching technology. In
Step 5, the proof mass is formed by the backside etching of the wafer by the ICP-RIE
process till the top position of the buried oxide layer. In Step 6, the Si3 N2 layer of 400–
500 nm is grown as an insulating layer by using Plasma Enhanced Chemical Vapor
Deposition (PECVD) [5]. In Step-7, the second Lithography is done to selectively
etch the top silicon nitride surface, followed by Step-8, where deposition of Titanium
and metallization by Au of 500 nm is done on the top surface of the substrate by a
vacuum evaporation/electron beam evaporation method for contact pad formation.
In Step-9, the top front side etching of the wafer is done till the position of the buried
oxide layer by the DRIE process to release the cantilever beam structure. Step 10,

Fig. 4 Fabrication process flow steps

384 N. Kusuma et al.

the last step of fabrication, Anodic Bonding of glass (Pyrex) to the Si substrate is
done for overload protection.
During the fabrication process, impurity concentration doping of the silicon should
be taken care of as it affects the Temperature characteristics related to the sensitivity
of the Accelerometer.

2.7 MEMS Accelerometer Packaging for Crash Testing

Applications

There are various types of packaging in MEMS such as ceramic type, plastic-type,
and metal types used for various MEMS applications, for MEMS Accelerometers
have free-standing microstructures and temperature-sensitive microelectronics and
hermetic sealing in the package is a must. After fabrication of the wafer, it is singu-
lated using a wafer dicer, and packaging is done by attaching the die using die attach,
wire bonding is done for connecting the contact pads of the die to the package, and
encapsulation or sealing is done for the whole package with a cap or lid at the top.
For the designed MEMS Accelerometer green anodized aluminum alloy material is
recommended to be used for housing during packaging for crash testing applications
as it protects against corrosion, enhances aesthetic qualities, resists scratching, and
has the most durable surface finish quality.

3 Modeling and Meshing

The Piezo-resistive MEMS Accelerometer was designed using Coventorware soft-

ware with a MEMS fabrication process and the same was validated using ANSYS
software. The following steps were considered.
The 2D structure was designed in Coventorware and considered for modeling with
Mass, beam, frame, and PZR components. The Accelerometer behavior was analyzed
under mechanical load (Acceleration/Force) conditions. The MEMS Accelerometer
sensor structure shown in Fig. 2 is partitioned by the edges of the structure to simulate
the clamping conditions. The frame structure was fixed with boundary conditions in
the software thereby restraining any movement in the XY-axis. The mass movement
was analyzed in the Z-axis.

3.1 2D Structure and 3D Modeling

2D structure and 3D modeling were done for the specified dimensions as shown in
Fig. 5 below.
Design and Modeling of Piezo-resistive MEMS Accelerometer … 385

Fig.5 2D structure and 3D model in Coventorware

3.2 Meshing

The structures of the MEMS Accelerometer were meshed during structural analysis
using FEA (Finite Element Analysis) method to get accurate results. Tetrahedral brick
mesh(A tetrahedron has 4 vertices, 6 edges, and is bounded by 4 triangular faces)
was selected in Coventorware Software to fit better complex geometry. Appropriate
element size is taken into consideration in each structure of the MEMS Accelerom-
eter. Meshing is done on the entire structure and minimum size is selected for PZRs
as shown in Fig. 6.
Element size of Mass, Beam, and Frame: 100 μm, PZR: 1 μm, and Mesh Type:
Tetrahedron was considered.

Fig. 6 Meshing in
coventorware
386 N. Kusuma et al.

mechDomain
Maximum Minimum
Node Displacement 4.230392E-01 5.827845E-52
Node X Displacement 3.482297E-02 -3.482297E-02
Node Y Displacement 3.482297E-02 -3.482297E-02
Node Z Displacement 4.230392E-01 -3.526732E-04

Fig. 7 Displacement in Coventorware

4 Simulation and Analysis

The simulation of the design of the Piezo-resistive-based MEMS Accelerometer

structure was carried out using Coventorware software for appropriate clamping
conditions. The Mass, beam deflection/displacement, and the Misses stress distri-
bution were measured for a maximum load of 1000 g as per the required technical
specifications.

4.1 Static Analysis—Displacement and Stress Analysis

The static study is used to simulate deflection, stress, and strain in solid mechanic
structures. When external force/acceleration is applied to the Proof mass of the
proposed accelerometer structure, the inertial force causes the proof mass to deflect
in the opposite direction of the applied force/acceleration. The displacement of the
proof mass under 1000 g acceleration in and out-of-plane direction was simulated
using FEM analysis [3]. Maximum displacement is found as 423 nm at the mass end.
Here it is observed that the maximum displacement is toward the center which is
on the proof mass of the MEMS Accelerometer as shown in Fig. 7.

4.2 Stress Analysis

In this research, Von Mises stress analysis is done for the Piezo-resistive-based
MEMS Accelerometer as it determines the yield strength of the material under load
conditions. Figure 8 shows the Von-mises stress obtained in Coventorware, here it is
observed that the stress is maximum at the beam, near the frame end as well as near
the proof mass end. These areas on the beam surface are very crucial to attain the
maximum sensitivity by placing the Piezo-resistive elements in precise locations. In
this design, the PZR sensing element is placed at both ends (frame end and Proof
mass end).
Design and Modeling of Piezo-resistive MEMS Accelerometer … 387

Fig. 8 Stress result in

Coventorware

4.3 Dynamic Analysis—Modal Analysis and Harmonic

Response

In this research, the dynamic characteristics of the designed Piezo-resistive-based

MEMS Accelerometer were studied. The most significant step of dynamic analysis
is the modal analysis which is used to determine the mode shapes and the natural
frequencies of the microstructures of the Accelerometer.

4.3.1 Modal Analysis

A modal analysis calculation computes the natural resonant frequencies of a mechan-

ical structure of the piezoresistive-based MEMS Accelerometer sensor at equilibrium
or steady-state conditions. At these resonant frequencies, an un-damped mechanical
structure of the Accelerometer responds to a force/acceleration exerted on the proof
mass. Improving resonant frequency reduces the displacement of the accelerometer
and also sensitivity; hence optimization of accelerometer geometrical parameters is
required [6].
The first resonant frequency of the accelerometer structure sets the operating
frequency range of the accelerometer. Along with the first mode, another 5 modes
of different shapes represent the overall deformation of the mechanical structure
oscillating at the associated modal frequency. These frequencies and their associated
mode shapes indicate when the Accelerometer will have its maximum response to
any input including noise. The resonant or natural frequency of the For the designed
MEMS Accelerometer mechanical structure, the modal analysis was carried out in
6 Modes at different resonant frequencies as shown in Table 2 (Values).
Modal analysis was done for the natural/resonant frequency of 26 kHz for the
designed accelerometer, the six different modes are obtained at 24.2, 41.5, 41.5,
54.5, 142.9, and 142.9 kHz as shown in Fig. 9.
388 N. Kusuma et al.

Table 2 Modal analysis

Mode domain
results in Coventorware
Frequency Centralized mass Damping
1 2.420923E04 8.303012E-06 0
2 4.151205E04 1.86805E-06 0
3 4.151205E04 1.86805E-06 0
4 5.450766E04 5.65593E-06 0
5 1.429731E05 1.291782E-05 0
6 1.429731E05 1.291782E-05 0

Mode-1: 24.2kHz Mode-2: 41.5KHz

Mode-3: 41.5KHz Mode-4: 54.5KHz

Mode-5: 142.9KHz Mode-6: 142.9KHz

Fig. 9 Modal analysis at 6 different modes in Coventorware

Design and Modeling of Piezo-resistive MEMS Accelerometer … 389

Fig. 10 Harmonic Analysis

4.3.2 Harmonic Response of Designed MEMS Accelerometer

The results of the modal analysis were verified by the structural response of the
designed piezoresistive-based MEMS Accelerometer to the applied acceleration/
force under different frequencies by using the harmonic response analysis. The
harmonic response is the graph between displacement and frequency response. Here
an impulse of output response of the MEMS Accelerometer is observed when the
operating frequency reaches 26 kHz (resonant frequency) as shown in Fig. 10. This
can be attributed to the resonance between the operating frequency and the natural
frequency of the structure. Thus, the natural frequency in modal analysis agrees well
with that of harmonic response analysis.
Damping effects are considered, i.e., Modal damping which is the function of
critical damping of the MEMS Accelerometer (nominally 0.7 critical).

4.4 PZR Analysis

PZR analysis was done using Coventorware, in PZR simulation MemPZR solver was
used to investigate the output voltage as a function of load/acceleration as shown in
Fig. 11a, b (Structure of PZR and 3D Model of PZR). PZR resistors were connected
in Wheatstone bridge configuration. Two input terminals of the bridge were subjected
to the potential difference (excitation voltage/ input voltage) of 5 V to simulate output
voltage conditions at two output terminals of the Wheatstone bridge circuit and the
sensitivity of the accelerometer was calculated.
Eight Piezo-resistors named R1m, R2m, R3m, R4m, R1f, R2f, R3f, and R4f at
the mass end and frame end, respectively, were used to simulate results, as shown in
Fig. 12.
Table 3 (piezoDomain) displays the current sensed through the PZR parts when
a 1000 g of force/load and 5 V excitation voltage is applied. The initial percentage
change in current is relative to the no-load case, further change in current is due to
the change in resistivity when the Piezo-resistor is subjected to stress due to the load
390 N. Kusuma et al.

a b

PZRs at both Proof Mass end

& frame end

Fig. 11 a Structure of PZR (PZR at Frame end is shown here) b 3D Model of PZR

Fig. 12 PZR connection in

R1m R1f
Wheatstone bridge
configuration R3m R3f

Vin = 5V Vout
R2f R2m

R4f R4m

being applied and full-scale voltage is calculated by setting patch query as shown in
Table3 (patch Query).
Full-scale voltage and sensitivity are calculated as follows.
Full-scale voltage for 1000 g [8] = 4.358 mV.
Sensitivity = 4.358 μV/g.

Table 3 Single beam on each side (4 beams) Piezo domain results and patch query
Piezo domain
Patch voltage Patch current % Change in current
R2_Vsp 5 2.501075E01 4.106151E-02
R3_Vsp 5.0E00 2.500953E01 3.606151E-02
R1_Vsp 0 −2.501033E01 4.328271E-02
R4_Vsp 0 −2.501116E01 4.666182E-02
Patch query
Step R3m_R2f_ R2f_R3m_ R3f_R2m_ R2m_R3f_
AvgVoltage_1 AvgVoltage_2 AvgVoltage_3 AvgVoltage_4
1 2.497686E00 2.497686E00 2.502044E00 2.502044E00
Design and Modeling of Piezo-resistive MEMS Accelerometer … 391

Fig. 13 2 beams on each side (8 Beams) MEMS Accelerometer

Table 4 Two beams on each side (8 beams) Piezo domain results and patch query
Piezo domain
Patch voltage Patch current % Change in current
R1m_Vsp 5.0E00 6.287849E00 3.051353E-01
R1f_Vsp 5.0E00 6.249487E00 −8.119128E-01
R8m_Vsn 0 −6.247018E00 4.130641E-01
R8f_Vsn 0 −6.218515E00 −3.075007E-01
Patch query
Step R7m_R2f_AvgVoltage_1 R7f_R2m_AvgVoltage_2
Sim1: 1 2.489819E00 2.499557E00

Further to improve the sensitivity of the sensor, the Piezo-resistive-based MEMS

Accelerometer structure was changed to two beams at each side of the device [1] as
shown in Fig. 13.
Full-scale voltage was calculated for 8-beam MEMS Accelerometer by setting
patch query as shown in Table 4.
Full-scale voltage and sensitivity are calculated as follows
Full-scale voltage for 1000 g = 9.747 mV.
Sensitivity = 9.747 μV/g.

4.5 Results and Validation

Results obtained in Coventorware software tools such as displacement, stress, and

resonant frequencies at different modes were validated using ANSYS software
tool by conducting similar analysis in ANSYS software and comparison has been
made with the theoretical value of Displacement or deformation, natural or resonant
392 N. Kusuma et al.

Table 5 Displacement validation

Displacement/deformation validation
Theoretical value Coventorware simulated ANSYS simulated
367.6 nm 423 nm 410 nm
Result: Simulated results are very close to the Theoretical Value

Table 6 Resonant frequency validation

Resonant frequency validation
Theoretical value Coventorware simulated ANSYS simulated
26 k Hz 24.2 kHz (Mode-1) 24.557 kHz (Mode-1)
Result: Simulated results are very close to the Theoretical Value

frequency, Full-scale output voltage and Sensitivity with their simulated values in
both the software as shown in Tables 5 and 6 below.
Further Full-Scale Voltage for 4-beam MEMS Accelerometer was 4.35 mV and
8-beam MEMS Accelerometer was 9.747 mV and the Sensitivity was 4.35 μV/g
and 9.747 μV/g, respectively. Optimization of the design parameters is required in
terms of dimensions of Beam, Mass, PZR, PZR positions, and resistivity value to get
the desired output. Full-scale output voltage also depends on the MEMS fabrication
processes selected, process equipment used at a foundry, and also MEMS packaging
processes. It can be further enhanced by using Signal conditioners to get the expected
output voltage and sensitivity.

5 Conclusions

In this research, Piezo-resistive-based MEMS Accelerometer was designed for an

automotive crash testing application, initially, theoretical/analytical calculation was
done to match technical specifications targeted in Sect. 1, later the same was
designed using Coventorware and validated using ANSYS software. The displace-
ment/deformation results, Stress analysis, Modal analysis for resonant frequencies,
and PZR analysis were done and full-scale voltage and Sensitivity were calculated
for 4-beam MEMS Accelerometer as well as 8-beam MEMS Accelerometer, 8-
beam accelerometer designs had better sensitivity. It was seen that the design can be
improved by varying beam and mass structure dimensions and the PZR position of
the MEMS Accelerometer.

Acknowledgements The authors would like to thank Dr. Nagahanumaiah, Director, CMTI, and
Mr. Shanmugaraj V, Center Head, Center for Sensors, Vision Technology and Controls for their
encouragement and support in carrying out this research work.
Design and Modeling of Piezo-resistive MEMS Accelerometer … 393

The state of the art is the base for any successful research work. In the current work, the literature
is inclined toward the design techniques/methodology of the Piezo-resistive MEMS Accelerom-
eter and its fabrication processes. The following are referred papers/journals from the preliminary
literature review.

References

1. Song J, Changde H, Wang R, Xue C, Zhang W. A mathematical model of a piezo-resistive

eight-beam three-axis accelerometer with simulation and experimental validation. Sensors.
2018;18:3641. https://doi.org/10.3390/s18113641.
2. Mohanty T. Design and analysis of MEMS-based piezoresistive accelerometer with low cross-
axis sensitivity. IOSR J Eng. 2013;03:54–59. https://doi.org/10.9790/3021-03415459.
3. Ping Y, Zhang C, Changde Z, He X, Jiaqi D, Yu C, Chunhui W, Du L, Juanting W, Zhang Z,
Xiujian Z, Chou W, Zhang W. An integrated MEMS piezoresistive tri-axis accelerometer. J
Semicond 2013;34.https://doi.org/10.1088/1674-4926/34/10/104009.
4. Ghemari Z Study and analysis of the piezoresistive accelerometer stability and improvement
of their performancesInt J Syst Assur Eng Manag. 2017;8:https://doi.org/10.1007/s13198-017-
0622-8.
5. Zhao X, Wang Y, Zhaoli Y, Wen D. Fabrication and characteristics of an SOI three-axis accel-
eration sensor based on MEMS technology. Micromachines. (2019);10:238. https://doi.org/10.
3390/mi10040238.
6. Liu Y, Zhao Y, Tian B. Analysis and design for piezoresistive accelerometer geometry
considering sensitivity, resonant frequency, and cross-axis sensitivity. Microsyst Technol.
2014;20:463–70. https://doi.org/10.1007/s00542-013-1894-9.
Classification of Orbits of Rotor
Supported on Squeeze Film Damper
by Supervised Learning Method

Anurag Kumar, Mayank Tiwari, and Akhilendra Singh

Abstract Squeeze film dampers (SFDs) along with ball bearings are used to suppress
the vibration of rotating machineries. Though a perfectly designed SFD shows stable
behavior, it can also have some undesirable nonsynchronous behavior for a range of
various parameters. The nonlinear equation of flexible rotor supported on SFD with
centering spring has been solved by ode15s methods. The time series data have been
collected and used with empirical mode decomposition (EMD) method. The modes
are extracted using the collected data. Further, the Shannon energy method is used for
finding the energy of each mode. The energy of different modes is used in the feature
matrix. The periodic and quasi-periodic vibration is taken as the response variable.
The feature matrix is then used for training and testing of data with SVM and KNN
models in MATLAB© . The Fine KNN model is more efficient in comparison with
Quadratic SVM. The Fine KNN model has 93.3% accuracy.

Keywords Squeeze film damper · Nonsynchronous · Shannon energy ·

Synchrosqueezing

Nomenclature

Parameter of bearing, m CμR

3
L
B S F D ωn c
3

c Radial clearance of damper, m

e Instantaneous eccentric position of the bearing geometric center
concerning the CSFD geometric center, m
g Acceleration due to gravity, m/s2

A. Kumar (B) · M. Tiwari · A. Singh

Department of Mechanical Engineering, IIT Patna, Bihar, India
e-mail: [email protected]
M. Tiwari
e-mail: [email protected]
A. Singh
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 395
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_26
396 A. Kumar et al.

K Stiffness ratio, kkRS

kR Shaft stiffness, N/m
kS Integral spring stiffness, N/m
L Axial length of CSFD, m
mD Rotor mass, kg
mC S F D Mass of CSFD, kg
R CSFD radius, m
t Time, s
U Nondimensional unbalance, uc
W Nondimensional gravity, ωg2 c
n
u Eccentricity of rotor center, m
m
α Mass ratio, ms Df d
ε Eccentricity ratio, ec
μ Dynamic viscosity of the lubricant, Ns/m2
Ω Nondimensional speed, ωωn
ω Rotor rotating speed, rad/s
ωn The critical speed of the rotor

1 Introduction

Squeeze film damper (SFD) is a well-established external damping device used along
with rolling element bearing in rotating machinery to suppress the vibration ampli-
tude of the system. The damping mechanism of SFD depends on the highly nonlinear
oil-film force. Due to the presence of this nonlinear oil film force, the SFD shows
nonlinear behavior. The criticality involved in the design of SFD is the control of
nonlinear vibration. It is observed that a critically designed SFD will also show
nonlinear vibration for some system parameters. In the past, a lot of research has
been conducted for analyzing the nonlinear behavior of SFD. The given references
follow the nonlinear study done on the centralized SFD. Barret [1] studied the steady
state and transient response of SFD. They calculated the stiffness and damping coef-
ficient for a stable base design. They also observed that for an eccentricity ratio
less than 0.4, the SFD will have optimum stability and low force transmissibility.
Holmes [2, 3] used the short bearing approximation and obtained whirl orbit and
pressure distribution for open-ended [2] and sealed [3] SFD numerically and exper-
imentally. The vibration characteristic [4] of the rotor supported on two SFD with
static misalignment was studied numerically and experimentally. It was observed that
the static eccentricity results in a bistable jump and induces subharmonic resonances.
Mclean [5, 6] investigated the stability of SFD on various design parameters such
as eccentricity ratio, rotational speed ratio, and bearing parameter. Zhao et al. [7,
8] studied the stability and bifurcation of a rigid rotor supported on eccentric SFD.
They used the short-bearing and pi-film cavitation approximation. The system shows
bistable jump, 2 T-periodic, nT-periodic, quasi-periodic, and even chaotic behavior
under large unbalance and static eccentricity limiting the operating speed. Bonello
Classification of Orbits of Rotor Supported on Squeeze Film Damper … 397

et al. [9, 10] studied the nonlinear behavior of rotor systems supported on SFD with
a retainer spring and without a retainer spring. They have used the reduced harmonic
balance method and numerical integration technique for predicting the motion of the
proposed rotor system. The predicted and experimental orbit and frequency spectrum
plot shows the presence of quasi-periodic motion in the system for some particular
speed. The frequency spectrum obtained from prediction and experimental results
shows the presence of fuzzy frequency component along with the first fundamental
and second fundamental frequency component. Jawaid [11] studied the bifurcation
of a flexible rotor system supported with a centralized squeeze film damper by using
the numerical technique (RK method). The study shows that an increase in stiffness
ratio causes an increase in the onset speed of bifurcation while the increase in mass
ratio reduces the onset speed of bifurcation. Chu [12–14] theoretically investigated
the effect of increasing the static eccentricity ratio of SFD with a retainer spring on
the resonance speed of the system. They used the harmonic balance method, simple
iteration, and fast integration method for comparing the results. They have also done
the experimental study for no damping and full damping case.
The fault diagnosis of rotating machinery supported on SFD is a challenging
task. Artificial intelligence (AI) and deep learning method is a recent data analysis
technique, which is being used in diagnostic and prognostic system control. The
orbit of the rotating machinery system contains a lot of information, which is being
used for monitoring and fault diagnosis of the system. A lot of research has been
done for denoising of orbit to acquire useful information. Haedong Jeong et al. [15]
implemented the convolution neural network for the generation and recognition of
orbit images of the rotor system. They also tested the implemented method with
the experimental data. Kankar et al. [16] implemented the artificial neural network
(ANN) and support vector machine (SVM) method for fault diagnosis of ball bear-
ings. The vibration data for various defects of the ball and races were analyzed and
orbit images were constructed by the above-mentioned method. The implemented
method was further used with experimental data.
This paper mainly focuses on the classification of synchronous and nonsyn-
chronous orbits of rotor systems supported by SFD. The machine learning method,
support vector machine has been used in this study. The vibration data are obtained
by solving the nonlinear equation of centralized SFD. The features are extracted from
this time series data. The extracted features have been used for training and testing
the data.

2 Mathematical Model

In this study, the SFD is centralized with an integral spring, which is termed a compact
SFD (CSFD). Figure 1a shows the schematic of a flexible rotor system supported
on CSFD and Fig. 1b is the kinematic representation of CSFD and disc. O and
OCSFD denote the center of the disc and the center of CSFD, respectively. The gravity
parameter W, mass ratio α and stiffness ratio K is the typical design parameter for
398 A. Kumar et al.

Fig. 1 a Schematic of rotor test rig supported on CSFD. b Kinematics of CSFD and Disc

the study of the synchronous response of this system. The vibration signal based on
these parameters has been obtained by solving the equation of motion of the system.
The ODE-15 s solver has been used for obtaining the vibration signal in MATLAB.
The equation of motion of the flexible rotor system supported on CSFD in
nondimensionalized [11] form is given in Eq. (1).

ẍ D + 2ζ ẋ D + (x D − xC S F D ) = U Ω2 cos(Ωτ )

ÿ D + 2ζ ẏ D + (y D − yC S F D ) + W = U Ω2 sin(Ωτ ) (1)

[x )] 1
CSF D ( ) yC S F D (
ẍC S F D + B I1 ε̇ + I2 εφ̇ − I2 ε̇ + I3 εφ̇ + (xC S F D − x D )
ε ε α
K
+ xC S F D = 0
α
[y )] 1
CSF D ( ) xC S F D (
ÿC S F D + B I1 ε̇ + I2 εφ̇ + I2 ε̇ + I3 εφ̇ + (yC S F D − y D )
ε ε α
K
+ yC S F D + W = 0
α
where,

θ1 +π cos 2 θ
I1 = ∫ dθ
θ1 (1 − εcosθ )3
θ1 +π sinθ cosθ
I2 = ∫ dθ
θ1 (1 − εcosθ )3
θ1 +π sin 2 θ
I3 = ∫ dθ
θ1 (1 − εcosθ )
3

( )
ε̇
θ1 = ar ctan −
εφ̇
Classification of Orbits of Rotor Supported on Squeeze Film Damper … 399

3 Empirical Mode Decomposition

Huang et al. [17] developed the empirical mode decomposition (EMD) method. This
method has been developed for the analysis of nonlinear and nonstationary data.
This method decomposes the complicated data set in a finite or often small number
of intrinsic mode functions (IMF).
Shannon energy (SE) calculates the energy of the local spectrum for each sample.
This shows the better detection range even in the presence of noise. This has a wider
detection domain with the least error. The SE is calculated as in eq. (2).
( )
S E[n] = −x 2 [n] × log x 2 [n] (2)

4 Results and Discussion

The nonlinear Eq. (1) of SFD has been solved by the ode15s method in MATLAB© .
The time-series data have been collected from the solution. The collected data have
been used in the EMD method and modes are extracted. The Shannon energy (SE)
has been calculated for all computed modes. The feature matrix has been obtained
from the extracted mode. The feature matrix is 121 × 5 in size with the fifth column
as the response variable. The labeling of the response variable has been done by
generating the bifurcation diagram. The data set has been divided into Training and
Test Data sets in the proportion of 75:25, respectively. From the Training Data Set,
25% of data have been kept for hold-out validation. The distribution of the 121 Data
Set is random and the proportion of the two categories for classification (Periodic
and Quasi-Periodic) has been kept close to equal. The training and testing have been
performed on MATLAB© . Different types of SVM and KNN classifiers have been
trained such as Linear, Cubic, Quadratic, and Gaussian.

Table 1 Accuracy obtained

Classifier Type Accuracy (%)
from different models on test
data SVM Linear 80
SVM Quadratic 83.3
SVM Fine Gaussian 73.3
SVM Medium Gaussian 80
SVM Coarse Gaussian 70
KNN Fine 93.3
KNN Medium 80
KNN Coarse 50
KNN Cosine 76.7
KNN Cubic 80
KNN Weighted 93.3
400 A. Kumar et al.

(a)

(b)

Fig. 2 a Scatter Plot for SE between MODE1 and MODE4 for quadratic SVM model performance
on test data b parallel coordinate plot for quadratic SVM model performance on test data
Classification of Orbits of Rotor Supported on Squeeze Film Damper … 401

4.1 Accuracy of SVM and KNN Models

The quadratic SVM model performs better than all other SVM types in terms of
accuracy on testing data and validation data. It achieved an accuracy of 83.3%
against evenly distributed 30 data sets on Testing Data and an accuracy of 73%
on validation data. The accuracy of other types of SVM models has been tabu-
lated in Table 1. Figure 2a represents the predictions made by the quadratic SVM
classifier between MODE1 and MODE4 SE values. There are four quasi-periodic
vibration data, which are labeled as periodic, and one periodic data, which is labeled
as quasi-periodic. Apart from that, all the classifications are correct and there is a
clear difference between them. The parallel coordinate plot in Fig. 2b also shows
that apart from feature DC1 rest are distinguishable for periodic and quasi-periodic
data sets. Figure 3a shows the confusion matrix for the quadratic SVM model where
it can be observed that a total of 5 out of 30 data sets are misclassified. Four quasi-
periodic data sets have been misclassified as periodic, and one periodic data set has
been misclassified as quasi-periodic. The ROC (Receiver Operating Characteristics)
curve shows AUC (Area Under the Cover) as 0.88.
The KNN Fine model outperforms all the other KNN types in terms of accuracy.
The model reached an accuracy of 93.3% on Test Data and 93% on Hold-Out Vali-
dation Data. The accuracy of all the KNN types has been listed in Table 1. Figure 4a
represents the predictions made by the Fine KNN classifier between MODE1 and
MODE4 SE values. There are two periodic data, which are labeled as quasi-periodic.
The parallel coordinate plot shows two incorrect predictions where periodic vibra-
tion is labeled as quasi-periodic in Fig. 4b. For all the other cases of Test Data, the
prediction is accurate. The Confusion Matrix in Fig. 5a clearly shows the correct
classification for 28 data sets and 2 misclassifications of periodic vibration into
quasi-periodic Vibration. The AUC in the ROC curve for the KNN Fine classifier is
0.93.

(a) (b)

Fig. 3 a Confusion matrix for testing data with quadratic SVM model b ROC curve for testing
result (SVM Model)
402 A. Kumar et al.

(a)

(b)

Fig. 4 a Scatter plot for SE between MODE1 and MODE4 for fine KNN model performance on
test data b parallel coordinate plot for fine KNN model performance on test data
Classification of Orbits of Rotor Supported on Squeeze Film Damper … 403

(a) (b)

Fig. 5 a Confusion matrix for testing data with KNN fine model b ROC curve for testing result
(KNN fine model)

5 Conclusion

The classification of SFD vibration in the Rotor system has been performed in this
study using SVM and KNN models. The study shows that the EMD along with
Shannon Energy calculation for the decomposed modes can be used for feature
generation to classify SFD vibration. Among the SVM models, the quadratic SVM
classifier achieves an accuracy of 83.3 on hold-out validation data and 73% on test
data. Among the KNN models, the fine KNN classifier achieves an accuracy of 93.3
on hold-out validation data and 93% on test data.
For the studied application, the fine KNN model will be effective for the
classification of SFD vibration.

References

1. Barrett LW, Gunter EJ. Steady-state and transient analysis of a squeeze film damper bearing
for rotor stability. 1975
2. Holmes R, Dogan M. Investigation of a rotor bearing assembly incorporating a squeeze-film
damper bearing. J Mech Eng Sci. 1982;24(3):129–37.
3. Holmes R, Dogan M. The performance of a sealed squeeze-film bearing in a flexible support
structure. Proc Inst Mech Eng Part C J Mech Eng Sci. 1985; 199(1):1–9
4. Holmes R, Eng D, Sykes JEH. the vibration of an aero-engine rotor incorporating two squeeze-
film dampers. 1996; 210
5. Mclean LJ, Hahn EJ. Unbalance behavior of squeeze film damped multi-mass flexible rotor
bearing systems. J Lubr Technol. 1983;105(1):22–8.
6. Mclean LJ, Hahn EJ. Stability of squeeze film damped multi-mass flexible rotor bearing
systems. J Tribol. 1985; 107(3), 402–409
7. Zhao JY, Linnett IW, Mclean LJ. Stability and bifurcation of unbalanced response of a squeeze
film damped flexible rotor. J. Tribol. 1994; 116(2):361–368
8. Zhao JY, Linnett IW, Mclean LJ. Subharmonic and quasi-periodic motions of an eccentric
squeeze film damper mounted rigid rotor. J Vib Acoust Trans ASME. 1994; 116(3):357–363
404 A. Kumar et al.

9. Bonello P, Brennan MJ, Holmes R. Non-linear modelling of rotor dynamic systems with
squeeze film dampers—an efficient integrated approach. J. Sound Vib. 2002; 249(4):743–773
10. Bonello P, Hai PM. 2009. A receptance harmonic balance technique for the computation of
the vibration of a whole aero-engine model with nonlinear bearings. J. Sound Vib. 2009;
324(1–2):221–242
11. Inayat-hussain JI. Bifurcations in the response of a flexible rotor in squeeze-film dampers with
retainer springs. Chaos, Solitons Fractals. 2009;39(2):519–32.
12. Chu FL, Holmes R. The effect of squeeze film damper parameters on the unbalance response
and stability of a flexible rotor. J. Eng. Gas Turbines Power. 120(1):140–148
13. Chu FL, Holmes R. Efficient computation on nonlinear responses of a rotating assembly incor-
porating the squeeze film damper. Comput. Methods Appl. Mech. Eng. 1998; 164(3):363–373
14. Chu FL, Holmes R. The damping capacity of the squeeze film damper in suppressing vibration
of a rotating assembly. Tribol. Int. 2000; 33(2):81–97
15. Jeong H, Park S, Woo S, Lee S. Rotating machinery diagnostics using deep learning on orbit
plot images. Procedia Manuf. 2016;5:1107–18.
16. Kankar PK, Sharma SC, Harsha SP. Expert systems with applications fault diagnosis of ball
bearings using machine learning methods. Expert Syst Appl. 2011;38(3):1876–86.
17. Huang NE, Shen Z, Long SR, Wu MC, Snin HH, Zheng Q, Yen NC, Tung CC, Liu HH. The
empirical mode decomposition and the hubert spectrum for nonlinear and non-stationary time
series analysis. Proc. R Soc A Math Phys Eng Sci. 1998; 454:903–995
Free Vibration Response of 3D Braided
Reinforced Composite Conoidal Shell
Using Third-Order Shear Deformation
Theory

Pabitra Maji and Brighu Nath Singh

Abstract The current computational work models the free vibration behavior of 3D
braided composite conoidal shells using the third-order shear deformation theory
(TSDT). Eight noded shell elements with twelve degrees of freedom per node are
utilized in the finite element method. The unique advantage of this theory is that the
transverse displacement depends on the thickness of the shell. As a result, it is more
able than the other shell theory to precisely predict the outcomes of the thick shell.
Four one-to-one braiding performance phases are offered to produce the 3D braided
conoidal shells. It is believed that the fibers exhibit transverse isotropy and matrix
isotropy. In order to determine the material properties of the composites, a volume
averaging method (VAM) is employed. The numerical code’s accuracy is assessed by
contrasting it with findings from earlier works of the literature. Numerous boundary
conditions, thickness ratios, height ratios, braided volume fractions, and braided
angles are investigated in this investigation.

Keywords Conoidal shells · Braided composite · TSDT · FEM · Free vibration

Nomenclature

a Length
b Width
h Thickness
HL. Lower height
HH . Higher height
[BL . ] Strain–displacement matrix

P. Maji (B) · B. N. Singh

Department of Aerospace Engineering, Indian Institute of Technology Kharagpur, Kharagpur,
India
e-mail: [email protected]
B. N. Singh
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 405
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_27
406 P. Maji and B. N. Singh

[K e ] Elemental stiffness matrix

[De ] Elasticity matrix
wi Braided pitch width
hi Braided pitch length
α Braid angle

1 Introduction

Due to its numerous benefits, an innovative 3D braided composite is becoming more

and more popular in various industries. They possess a number of unique qualities,
including stiffness in the thickness direction, lightness, anti-delamination ability,
increased strength, etc. The 3D braided composite was successfully applied in the
aerospace, architectural, civil, mechanical, etc. industries. Conoidal shell structures
are simple to construct and are frequently utilized in beautiful architecture to let the
northern lights in. In comparison to other composites, the microstructure of textile
composites is highly complex. Fewer fibers are aligned in the thickness direction in
this composite, which results in anti-delamination and boosts stiffness. In the previous
few decades, 3D braided composites have quickly found use in many different
fields. Specific strength, thickness strength, anti-delamination capabilities, etc. are
all quite high. Since then, 3D braided composites have seen widespread application
in industries like aerospace, automotive, marine, and medicine. The braided struc-
ture features a more complex geometrical design than regular composites. Because
of this, determining the 3D braided composite’s corresponding material qualities
requires a lot of care. To guarantee risk-free operation under dynamic situations, a
thorough familiarity with their vibration characteristics is essential.
The material behavior of 3D multidirectional textile composites exhibits non-
linear elastic behavior and is highly anisotropic. There is little information available
regarding mechanical characteristics because the majority of research into predicting
the behavior of multidirectional C–C composites was based on tests. The braided
materials’ microstructure arrangement is quite tangled. Only a few numerical models
have been created thus far to simulate the real behavior of the 3D braided composites.
An accurate knowledge of these materials’ mechanical behavior is necessary given
their growing application. Studies on changes in mechanical properties and struc-
tures are crucial to determining the durability of structural components composed of
textile composites. Therefore, developing a precise mathematical model to calculate
the corresponding mechanical properties of the braided structure is essential. The
mechanical behavior of textile composites can be predicted using computationally
efficient mathematical techniques.
The effective Young moduli of a 3D braided composite were examined by
Kalidindi et al. [1] utilizing experimental studies and numerical simulation. The
homogenization method was used by Zurong et al. [2] to determine the effective
Free Vibration Response of 3D Braided Reinforced Composite … 407

Young moduli of a 3D braided composite. Dong and Huo [3] investigated the mechan-
ical behavior of braided composite and computed the braided structure’s mechanical
modulus and micro stress using two-scale FE models. Shokrieh and Mazloomi [4]
developed a multi-unit cell model to investigate the mechanical properties of the 3D
braided four-directional composite based on the Bridging model using the volume
averaging method. Sun et al. [5] measured the actual material moduli of 3D braided
composites based on finite element methods (FEM). For the purpose of calculating
the longitudinal and in-plane material characteristics of a 3D braided composite, Ai
et al. [6] used FEM studies. Using the FEM, Chakravorty et al. [7] examined the modal
behavior of layered conoidal shells. By using FEM and first-order shear deformation
theory, Das et al. [8] examined the bending response of corners supported by lami-
nated conoidal shells (FSDT). Higher order shear deformation theory was used by
Pradyumna and Bandyopadhyay [9] to calculate the modal response of layered mate-
rials (HSDT). Using third-order shear deformation theory, Kant and Swaminathan
[10] constructed the vibration response of thick panels.
While Li [11] examined the buckling description of plates formed of the 3D
braided composite under the thermal fields based on an analytical method, Yang
and Haung [12] examined the dynamic properties of the 3D braided composite plate
using higher order shear deformation theory. Wang et al. [13] investigated damage
and failure behavior as well as the mechanical performance of braided composite
structures. Using IHSDT and the finite element method, Singh and Singh [14] deter-
mined the buckling performance of plates composed of 3D braided composite. Using
the new shear deformation theory based on the FEM, Singh and Singh [15] also eval-
uated the static and modal characteristics of the 3D braided plate. Maji et al. [16]
used TSDT to examine the free vibrations of curved panels constructed of 3D braided
composite. The modal response of 3D braided composite cantilever-supported cylin-
drical and conical shells under rotations was recently computed by Maji and Singh
[17, 18] using FEM (TSDT).
The free vibration properties of laminated and 3D braided composite plates and
shells have been widely explored, as evidenced by the aforementioned references.
The authors report that they have been unable to locate any previous works dedicated
to the topic of rotating 3D braided composite conoidal shells. To address this gap
in the literature, the authors conduct analyses they believe to be the first attempt at
characterizing the free vibration properties of 3D braided composite conoidal shells.
The free vibration behavior of 3D braided composite conoidal panels based on the
TSDT theory is thus accomplished in this case. In contrast to the other shell theories,
it has more potential for calculating the precise consequences of the thick shell. Finite
element methods at braided angles, braided volume fractions, aspect ratios, height
ratios, and thickness ratios were used to achieve the current results.
408 P. Maji and B. N. Singh

2 Mathematical Formulations

A conoidal shell panel with various dimensions is illustrated in Fig. 1. The dimensions
of conoidal shells in the Cartesian coordinate systems are length (a), width (b),
thickness (h), lower height (H L ), and higher height (H H ).

2.1 Displacement Fields

The elements of the conoidal shells’ displacement fields [u, v, and w] in the three
distinct directions are described as [10]

∑
4
û(x, y, z, t) = z i−1 u i (x, y, t) (1)
i=1

∑
4
v̂(x, y, z, t) = z i−1 vi (x, y, t) (2)
i=1

∑
4
ŵ(x, y, z, t) = z i−1 wi (x, y, t) (3)
i=1

where u and v have denoted the in-plane displacements, and w represents the
transverse displacement.
The braided composite and matrix are assumed linear in nature. The linear nature
of strain–displacement relationships of the conoidal shells are articulated as [10]

{ε}x yz = {ε0 } + z{k1 } + z 2 {k2 } + z 3 {k3 } (4)

Fig. 1 Geometric
configuration of conoidal
shells
Free Vibration Response of 3D Braided Reinforced Composite … 409

where {ε0 } are the mid-plane strain components. More specifics are given in
Appendix.
The linear strain–displacement relationship of the conoidal textile shells can be
articulated as

{ε}x yz = [B L ]{de } (5)

where [BL ] is defined as the strain–displacement matrix. More specifics are given in
Appendix.
The stress–strain for conoidal shells can be expressed as
_
{σ }x yz = [ Q]{ε}x yz = [T ][Q][T ]T {ε}x yz (6)

{ }T
{σ }Tx yz = σx x σ yy σzz σ yz σx z σx y (7)

2.2 The Strain Energy of Conoidal Shells

The strain energy (U 1 ) of conoidal shells due to conservative force is given as

{
1
U1 = {ε}T {σ }dv (8)
2 v
{ {
1
= {de }T [B L ]T [De ] [B L ] {de } d x d y (9)
2 A
{ {
[K e ] = [B L ]T [De ] [B L ] d x d y (10)
A

where [K e ] is defined as the elemental stiffness. [De ] is the elasticity matrix.

2.3 The Kinetic Energy of Conoidal Shells

The kinetic energy, U 2 , of the conoidal shells element is defined as

{
1
U2 = ρ V→ .V→ dv (11)
2 v

where V→ is the velocity vector and ρ is the density of the conoidal shells,
410 P. Maji and B. N. Singh
{
1 { }T T { }
U2 = ρ ḋe [Ne ] [Ne ] ḋe ∀ (12)
2 ∀

1 { }T { }
= ḋe [Me ] ḋe (13)
2
{
[Me ] = [N ]T [m][N ]d xd y (14)
A

where [Me ] is the defined elemental mass matrix of the conoidal shells. [Ne ] is the
shape function matrix.

2.4 FE Formulations

An isoparametric finite element formulation is implemented to the model’s textile

conoidal shells. In the nodal coordinate system, the generalized field variables can
be written as

[ ] ∑ 8 { }
δ̂ = [Ni ] d̂ei (15)
i=1

∑
8 ∑
8
x= [Ni ]{xi } and y = [Ni ]{yi } (16)
i=1 i=1

The elemental stiffness [K e ] and mass matrices [M e ] in the natural coordinates

system are articulated as
{ +1 { +1
[K e ] = [B L ]T [De ] [B L ] |J | dξ dη (17)
−1 −1
{ +1 { +1
[Me ] = [Ne ]T [m] [Ne ] |J | dξ dη (18)
−1 −1

2.5 Micromechanical Approach of 3D Braided Conoidal

Shells

Figure 2 illustrates the yarn configuration of the 3D braid conoidal composite shell
[4]. The interior surface, exterior surface, surface, and corner regions are the four
distinct segments that make up each unit cell configuration [4]. Wi, hi, and α braid
angle are the terms used to describe the braided pitch:
Free Vibration Response of 3D Braided Reinforced Composite … 411

a b
Interior region
Surface region
Corner region

Fig. 2 a Interior, surface, and corner regions [4]. b Fiber configuration of interior region unit cell
[4]

√
tan γ = 2 tan α (19)

tan β = tan θ = 1/2 tan α (20)

The total fiber volume fraction can be calculated from the following equation as
follows:

V f = Ci Vi + Css Vss + Cis Vis + Cc Vc (21)

where C i , C is , C ss , and C c have defined the volume proportions of the four different
control volumes while V i , V is , V ss , and V cc are denoted as fiber volume fraction of
the above regions, respectively.

2.5.1 Homogenization Procedure for Material Property

In this technique, the fibers are considered transversely isotropic, and the matrix is
assumed as isotropic materials, which is anticipated by Haung [12]. The stiffness
matrix is characterized as [C] f = [S]−1
f where [S] f is the compliance matrix for the
fiber.
The stiffness matrix of the yearn matrix can be well-defined as [C]m = [S]−1 m
where [S]m is defined as the compliance matrix.
The compliance matrix for all unidirectional 3D braided conoidal shells is
characterized as
[ ] ( )( )−1
Si j n
= Vnf [Si j ] f + Vnm [Si j ]m [ai j ] Vnf [I ] + Vnm [ai j ] (22)
412 P. Maji and B. N. Singh

f
where Vn and Vnm indicate the volume fraction of fibers and matrix of various unidi-
f
rectional braided conoidal shells, respectively, and are related as Vn + Vnm = 1. [ai j ]
Indicates bridging matrix.
Finally, the total stiffness matrix of the conoidal shell based on the volume
averaging techniques is expressed as [4]
∑
[C] = Vn [C i j ]n n = (i1, i2, ......., c3, c4) (23)

For even number of m and n, the volume fraction of the four different regions can be
represented as

2(m − 1)(n − 1) + 2
Vi1 = Vi2 = Vi3 = Vi4 = (24)
4(2mn + m + n)
(m + n − 4)
Vis1 = Vis2 = Vis3 = Vis4 = (25)
4(2mn + m + n)
(m + n − 4)
Vss1 = Vss2 = Vss3 = Vss4 = (26)
2(2mn + m + n)
2
Vc1 = Vc2 = Vc3 = Vc4 = (27)
(2mn + m + n)

The equivalent mechanical properties of the 3D braided conoidal shells are

calculated as
1 1 1
Ex x = , E yy = , E zz = , (28)
S 11 S 22 S 33
1 1 1
Gxy = , Gxz = , G yz = , (29)
S 55 S 66 S 44
S 12 S 13 S 23
νx y = − , νx z = − and ν yz = − (30)
S 11 S 11 S 22

2.6 Free Vibration Analysis

Based on Hamilton’s principle, the equation of motion of the conoidal panels is

known by
{ } { } { }
d ∂U2 ∂U2 ∂
− + (U1 + U3 ) = 0 (31)
dt ∂ ḋe ∂de ∂de
Free Vibration Response of 3D Braided Reinforced Composite … 413

Neglecting the damping results, the free vibration equations in the global
coordination are known by
{ }
[M] d̈ + ([K ]){d} = 0 (32)

Considering harmonic vibrations, {d} = {d}eiωn t , we have

[ ]
[K ] − ωn2 [M] {d} = 0 (33)

This is an Eigen solution and is worked out for the eigenvalues and eigenvectors

[A]{d} = λ{d} (34)

where, [A] = [K ]−1 [M] and λ = 1/ωn2 .

2.7 Boundary Conditions

For four-edged simply supported (SSSS) cases:

x = 0, a; v1 = w1 = v2 = w2 = v3 = w3 = v4 = w4 = 0, (35)

y = 0, b; u 1 = w1 = u 2 = w2 = u 3 = w3 = u 4 = w4 = 0 (36)

For four-edged clamp-supported (CCCC) cases (Table 1):

x = 0, a; y = 0, b; u 1 = v1 = w1 = u 2 = v2 = w2 = u 3 = v3 = w3 = u 4 = v4 = w4 = 0 (37)

Material-3: E 1 /E 2 = 25, G12 = G13 = 0.5E 2 , G23 = 0.2E 2 , ν12 = ν13 = ν23 =
0.25.

Table 1 Material property

AS4 E 11 f E 22 f G 12 f G 23 f ν12 f Em νm
Material-1 234.6 13.8 13.8 5.5 0.2 2.94 0.35
Material-2 230.0 40.0 24.3 14.3 0.35 2.94 0.35
414 P. Maji and B. N. Singh

3 Results and Discussions

The isoparametric finite element approach is used to generate current numerical

results. Software called MATLAB was used to produce the current numerical results.
The numerous evaluations are taken into account while evaluating the correctness of
the MATLAB code, which is provided in Tables 2 and 3. The outcomes of laminated
conoidal shells for various lamina configurations with clamp support are shown
in Table 2. The results of Yang and Haung and Singh and Singh are shown in
Table 3 together with the NDFF confirmation of the 3D braided composite SSSS
plate. The results are clearly well-converged with mesh size (8 × 8). Using material-
2, the current numerical findings are produced, and √non-dimensional fundamental
frequencies (NDFF) are articulated as ϖ = ωa 2 / h 12ρ f /E 22 f .

3.1 Effect of Braid Angle (α)

The natural frequency (NF) results of conoidal shells are generated by varying braided
angles with SSSS- and CCCC-supported boundary conditions (BCs) and illustrated in
Fig. 3. The used parameters are a/b = 1, a/H H = 10.0, and a/h = 100, taken constantly
for this example. In this study, four sets of braided angles are taken into consideration,
such as 20° , 25°, 30°, and 35°, respectively. It is observed that the fundamental
frequency (FF) of the conoidal shells drops with the enhancement in braided angle (α)
with CCCC BCs, and the secondary frequency (SF) of panels increases except for α =
350 with CCCC BCs. The corresponding elastic properties of the 3D braided conoidal
shell panels decrease with the increase in braid angle. For that reason, the stiffness

Table 2 Non-dimensional frequency of laminated conoidal shells for different lamina configura-
tions with clamp support. a/b = 1.0, a/H H = 5, a/h = 100. Material-3
θ0 H L /H H
0.2 0.15 0.1 0.05 0.00
15/−15 Present FEM 65.5669 63.6969 61.8428 60.0567 58.3841
Pradyumna and 65.6122 63.7285 61.8627 60.0667 58.3857
Bandyopadhyay [9]
30/−30 Present FEM 71.0136 67.8482 64.9079 62.2264 59.8275
Pradyumna and 71.3619 68.1776 65.3564 62.5348 60.1323
Bandyopadhyay [9]

Table 3 The √ NDFF

Braid angle (α) Singh and Singh [16] Present FEM
(ϖ = ωa 2 / h 12ρ f /E 22 f )
of the simply supported 3D 20 16.1892 15.6445
braided composite plate at 30 16.1464 15.3837
V f , Material-1
Free Vibration Response of 3D Braided Reinforced Composite … 415

of the conoidal shells decreases, and the natural frequency decreases. It can seem
that both natural frequencies also are increased with an increase in H L for the SSSS-
supported conoidal shells. The gradual increase in SF is observed with the rise in H L .
Similarly, for the CCCC cases, the fundamental frequency (FF) of the conoidal also is
increased with an increase in H L. Interestingly, for the comparative study, the opposite
nature of the curve is identified for the secondary frequency. The greater value of
natural frequencies is found with Clamp boundary conditions, as CCCC supports
have more constraints than simple support. The highest percentage rise in natural
frequency is identified at higher values of H L with CCCC boundary conditions. Its
more pronounced effect is identified at lower braided angles. Furthermore, it is also
observed that the percentage decrease of NF of braided composite conoidal shells is
higher with the CCCC BCs for intensification in braided angle.

20 22
=200 =200
=250
=250
=300
18 21
=350 =300
SSSS
=350
FF

SSSS
16 20
SF

14 19

12 18
0 1 2 3 4 0 1 2 3 4
HL HL

(a) (b)
30.0 33.00
=200 =200
=250 =250
=300 =300
27.5 32.75
=350 =350
CCCC
CCCC

25.0 32.50
FF

SF

22.5 32.25

20.0 32.00
0 1 2 3 4 0 1 2 3 4
HL HL
(c) (d)

Fig. 3 The natural frequency (NF) results of conoidal shells by varying α

416 P. Maji and B. N. Singh

3.2 Effect of Braided Volume Fraction (Vf )

The natural frequency (NF) results of conoidal shells are generated by varying braided
volume fractions with all edges SSSS and CCCC BCs and illustrated in Fig. 4. The
used parameters are a/b = 1, a/H H = 10.0, and a/h = 100 kept constant for this
example. In this study, five sets of braided volume fractions (V f ) are taken into
consideration, such as 0.3, 0.35, 0.4, 0.45, and 0.5, respectively. It is observed that
both the fundamental frequency (FF) and secondary frequency (SF) of the conoidal
shells enhance with the enhancement in V f . The corresponding elastic properties of
the panels increase with the increase in volume fractions (V f ). For that reason, the
stiffness of the conoidal shells always increases, and the natural frequency increases.
It can seem that both natural frequencies also are increased with an increase in H L
for the SSSS-supported conoidal shells. The gradual increase in SF is observed with
the rise in H L with SSSS support. Similarly, for the CCCC cases, the fundamental
frequency (FF) of the conoidal shells are also increased with an increase in H L.
Interestingly, for the comparative study, the opposite nature is identified for the
secondary frequency. The greater value of natural frequencies is found with Clamp
boundary conditions, as CCCC supports have more constraints than simply supported
boundary condition.

3.3 Effect of Height Ratio (HL /HH ) on Natural Frequency

The fundamental frequency (FF) results of conoidal shells are generated by varying
height ratio (H L /H H ) with all edges SSSS and CCCC BCs and illustrated in Tables 4
and 5. The used parameters are a/b = 1, a/H H = 5.0, and a/h = 100 kept constant
for this example. Here, five different values of height ratio (H L /H H ) are considered,
such as 0.2, 0.15, 0.1, 0.05, and 0.0, respectively. It is noticed that the FF value
always increases due to the increase in height ratios because the higher height ratio
(H L /H H ) provides a stiffer structure. Furthermore, it is also observed that the highest
percentage decrease in frequency parameters of 3D braided conoidal shells with
the higher volume fractions is due to the drop in higher height ratio (H L /H H ). The
percentage decrease of the FF of conoidal shells due to the drop in height ratio (H L /
H H ) is depicted in Tables 6 and 7. The percentage decrease rate of FF due to the
drop in height ratio is more with higher volume fractions and braid angles. It can
be seen that percentage decrease rate of FF effect is more pronounced with CCCC
supported.
Free Vibration Response of 3D Braided Reinforced Composite … 417

22 24

SSSS
20

18 20

16
FF

SF
Vf=0.30
14 Vf=0.30
Vf=0.35 16
Vf=0.35
Vf=0.40 SSSS
Vf=0.40
12
Vf=0.45 Vf=0.45
Vf=0.50 Vf=0.50
10 12
0 1 2 3 4 0 1 2 3 4
HL HL
(a) (b)
33 45
Vf=0.30 Vf=0.30
CCCC
Vf=0.35 Vf=0.35
30 CCCC
Vf=0.40 40 Vf=0.40
Vf=0.45 Vf=0.45
Vf=0.50 Vf=0.50
27 35
FF

SF

24 30

21 25

18 20
0 1 2 3 4 0 1 2 3 4
HL HL

(c) (d)

Fig. 4 The natural frequency (NF) results of conoidal shells by varying V f

Table 4 The FF of conoidal shells due to the drop in H L /H H with braided volume fractions
BCs H L /H H Vf
0.30 0.35 0.40 0.45 0.50
SSSS 0.20 15.9135 16.9293 17.9691 19.0539 20.2112
0.15 15.5224 16.5142 17.5270 18.5821 19.7064
0.10 14.5808 15.5220 16.4770 17.4667 18.5167
0.05 13.3107 14.1793 15.0565 15.9619 16.9187
0.00 12.0768 12.8727 13.6734 14.4969 15.3640
CCCC 0.20 23.2008 24.6166 26.0708 27.5944 29.2263
0.15 22.4171 23.7743 25.1703 26.6346 28.2042
0.10 21.6152 22.9175 24.2577 25.6642 27.1722
0.05 20.8203 22.0722 23.3604 24.7120 26.1606
0.00 20.0563 21.2632 22.5040 23.8046 25.1974
418 P. Maji and B. N. Singh

Table 5 The FF of conoidal shells due to the drop in H L /H H with braid angles
BCs H L /H H α
20 25 30 35
SSSS 0.20 19.0873 19.0539 18.8059 18.4709
0.15 18.6797 18.5821 18.2872 17.9108
0.10 17.6758 17.4667 17.0906 16.6472
0.05 16.2457 15.9619 15.5314 15.0490
0.00 14.8472 14.4969 14.0206 13.5086
CCCC 0.20 27.9891 27.5944 26.9926 26.4071
0.15 27.1601 26.6346 25.9645 25.3585
0.10 26.3045 25.6642 24.9308 24.2983
0.05 25.4484 24.7120 23.9220 23.2597
0.00 24.6180 23.8046 22.9664 22.2736

Table 6 The percentage decrease of FF of conoidal shells due to the drop in H L /H H with braided
volume fractions
BCs H L /H H Vf
0.30 0.35 0.40 0.45 0.50
SSSS 0.15 2.4573 2.4524 2.4604 2.4762 2.4977
0.10 8.3747 8.3128 8.3038 8.3301 8.3839
0.05 16.3559 16.2443 16.2090 16.2281 16.2907
0.00 24.1096 23.9624 23.9059 23.9167 23.9827
CCCC 0.15 3.3781 3.4216 3.4540 3.4783 3.4970
0.10 6.8342 6.9023 6.9542 6.9949 7.0280
0.05 10.2607 10.3362 10.3960 10.4457 10.4894
0.00 13.5536 13.6226 13.6810 13.7338 13.7850

Table 7 The percentage decrease of FF of conoidal shells due to the drop in H L /H H with braid
angles
BCs H L /H H α
20 25 30 35
SSSS 0.15 2.1351 2.4762 2.7583 3.0321
0.10 7.3945 8.3301 9.1213 9.8731
0.05 14.8872 16.2281 17.4122 18.5256
0.00 22.2138 23.9167 25.4459 26.8655
CCCC 0.15 2.9619 3.4783 3.8085 3.9709
0.10 6.0190 6.9949 7.6382 7.9857
0.05 9.0774 10.4457 11.3755 11.9189
0.00 12.0443 13.7338 14.9159 15.6529
Free Vibration Response of 3D Braided Reinforced Composite … 419

3.4 Influence of Thickness Ratio (a/h)

The fundamental frequency (FF) results of conoidal shells are generated by varying
the thickness ratio (a/h) with different a/H H and are illustrated in Tables 8 and 10. It
is observed that the FF values of conoidal panels are reduced due to an increase in
a/H H . The more pronounced an effect of a/H H ratios identifies with higher thickness
ratios. It is also clear that the FF value enhances with the increase of a/h values,
irrespective of a/H H . Interestingly, the more pronounced effect of a/H H is identified
with a lower braided angle, and similar observations of the more pronounced effect
are found with higher volume fraction (Tables 9, 10 and 11).

Table 8 FF of 3D braided conoidal shells with CCCC supported by varying braided angles. a/b =
1.0, H L /H H = 0.2
a/H H a/h α
20 25 30 35
2.5 10 11.5999 11.3865 11.1095 10.8483
100 40.3857 40.0498 39.2570 38.3655
5.0 10 9.3355 8.9784 8.5909 8.2566
100 27.9891 27.5943 26.9926 26.4071
7.5 10 8.5451 8.1320 7.6940 7.3186
100 20.2609 19.8749 19.4464 19.0660
10.0 10 8.3275 7.8975 7.4436 7.0543
100 14.2197 13.7677 13.3598 13.0230

Table 9 The percentage change of FF of 3D braided conoidal shells with CCCC supported by
varying braided angles. a/b = 1.0, H L /H H = 0.2
a/H H a/h α
20 25 30 35
5.0 10 −19.5208 −21.1490 −22.6703 −23.8903
100 −30.6955 −31.1001 −31.2415 −31.1697
7.5 10 −26.3346 −28.5824 −30.7434 −32.5372
100 −49.8315 −50.3745 −50.4639 −50.3043
10.0 10 −28.2112 −30.6415 −32.9975 −34.9732
100 −64.7903 −65.6236 −65.9684 −66.0554
420 P. Maji and B. N. Singh

Table 10 FF of 3D braided conoidal shells with CCCC supported by varying braided volume
fractions. a/b = 1.0, H L /H H = 0.2
a/H H a/h Vf
0.30 0.35 0.40 0.45 0.50
2.5 10 9.5480 10.1427 10.7514 11.3866 12.0641
100 33.6536 35.7199 37.8366 40.0500 42.4164
5.0 10 7.5523 8.0419 8.5352 8.9784 9.5764
100 23.9433 25.4195 26.9328 27.5943 30.2088
7.5 10 6.7687 7.2218 7.6731 8.1320 8.6098
100 16.7227 17.7321 18.7756 19.8749 21.0582
10.0 10 6.5665 7.0104 7.4510 7.8975 8.3610
100 11.5711 12.2783 13.0055 13.7677 14.5843

Table 11 Percentage change of FF of 3D braided conoidal shells with CCCC supported by varying
braided volume fractions. a/b = 1.0, H L /H H = 0.2
a/H H a/h Vf
0.30 0.35 0.40 0.45 0.50
5.0 10 −20.9017 −20.7133 −20.6135 −21.1494 −20.6211
100 −28.8537 −28.8365 −28.8182 −31.1004 −28.7804
7.5 10 −29.1084 −28.7985 −28.6320 −28.5828 −28.6326
100 −50.3093 −50.3581 −50.3773 −50.3748 −50.3535
10.0 10 −31.2262 −30.8824 −30.6974 −30.6418 −30.6956
100 −65.6170 −65.6262 −65.6273 −65.6237 −65.6163

4 Conclusions

The present research is carried out for the free vibration nature of 3D textile composite
conoidal shells based on the third-order shear deformation theory (TSDT) using the
finite element method. Different evaluations are carried out to confirm the accuracy
and precision of the models. It is observed that the fundamental frequency (FF) of
the conoidal shells drops with the enhancement in braided angle (α) with CCCC
BCs, and the secondary frequency (SF) of panels increases except for α = 35° with
CCCC BCs. The corresponding elastic properties of the 3D braided conoidal shell
panels decrease with the increase in braid angle. For that reason, the stiffness of
the conoidal shells decreases, and the natural frequency decreases. It can seem that
both natural frequencies also are increased with an increase in H L for the SSSS-
supported conoidal shells. The gradual increase in SF is observed with the rise in
H L. Similarly, for the CCCC cases, the fundamental frequency (FF) of the conoidal
also increased with an increase in H L. The greater value of natural frequencies is
found with Clamp boundary conditions, as CCCC supports have more constraints
Free Vibration Response of 3D Braided Reinforced Composite … 421

than simply support. It is observed that both the fundamental frequency (FF) and
secondary frequency (SF) of the conoidal shells enhance with the enhancement in
V f . The corresponding elastic properties of the panels increase with the increase in
volume fractions (V f ). For that reason, the stiffness of the conoidal shells always
increases, and the natural frequency increases. It is noticed that the FF value always
increases due to the increase in height ratios because the higher height ratio (H L /H H )
provides a stiffer structure. It is observed that the FF values of conoidal panels are
reduced due to an increase in a/H H . The more pronounced an effect of a/H H ratios
identified with higher thickness ratios.

Acknowledgements The authors are pleased with the ISRO for providing financial support.

Appendix
( ) ( ) ( ) ( )
ε x x = ∂∂ux + Rwx , ε yy = ∂∂vy + Rwy , ε zz = ∂w ∂z
, ε yz = ∂w
− v
+
( ∂v ) ( ) ( ) ( ) ( )∂ y { R}y
, ε x z = ∂w − Rux + ∂u , ε x y = ∂u + Rwx y + ∂∂vx + Rwx y , ε0
T
∂z ∂x ∂z ∂y
=
{ }T { } { }T { }
T T
εx0 x ε0yy εzz0
ε0yz εx0z εx0y , k 1 = k x1x k 1yy k zz
1
k 1yz k x1z k x1y , k 2 =
{ }T { } { }T
T
k x2x k 2yy k zz
2
k 2yz k x2z k x2y , k 3 = k x3x k 3yy 0 k 3yz k x3z k x3y .

References

1. Kalidindi SR, Abusafieh A. Longitudinal and transverse moduli and strengths of low angle 3-D
braided composites. J Comp Mater. 1196; 30(8):885–905
2. Zuorong C, Dechao Z, Lu M, Ye L. Evaluation of elastic properties of 3-D (4-step) regular
braided composites by a homogenisation method. Compos Struct. 1999;47:477–82. https://doi.
org/10.1016/S0263-8223(00)00026-X.
3. Dong J, Huo N. A two-scale method for predicting the mechanical properties of 3D braided
composites with internal defects. Compos Struct. 2016;152:1–10. https://doi.org/10.1016/j.
compstruct.2016.05.025.
4. Shokrieh MM, Mazloomi MS. A new analytical model for calculation of stiffness of three-
dimensional four-directional braided composites. Compos Struct. 2012;94:1005–15. https://
doi.org/10.1016/j.compstruct.2011.09.010.
5. Sun H, Di S, Zhang N, Pan N, Wu C. Micromechanics of braided composites via multivariable
FEM. Comput Struct. 2003;81:2021–7. https://doi.org/10.1016/S0045-7949(03)00228-1.
6. Ai J, Du X, Li D, Jiang L. Parametric study on longitudinal and out-of-plane compressive prop-
erties, progressive damage and failure of 3D five-directional braided composites. Compos. Part
A Appl. Sci. Manuf. 2022; 156:106840. https://doi.org/10.1016/j.compositesa.2022.106840
7. Chakravorty D, Bandyopadhyay JN, Sinha PK. Finite element free vibration analysis of doubly
curved laminated composite shells. J Sound Vib. 1996;191:491–504. https://doi.org/10.1006/
jsvi.1996.0136.
422 P. Maji and B. N. Singh

8. Das HS, Chakravorty D. A finite element application in the analysis and design of point-
supported composite conoidal shell roofs: suggesting selection guidelines. J Strain Anal Eng
Des. 2010;45:165–77. https://doi.org/10.1243/03093247JSA582.
9. Pradyumna S, Bandyopadhyay JN. Static and free vibration analyses of laminated shells using
a higher-order theory. J Reinf Plast Compos. 2008; 27:167–186. https://doi.org/10.1177/073
1684407081385
10. Kant T, Swaminathan K. Analytical solutions for free vibration of laminated composite and
sandwich plates based on a higher-order refined theory. Compos Struct. 2001;53:73–85. https://
doi.org/10.1016/S0263-8223(00)00180-X.
11. Li ZM. Thermal postbuckling behavior of 3D braided rectangular plates. J Therm Stress.
2011;34:626–49. https://doi.org/10.1080/01495739.2011.581050.
12. Yang J, Huang X. Dynamic stability behavior of 3D braided composite plates integrated with
piezoelectric layers. J Compos Mater. 2009;43:2223–38. https://doi.org/10.1177/002199830
9339219.
13. Wang B, Zhang G, Nie X, Wu C. A multi-scale finite element approach for the mechanical
behavior analysis of 3D braided composite structures. Compos Struct. 2022;279: 114711.
https://doi.org/10.1016/j.compstruct.2021.114711.
14. Singh DB, Singh BN. Buckling analysis of three dimensional braided composite plates
under uniaxial loading using inverse hyperbolic shear deformation theory. Compos Struct.
2016;157:360–5. https://doi.org/10.1016/j.compstruct.2016.08.029.
15. Singh DB, Singh BN. Assessment and accuracy of new nonpolynomial shear deformation
theories for static analysis of laminated and braided composite plates. J Aerosp Eng. 2017; 30.
https://doi.org/10.1061/(ASCE)AS.1943-5525.0000768
16. Maji P, Singh BN, Singh DB. A third-order polynomial for the free vibration response of 3D
braided curved panels using various boundary conditions. Mech Based Des Struct Mach 2021;
1–23. https://doi.org/10.1080/15397734.2021.1894574
17. Maji P, Singh BN. Free vibration responses of 3D braided rotating cylindrical shells based
on third-order shear deformation. Compos Struct. 2020;260: 113255. https://doi.org/10.1016/
j.compstruct.2020.113255.
18. Maji P, Singh BN. Shear deformation theory for free vibration responses of 3D braided pre-
twisted conical shells under rotation. Int J Comput Methods Eng Sci Mech. 2021; 1–20. https://
doi.org/10.1080/15502287.2021.1916792
A Dynamic Model for Polymer Draft
Gears

Om Prakash Yadav and Nalinaksh S. Vyas

Abstract A number of rail vehicles (coaches or wagons) are inter-connected through

coupling mechanisms to form a train. Center Buffer Coupling is widely used in
mainline passenger trains in Indian Railways. This coupling system includes an
AAR coupler and a polymer-based balanced draft gear, which consists of two sepa-
rate packs of polymer springs for buff and draft energy absorption. Unlike metal
springs, polymer springs attribute nonlinear stiffness and damping characteristics.
Several mathematical models of polymer draft gears are developed based on static and
dynamic experimental curves. Most commonly used models include either lookup
table methods or polynomial fitting methods. Despite their potential to simulate close
dynamic characteristics of polymer draft gears, these models suffer from numerous
modeling and parameter tuning constraints due to limited experimental data, particu-
larly dynamic characteristics at different loading rates. This study introduces a simpli-
fied polymer draft gear model with improved incorporation of loading rate depen-
dency. Polynomial fitting has been replaced with exponential fitting in both static and
dynamic force components, to accurately describe the rise or decay in force compo-
nents which naturally follow an exponential trend with deflection. The proposed
fitting reduced the requirement of a number of tuning parameters and provided a
consistent trend even with an insufficient amount of measured data points. Thus the
proposed method graciously deals with the limitation of the academic researchers,
the unavailability of in-house testing facilities, and the requirement of adequate test
data in raw form from the manufacturers. Also, one of the dynamic force component
parameters has been identified as a controlling parameter for loading rate dependency.
By constraining this parameter, dissimilar effects of loading rate in dynamic loading
and unloading responses can be incorporated to simulate realistic polymer behavior.
This parameter is also attributed to the establishment of quasi-static hysteresis which
is usually overlooked. This model requires less number of tuning parameters and
less number of experimental characteristics, while closely describing the realistic

O. P. Yadav (B) · N. S. Vyas

Department of Mechanical Engineering, IIT Kanpur, Kanpur, Uttar Pradesh, India
e-mail: [email protected]
N. S. Vyas
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 423
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_28
424 O. P. Yadav and N. S. Vyas

characteristics of polymer draft gear. A good match of simulated dynamic character-

istics was found with drop-hammer impact test data. The model was then utilized in
simulating the shunting dynamics of rail vehicles to estimate safe shunting velocity
to avoid structural damage of vehicles.

Keywords Center buffer coupler · Balanced draft gear · Polymer hysteresis ·

Nonlinear damping · Longitudinal train dynamics

1 Introduction

Longitudinal train dynamics (LTD) is an important aspect of railway engineering.

LTD studies involve the estimation of relative motions and forces between connected
vehicles along the track (longitudinal direction). Transmission of these in-train longi-
tudinal forces between rail vehicles has a great influence on passenger comfort and
structural safety of vehicles [1]. Since a number of rail vehicles are inter-connected,
the dynamics of a train as a whole is critically dependent on the coupling mechanism
in addition to vehicle and track parameters [2]. The coupling mechanism between
rail vehicles involves a coupler and a draft gear to transmit and damp out the in-train
forces, respectively. Figure 1a shows a typical coupling mechanism with a balanced
type draft gear, and Fig. 1b shows the schematic of elements involved in transmission
of longitudinal forces.

Fig. 1 a A typical coupling mechanism and b schematic diagram of longitudinal force transmission
A Dynamic Model for Polymer Draft Gears 425

Indian railway widely uses Center Buffer Coupling (CBC) in its mainline
passenger and freight trains. AAR H type couplers with polymer-based balanced
draft gear are used in passenger coaches. Despite the high-capacity performance
of AAR couplers and balanced draft gears, longitudinal jerks have always been
major issues with such coupling mechanisms. Based on scientific contributions and
recommendations of Research Design and Standards Organisation (RDSO), India,
jerk issues have been minimized to a certain extent [3]. An in-depth understanding
of LTD of these rail vehicles can help further minimize the jerks and recommend
design enhancements to achieve better performance. Therefore, modeling accurate
coupling mechanisms for such LTD simulations is vital.
While a coupler can be readily modeled using its body stiffness and a dead-zone
(to represent coupler slack), modeling of draft gear involves proper representation of
nonlinear stiffening and damping action. Based on the major damping mechanism,
draft gears can be broadly categorized into (a) friction and (b) polymer draft gears. In
friction draft gears, damping action is mainly achieved by energy dissipation through
frictional action between wedges and stiffening can be achieved by steel springs or
rubber springs. On the other hand, in polymer draft gears both the spring and damping
actions are provided by the polymer [4]. A polymer-based balanced draft gear is used
in coaches of Indian railway in which two separate sets of polymer rings are used to
attenuate buff and draft forces.
Since accurate analytical dynamic models of polymer materials are not prop-
erly established, modeling of dynamic characteristics of draft gears is solely based
on experimental data. Moreover, the availability of limited test data gives rise to
numerous difficulties in model development and parameter tuning [1]. The most
common approach to model polymer draft gear characteristics is fitting experimental
data with higher order polynomials. Such an approach was earlier developed by [5],
which involves fitting of sixth-order polynomial on static characteristics of draft
gear. This characteristic is then scaled to fit loading and unloading characteristics
through a power function incorporating deflection and velocity variables. Further,
based on the polynomial fitting method a new approach was developed by Cole et al.
[6], in which the loading and unloading curves are separately modeled using sixth-
order polynomials. In this approach, static characteristics are modeled by a mean
or weighted mean curve between loading and unloading states. Instead of directly
scaling the static characteristics, they separately modeled the dynamic force compo-
nent and added it to the static force component to achieve total force. The dynamic
force component consists of a combination of (a) deflection and velocity-dependent
power function and (b) velocity-dependent exponential function.
Another common approach to modeling polymer draft gear characteristics is the
lookup table method, as developed by [7–9]. Despite its advantage of interpolation or
extrapolation of dynamic force characteristics using standard algorithms, the discon-
tinuity between loading and unloading during velocity reversal may cause integration
issues. To solve this discontinuity issue during the transition, various approximation
methods such as velocity-dependent, intermediate slope, and smoothing approxima-
tion approaches are required [10]. A few other approaches for modeling dynamic
426 O. P. Yadav and N. S. Vyas

characteristics of rubber or polymer materials include modeling it as a combination

of elastic, friction, and viscous elements [11, 12].
This paper presents a non-polynomial fitting approach to model polymer draft
gears. The proposed model requires a lesser number of parameters with the capability
of simulating more realistic characteristics of draft gear with improved velocity
dependence. The viability of the proposed model has been proved by comparing it
with experimental data. Finally, shunting simulations have been performed to show
its LTD application.

2 Brief Review of Existing Models

Various mathematical models are available to describe the dynamic characteristics

of polymer components. The models based on the assumption of elastomeric char-
acteristics as a combination of elastic, friction, and viscous elements [11, 12] are not
much suitable for modeling draft gear behavior. In this section, we therefore limit
ourselves to the lookup table [7–9] and polynomial fitting methods [5, 6].

2.1 Nonlinear Hysteresis Characteristics

Figure 2 illustrates various characteristics of polymer draft gears which exhibit

nonlinear hysteresis behavior. In the figure, larger hysteresis envelopes correspond
to dynamic condition and smaller ones correspond to quasi-static conditions. While
modeling dynamic hysteresis characteristics, often a hypothetical static curve is
referred to as developing both dynamic loading and unloading curves, and the
hysteresis at quasi-static velocities is not properly modeled. As seen in the figure
(Fig. 2), even in quasi-static conditions, the current approaches [6] model loading
and unloading curves that are roughly equidistant. Nevertheless, in practice [13], the
unloading curves follow trajectories that are quite close to one another across the
whole range of loading rates (quasi-static to dynamic), and the difference between
the two curves is noticeably smaller than the difference between the loading curves.
Dynamic hysteresis envelopes are usually symmetric for friction draft gears and
single-pack polymer draft gears, while they are asymmetric for twin-pack and
balanced-type draft gears [14]. In the case of friction draft gears, the hysteresis
model can be completely defined through its well-defined physical attributes (such
as friction coefficient and wedge angles) without the requirement of experimental
data [15]. However, due to the lack of understanding of the accurate damping mecha-
nism of polymers, their hysteresis models are developed using experimental data [1].
As discussed in the introduction section, mathematical modeling of the hysteresis
behavior of polymer draft gears often requires separate modeling of important curves:
(a) quasi-static loading and unloading curves, (b) dynamic loading and unloading
A Dynamic Model for Polymer Draft Gears 427

Fig. 2 Ideal draft gear hysteresis characteristics

curves, and (c) transition curves (in lookup table approaches). Static curves are gener-
ated experimentally by applying loads under quasi-static conditions. Dynamic curves
are generated through drop-hammer impact tests or track impact tests. By fitting
mathematical functions on these data, hysteresis characteristics can be modeled.
Hysteresis modeling of polymer draft gears generally includes lookup tables and
polynomial fitting methods [5–9].

2.2 Lookup Table Methods

In lookup table methods, experimental force–deflection data of static curves and

dynamic curves is fed into fixed tables. Draft gear force at a certain deflection and
velocity state is estimated through standard algorithms by interpolating or extrap-
olating data from these tables. For transition continuity, methods such as velocity-
dependent, intermediate slope, and smoothing approximation are required to connect
loading and unloading paths. A general lookup table model with velocity-dependent
transition approach [7] can be presented as Eq. (1)
( )
f t +( FD' − )f t |∆v| ≥ ve
FD (x, v) =
f t + FD' − f t ∆v ve
|∆v| < ve
(1)
f l (x) ∆v∆x ≥ 0
f t (x) =
f u (x) ∆v∆x < 0

where F D and F' D are current and previous time step draft gear forces, respectively;
f t is the force estimated from lookup tables; f l and f u are loading and unloading
functions, respectively; and ∆x and ∆v are changes in deflection and velocity,
respectively, at each time step.
428 O. P. Yadav and N. S. Vyas

2.3 Polynomial Fitting Methods

Polynomial fitting is the preferred method in which experimental force–deflection

data of static curves are usually fitted with higher order polynomials. A polynomial
function can be described as Eq. (2)

∑
j
f (x) = an x n (2)
n=1

where x is the draft gear deflection, j is the degree of polynomial, and coefficients
an are determined from data fitting. As discussed in the introduction section, two
different approaches of polynomial fitting to model dynamic force can be used. The
approach of Belousov [5] requires fitting a polynomial on a static curve, thus static
force component F s can be defined as Eq. (3)

∑
j
Fs (x) = an x n (3)
n=1

The static curve is then directly scaled to obtain total dynamic force during loading
or unloading state, using Eq. (4) as
[ ]
F(x, v) = bF p + Fs (x) 1 + c(x − x0 )d(|v|/vmax ) (4)

where F is the total force under dynamic condition, F p is the preload, F s is the static
force obtained from Eq. (3), x 0 is the preload deflection, and b, c, d are various
coefficients which are determined from fitting Eq. (4) on experimental dynamic
curves.
The approach of Cole et al. [6] gives more flexible tuning ability as the total
dynamic force is not obtained by direct scaling of static force component; instead
total dynamic force is considered to be the sum of static and dynamic compo-
nents. Static and dynamic force components are computed separately. This approach
requires polynomial fitting on dynamic loading and unloading curves, and the mean
or weighted mean of these curves is assumed to describe the static characteristic.
Thus, the resultant static force component can be written as Eq. (5)

∑
j
∑
j
Fs (x) = g1 an x + g2
n
bn x n (5)
n=1 n=1

where g1 + g2 = 1.
Dynamic force component is modeled separately, as a function of both velocity
and deflection, defined as Eq. (6)
A Dynamic Model for Polymer Draft Gears 429
[ | | | | ][ ( | | )]
| x | p | v |q | v |r
Fd (x, v) = C1 + C2 || | | | 1 − exp −λ | | (6)
xmax | | vmax | |v |
max

where F d is the dynamic force component and C 1 , C 2 , p, q, r, λ are various coefficients

which are determined from fitting Eq. (6) to match with experimental dynamic curves,
which are the summation of F s and F d given as [Eq. (7)]

F(x, v) = Fs (x) + Fd (x, v) (7)

3 Model Development and Validation

Model development in this study requires measured dynamic characteristics only. The
static curve has been approximated by the mean of loading and unloading dynamic
curves, as suggested by Cole et al. [6]. The reason for using Cole et al.’s suggestion is
to overcome experimental limitations. Static characteristics are obtained separately
on a static test rig, while dynamic characteristics are obtained through a drop-hammer
impact test. These characteristics do not usually fit together as presented in an ideal
situation in Fig. 2. Dynamic force components are modeled as deflection and velocity-
dependent functions. Total draft gear force under dynamic conditions is taken as the
sum of static and dynamic components.
While most of the models ignore the accuracy of the quasi-static unloading
component, it has a significant contribution in polymer draft gears. Quasi-static
hysteresis exists even in the glassy state of polymers, at low temperatures. Polymer
chains have inhibited movement, which store elastic energy and also stiffen the
polymer. While unloading, due to relaxation delay the reaction from the polymer
drops drastically, even at quasi-static rates [16]. Evidence of static hysteresis of poly-
mers used in energy-absorbing components can be found in polymer manufacturers’
documentation [13] as well as in experimental studies [17].
In this study, we approximate the quasi-static characteristics through a quasi-static
unloading curve in conjunction with a quasi-static loading curve to form the hysteresis
envelope. Experimental observations show that the effect of the loading rate in
dynamic unloading curves is not the same as in loading curves. This phenomenon also
has been incorporated into the proposed model. The details of the applied approach
can be referred to in Sect. 3.4.
As discussed previously, higher order polynomial fitting is a widely used function
to fit static as well as dynamic characteristics. We propose exponential functions of
deflection in both static and dynamic force component models, arguing that the rise
or decay in force components follows an exponential trend with deflection. Polymer
chains undergo inhibited movement, and the rate of accumulation of polymer chains
during compression is exponentially proportional to the amount of compression (or
deflection) [18]. Exponential fit has additional advantages such as a reduced number
of parameters and a consistent trend. Despite the advantage of close-fitting higher
430 O. P. Yadav and N. S. Vyas

Fig. 3 Typical measured dynamic characteristics of balanced draft gear

order polynomial functions in general, the polynomial fitting technique suffers from
over- or under-estimation, when a number of data points is not sufficient [19]. This is
a general limitation with most of the academic researchers, due to the unavailability
of in-house testing facilities and proper test data in raw form from the manufacturers.
A typical measured characteristic of balanced type polymer draft gear, used in this
study, provided by M/s Frontier Alloys Steels Ltd., Kanpur, is shown in Fig. 3.

3.1 Static Force Component

To model static components, experimental data points of dynamic curves were

directly averaged to approximate static loading characteristics. Then, a two-term
exponential function was fitted as given below in Eq. (8)

Fs (x) = F0 + F1 exp(γ1 x) + F2 exp(γ2 x) (8)

where F 0 , F 1 , F 2 , γ1 , and γ2 are the tuning parameters. This function requires only
five parameters in contrast to the sixth-order polynomial fit which requires seven
parameters. Exponential fitting on approximated static characteristics shows a close
agreement as seen in Fig. 4. Corresponding parameters are listed in Table 1.

3.2 Dynamic Force Component

The model for the dynamic force component is based on the assumption that nonlinear
damping is exponentially proportional to deflection as internal friction (i.e., damping)
A Dynamic Model for Polymer Draft Gears 431

Fig. 4 Static force components

Table 1 Fitting parameters

Parameter Buff Draft
for static force components
F0 −3.056e + 05 −2.699e + 08
F1 3.056e + 05 2.699e + 08
F2 2.693e-07 −0.04206
γ1 16.12 0.03279
γ2 419.5 −340.5

is also dependent on the rate of inhibition of polymer chains. We also assume that
damping force is a nonlinear function of velocity. Thus, a final form of the proposed
dynamic force component is defined as Eq. (9)
[ ]
Fd (x, v) = F3 1 − exp(γ3 |x|) |v|α (9)

where F 3 , γ3 , and α are the tuning parameters. Figure 5 shows the fitting of Eq. (9)
using a time-varying harmonic deflection curve with an amplitude of 0.068 m on buff
mode and an amplitude of 0.048 m on draft mode force components at a frequency
of 10 Hz. Corresponding parameters are listed in Table 2. A reasonable fit can be
observed.

3.3 Total Force

Total draft gear force, F, is the sum of static and dynamic force components, given
as Eq. (10)

F(x, v) = Fs (x) + Fd (x, v) (10)

432 O. P. Yadav and N. S. Vyas

Fig. 5 Dynamic force components at 10 Hz

Table 2 Fitting parameters for dynamic force components

Parameter Buff Draft
Loading Unloading Loading Unloading
F3 52e3 −32e3 −68e3 50e3
γ3 21 38 21 35
α 0.8 0.2 0.8 0.15

All required parameters for simulating total force at 10 Hz frequency have been
presented combined in Table 3. The corresponding dynamic response of Eq. (10)
has been compared with measured drop-hammer test data, in Fig. 6. A reasonable fit
with the goodness of consistency in the trend because of exponential function can
be observed.

Table 3 Fitting parameters for total dynamic force

Parameter Buff Draft
Loading Unloading Loading Unloading
Static parameters F0 −3.056e + 05 −2.699e + 08
F1 3.056e + 05 2.699e + 08
F2 2.693e-07 −0.04206
γ1 16.12 0.03279
γ2 419.5 −340.5
Dynamic parameters F3 52e3 −32e3 −68e3 50e3
γ3 21 38 21 35
α 0.8 0.2 0.8 0.15
A Dynamic Model for Polymer Draft Gears 433

Fig. 6 Total force at 10 Hz

3.4 Controlling Unloading Response

While there can be various combinations of parameters which can give closely
similar curves, to incorporate realistic differences in the variation in loading and
unloading responses at various loading rates, parameter α was restricted in some
specific range, while tuning other parameters. In order to keep smaller variation in
unloading response with variation in frequency, compared to loading response and
also to include the quasi-static hysteresis effect, the parameter α has been selected
such that the term |v|α has low influence of frequency during unloading. The value
of α was restricted between 0.1 and 0.2 (values closer to zero) for unloading curves
and between 0.6 and 1.2 (values closer to one) for loading curves while tuning other
parameters. Figure 7 shows the response at various frequencies ranging from 0.1 Hz
(quasi-static) to 10.0 Hz. It can be observed that the variation in loading responses at
10.0 and 1.0 Hz is considerably higher than that in unloading responses (highlighted
by marking arrows 1 and 2). Additionally, the unloading curve under quasi-static
loading (0.1 Hz) can be seen to be significantly different than the quasi-static loading
curve, thus creating a realistic quasi-static hysteresis envelope.

4 Shunting Impact Simulation

Shunting of rail cars is an important operation, where high magnitude impacts

between cars occur at a low loading rate. During the course, if the transmitted
longitudinal force reaches beyond the strength of the car body, it will get damaged.
Therefore, knowing the maximum allowable shunting speed becomes important for
the structural safety of rail vehicles. In this section, the impact performance of draft
434 O. P. Yadav and N. S. Vyas

Fig. 7 Effect of loading rate on model response

gears is studied in a one-to-one vehicle impact scenario. The mass of both vehicles
is 100-tonne. One vehicle with an initial velocity impacts with another stationary
vehicle on a tangent track. Under this condition, traction forces, braking forces, and
track resistances such as grade resistance are not present [19]. Resistances from
track irregularity, suspension elements, and negligible air drag are considered which
are combined with friction for simplicity. An effective friction coefficient of 0.2 is
considered. The test setup is shown in Fig. 8.
The moment the moving vehicle impacts stationary vehicle, the coupling mech-
anism gets locked and both vehicles start moving, in a coupled state. The equations
of motion of the system can be derived as Eq. (11)

m 1 ẍ1 = −2Fdg (∆x, ∆v) − μm 1 g.sgn(v1 )

(11)
m 2 ẍ2 = 2Fdg (∆x, ∆v) − μm 2 g.sgn(v2 )

where ∆x and ∆v are the relative deflections and velocities, defined as

Fig. 8 One-to-one shunting impact model

A Dynamic Model for Polymer Draft Gears 435

x1 − x2 v1 − v2
∆x = , ∆v =
2 2
Shunting impact simulations are performed at different initial velocities of the
impacting vehicle. The comparison of the impact performance of draft gear at various
impact speeds is shown in Fig. 9. The hysteresis envelope of the drop-hammer test
has been overlaid for reference in observing the results. Both force and deflection
of draft gear increase with an increase in impact speed as the requirement of energy
absorption has increased. Also, the loading and unloading curves remain inside the
drop-hammer impact test envelope because the loading rate is low. Figure 9a shows
that, at a low impact speed of 4 km/hr, only buff mode (half cycle) is sufficient to
dissipate the whole kinetic energy. Figure 9b–d shows full oscillation cycles with an
increasing number of oscillations, with an increase in impact speed. Peak buff force
also increases and reaches up to 1500 kN at 7 km/hr, while at 8 km/hr, it goes beyond
2000 kN. It can thus be stated that in this case, the safe shunting speed is 7 km/hr.

Fig. 9 Draft gear impact performance at various velocities

436 O. P. Yadav and N. S. Vyas

5 Conclusions

The longitudinal dynamics of rail vehicles is influenced by the coupling mecha-

nism. This study presents a draft gear model which can be easily developed with
measured data and can closely simulate the behavior of polymer draft gears. The
proposed model requires a lesser number of parameters than the existing ones,
and thus, its tuning is comparatively easy. This model also improves the incorpo-
ration of unloading characteristics and quasi-static hysteresis. Simulated dynamic
characteristic shows a good match with experimental drop-hammer characteristics.
The draft gear model developed is utilized to simulate its impact characteristics for
the one-to-one shunting operation of two railway vehicles. It is observed that track
impact characteristics of draft gears lie within drop-hammer impact characteristics
for low-impact speeds and peak forces are lower. Beyond a specific impact speed, the
peak forces at higher impact speed are considerably higher than those characterized
through the drop-hammer test.

Acknowledgements The authors are thankful to the staff from M/s Frontier Alloy Steels Ltd.,
Kanpur, for providing experimental data and sharing their experience and knowledge.

References

1. Cole C, Spiryagin M, Wu Q, Bosomworth C. Modelling issues in passenger draft gear connec-

tions. In: Proceedings of the 24th symposium of the international association for vehicle system
dynamics. 2015.
2. Iwnicki S, Spiryagin M, Cole C, McSweeney T, editors. Handbook of railway vehicle dynamics.
Boca Raton: CRC Press;2010.
3. Swaroop A. A technical note in continuation of reasoned document of EOI of AAR ‘H’ type
tight lock couplers having balanced type draft gear, RDSO, Editor. Lucknow India;2011.
4. Qing W, Luo S, Tianwei Q, Yang X. Comparisons of draft gear damping mechanisms. Veh
Syst Dyn. 2017;55(4):501–16. https://doi.org/10.1080/00423114.2016.1252049.
5. Belousov AG. Development of design and numerical modelling of friction-polymer draft gears,
PhD thesis, Bryansk State University, Bryansk, Russia;2006.
6. Cole C, Spiryagin M, Wu Q, Bosomworth C. Practical modelling and simulation of polymer
draft gear connections. In: ICRT 2017: Railway development, operations, and maintenance.
Reston, VA: American Society of Civil Engineers; 2018. p. 413–23.
7. Wu Q, Luo S, Xu Z, Ma W. Coupler jackknifing and derailments of locomotives on tangent
track. Veh Syst Dyn. 2013;51(11):1784–800. https://doi.org/10.1080/00423114.2013.830184.
8. Xu Z, Ma W, Wu Q, Luo S. Coupler rotation behaviour and its effect on heavy haul trains. Veh
Syst Dyn. 2013;51(12):1818–38. https://doi.org/10.1080/00423114.2013.834369.
9. Qing W, Yang X, Cole C, Luo S. Modelling polymer draft gears. Veh Syst Dyn.
2016;54(9):1208–25. https://doi.org/10.1080/00423114.2016.1196822.
10. Zhou Y, Shi Z, Hecht M. Comparison of friction draft gear models for simulation in longitudinal
train dynamics. Veh Syst Dyn. 2021. https://doi.org/10.1080/00423114.2021.1901943.
11. Berg M. A model for rubber springs in the dynamic analysis of rail vehicles. Proc Instit Mech
Eng Pt F J Rail Rapid Transit. 1997;211(2):95–108. https://doi.org/10.1243/095440997153
0941.
A Dynamic Model for Polymer Draft Gears 437

12. Dumont E, Maurer W. Modelling elastomer buffers with DyMoRail. In: Proceedings of the
10th International modelica conference. Lund: Sweden;2004. https://doi.org/10.3384/ecp140
96923.
13. Buffer Springs. https://www.minerent.com/products/tp_Buffer_Springs-catA.php. Accessed
15 Mar 2021.
14. Wagner S, Cole C, Spiryagin M. A review on design and testing methodologies of modern
freight train draft gear system. Rail Eng Sci. 2021;29:127–51. https://doi.org/10.1007/s40534-
021-00237-y.
15. Qing W, Spiryagin M, Cole C. Advanced dynamic modelling for friction draft gears. Veh Syst
Dyn. 2015;53(4):475–92. https://doi.org/10.1080/00423114.2014.1002504.
16. Chen H, Trivedi A, Siviour C. Application of linear viscoelastic continuum damage theory to the
low and high strain rate response of thermoplastic polyurethane. Exp Mech. 2020;60:925–36.
https://doi.org/10.1007/s11340-020-00608-2.
17. Mickoski H, Mickoski I, Djidrov M, Zdraveski F. Mathematical model of new type of train
buffer made of polymer absorber—determination of dynamic impact curve for different
temperatures. Machines. 2018;6(4):47.
18. Kothari K, Hu Y, Gupta S, Elbanna A. Mechanical response of two-dimensional polymer
networks: role of topology, rate dependence, and damage accumulation. J Appl Mech.
2018;85(3).
19. Andrew PK, Robert JE. Statistics for biomedical engineers and scientists, Chapter 2. Academic
Press;2019. p 23–56. https://doi.org/10.1016/B978-0-08-102939-8.00011-6.
Modelling and Dynamic Analysis
of Tapered Viscoelastic Laminated
Composite Beam

Krishanu Ganguly and Haraprasad Roy

Abstract Modern engineering eventually suggests the substitution of heavy beams

made of conventional materials with light but stronger beams made of advanced mate-
rials such as composites because of their efficient properties such as high strength to
weight proportion, quality and better execution. Further, the potency of composite
structures may be optimized by altering their stacking series and orientation of fibres.
Hence, the study of a stacked composite beam subjected to varying loading condi-
tions would find several realistic applications and is of greater interest in engineering
design sectors. In this study, a viscoelastic tapered composite beam with rectangular
cross-section is modelled using Equivalent Single Layer Theory (ESLT). Further, a
finite element formulation considering Euler–Bernoulli beam theory is developed
to carry out the dynamic analysis of the composite beam for different stacking
sequences. The viscoelastic properties of carbon-epoxy composite material deduced
from the storage modulus and loss factor in the frequency domain are incorporated
using the operator-based approach in the time domain leading to a higher-order
beam model. The ESLT is made to use the viscoelastic material property of a single
lamina of the composite and further deduce the equivalent material property for the
whole tapered finite element composite beam laminate. Numerical results in terms
of frequency responses are presented to compare the influence of various stacking
sequences over varying taper conditions of the beam.

Keywords Laminated composite beam · Tapered cross-section · Equivalent single

layer theory · Viscoelasticity · Operator based approach · Dynamic analysis

K. Ganguly (B)
Department of Mechanical Engineering, IIT Delhi, New Delhi, India
e-mail: [email protected]
H. Roy
Department of Mechanical Engineering, NIT Rourkela, Orissa, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 439
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_29
440 K. Ganguly and H. Roy

1 Introduction

Beams are the common structural components which are usually classified based on
their geometric configuration, i.e. uniform or tapered, and slender or thick. On the
basis of their several engineering applications, a large number of studies can be found
in literature initially considering uniform isotropic beams using both the classical
and Finite Element (FE) approach. Nickel and Secor [1] used variation approach to
model the Timoshenko beam element with different degrees of freedom (Lateral and
rotational). Thomas and Abbas [2] presented an analysis over the most basic finite
element model of a Timoshenko beam. Dawe [3] presented a similar analysis over
three noded Timoshenko beam of uniform cross-section.
At the same time, the research work shifted to the modelling of simple beam-
based structures made of anisotropic materials such as composites due to its high
strength/stiffness to weight ratio. Abarcar and Cunniff [4] performed experiments
over uniform cross-sectioned graphite-epoxy and Boron-Epoxy composites with
varying fibre orientations and compared with the numerical results for validation.
Chen and Yang [5] developed a microcomputer program for analysing a symmet-
rically laminated beam, discretized into two noded elements using finite element
approach with each node having 6 degrees of freedom. Abramovich [6] performed
free vibration analysis of a symmetrically stacked composite beam considering rotary
inertia and shear deformation effect. Nei and Cai [7] applied equilibrium and curva-
ture compatibility relations to include the shear slip effect on the deformation analysis
of steel–concrete composite beams. Ke et al. [8] performed non-linear vibrational
analysis of carbon nano-tube reinforced functionally graded nano-composite beams.
Hajianmaleki and Qatu [9] presented a thorough review on the vibrational anal-
ysis performed over composite beams modelled using different theories in the last
2 decades. Pagani and Carrera [10] explored the capabilities of applying Carrera
Unified Formulation to deal with the large deflections and post-buckling results of
the composite laminated beam.
Soon it was realized that non-uniform beams may provide a better or more suit-
able distribution of mass and strength than uniform beams and therefore can meet
special functional requirements in architecture, aeronautics, robotics and other inno-
vative engineering applications. Considering this, relatively few researchers have
carried out studies over non-uniform and tapered composite beams. Farghaly and
Gadelrab [11] performed free vibration analysis of stepped Timoshenko composite
beams to obtain the natural frequencies. Rao and Ganesan [12] applied finite element-
based higher-order shear deformation theory to examine the harmonic response of
tapered composite beams. Further, several authors carried out free vibration analysis
over linearly tapered beam [13–15]. Ganesan and Zabihollah [16] performed free
undamped vibrational analysis over tapered composite beam through higher order
finite element formulation in order to overcome geometric and material disconti-
nuity. Gayen and Roy [17] applied conventional laminated theory to model a tapered
symmetrically and asymmetrically laminated composite beam and perform stress
distribution analysis under hygro-thermal loading conditions. Nijgh et al. [18] carried
Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 441

out experimental analysis on demountability and reusability of tapered composite

beams, i.e. tapered steel beam prefabricated into concrete decks.
The current study presents FE model of a viscoelastic tapered viscoelastic
composite beam formulated using Equivalent Single Layer Theory (ESLT). The
tapered parameters are included in the formulation and the stiffness and mass matrix
of each discretized element are derived on the basis of the Euler–Bernoulli beam
theory. An operator-based constitutive relationship is employed to incorporate the
material damping and higher order FE equation is developed. The derived higher
order equation of motion is used to perform analysis over the composite beam which
includes the material damping parameters as well as varying stacking sequences.
The numerical results represent a parametric evaluation of the frequency response
analysis of the cantilever composite beam based on the changing tapered conditions
over varying stacking sequence of the composite beam.

2 Mathematical Modelling: Viscoelastic Tapered

Composite Beam

This section deals with the derivation of the FE equation of motion of a tapered
viscoelastic composite beam as shown in Fig. 1 applying Euler–Bernoulli beam
theory. The beam is configured to taper along its breadth (b) throughout the length,
while the height of the beam (h) remains constant. This results in constant height of
each lamina of the composite throughout the length. Each node of a 2-noded beam
element having 2 degrees of freedom (translation and rotation) per node and stiffness
and mass matrix of each tapered beam are derived accordingly.
The undamped FE equation of motion of a uniform cross-sectional beam can be
written as

[M]{q̈} + [K]{q} = {F} (1)

Fig. 1 Front and top view of the breadthwise tapered composite beam
442 K. Ganguly and H. Roy

where {F} and {q} are the force vector and nodal coordinate respectively. [K] and
[M] are the stiffness and mass matrix. Following ESLT from [19], the expression of
stiffness and mass matrix of the beam element with n layers are respectively given
as
l
∑
n
''
[K] = E i Ii N (x)'' N (x)T d x (2)
i=1 0

l
∑
n
[M] = ρi Ai N (x)N (x)T d x (3)
i=1 0

Here, E i and ρi are the elastic modulus and density of ith layer respectively. l is
the element length and N (x) is the Hermite Shape function. Considering rectangular
cross section, the ith layer moment of inertia and cross-sectional area are written as

bh i3
Ii = + Ai di2 (4)
12

Ai = bh i (5)

In the preceding equation, di is the distance of ith layer neutral axis from the
reference axis. For first layer say i = 1, d1 = 0, as the reference axis passes through
its centroid, while for i ≥ 2, di is given as
n (
∑ )
h i−1 + h i
di = (6)
i=2
2

For a tapered beam with varying breadth along the length, it is understood that the
moment of inertia and cross-sectional area of the beam are the function of x and
varies with breadth (b). Hence for tapered beam, those are written as

b(x)h i3
Ii (x) = + Ai (x)di2 (7)
12

Ai (x) = b(x)h i (8)

Simultaneously, the stiffness and the mass matrix of the tapered composite beam
can be written as
l
∑
n
''
[K] = Ei N (x)'' N (x)T Ii (x)d x (9)
i=1 0
Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 443

l
∑
n
[M] = ρi N (x)N (x)T Ai (x)d x (10)
i=1 0

The stiffness matrix for the tapered composite beam is further re-written as

∑
n l ( )
'' b(x)h i3
[K] = Ei N (x)'' N (x)T + b(x)h i di2 d x (11)
i=1
12
0

After substituting di , the above equation is simplified as

∑
n
[ ]
[K] = Ei I i K (12)
i=1

where
l
[ ] ''
K = N (x)'' N (x)T b(x)d x
0

n (
∑ )2
h i3 h i3 h i−1 +h i
Ii = 12
for i = 1 and I i = 12
+ hi 2
for i ≥ 2.
i=2
Similarly, the mass matrix of the beam element of the whole composite laminate
can be directly derived as a function of breadth from Eqs. (7) and (8),

l
∑
n
[M] = ρi h i N (x)N (x)T b(x)d x (13)
i=1 0

These elemental stiffness and mass matrices are derived for each finite element
of the beam as they vary in geometry due to the tapered condition. Ultimately, the
global stiffness and mass matrix are formed following normal finite element model
formulation.
Hence on substitution, the equation of motion of the tapered beam is given as

∑
n
[ ]
[M]{q̈}+ E i Ī i K̄ {q}={F} (14)
i=1

In the above relation, the elastic modulus of the relatively long composite beam
structure is taken considering normal strain along x-direction. Following Ganguly
et al. [20], its expression is written as
444 K. Ganguly and H. Roy
( ) ( )
1 1
E = E1 cos4 α + E2 sin4 α
1 − υ12 υ21 1 − υ12 υ21
( )
2υ12 E2
+ cos2 α sin2 α + 4G 12 (15)
1 − υ12 υ21

On modelling the beam with viscoelastic material using operator-based approach,

the material modulus in the above equation of motion is represented as
( ) ( )
1 1
E() = E1 () cos (α) 4
+ E2 () sin (α)
4
1 − υ12 υ21 1 − υ12 υ21
( )
2υ E
12 2 ()
+ cos2 (α) sin2 (α) + 4G 12 () (16)
1 − υ12 υ21

These directional modulus operators E1 (), E2 () and G 12 () are derived after

carrying out frequency-dependent storage and loss modulus Dynamic Mechan-
ical Analyser (DMA) experiments over samples of composite for and further
matching them with proper spring-dashpot rheological models to extract the material
viscoelastic parameters [20].
For a material which match a 4-element viscoelastic model, the relations for
directional modulus operators are given as

a0l +a1l D + a2l D2 a0t +a1t D + a2t D2 a0s +a1s D + a2s D2

E1 () = b0l +b1l D
, E2 () = b0t +b1t D
, G 12 () = b0s +b1s D
(17)

where the viscoelastic parameters a0 ,a1 ,a2 ,b0 and b1 can be derived ensuing [21]
and the property along transverse, longitudinal and shear direction are denoted by
superscripts l, t and s respectively.
On substitution of the above relations and solving, the equation of motion of the
viscoelastic composite tapered beam can be written as
( )
m 0 + m 1 D + m 2 D2 + m 3 D3 + m 4 D4 [ ]
[M]{q̈}+ ( ) K {q}={F} (18)
n 0 + n 1 D + n 2 D2 +n 3 D3

where

D = d/dt
⎧ ( ) ( )( )⎫
∑n ⎪⎨ cos4 αi I i b0t bs0 al0 + sin4 αi I i + 8ν12 cos2 αi sin2 αi I i a0t b0l b0s ⎪ ⎬
1
m0 = ( )
4(1 − ν12 ν21 ) ⎪ ⎪
i=1 ⎩ +16(1 − ν12 ν21 ) cos αi sin αi I i a0 b0 b0 ⎭
2 2 s l t
⎧ ( ) ⎫
⎪
⎪ cos4 αi I b1t bs0 al0 + b0t bs0 al1 + b0t bs1 al0 ⎪
⎪
n ⎪
⎪ ⎪
1 ∑ ⎨ ( )( )⎪
⎬
m1 = 4 2 2 t l s t l s
+ sin αi I i + 8ν12 cos αi sin αi I i a0 b1 b0 + a1 b0 b0 + a0 b0 b1 t l s
4(1 − ν12 ν21 ) ⎪ ⎪
i=1 ⎪
⎪
⎪ ( ) ⎪
⎪
⎪
⎩ +16(1 − ν12 ν21 ) cos2 αi sin2 αi I i a0s b0l b1t + a0s b0l b0t + a1s b0t b0l ⎭
Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 445
⎧ 4 ( ) ⎫
⎪
⎪ cos αi I a0t as0 al2 + a0t as1 al1 + a1t bs0 al1 + a1t bs1 al0 ⎪
⎪
⎪
⎪ ⎪
1 ∑n ⎨ ( )( )⎪
⎬
m2 = 4 2 2 t l s t l s t l
+ sin αi I i + 8ν12 cos αi sin αi I i a0 b1 b1 + a1 b0 b1 + a1 b1 b0 + a2 b0 b0 s t l s
4(1 − ν12 ν21 ) ⎪
⎪ ⎪
⎪
⎪
i=1 ⎪ ( ) ⎪
⎪
⎩ ⎭
+16(1 − ν12 ν21 ) cos2 αi sin2 αi I i a0s b1l b1t + a1s b0l b1t + a1s b0t b1l + a2s b0l b0t
⎧ 4 ( ) ⎫
⎪
⎪ cos αi I i b0t bs1 al2 + b1t bs0 al2 + b0t bs1 al1 ⎪
⎪
n ⎨
⎪
⎪ ( ⎪
⎪
1 ∑ )( )⎬
m3 = + sin4 αi I i + 8ν12 cos2 αi sin2 αi I i a1t b1l b1s + a2t b0l b1s + a2t b1l b0s
4(1 − ν12 ν21 ) ⎪ ⎪
i=1 ⎪
⎪
⎪ ( ) ⎪
⎪
⎪
⎩ ⎭
+16(1 − ν12 ν21 ) cos2 αi sin2 αi I i a1s b1l b1t + a2s b0l b1t + a2s b1l a0t
⎧ ( ) ( )( )⎫
∑n ⎪⎨ cos4 αi I i b1t bs1 al2 + sin4 αi I i + 8ν12 cos2 αi sin2 αi I i a2t b1l b1s ⎪ ⎬
1
m4 = ( )
4(1 − ν12 ν21 ) ⎪ ⎪
i=1 ⎩ +16(1 − ν12 ν21 ) cos αi sin αi I i a b b ⎭
2 2 s l t
2 1 1
n 0 = b0l b0t b0s

n 1 = b0l b0t b1s + b0l b0s b1t + b0s b0t b1l

n 2 = b0l b1t b1s + b0t b1l b1s + b0s b1l b1t

n 3 = b1l b1t b1s

On solving the above equation for the time differential operator ‘D’, the above
equation can be written as

1 [ {....} {...}
[M]{q̈}+ ( ) m 4 q + m 3 q + m 2 {q̈} + m 1 {q̇}
n0 + n1D + n2D + n3D
2 3
][ ]
+m 0 {q} K̄ = {F} (19)

Further simplifying Eq. (14), the equation of motion in its higher-order form is
obtained as
{. . . . .} {....} {...}
[A5 ] q +[A4 ] q +[A3 ] q +[A2 ]{R q}+[A1 ]{q̇}+[A0 ]{q} = [B]{u} (20)

where

[A5 ]=[M]n 3

{ [ ]}
[A4 ] = [M]n 2 + m 4 K

{ [ ]}
[A3 ] = [M]n 1 + m 3 K

{ [ ]}
[A2 ] = [M]n 0 + m 2 K
446 K. Ganguly and H. Roy

{ [ ]}
[A1 ] = m 1 K

[ ]
[A0 ] = m 0 K

( )
[B] = n 0 + n 1 D + n 2 D2 +n 3 D3 [P]

Here, [P] is the position of the external force while {u} is defined as the force vector
originated equivalent to any external disturbance.

3 Results and Discussions

A discretized cantilever beam with rectangular cross-section shown schematically

in Fig. 1 is considered for analysis. The height (h) of the beam consists of 8 uni-
directional carbon-epoxy layers with each layer having equal volume % of carbon
fibre and aspect ratio. The material data of the composite are given in Table 1.
The constant geometric parameters of the beam, i.e., the length L of the beam is
1 m and L/ h is 10. However the breadth of the beam is changing following three
different taper types, i.e. linear, parabolic and exponential given in Fig. 2. For all
cases, the root dimension of breadth is chosen to be same and is equal to br .
The above table represents the taper conditions and respective breadth equations.
The term δ is the taper parameter whose increment results in to increase in taper
angle. For analysis purpose, the value of the taper parameter δ is considered in such
a manner that the tip value of breadth at the free end (bt ) of the beam remains same
for any taper condition.
For analysis purpose, 4 different stacking sequences are considered as given in
Table 2 which comprises the height h of the beam.
Analysis is carried out to determine the best taper with best sequence in terms of
frequency response analysis. The frequency responses are compared among various
taper conditions by varying the taper parameter for all the sequences.
Figures 3, 4, 5 and 6 represent the unbalance response plot for 4 different sequences
considered in Table 2. Each figure compares the response plot among the 3 different
tapered conditions. It is observed that although for initial values of taper parameter,

Table 1 Composite constituent properties [20]

Epoxy Carbon fibre
Density Poisson’s Young’s Volume Density Poisson’s Young’s Volume
ρm (kg/ Ratio modulus fraction ρ f (kg/ Ratio modulus fraction
m3) (GPa) (%) m3) (%)
1100 0.31 3.5 83 1750 0.26 310 17
Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 447

Taper type Taper equation Breadth cross-section of beam

x
Linear br 1
L
(a)

2
x
Parabolic br 1
L
(b)

1
x 2
Exponential br *exp
L
(c)

Fig. 2 a Linear taper. b Parabolic taper. c Exponential taper

Table 2 Stacking sequence

Sequence no Stacking sequence
[21]
1 [ 90, 90, 45, 0]s
2 [90, 45, 45, 0]s
3 [90, 90, 0, 45]s
4 [45, 0, 45, 0, 90, 0, 90, 0]

there is no much difference in the first natural frequency (FNF), but gradual incre-
ments in the value of δ suggests the highest FNF for linear taper followed by parabolic
and then exponential pattern. It is also observed that with increase in taper angle, the
response amplitude also increases for any taper pattern or stacking sequence.
Since each sequence is observed to give maximum FNF value for δ = 0.8, Table 3
is presented to compare the FNF value among all 4 sequences to understand the best
sequence with certain taper pattern. On comparison, Sequence 4 tends to show best
result in terms of FNF with linear taper pattern because of increased stiffness with
highest 0˚ fibre orientation at the outer layers.
448 K. Ganguly and H. Roy

Sequence 1

Fig. 3 Frequency response plot for Sequence 1

Sequence 2

Fig. 4 Frequency response plot for Sequence 2

Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 449

Sequence 3

Fig. 5 Frequency response plot for Sequence 3

4 Conclusions

A FE model of a tapered composite beam is formulated using Equivalent Single

Layer Theory (ESLT). The stiffness and mass matrix of the taper element are formu-
lated using Euler–Bernoulli beam theory. The inclusion of material damping using
operator-based approach leads to FE of higher-order form which is further used
to perform analysis of the composite beam. On comparing different taper patterns,
it can be concluded that the linear taper provides maximum stiffness resulting in
higher natural frequency compared to parabolic and exponential patterns. The natural
frequency is also seen to be increasing with increase in the taper angle; however the
response amplitude also tends to increase.
450 K. Ganguly and H. Roy

Sequence 4

Fig. 6 Frequency response plot for Sequence 4

Table 3 FNF for all

Stacking sequence at δ = 0.8 FNF (Hz)
sequences and patterns at
δ = 0.8 Linear Parabolic Exponential
Sequence 1 981 918 771
Sequence 2 1248 1171 988
Sequence 3 1535 1444 1213
Sequence 4 1837 1724 1451

References

1. Nickel RE, Secor GA. Convergence of consistently derived Timoshenko beam finite elements.
Int J Numer Meth Eng. 1972;5(2):243–52.
2. Thomas J, Abbas BA. Finite element model for dynamic analysis of Timoshenko beam. J Sound
Vib. 1975;41(3):291–9.
3. Dawe DJ. A finite element for the vibration analysis of Timoshenko beams. J Sound Vib.
1978;60(1):11–20.
4. Abarcar RB, Cunniff PF. The vibration of cantilever beams of fiber reinforced material. J
Compos Mater. 1972;6(4):504–17.
5. Chen AT, Yang TY. Static and dynamic formulation of a symmetrically laminated beam finite
element for a microcomputer. J Compos Mater. 1985;19(5):459–75.
Modelling and Dynamic Analysis of Tapered Viscoelastic Laminated … 451

6. Abramovich H. Shear deformation and rotary inertia effects of vibrating composite beams.
Compos Struct. 1992;20(3):165–73.
7. Nie J, Cai CS. Steel–concrete composite beams considering shear slip effects. J Struct Eng.
2003;129(4):495–506.
8. Ke LL, Yang J, Kitipornchai S. Nonlinear free vibration of functionally graded carbon nanotube-
reinforced composite beams. Compos Struct. 2010;92(3):676–83.
9. Hajianmaleki M, Qatu MS. Vibrations of straight and curved composite beams: a review.
Compos Struct. 2013;100:218–32.
10. Pagani A, Carrera E. Large-deflection and post-buckling analyses of laminated composite
beams by Carrera unified formulation. Compos Struct. 2017;170:40–52.
11. Farghaly SH, Gadelrab RM. Free vibration of a stepped composite Timoshenko cantilever
beam. J Sound Vib. 1995;187(5):886–96.
12. Rao SR, Ganesan N. Dynamic response of non-uniform composite beams. J Sound Vib.
1997;200(5):563–77.
13. Tong X, Tabarrok B, Yeh KY. Vibration analysis of Timoshenko beams with non-homogeneity
and varying cross-section. J Sound Vib. 1995;186(5):821–35.
14. Cleghorn WL, Tabarrok B. Finite element formulation of a tapered Timoshenko beam for free
lateral vibration analysis. J Sound Vib. 1992;152(3):461–70.
15. Polyzois D, Raftoyiannis IG, Ibrahim S. Finite elements method for the dynamic analysis of
tapered composite poles. Compos Struct. 1998;43(1):25–34.
16. Ganesan R, Zabihollah A. Vibration analysis of tapered composite beams using a higher-order
finite element. Part I: Formulation. Comp struct. 2007;77(3):306–18.
17. Gayen D, Roy T. Hygro-thermal effects on stress analysis of tapered laminated composite
beam. Int J Compos Mater. 2013;3(3):46–55.
18. Nijgh MP, Gîrbacea IA, Veljkovic M. Elastic behaviour of a tapered steel-concrete composite
beam optimized for reuse. Eng Struct. 2019;183:366–74.
19. Reddy JN. An evaluation of equivalent-single-layer and layerwise theories of composite
laminates. Compos Struct. 1993;25(1–4):21–35.
20. Ganguly K, Raj A, Roy H. Modelling and comparative study of viscoelastic laminated
composite beam–an operator based finite element approach. Mech Time-Dependent Mater.
2021;25:691–710.
21. Dutt JK, Roy H. Viscoelastic modelling of rotor—shaft systems using an operator-based
approach. Proc Inst Mech Eng C J Mech Eng Sci. 2011;225(1):73–87.
Analysis of Limiting Load Capacity
and Stiffness Coefficients
of Porous-Bump-Recess Foil Journal
Bearing

S. Arokya Agustin, C. Shravankumar, A. Arul Jeya Kumar,

and T. V. V. L. N. Rao

Abstract The analytical model of limiting load carrying capacity and limiting
coefficients stiffness for porous-bump-recess configuration on top foil of air foil
journal bearing is presented. Large bearing number is considered for the analysis.
A simplified steady-state compressible Reynold’s equation for porous-bump-recess
foil bearing is analyzed based on porous-bump-recess and plain configuration. It is
assumed that the porous substance of the top foil has uniform permeability. Addi-
tionally, it is presumable that the fluid’s pressure gradient will adhere to Darcy’s law
in the porous-bump-recess arrangement of the material. The following parameters
are included in the model: Bearing number, bump height, porosity ratio, eccentricity
ratio, and bump compliance. The results of foil journal bearing are compared for
porous-bump-recess and plain bearing.

Keywords Foil journal bearing · FGM · Bump foil · Load capacity

Nomenclature

C Radial clearance, m
fα Effective compliance factor
h, H Thickness of fluid film, m; H = Ch
Hb Texture bump height
ki j , K i j Coefficients (Stiffness), Ns/m; K i j = K i j /W , i, j = x, y, K i j =
ki j C/ pa L R
L Length of the bearing, m
pa Ambient pressure, N/m2

S. A. Agustin · C. Shravankumar · A. A. J. Kumar

SRM Institute of Science and Technology, Kattankulathur 603203, India
T. V. V. L. N. Rao (B)
Faculty of Engineering, Assam Down Town University, Guwahati, Assam 781026, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 453
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_30
454 S. A. Agustin et al.

p, P Bearing pressure, N/m2 ; p = ppa

R Radius of the journal, m
x, y, X, Y Coordinates at the center of the bearing, m; X = x/C, Y = y/C
∆X, ∆Y Displacements (nondimensional) in x and y directions respectively
α Compliance coefficient for bump foil
ε Eccentricity ratio
θ Angular measurement from the X-axis (negative) fixed location
φ Bearing attitude angle
ω Velocity (angular), rad/s
R2
Λ Bearing number; Λ = 6μω pa C 2
Λr Foil porosity parameter to bearing number
w, W Static load, N; W = pawL R
Wε , Wφ Nondimensional radial and tangential load
θt Angle of coordinate measured from the negative X-axis fixed location
for texture bump region

1 Introduction

In gas foil bearings that are lubricated with air or gas, the hydrodynamic pressure
that is created between the top foil and the journal causes the foil structure to bend.
The desired gas foil bearing performance characteristics obtained by the deformation
will lead to a change in the film thickness and hence the hydrodynamic pressure.
Heshmat et al. [1] presented desirable design features (structural, geometric, and
operational variables) of a gas foil journal bearing. A gas journal bearing’s perfor-
mance was assessed by Peng and Carpino [2] using Lund’s [3] perturbation solution
approach. The influence of bearing compliance on dynamic coefficients is examined.
In order to forecast the hydrodynamic performance of a foil journal bearing under
limiting conditions, Peng and Khonsari [4, 5] created an analytical model. Based
on the numerical results, they then showed the load capacity over a wide range of
operating speeds. Sawicki and Rao [6] presented the limiting dynamic performance
characteristics of foil bearing owing to the analysis by Peng et al. [4].
Radil et al. [7] determined the direct impact of clearance in radial direction on
the limiting load capacity coefficient of the foil bearing. Rubio and San Andrés [8]
and Kim and San Andrés [9] presented an analytical model for predicting the foil
bearing structural stiffness with notable stiffness changes resulted with preload, in
particular for small radial loads. Yu et al. [10] developed a numerical model of foil
journal bearing with protuberant foil structure and analyzed the bearing performance
characteristics. Mahner et al. [11] presented the influence of a preload on the elastic
foil structural stiffness and the bearing damping. In order to improve the load capacity
and stability performance of foil bearings, Yan et al. [12] proposed a surface micro-
groove structure to be applied to the top foil. Samantha et al. [13] presented a review
on the development, functionality, and safe operation of the foil bearing technology.
Agustin et al. [14] developed a method to calculate the linearized nondimensional
Analysis of Limiting Load Capacity and Stiffness Coefficients … 455

stiffness coefficients of a foil journal bearing that had a top foil texture bump profile.
Jiang et al. [15] investigated the influence of porosity parameter on the static and
dynamic coefficients of an externally pressurized porous gas journal bearing.
The intent of this analysis is to explore load capacity (limiting) and coefficients
of stiffness (limiting) of porous bump texture journal foil bearing. The influence of
extent and height of top foil texture bump along with foil porosity parameter and
effective compliance factor of the parameters of a journal foil bearing are inspected.

2 Methodology

The following equation is the simplified one from Reynolds equation for compress-
ible flow of porous foil journal bearing with large bearing number.

∂(P H ) ( )
+ Λr P 2 − 1 = 0 (1)
∂θ
Figure 1 presents a diagrammatic representation of texture bump foil. Beginning
with a point that is fixed along the X-axis to the negative, the bump region of the
texture expands outward in a circumferential direction (Fig. 2 represents load line).
It is assumed that the top foil surface is having texture bump. The following equation
can be used to calculate the film thickness non-dimensionally for a journal bearing
with porous foil texture bump:
Λp
H = 1 + ε cos(θ − φ) − Hb + α(P − 1), Λr = 2Λ
(2)

A simplified perturbation method is used to obtain the pressure gradients and

steady-state pressure. During the Journal steady-state, for infinitesimally small
perturbation, film thickness and the pressure distribution are given as

H = Ho + Hx ∆X + Hy ∆Y, P = Po + Px ∆X + Py ∆Y (3)

where

Fig. 1 Diagrammatic
representation of bump foil
texture
456 S. A. Agustin et al.

Fig. 2 Diagrammatic view

of journal bearing with foil
texture

Ho = 1 + εo cos(θ − ϕo ) − Hb + α(Po − 1), Hx = cos θ + α Px , H y = sin θ + α Py

(4)

By substituting Eqs. 3 and 4 in Eq. 1, the simplified Reynolds equation for steady-
state and dynamic conditions are given as

∂(Po Ho ) ( )
+ Λr Po2 − 1 = 0 (5)
∂θ
∂(Po Hx + Px Ho )
+ 2Λr Po Px = 0 (6)
∂θ
( )
∂ Po Hy + Py Ho
+ 2Λr Po Py = 0 (7)
∂θ
Integrating Eq. (5) results in
θ( )
Po Ho + Λr Po2 − 1 dθ = C1 (8)
0

The boundary conditions at inlet (θ = 0) is

Po = 1 at θ = 0 (9)

Solving the limiting pressure for porous foil journal bearing with texture bump in
Eq. (4) by substituting pressure boundary conditions in Eq. (9) results in
Analysis of Limiting Load Capacity and Stiffness Coefficients … 457

θ( )
1 + εo cosφo − Hb − Λr 0 Po2 − 1 dθ
Po = (10)
1 + εo cos(θo − φo ) − Hb + α(Po − 1)

Taking average pressure into consideration, Eq. (10) is simplified as

( 2 )
1 + εo cosφo − Hb − Λr Po,av −1 θ
Po = (11)
1 + εo cos(θo − φo ) − Hb + α(Po − 1)

By solving Eq. 11, the limiting pressure (Po ) is obtained as

[
|
| (1 +(ε − Hb − α)(1 + εo cos(θo( − φo ) −)H)b − α)2
−(1 + εo cos(θo − φo ) − Hb − α) + | 2 −1 θ
+4α 1 + εo cos φo − Hb − Λr Pav
Po =
2α
(12)

Substituting Eqs. (3–4) in Eqs. (6), reduces to

∂(Po cosθ + Px (α P o + Ho ))
+ 2Λr Po Px = 0 (13)
∂θ
The bump texture foil bearing pressure gradient distribution under limiting condi-
tions is found by Eq. 13 using Po = 1 and pressure gradient (Px = Py = 0) and inlet
boundary conditions (θ = 0) as
θ
Po cosθ + Px (α P o + Ho ) + 2Λr Po Px dθ = C1 (14)
0

Taking average pressure into consideration, Eq. (14) is simplified as

Po cosθ + Px (α P o + Ho ) + 2Λr Po,av Px,av θ = C1 (15)

Simplifying Eq. (16) results in

1 − Po cosθ − 2Λr Po,av Px,av θ

Px = (16)
1 − α + 2α Po + εo cos(θ − φo ) − Hb

Substituting Eqs. (3–4) in Eqs. (7), reduces to

( )
∂ Po sinθ + Py (α P o + Ho )
+ 2Λr Po Py = 0 (17)
∂θ
The bump texture foil bearing pressure gradient distribution under limiting condi-
tions is derived by solving Eq. 17 using Po = 1 and Px = Py = 0 and θ = 0 as and
taking average pressure into consideration, Eq. (17) is simplified as
458 S. A. Agustin et al.

θ
Po sinθ + Py (α P o + Ho ) + 2Λr Po Py dθ = C2 (18)
0

Po sinθ + Py (α P o + Ho ) + 2Λr Po,av Py,av θ = C2 (19)

Simplifying Eq. (19) results in

−Po sinθ − 2Λr Po,av Px,av θ

Py = (20)
1 − α + 2α Po + εo cos(θ − φo ) − Hb

The limiting nondimensional load and limiting nondimensional bearing stiffness

coefficients K x x , K yx , K x y , K yy are evaluated by integration of pressure and pressure
gradients as
θ=2π θ =2π
Fx cos θ Kx j cos θ
= Po dθ, = Pj dθ for j = x, y
Fy θ =0 sin θ Kyj θ =0 sin θ
(21)

3 Results and Discussion

The nondimensional load limiting (W) and coefficients (stiffness) (Kij) of foil porous
bearing are reported. The analysis uses the following parameters: nondimensional
bump texture height (H b ) = 0.5–1; circumferential coordinate (from the X-axis nega-
tive fixed location) for bump texture region (θ t ) = 0–240°; foil porosity parameter
to bearing number (Λr ) = 0–0.3; reference foil compliance coefficient (α) = 1, 10;
eccentricity ratio (ε) = 0.2, 0.8.
From Fig. 3, the nondimensional limiting load of journal bearing with porous foil
can be seen. As shown in Fig. 3i, the nondimensional load (W ) remains constant until
a threshold circumferential coordinate (from the negative X-axis fixed location) for
bump texture region (θ t ) and subsequently grows with increasing bump texture height
(Hb ). As depicted in Fig. 3ii for the parameters considered in the analysis, for a higher
value of compliance coefficient (α = 10), the nondimensional load (W) is independent
of the nondimensional bump texture height (Hb ). From Fig. 3iii, the nondimensional
load (W) falls when the foil porosity parameter (Λr ) increases from 0 to 0.3.
Figure 4 depicts the limiting (nondimensional) coefficients of stiffness of
foil porous bearing. When demonstrated in Fig. 4i, the nondimensional stiffness
coefficients (Kij) exhibit greater fluctuation as the nondimensional bump texture
height (Hb ) increases from 0.5 to 1.0. The circumferential coordinate (from the
X-axis negative fixed location) for bump texture region (θ t ) significantly influences
the nondimensional pressure gradients (Py ). The nondimensional coefficients of
stiffness (K yj ) perpendicular to load line obtained from nondimensional pressure
gradients (Py ) showed significant variation with increasing circumferential coordi-
nate (from the negative X-axis fixed location) (θ t ) for the texture extent of region
Analysis of Limiting Load Capacity and Stiffness Coefficients … 459

(i) α=1, Λr=0.3

(ii) θt=180°, Hb=1 (iii) θt=180°, Hb=1

Fig. 3 Load (nondimensional) capacity

of the bump. The stiffness (nondimensional) coefficients (K ij ) are invariant with

increasing foil porosity parameter to bearing number ratio (Λr ) = 0–0.3 as plotted
in Fig. 4(iii) and (iv).

4 Conclusions

This study examines the effects of porous foil journal bearings with bump-textured
top foils on stiffness coefficients and limiting load capacities. For a foil bearing
working at high speed, the compressible Reynolds equation is used to determine
460 S. A. Agustin et al.

(i) Kij (ε=0.2, α=1, Λr=0.3). (ii) Kij (ε=0.8, α=1, Λr=0.3).

(iii) Kij (ε=0.2, θt=180°, Hb=1). (iv) Kij (ε=0.8, θt=180°, Hb=1).

Fig. 4 Nondimensional stiffness coefficients

the limiting load capacity and stiffness coefficients (large bearing numbers). The
increasing bump texture height of porous foil journal bearing enhances load capacity
and influences higher variation of the nondimensional stiffness.
The bump texture porous top foil with augmented sub-foil compliance has promise
to enrich the foil bearing features.

Acknowledgements The authors are grateful to the support of SRMIST and AdtU.
Analysis of Limiting Load Capacity and Stiffness Coefficients … 461

References

1. Heshmat H, Walowit JA, Pinkus O. Analysis of gas-lubricated foil journal bearings. J Lubr
Technol. 1983;105:647–55.
2. Peng J-P, Carpino M. Calculation of stiffness and damping coefficients for elastically supported
gas foil bearings. J Tribol. 1993;115(1):20–7.
3. Lund JW. Review of the concept of dynamic coefficients for fluid film journal bearings. ASME
J Lubric Technol. 1987;109(1):37–41.
4. Peng ZC, Khonsari MM. On the limiting load carrying capacity of foil bearings. ASME J
Tribol. 2004;126:817–8.
5. Peng ZC, Khonsari MM. Hydrodynamic analysis of compliant foil bearing with compressible
air flow. ASME J Tribol. 2004;126:542–6.
6. Sawicki JT, Rao TVVLN. Limiting stiffness and damping coefficients of foil bearing. In:
Proceedings of IDETC/CIE. 2005. p. 1069–73.
7. Radil K, Howard S, Dykas B. The role of radial clearance on the performance of foil air
bearings. Tribol Trans. 2002;45(4):485–90.
8. Rubio D, San AL. Bump-type foil bearing structural stiffness: experiments and predictions. J
Eng Gas Turbines Power. 2006;128:653–60.
9. Kim TH, San AL. Analysis of advanced gas foil bearings with piecewise linear elastic supports.
Tribol Int. 2007;40(8):1239–45.
10. Hou Y, Shuangtao C, Rugang C, Qiaoyu Z, Hongli Z. Numerical study on foil journal bearings
with protuberant foil structure. Tribol Int. 2011;44(9):1061–70.
11. Marcel Mahner P, Li AL, Schweizer B. Numerical and experimental investigations on preload
effects in air foil journal bearings. J Eng Gas Turbines Power. 2018;140(3): 032505.
12. Yan J, Zhang G, Liu Z, Zhao J, Liang X. Performance of a novel foil journal bearing with
surface micro-grooved top foil. Proc Instit Mech Eng Pt J J Eng Tribol. 2018;232(9):1126–39.
13. Samanta P, Murmu NC, Khonsari MM. The evolution of foil bearing technology. Tribol Int.
2019;135:305–23.
14. Arokya Agustin S, Shravankumar C, Rao TVVLN. Limiting stiffness coefficients analysis of
texture foil journal bearing. IOP Conf. Ser. Mater. Sci. Eng. 2020;912:022031.
15. Jiang S, Lin S, Xu C. Static and dynamic characteristics of externally pressurized porous gas
journal bearing with four degrees-of-freedom. ASME J Tribol. 2018;140: 011702.
Limiting Load Capacity and Stiffness
Coefficients of Bump Recess Foil Journal
Bearing with FGM

S. Arokya Agustin, C. Shravankumar, A. Arul Jeya Kumar,

and T. V. V. L. N. Rao

Abstract This paper combines the bump-recess configuration on top foil of air
foil journal bearing using Functionally Graded Material (FGM) in order to discover
the journal bearing’s limiting load capacity and coefficients of limiting stiffness.
An analytical model for limiting load capacity and limiting stiffness coefficients of
the bump-recess configuration on top foil of air foil journal bearing using FGM is
presented. Estimates of the limiting load capacity and limiting stiffness coefficients
of the bump-recess configuration on the top foil of an air foil journal bearing using
FGM are based on a simplified version of the compressible Reynolds equation.
These estimates are made for higher bearing numbers coupled with higher speeds
for the journal foil bearing. The theoretical model accounts for the bump-recess
configuration with FGM and bump foil compliance. A textural bump of the top foil
bearing’s limiting pressure gradient solution is examined. Distribution of pressure
is calculated using high operating speed bearing approximation. We compare the
results of the limiting load capacity and limiting coefficients of stiffness of bump-
type journal foil bearings with FGM for different bump-recess heights. Different top
foil texture bump patterns are examined for nondimensional coefficients of stiffness
produced using the procedure based on infinitesimally small disturbances called
perturbation method. A journal (foil) bearing’s limiting stiffness coefficients are
explored in relation to the extent and height of the top foil texture bump.

Keywords Foil journal bearing · FGM · Bump foil · Load capacity

S. A. Agustin · C. Shravankumar · A. A. Jeya Kumar

SRM Institute of Science and Technology, Kattankulathur 603203, India
T. V. V. L. N. Rao (B)
Faculty of Engineering, Assam Down Town University, Guwahati, Assam 781026, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 463
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_31
464 S. A. Agustin et al.

Nomenclature

C Clearance (radial), m
E fg Modulus of elasticity of functionally graded bottom foil
E 1, E 2 Modulus of elasticity of base material and Modulus of elasticity of
reference material of functionally graded bottom foil
Er Ratio of modulus of elasticity of reference to base material (E 2 /E 1 ) of
functionally graded bottom foil
fα Effective compliance factor
h, H Thickness of fluid film, m; H = Ch
Hb Texture bump height
ki j , K i j Coefficients of Stiffness, N/m; K i j = K i j /W , i, j = x, y, K i j =
ki j C/ pa L R
L Length of the bearing, m
pa Ambient pressure, N/m2
p, P Bearing pressure, N/m2 ; P = ppa
R Radius of the journal, m
x, y, X, Y Bearing coordinates; X = x/C, Y = y/C
∆X, ∆Y x and y direction displacements (nondimensional)
α Compliance coefficient of bump foil
ε Ratio of eccentricity
θ Angle measured from X-axis on negative direction
φ Attitude (bearing) angle
ω Circular velocity, rad/s
λo Gradient index
R2
Λ Bearing number; = 6μω pa C 2
w, W Load at Static, N; W = pawL R
Wε , Wφ RAdial load and tangential load (Nondimensional)
θt Angle measured from X-axis for bump texture region on negative
direction

1 Introduction

Gas foil bearings with top, sub, and bearing housing are compliant members that
are air/gas lubricated. The top foil deforms due to the hydrodynamic pressure that is
created between the journal and that foil. The film thickness will change as a result
of the deformation, which will also affect the hydrodynamic pressure. By controlling
the relationship between fluid film pressure and foil deformation, the desired gas foil
bearing performance characteristics are accomplished.
Desirable design elements (structural, geometric, and operational conditions) of
air based journal foil bearing were discussed by Heshmat et al. [1]. A compliant
foil supported by a spring is used in a gas journal bearing, and its performance is
Limiting Load Capacity and Stiffness Coefficients of Bump Recess Foil … 465

assessed. A gas foil bearing’s linearized dynamic coefficients were computed by

Peng and Carpino [2] using Lund’s [3] perturbation solution technique. The effect of
the bearing compliance on the dynamic coefficients is discussed. Peng and Khonsari
[4] developed an analytical model to predict the hydrodynamic performance of a
foil journal bearing under limiting conditions. Reynolds equation of compressible
fluid flow for a bearing running at high speed is taken into consideration by the
analytical model for high bearing numbers. Established on the numerical results,
Peng and Khonsari [5] provided the load (nondimensional) capacity at a variety of
operating speeds. Sawicki and Rao [6] presented the performance characteristics
under dynamic load condition of foil bearing based on the analysis done by Peng
and Khonsari [4].
The impact of clearance in radial direction on the foil air bearings’ load capacity
coefficient was identified by Radil et al. [7]. The foil bearing’s load capacity coef-
ficient is directly influenced by the radial clearance. Bearing seizure is more likely
in bearings that are operating with less-than-optimal radial clearances. Although
there were no issues with thermal management when bearings operated with radial
clearance twice the ideal, their maximum load capacity coefficient degraded by 20%.
Rubio and San Andrés [8] presented an analytical model for predicting the foil bearing
structural stiffness. Notable stiffness changes resulted with preload, in particular for
small radial loads. Kim and San Andrés [9] presented rotordynamic coefficients of
a double bump strip layer gas foil bearing. Within a double bump strip layer gas
foil bearing, the structural damping contributes to an increase in the bearing’s direct
damping force coefficients. Yu et al. [10] built a computational model of a foil journal
bearing with a protuberant foil structure and performed an analysis of the bearing’s
performance characteristics. Mahner et al. [11] presented the findings of their study
on the effect of a preloaded three-pad air foil journal bearing. The assembly preload
increases the elastic foil structural stiffness and the bearing damping. When the bear-
ings are just minimally preloaded, the fluid film has a significant impact on the total
bearing stiffness. The overall bearing stiffness is dependent on the assembly preload.
Yan et al. [12] proposed the surface micro-groove structure on top foil to improve the
load capacity and stability performance of foil bearing. When compared with conven-
tional foil journal bearings, the load capacity and direct stiffness are improved with
increasing micro groove depth. Samantha et al. [13] provided a review on the evolu-
tion, functionality, and risk-free operation of the foil bearing technology. They also
brought attention to the difficulties in the design, analysis, and performance charac-
teristics that need to be taken into account. Agustin et al. [14] found the coefficients
of stiffness with nondimensional linearized model of a journal foil bearing with a top
foil bump texture profile. In addition, they derived the limiting pressure gradient solu-
tion. It was determined how much of an effect the top foil texture bump had, as well as
how high it was, on the coefficients of stiffness of a journal foil bearing. Nijssen and
van Ostayen [15] showed an improved potential of compliant-hydrostatic bearings
with the functionally graded material supports as compared to conventional elastic
supports. Birman and Byrd [16] presented a review of the principal developments in
theory and applications of functionally graded materials (FGMs).
466 S. A. Agustin et al.

The novelty of this study of air foil journal bearing is the combination of the
bump-recess configuration on top foil using functionally graded (FG) compliant
bottom foil. The primary objective of this research is to determine the coefficients of
stiffness of FG bump texture journal foil bearings. It is explored how the coefficients
of stiffness of a foil journal bearing are affected by the length and height of inner foil
texture bump, as well as the FG material gradient index and effective compliance
factor.

2 Methodology

For the pressure field to remain bounded for high bearing number operation, the
compressible Reynolds equation can be simplified to look like this: [4], which will
result in the necessary conditions being met.

∂(P H )
=0 (1)
∂θ
Based on Reddy et al. [17], the modulus of elasticity of functionally graded bottom
foil model used in the analysis is
( )
E f g = E 1 (0.5 − ζ )λ + E 2 1 − (0.5 − ζ )λ (2)

where ζ refers to the axis at the mid-surface of the functionally graded bottom foil.
The gradient index (λ) = 0 yields the modulus of elasticity of the base material. The
gradient index (λ) increases with increasing reference material portion of functionally
graded bottom foil.
Figure 1 offers a diagrammatic representation of the texture (bump) foil. Begin-
ning at a point that is fixed to the X-axis in the negative direction, the bump region
of the texture expands circumferentially (load line is shown in Fig. 2). The bump
texture is placed on top of the foil surface. The following equation can be used to
determine the nondimensional film thickness for a FG bump texture foil bearing:

H = 1 + ε cos(θ − ϕ) − Hb + f α α(P − 1) (3)

where the effective compliance factor of the functionally graded bottom foil is
considered as
( θ
)
f α = 0.5 1
, λ = λ0 1 − 2π (4)
−0.5 [(0.5−ζ ) +E r (1−(0.5−ζ ) )]dζ
λ λ

When the simplified compressible Reynolds equation is subjected to an infinites-

imally small perturbation, [3] the pressure (steady-state) and pressure (dynamic)
gradients can be calculated. When there are infinitesimally small disturbances in
relation to the journal steady-state, the pressure dispersion and film thickness are as
Limiting Load Capacity and Stiffness Coefficients of Bump Recess Foil … 467

Fig. 1 Bump foil diagram

Fig. 2 Texture foil bearing

schematic

follows:

P = Po + Px ∆X + Py ∆Y, H = Ho + Hx ∆X + Hy ∆Y (5)

where

Ho = 1 + εo cos(θ − ϕo ) − Hb + f α α(Po − 1), Hx = cos θ + f α α Px ,

Hy = sin θ + f α α Py (6)

Substituting Eqs. (5, 6) in Eq. (1), Reynolds equation for compressible is simplified
for steady-state and dynamic condition and given as:

∂(Po Ho )
=0 (7)
∂θ
468 S. A. Agustin et al.

∂ ∂ ( )
(Po Hx + Px Ho ) = 0, Po Hy + Py Ho = 0 (8)
∂θ ∂θ
The limiting pressure distribution (PO ) for texture (bump) foil bearing is got by
resolving Eq. (7) by treating pressure equal to 1 at inlet angle equal to 0 at the inlet
as [4]

1 + εo cos ϕo − Hb
Po = (9)
1 + εo cos(θ − ϕo ) − Hb + f α α(Po − 1)

The limiting pressure can be expressed more simply as Eq. (9) as

( ) /( )2 ( )
− 1 + εo cos(θ − ϕo ) − Hb − f α α + 1 + εo cos(θ − ϕo ) − Hb − f α α + 4 f α α 1 + εo cos ϕo − Hb
Po =
2 fα α
(10)

Substituting Eqs. (5, 6) in Eqs. (8), reduces to

∂ ∂ ( )
(Po cos θ + Px ( f α α Po + Ho )) = 0, Po sin θ + Py ( f α α Po + Ho ) = 0
∂θ ∂θ
(11)

By substituting Po = 1 and Px = Py = 0 at inlet θ = 0, the pressure gradients

are obtained as [14]

1 − Po cos θ
Px = (12)
1 − f α α + 2 f α α Po + εo cos(θ − ϕo ) − Hb
−Po sin θ
Py = (13)
1 − f α α + 2 f α α Po + εo cos(θ − ϕo ) − Hb

The limiting nondimensional load and limiting nondimensional bearing (stiff-

ness) coefficients K x x , K yx , K x y , K yy are assessed by the integration of pressure and
pressure gradients as
θ=2π θ =2π
Fx cos θ Kx j cos θ
= Po dθ, = Pj dθ for j = x, y
Fy θ =0 sin θ Kyj θ =0 sin θ
(14)
Limiting Load Capacity and Stiffness Coefficients of Bump Recess Foil … 469

3 Results and Discussion

The limiting nondimensional load (W ) and nondimensional (stiffness) coefficients

(K ij ) of functionally graded (FG) bump texture foil journal bearing are shown. The
variables used in the analysis are: ratio of modulus of elasticity of reference to
base material (E 2 /E 1 ) of functionally graded bottom foil (E r ) = 1, 2; nondimen-
sional bump texture height (H b ) = 0–1.5; circumferential coordinate (measured from
X-axis on negative direction) for texture bump region (θ t ) = 0–240°; gradient index
(λo ) = 0–6; base foil compliance coefficient (α) = 1,10; eccentricity ratio (ε) = 0.2,
0.8.
From Fig. 3, the nondimensional load (W ) of functionally graded (FG) bump
texture journal foil bearing. As shown in Fig. 3i, the nondimensional load capacity
(W ) remains constant until a threshold circumferential coordinate (measured from
X-axis in negative direction) for bump texture region (θ t ) and then increases with
increasing nondimensional bump texture height (H b ). The nondimensional load (W)
is independent of the nondimensional bump texture height (Hb ) for larger base foil
compliance coefficients (α = 10), as shown in Fig. 3ii for the parameters that
were examined in the investigation. The nondimensional load (W ) increases with
increasing gradient index (λo ) until a critical value and remains nearly constant with
increasing gradient index (λo ) as plotted in Fig. 3iii. Increasing the ratio of modulus
of elasticity of reference to base material (E 2 /E 1 ) of functionally graded bottom foil
(E r ) increases (Fig. 3iv) the nondimensional load capacity (W ).
Figure 4 shows the coefficients of stiffness of functionally graded (FG) bump
texture foil journal bearing. As shown in Fig. 4i, higher variation of the nondimen-
sional stiffness coefficients (K ij ) is observed with increasing nondimensional bump
texture height (H b ). Significantly affecting the nondimensional pressure gradients
(Py ) is the circumferential coordinate for the texture bump region, which is denoted
by θ t . The coefficients of nondimensional stiffness (K yj ) perpendicular to line of
load obtained from nondimensional pressure gradients (Py ) showed significant vari-
ation with increasing circumferential coordinate (Measured from X-axis in negative
direction) for texture bump region (θ t ). As depicted in Fig. 4ii, the trends of the nondi-
mensional stiffness coefficients (K ij ) at higher eccentricity ratio (ε = 0.8) are similar
to those obtained for lower eccentricity ratio (ε = 0.2) in Fig. 4i. The coefficients of
nondimensional stiffness (K ij ) are invariant with gradient index (λo ) (Fig. 4iii) and
reference foil compliance coefficient (α) (Fig. 4iv).

4 Conclusions

The stiffness and load capacity limits of functionally graded (FG) foil journal bearings
with bump texture top foils are studied in this work. The configuration of the gas
foil bearing consists of bump texture profile of the top foil and functionally graded
(FG) sub-foil compliance in the bump texture region. The limiting load capacity and
470 S. A. Agustin et al.

(i) α=1, Er=2, λo=4 (ii) θt=180°, Er=2, λo=4

(iii) θt=180°, Er=2, Hb=1 (iv) θt=180°, λo=4, Hb=1

Fig. 3 Nondimensional based load capacity

coefficients of stiffness are based on the Reynolds equation of compressible fluid

flow for foil bearing operating at high speed (large bearing numbers). A functionally
graded (FG) material of the sub-foil with a bump texture top foil extending along the
bearing circumference provides enhanced load capacity for higher extents of texture
bump region (θ t = 120–240°) and lower base foil compliance coefficient (α = 1).
The extent of bump texture top foil with a functionally graded (FG) material of the
sub-foil influences the nondimensional stiffness coefficients.
Enhanced sub-foil compliance and bump texture on the top foil could improve
foil bearing properties. The scope of future work includes analysis of bump-recess
foil journal bearing with FGM for a range of low to high bearing number operations.
Limiting Load Capacity and Stiffness Coefficients of Bump Recess Foil … 471

(i) Kij (ε=0.2, α=1, Er=2, λo=4). (ii) Kij (ε=0.8, α=1, Er=2, λo=4).

(iii) Kij (ε=0.2, θt=180°, Er=2, Hb=1). (iv) Kij (ε=0.8, θt=180°, λo=4, Hb=1).

Fig. 4 Nondimensional stiffness coefficients

Acknowledgements The authors are grateful to the support of SRMIST and AdtU.

References

1. Heshmat H, Walowit JA, Pinkus O. Analysis of gas-lubricated foil journal bearings. J Lubr
Technol. 1983;105:647–55.
2. Peng J-P, Carpino M. Calculation of stiffness and damping coefficients for elastically supported
gas foil bearings. J Tribol. 1993;115(1):20–7.
3. Lund JW. Review of the concept of dynamic coefficients for fluid film journal bearings. ASME
J Lubric Technol. 1987;109(1):37–41.
472 S. A. Agustin et al.

4. Peng ZC, Khonsari MM. On the limiting load carrying capacity of foil bearings. ASME J
Tribol. 2004;126:817–8.
5. Peng ZC, Khonsari MM. Hydrodynamic analysis of compliant foil bearing with compressible
air flow. ASME J Tribol. 2004;126:542–6.
6. Sawicki JT, Rao TVVLN. Limiting stiffness and damping coefficients of foil bearing. In:
Proceedings of IDETC/CIE. 2005. p. 1069–73.
7. Radil K, Howard S, Dykas B. The role of radial clearance on the performance of foil air
bearings. Tribol Trans. 2002;45(4):485–90.
8. Rubio D, San AL. Bump-type foil bearing structural stiffness: experiments and predictions. J
Eng Gas Turbines Power. 2006;128:653–60.
9. Kim TH, San AL. Analysis of advanced gas foil bearings with piecewise linear elastic supports.
Tribol Int. 2007;40(8):1239–45.
10. Hou Y, Shuangtao C, Rugang C, Qiaoyu Z, Hongli Z. Numerical study on foil journal bearings
with protuberant foil structure. Tribol Int. 2011;44(9):1061–70.
11. Marcel Mahner P, Li AL, Schweizer B. Numerical and experimental investigations on preload
effects in air foil journal bearings. J Eng Gas Turbines Power. 2018;140(3): 032505.
12. Yan J, Zhang G, Liu Z, Zhao J, Liang X. Performance of a novel foil journal bearing with
surface micro-grooved top foil. Proc Inst Mech Eng Pt J J Eng Tribol. 2018;232(9):1126–39.
13. Samanta P, Murmu NC, Khonsari MM. The evolution of foil bearing technology. Tribol Int.
2019;135:305–23.
14. Arokya Agustin S, Shravankumar C, Rao TVVLN. Limiting stiffness coefficients analysis of
texture foil journal bearing. IOP Conf. Ser. Mater. Sci. Eng. 2020;912:022031
15. Nijssen JPA, van Ostayen RAJ. Compliant hydrostatic bearings utilizing functionally graded
materials. J Tribol. 2020;142(11):111801
16. Birman V, Byrd LW. Modeling and analysis of functionally graded materials and structures.
Appl Mech Rev. 2007;60(5):195–216.
17. Reddy JN, Wang CM, Kitipornchai S. Axisymmetric bending of functionally graded circular
and annular plates. Euro J Mech A Solids. 1999;18:185–99.
Modal Analysis of 3D Braided Composite
Reinforced Twisted Spherical Shell Using
3D Finite Element Method

Sambhaji Lore, Pabitra Maji, and Brighu Nath Singh

Abstract The utilization of 3D braided composites has witnessed a surge in popu-

larity across diverse industries such as aerospace, naval, nuclear, automobile, and
other high-performance sectors. This widespread adoption can be attributed to their
exceptional specific stiffness, strength, and notably low specific weight. This research
endeavor aims to forecast the corresponding material properties of braided compos-
ites utilizing the volume averaging method. In this model, it is assumed that the
fibers possess transverse isotropy while the matrix maintains homogeneous isotropy.
To investigate the characteristics of free vibration in spherical shell panels, the third-
order shear deformation theory (TSDT) is employed. An eight-noded isoparametric
finite element is utilized, and the generalized dynamic equilibrium equation is derived
from Lagrange’s equation of motion. The accuracy of the finite element model is
assessed by comparing it against existing results. The study encompasses an exami-
nation of the free vibration response of the spherical shell under varying conditions,
including aspect ratios, braided volume fractions, braided angles, thickness ratios,
curvature ratios, and boundary conditions. The outcomes reveal substantial influences
of aspect ratios, volume fractions, twist angles, braided angles, curvature ratios, and
thickness ratios on the frequency parameters.

Keywords Spherical shells · Modal response · TSDT · 3D finite element ·

Braided composites

S. Lore (B)
Department of Aeronautical Engineering, NITTE Meenakshi Institute of Technology, Bangalore,
India
e-mail: [email protected]
P. Maji · B. N. Singh
Department of Aerospace Engineering, Indian Institute of Technology, Kharagpur, West Bengal,
India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 473
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_32
474 S. Lore et al.

1 Introduction

Composite materials have experienced a surge in popularity across diverse indus-

tries, including aerospace, marine, civil, and automobile, owing to their remark-
able mechanical properties. These properties encompass outstanding strength-to-
weight and stiffness-to-weight ratios, cost-effective maintenance, and exceptional
resistance to fatigue and corrosion. In the examination of thin-walled composite
shell structures, the vibration characteristics, such as natural frequencies and mode
shapes, play a pivotal role. An innovative type of composite material, referred to as
textile 3D composites, has been developed utilizing fabrication techniques commonly
employed in the textile industry. The interlacing of tows in the thickness direction
within textile composites effectively eliminates the risk of delamination. However,
the geometric structural design of textile composites proves more intricate compared
to laminated composites, necessitating significant effort to accurately determine
the equivalent material properties. Extensive research exists in the literature that
addresses the determination of physical and mechanical properties specific to 3D
textile composites.
The growing utilization of composite shell panels in numerous engineering and
industrial applications has sparked significant research interest in understanding the
dynamic behavior of these structures. Maji et al. [1] conducted an investigation
focusing on the free vibration analysis of 3D braided curved shell panels using
the third-order shear deformation theory. Their approach employed an eight-noded
isoparametric finite element method, with twelve degrees of freedom per node. The
team utilized the volume averaging method (VAM) to determine the equivalent mate-
rial properties. Singh and Singh [2] employed non-polynomial higher order shear
deformation theories to analyze the free vibration and buckling behavior of braided
and laminated composite plates. Reddy and Asce [3] presented exact solutions for
the static and dynamic equations of simply supported, cross-ply laminated, doubly
curved shells subjected to uniformly distributed, sinusoidal, and point loads. Li and
Wang [4] conducted an analysis of large amplitude free vibrations in a 3D braided
composite cylindrical shell subjected to axial forces. They utilized the two-step
perturbation technique to determine the frequencies and parametric resonances. Zhao
et al. [5] investigated the nonlinear bending characteristics of a three-dimensional
(3D) braided composite cylindrical shell panel. Singh and Singh [6] studied the influ-
ence of various parameters on the buckling behavior of 3D braided composites. Javed
et al. [7] conducted a free vibration analysis of conical shells with variable thickness,
examining their behavior under both simply supported and clamped boundary condi-
tions using collocation with spline approximation. Haichao Li et al. [8] analyzed the
free vibration behavior of functionally graded doubly curved shells with variable
thickness. They employed the Ritz method for their analysis. Fard and Baghestani
[9] developed a semi-analytical method based on the first-order shear deformation
theory to analyze the vibration characteristics of doubly curved moderately thick
shells.
Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 475

Huiyu Sun et al. [10] proposed a methodology for determining the microme-
chanical properties of 3D braided composites by employing homogenization theory
in conjunction with finite element analysis (FEM). Maji and Singh [11] computed
the equivalent material properties of 3D braided composites using bridging models
based on the volume averaging method. They utilized these properties to analyze
the free vibration response of rotating cylindrical shells. Mahmood and Mohammad
[12, 13] introduced analytical models to predict the stiffness of 2D tri-axial and
3D four-directional braided composites. They defined a unit cell as a representa-
tive volume element for braided composites and its constituent elements. Xuhao
et al. [14] developed a constitutive and finite element model for 3D braided compos-
ites with voids. They predicted the evolution of material properties in the presence
of defects within 3D braided composite materials. Jingran et al. [15] formulated a
coupled elastic–plastic damage model to analyze the nonlinear mechanical behavior
of 3D braided composites. Wang and Wang [16] demonstrated an analytical link
between microstructure, manufacturing processes, and material properties of 3D
braided composites. Sun and Sun [17] employed the volume-average-compliance
method to predict the mechanical properties of 3D braided composites. Han and
Jiang [18] investigated the failure mechanisms and mechanical properties of 3D six-
directional braided composites under various temperature conditions. Chen et al.
[19] established a homogenization method to evaluate the elastic properties of 3D
regular braided composites.
From the existing literature, it becomes apparent that the majority of research
efforts have been directed toward the vibration analysis of laminated and 3D braided
composite plates and shells, primarily employing FSDT and 2D shear deformation
theories. Nevertheless, in this particular study, the authors have adopted a highly effi-
cient approach by utilizing a 3D third-order shear deformation theory to investigate
the free vibration characteristics of 3D braided composite spherical shell panels. To
determine the equivalent material properties of these composites, the volume aver-
aging method was employed. Additionally, the study comprehensively examined
the influence of various factors, including braided angles, braided volume fractions,
aspect ratios, thickness ratios, curvature ratios, and boundary conditions, on the free
vibration response of spherical shells.

2 Theoretical Formulation

The analysis focuses on a doubly curved shell panel composed of braided composite
material, as depicted in Fig. 1. The geometric dimensions of the shell panel, including
length (a), width (b), and thickness (h), along with the radii of curvature in the x
(Rx ) and y (Ry ) directions, are defined using Cartesian coordinates. In the case of
spherical shells, the values for Rx and Ry are equivalent, denoted as R.
476 S. Lore et al.

Fig. 1 Geometric
configuration of doubly
curved shell panel

2.1 Displacement Field

The displacement field for the modal analysis of the spherical shell is assumed as

u(x, y, z) = u 1 (x, y) + zu 2 (x, y) + z 2 u 3 (x, y) + z 3 u 4 (x, y)

v(x, y, z) = v1 (x, y) + zv2 (x, y) + z 2 v3 (x, y) + z 3 v4 (x, y) (1)

w(x, y, z) = w1 (x, y) + zw2 (x, y) + z 2 w3 (x, y) + z 3 w4 (x, y)

where u, v, and w denote the displacements of any point in the x, y, and z directions,
respectively.

2.2 Strain–Displacement Relationship

The linear strain–displacement relationship for the spherical shell is expressed as

{ε}x yz = {ε0 } + z{k1 } + z 2 {k2 } + z 3 {k3 } (2)

where {ε0 } are the mid-plane strain components and {k1 }, {k2 }, {k3 } are the higher
order strain components.
The strain–displacement relationship can be written as

{ε}x yz = [B L ]{de } (3)

where [BL ] is the strain–displacement matrix.

Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 477

2.3 Constitutive Relations

The stress–strain relations for the spherical shells can be expressed as

_
{σ }x yz = [ Q]{ε}x yz = [T ][Q][T ]T {ε}x yz (4)
{ }T
where {σ }x yz = σx x σ yy σzz τ yz τx z τx y .

2.4 Strain Energy

The strain energy (U) of spherical shells due to conservative force is given as

1
U= {ε}T {σ }dv
2
v

1
= {de }T [B L ]T [De ] [B L ] {de } d x d y (5)
2
A

[ ]
Ke = [B L ]T [De ] [B L ] d x d y (6)
A

where [K e ] and [De ] are defined as the elemental stiffness and elasticity matrices,
respectively.

2.5 Kinetic Energy

The kinetic energy, (T) of spherical shell element, is expressed as

1
T = ρ V .V dv (7)
2
v

where V is the velocity vector and ρ is the density of the shell material,

1 { }T T { }
T = ρ ḋe [Ne ] [Ne ] ḋe ∀
2
∀

1 { }T { }
= ḋe [Me ] ḋe (8)
2
478 S. Lore et al.

[Me ] = [N ]T [m][N ]d xd y (9)

A

where [Me ] is the elemental mass matrix of a doubly curved shell and [Ne ] is the
shape function matrix.

2.6 Finite Element Formulation

An eight-noded isoparametric serendipity shell element is used for the present finite
element analysis. The elemental stiffness matrix [K e ] and elemental mass matrix
[M e ] in the natural co-ordinate system are expressed as

+1 +1
[ ]
Ke = [B L ]T [De ] [B L ] |J | dξ dς (10)
−1 −1

+1 +1
[ ]
Me = [Ne ]T [m] [Ne ] |J | dξ dς (11)
−1 −1

3 Micromechanical Approach of 3D Braided Spherical

Shells

The yarn pattern of 3D braided composite [13] is demonstrated in Fig. 2. Every

unit cell configuration is discretized into four different control volumes (CVs), viz.
interior surface, corner regions, surface, and exterior surface [13]. hi is the braiding
pitch length, wi is the braiding pitch width, and α is the braiding angle:
√
tan γ = 2 tan α (12)

tan β = tan θ = 1/2 tan α (13)

The total fiber volume fraction is computed from the following relations:

V f = Ci Vi + Css Vss + Cis Vis + Cc Vc (14)

Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 479

a b
Interior region
Surface region
Corner region

Fig. 2 a Interior, surface, and corner regions [13]. b Fiber configuration of Interior region unit cell
[13]

where C i , C is , C ss , and C c are defined as the volume proportions of the four different
control volumes while V i , V is , V ss , and V cc denote the fiber volume fraction of the
above four regions, respectively.

3.1 Mechanical/Material Properties of Unidirectional

Braided Spherical Shell Panels

In this technique, the fibers are considered as transversely isotropic, and the matrix
is considered as isotropic material. The stiffness matrix can be characterized as
[C] f = [S]−1
f where [S] f is the compliance matrix of the fibers.
The stiffness matrix for the matrix material can be defined as [C]m = [S]−1
m where
[S]m is the compliance matrix of the matrix.
The compliance matrix for each unidirectional 3D braided spherical shell is
characterized as
[ ] ( )( )−1
Si j n
= Vnf [Si j ] f + Vnm [Si j ]m [ai j ] Vnf [I ] + Vnm [ai j ] (15)

f
where Vn and Vnm indicate the volume fractions of fibers and matrix of unidirectional
f
braided spherical shell panels respectively and are related as Vn + Vnm = 1. [ai j ]
indicates bridging matrix.
Finally, the total stiffness matrix of the braided composite based on the volume
averaging method is expressed as [13]
∑
[C] = Vn [C i j ]n n = (i1, i2, ......., c3, c4) (16)
480 S. Lore et al.

For even number of m and n, the volume fraction of the four different regions can be
represented as

2(m − 1) (n − 1) + 2
Vi1 = Vi2 = Vi3 = Vi4 =
4 (2mn + m + n)
(m + n − 4)
Vis1 = Vis2 = Vis3 = Vis4 =
4 (2mn + m + n)
(m + n − 4)
Vss1 = Vss2 = Vss3 = Vss4 = (17)
2 (2mn + m + n)
2
Vc1 = Vc2 = Vc3 = Vc4 =
(2 mn + m + n)

The equivalent elastic properties of the 3D braided spherical shells are calculated
as
1 1 1
Ex x = , E yy = , E zz =
S 11 S 22 S 33
1 1 1
Gxy = , Gxz = , G yz = (18)
S 55 S 66 S 44
S 12 S 13 S 23
νx y = − , νx z = − and ν yz = −
S 11 S 11 S 22

4 Free Vibration Analysis

Based on Hamilton’s principle, the equations of motion of the spherical shell panels
are given by

d ∂U2 ∂U2 ∂
− + (U1 + U3 ) = 0 (19)
dt ∂ ḋe ∂de ∂de

Neglecting the damping effects, the free vibration equations in the global
coordination can be expressed as
{ }
[M] d̈ + ( [K ] ) {d} = 0 (20)

Considering harmonic vibrations, {d} = {d} ei ωn t , we have

[ ]
[K ] − ωn2 [M] {d} = 0 (21)
Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 481

This is an eigenvalue problem and is solved for the eigenvalues and eigenvectors

[A] {d} = λ {d} (22)

where [A] = [K ]−1 [M] and λ = 1/ωn2 .

5 Boundary Conditions

Different types of boundary conditions (BCs) are used for the present analysis:
A. All Edges Simply Supported (SSSS):

v1 = v2 = v3 = v4 = w1 = w2 = w3 = w4 = 0 x = 0, a;
u 1 = v1 = w1 = u 2 = v2 = w2 = u 3 = v3 = w3 = u 4 = v4 = w4 = 0 x = 0, a; y = 0, b.

B. All Edges Clamped (CCCC):

u 1 = v1 = w1 = u 2 = v2 = w2 = u 3 = v3 = w3 = u 4 = v4 = w4 = 0 x = 0, a; y = 0, b.

6 Material Properties

The following materials are used in this study:

AS4 E 11 f E 22 f G 12 f G 23 f ν12 f Em νm
Material-1 234.6 13.8 13.8 5.5 (GPa) 0.2 2.94 0.35
(GPa) (GPa) (GPa) (GPa)
Material-2 230.0 40.0 24.3 14.3 0.35 2.94 0.35
(GPa) (GPa) (GPa) (GPa) (GPa)

Material-3: E 1 /E 2 = 25, G12 = G13 = 0.5E 2 , G23 = 0.2E 2 , ν12 = ν13 = ν23 =
0.25.

7 Results and Discussions

The present free vibration analysis of 3D braided composite spherical shell panels is
carried out by the eight-noded isoparametric finite element method in a MATLAB
environment. The accuracy and efficacy of the present finite element model are vali-
dated by comparing the obtained results with the available literature results and are
482 S. Lore et al.

√
Table 1 Variation of non-dimensional fundamental frequencies 𝛡 = (ωb2 / h) ρ/E 2 of a simply
supported laminated spherical shell with a/b = 1.0, a/h = 100, and Material-3
Ply Sequence Source R/a
5 10 20 50 100
[0° /90° ] Present 28.8364 16.7201 11.8601 10.0849 9.8050
J. Reddy [3] 28.840 16.710 11.840 10.06 9.784
[0° /90° /0° ] Present 31.0096 20.3688 16.6532 15.4514 15.2720
J. Reddy [3] 31.020 20.350 16.62 15.42 15.2117

√
Table 2 The NDFF (𝛡 = ωa 2 / h 12ρ f /E 22 f ) of the simply supported 3D braided composite
plate, Material-1
Volume fraction Braid angle (α) Yang and Haung Singh and Singh [2] Present FEM
(V f ) [20]
0.4 20 16.272 16.1892 15.6445
30 16.237 16.1464 15.3837
0.45 20 16.523 16.4816 16.5481
30 16.452 16.3748 15.9131

presented in Tables 1 and 2. Table 1 demonstrates the results of laminated spher-

ical shells for different lamina configurations and radius of curvatures with simply
supported all edges. The confirmation of the accuracy of non-dimensional funda-
mental frequencies (NDFF) of the 3D braided composite SSSS plate with the results
of Yang and Haung [20] and Singh and Singh [2] is presented in Table 2. It is clear
that the results are well converged with a mesh size of (8 × 8). The present numerical
results are generated
√ using material models mentioned above, and NDFFs are given
as 𝛡 = ωa 2 / h 12ρ f /E 22 f .

7.1 Effect of Braided Fiber Volume Fraction (Vf )

and Curvature Ratios (r/a) on Natural Frequency

The non-dimensional fundamental frequencies (NDFFs) of spherical shells are

analyzed by varying braided fiber volume fractions with SSSS and CCCC boundary
conditions (BCs) and illustrated in Fig. 3. The five different fiber volume fractions
are employed for the analysis, i.e., 0.30, 0.35, 0.40, 0.45, and 0.50. It is observed that
the natural frequency of the spherical shells increases with the increase in braided
volume fractions, irrespective of BCs. The NDFF values of the spherical shells always
enhance with the enhancement in fiber volume fraction (V f ). The equivalent elastic
stiffness of spherical panels always boosts with the increase in V f , because fibers have
higher elastic stiffness compared to matrix. It is obvious that the natural frequency
decreases due to an increase in curvature ratios (R/a). In addition, the percentage
Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 483

10 16
SSSS V f=0.30
CCCC V f =0.30
a/h=10 V f=0.35 a/h=10 V f =0.35
V f=0.40 14 V f =0.40

Natural frequency
Natural frequency

V f=0.45 V f =0.45
8
V f=0.50 V f =0.50
12

6 10

8
4
1 2 3 4 5 1 2 3 4 5
R/a R/a
(a) (b)

Fig. 3 Variation of NDFF of 3D braided spherical shells against curvature with different values of
Vf

Table 3 The percentage decrease of NDFF of spherical shells due to the increase in curvature with
braided volume fractions
BCs Rx /a = Ry /a = R/a Vf
0.30 0.35 0.40 0.45 0.50
SSSS 2.0 28.5571 28.3404 28.2031 28.1376 28.1367
3.0 35.7766 35.5047 35.3355 35.2588 35.2655
4.0 38.5479 38.2535 38.0715 37.9907 38.0011
5.0 39.8819 39.5763 39.3880 39.3052 39.3174
CCCC 2.0 27.3970 27.3349 27.2833 27.2402 27.2033
3.0 33.9851 33.9021 33.8339 33.7779 33.7310
4.0 36.4721 36.3804 36.3053 36.2440 36.1930
5.0 37.6604 37.5643 37.4858 37.4218 37.3689

decrease of NDFF values of 3D braided composite spherical shells is shown in

Table 3. The highest percentage decrease in natural frequency is identified at the
lowest values of V f . Its more pronounced effect is identified with SSSS boundary
conditions.

7.2 Effect of Braid Angle (α) and Boundary

√ Conditions
on Natural Frequency (𝛡 = ωa2 / h ρ f /E22 f )

The non-dimensional fundamental frequency (NDFF) results of spherical shells are

obtained by varying braided angle (α) with SSSS and CCCC boundary conditions
(BCs) and presented in Table 4. For this analysis, four sets of braided angles are
considered, i.e., 20° , 25° , 30° , and 35° , respectively. The equivalent elastic properties
484 S. Lore et al.

of the 3D braided spherical shells decrease with the increase in braid angle. So, the
stiffness of the panels decreases, and the natural frequency also decreases. From
the results, it is clear that the frequency parameters decrease due to an increase in
curvature ratios (R/a). The NDFFs of the spherical shell panels are higher with CCCC
BCs in comparison with SSSS BCs, because CCCC BCs are stiffer in comparison
with the SSSS-supported BSs. In addition, the percentage decrease of NDFF values
of 3D braided composite spherical shells is shown in Table 5. The highest percentage
decrease in natural frequency is identified at lower values of α. Its more pronounced
effect is identified with CCCC boundary conditions.

Table 4 The NDFF of spherical shells due to the increase in braided angles by varying curvatures
Boundary conditions R/a α
20 25 30 35
SSSS 1.0 8.7627 9.0282 9.1759 9.2019
2.0 6.4669 6.4479 6.4033 6.3573
3.0 5.8997 5.8450 5.7559 5.6301
4.0 5.6842 5.5983 5.4868 5.3489
5.0 5.5810 5.4796 5.3570 5.2131
CCCC 1.0 15.7697 14.4885 13.3213 12.3764
2.0 11.2936 10.5418 9.8173 9.2146
3.0 10.1918 9.5946 8.9915 8.4793
4.0 9.7725 9.2373 8.6818 8.2048
5.0 9.5716 9.0666 8.5344 8.0743

Table 5 The percentage decrease in NDFF of spherical shells in the increase in braided angles by
varying curvatures
Boundary conditions R/a α
20 25 30 35
SSSS 2.0 26.1999 25.0960 23.5510 22.4904
3.0 32.6721 32.1000 31.2798 31.3566
4.0 35.1316 34.9652 34.4925 34.7843
5.0 36.3097 36.3439 36.0429 36.4405
CCCC 2.0 28.3838 27.2402 26.3036 25.5468
3.0 35.3711 33.7779 32.5030 31.4886
4.0 38.0295 36.2440 34.8273 33.7064
5.0 39.3040 37.4218 35.9343 34.7606
Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 485

7.5 7.5

6.0 SSSS, a/h=10, 6.0 SSSS,a/b=1,

R/a=5 a/h=10,R/a=5

4.5 4.5

NDFF
NDFF

3.0 3.0

1.5 1.5

0.0 0.0
0.30 0.35 0.40 0.45 0.50 20 25 30 35
Volume fraction Braided Angle
(a) ( b)

Fig. 4 Variation of NDFF of 3D braided twisted spherical shell against a volume fractions (V f )
and b braid angle (α)

7.3 Effect of Twist√Angle (ϕ) on Natural Frequency

(𝛡 = ωa2 / h ρ f /E22 f )

In Fig. 4, the NDFF results of spherical shells are plotted against braided volume
fractions and braided angles by varying twist angles (ϕ). It is noticed that the NDFF
parameters decrease due to the increase in twist angle (ϕ) because the higher twist
angle (ϕ) provides a less stiff structure. It is also observed that the highest percentage
decrease in NDFF of 3D braided spherical shells due to a rise in ϕ is identified at the
higher volume fractions.

7.4 Effect of Aspect

√ Ratios on Natural Frequency
(𝛡 = ωa2 / h ρ f /E22 f )

The influence of aspect ratio (a/b) on the NDFF of 3D braided spherical shells
obtained by varying braided angles (α) with SSSS and CCCC boundary conditions
(BCs) is represented in Table 6. The aspect ratio is the structural parameter. It is
observed that the NDFF values of spherical shell panels increase due to the increase
in aspect ratio (a/b). The more pronounced effect of the a/b ratio is identified in
CCCC BCs. It is also observed that the NDFF values of spherical panels increase
due to an increase in thickness ratios (a/h). The more pronounced effect of a/h ratios
is identified in CCCC BCs.
486 S. Lore et al.

Table 6 The NDFF of spherical shells in the increase in aspect ratios (a/b) and thickness ratios (a/
h)
BCs α Aspect ratios and thickness ratios (a/h)
a/b = 0.5,h = 0.1 a/b = 1.0; h = 0.15 a/b = 1.5; h = 0.20 a/b = 2.0; h = 0.25
SSSS 20 4.1613 5.4396 7.3543 9.8087
25 3.7859 5.3165 7.2578 9.7130
CCCC 20 8.1531 9.2961 11.7318 15.3869
25 7.4000 8.8335 11.6865 15.2315

8 Conclusions

In this study, an analysis of the free vibrations of 3D braided spherical shells is

presented, employing the third-order shear deformation theory (TSDT) in conjunc-
tion with an isoparametric finite element method. Rigorous evaluations were
performed to validate and ascertain the accuracy and efficacy of the finite element
models utilized. The findings of this study yield the following significant conclusions:
1. The investigation revealed that the natural frequency of spherical shells demon-
strates an increase as the braided volume fractions are raised. However, in similar
experimental conditions, the Normalized Dynamic Flexibility Factor (NDFF)
values decrease due to an escalation in braided angles.
2. The analysis demonstrated a decrease in natural frequencies as the curvature
ratios (R/a) increased.
3. The NDFF parameters exhibited a decrease as the twist angle (ϕ) increased.
4. A rise in aspect ratio (a/b) resulted in increased NDFF values for spherical panels.
The most notable effect of a/b ratios was observed in cases with Clamped–
Clamped–Clamped–Clamped (CCCC) boundary conditions.
5. The fundamental frequency experienced an increase with higher a/h values
(length-to-thickness ratios).
6. The NDFF values consistently displayed higher magnitudes for spherical shell
panels with CCCC boundary conditions. This outcome can be attributed to
the stiffer nature of CCCC supports compared to Simply Supported–Simply
Supported–Simply Supported–Simply Supported (SSSS) boundary conditions.

Acknowledgements The authors are delighted in ISRO for providing economic support.
Modal Analysis of 3D Braided Composite Reinforced Twisted Spherical … 487

Appendix
⎡ ⎤
1 0 0 Z 0 0 Z2 0 0 Z3 0 0
[Tc ] = ⎣ 0 1 0 0 Z 0 0 Z 2 0 0 Z 3 0 ⎦
0 0 1 0 0 Z 0 0 Z2 0 0 Z3
( ) ( ) ( ) ( )
∂u ∂v
εx x = ∂x
+ Rwx ,ε yy = ∂y
+ Rwy ,ε zz = ∂w ∂z
,ε yz = ∂w
∂y
− Rvy +
( ∂v ) ( ) ( ∂u ) ( ) ( ){ }
∂w ∂u ∂v T
∂z
,ε x z = ∂x
− u
Rx
+ ∂z
,ε x y = ∂y
+ w
Rx y
+ ∂x
+ w
Rx y
, ε0 =
{ }T { } { }T { }
T T
εx0 x ε0yy εzz
0
ε0yz εx0z εx0y , k 1 = k x1x k 1yy k zz
1
k 1yz k x1z k x1y , k 2 =
{ }T { } { }T
T
k x2x k 2yy k zz
2
k 2yz k x2z k x2y , k 3 = k x3x k 3yy 0 k 3yz k x3z k x3y .

References

1. Maji P, Singh BN, Singh DB. A third-order polynomial for the free vibration response of
3D braided curved panels using various boundary conditions Mech Based Des Struct Mach.
2021;1–23. https://doi.org/10.1080/15397734.2021.1894574
2. Singh DB, Singh BN. New higher order shear deformation theories for free vibration and buck-
ling analysis of laminated and braided composite plates. Int J Mech Sci. 2017;131–132:265–77.
https://doi.org/10.1016/j.ijmecsci.2017.06.053.
3. Reddy JN. Exact solutions of moderately thick laminated shells. J Eng Mech. 1984;110(5):794–
809. https://doi.org/10.1061/(asce)0733-9399(1984)110:5(794).
4. Li Z-M, Wang M. Large-amplitude vibration analysis of 3D braided composite cylindrical
shells in an elastic medium. J Aerosp Eng. 2016;29(1):04015029. https://doi.org/10.1061/(asc
e)as.1943-5525.0000515.
5. Zhao YX, Liu T, Li ZM. Nonlinear bending analysis of a 3D braided composite cylindrical panel
subjected to transverse loads in thermal environments. Chinese J Aeronaut. 2018;31(8):1716–
27. https://doi.org/10.1016/j.cja.2018.03.022.
6. Singh DB, Singh BN. Buckling analysis of three dimensional braided composite plates
under uniaxial loading using inverse hyperbolic shear deformation theory. Compos Struct.
2016;157:360–5. https://doi.org/10.1016/j.compstruct.2016.08.029.
7. Javed S, Al Mukahal FHH, Salama MA. Free vibration analysis of composite conical shells
with variable thickness. Shock Vib. 2020;2020. https://doi.org/10.1155/2020/4028607
8. Li H, Pang F, Gong Q, Teng Y. Free vibration analysis of axisymmetric functionally graded
doubly-curved shells with un-uniform thickness distribution based on Ritz method. Compos
Struct. 2019;225:111145. https://doi.org/10.1016/j.compstruct.2019.111145
9. Malekzadeh Fard K, Baghestani AM. Free vibration analysis of deep doubly curved open
shells using the Ritz method. Aerosp Sci Technol. 2017;69:136–148. https://doi.org/10.1016/
j.ast.2017.06.021
10. Sun H, Di S, Zhang N, Pan N, Wu C. Micromechanics of braided composites via multivariable
FEM. Comput Struct. 2003;81(20):2021–7. https://doi.org/10.1016/S0045-7949(03)00228-1.
11. Maji P, Singh BN. Free vibration responses of 3D braided rotating cylindrical shells based
on third-order shear deformation. Compos Struct. 2021;260:113255. https://doi.org/10.1016/
j.compstruct.2020.113255
12. Shokrieh MM, Mazloomi MS. An analytical method for calculating stiffness of two-
dimensional tri-axial braided composites. Compos Struct. 2010;92(12):2901–5. https://doi.
org/10.1016/j.compstruct.2010.04.016.
488 S. Lore et al.

13. Shokrieh MM, Mazloomi MS. A new analytical model for calculation of stiffness of three-
dimensional four-directional braided composites. Compos Struct. 2012;94(3):1005–15. https://
doi.org/10.1016/j.compstruct.2011.09.010.
14. Gao X, Yuan L, Fu Y, Yao X, Yang H. Prediction of mechanical properties on 3D braided
composites with void defects. Compos Part B Eng. 2020;197:108164. https://doi.org/10.1016/
j.compositesb.2020.108164
15. Ge J, He C, Liang J, Chen Y, Fang D. A coupled elastic-plastic damage model for the mechanical
behavior of three-dimensional (3D) braided composites. Compos Sci Technol. 2018;157:86–98.
https://doi.org/10.1016/j.compscitech.2018.01.027.
16. Wang YQ, Wang ASD. Microstructure/property relationships in three-dimensionally braided
fiber composites. Compos Sci Technol. 1995;53(2):213–22. https://doi.org/10.1016/0266-353
8(95)00021-6.
17. Sun X, Sun C. Mechanical properties of three-dimensional braided composites. Compos Struct.
2004;65(3–4):485–92. https://doi.org/10.1016/j.compstruct.2003.12.009.
18. Han WF, Li DS, Jiang L. Mechanical properties and failure mechanisms of 3D six-directional
braided composites at elevated and liquid nitrogen temperatures. Mater Lett X. 2020;6:100035.
https://doi.org/10.1016/j.mlblux.2019.100035
19. Zuorong C, Dechao Z, Lu M, Ye L. Evaluation of elastic properties of 3-D (4-step) regular
braided composites by a homogenisation method. Compos Struct. 1999;47(1–4):477–82.
https://doi.org/10.1016/S0263-8223(00)00026-X.
20. Yang J, Huang X. Dynamic stability behavior of 3D braided composite plates integrated with
piezoelectric layers. J Compos Mater. 2009;43(20):2223–38. https://doi.org/10.1177/002199
8309339219.
Dynamic Analysis of Flexible-Joint
and Single Flexible-Link Manipulator
by Using Finite Element Analysis

Rajesh Ranjan, Saleendra Hari Babu, and Santosha Kumar Dwivedy

Abstract To improve productivity by operating the robotic manipulators at a higher

speed, one may use a flexible-link manipulator. Sometimes, the joints of the manip-
ulator can also be considered flexible. However, these flexible links and joints create
unwanted vibration which reduces the efficacy of the manipulators. Due to this,
many researchers are working to study the dynamics of these systems to reduce
the problem of vibration in the systems. The present work focuses on the dynamic
modeling and modal analysis of a cantilever beam-based curved flexible single link
and joint manipulator. The analyses are carried out by using finite element analysis
software ANSYS. The effects of joint and link flexibility on the system frequencies
have been investigated. In the analyses, rectangular and tapered cross-sections for the
links are considered. Furthermore, the findings of this study are compared to those
of previous experimental studies published in the literature.

Keywords Curved flexible link · Flexible joint · Mode shapes · Natural

frequencies

Nomenclature

a Width of the tapered link at the left end

b Width of the tapered link at the right end
B Width of the link
E Young’s modules

R. Ranjan (B) · S. H. Babu

Department of Mechanical Engineering, GITAM Deemed to Be University,
Bangalore PIN-561203, India
S. H. Babu
e-mail: [email protected]
S. K. Dwivedy
Department of Mechanical Engineering, IIT Guwahati, Assam 781039, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 489
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_33
490 R. Ranjan et al.

H Thickness of the link

I Moment of inertia
Jr Rotor inertia
Ja Payload of mass moment of inertia
K Joint stiffness
L Length of the flexible link
Mp Payload mass
N Shape function
r Radius of the hole
τ0 External torque
α Angle of link
ρ Density of material

1 Introduction

In recent years, the research on lightweight flexible manipulators has seen height-
ened interest in the scientific community. The flexible manipulators have the advan-
tages of higher operational speed, large working volume with greater payload-to-
manipulator-weight ratio, and low cost. However, they suffer from their inherent
flexibilities that need to be tackled in order to effectively utilize their potential advan-
tages. The link and joint flexibilities are the two main reasons for the vibration of
light manipulators, which can be proved costly to neglect them in dynamic modeling
of the manipulator. A common method of dynamic modeling of a flexible robot
manipulator for controller design is to use the Euler–Lagrange formulation with the
assumed mode method, which gives the dynamic model of the manipulator in closed
form. Most of the studies of flexible-link manipulators (FLMs) in the literature have
considered the flexible links of manipulators as the Euler–Bernoulli beam.
Extensive research has been carried out on the modeling and the control of flexible
manipulators. Dwivedy and Eberhard [1] reviewed the literature for the period 1974
to 2005 on dynamic analysis of flexible manipulators and classified them based on
the study of manipulators with a different number of flexible links, methods used for
dynamic analysis, and the control techniques applied. Kiang et al. [2] did a survey of
the control techniques used in controlling the flexible manipulators during the period
from 1990 to 2002. Ankarali and Diken [3] did the vibration control of an elastic
link of a manipulator. Wang and Guan [4] examined the effects of different payload
masses on the fundamental frequencies of flexible manipulators. They found that
shear deformation and rotary inertia of the flexible link don’t have much effect on
the fundamental frequency of vibration. Wang and Russell [5, 6] applied the shape
optimization techniques for vibration control of a flexible manipulator. KhalilIbrahim
et al. [7] did the mode shape analysis of a single-link flexible manipulator. Shin and
Brennan [8] applied control algorithms for residual vibration control of flexible
manipulators based on the shock response spectrum. Korayem and Nikoobin [9],
Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 491

Malgaca and Karagülle [10], and Malgaca et al. [11] used the finite element method
for residual vibration control of flexible-link flexible-joint manipulators. De Luca
and Siciliano [12] did the trajectory control of a single-link flexible manipulator.
Abe [13] used the particle swarm optimization technique for trajectory planning for
the residual vibration control of a two-link rigid–flexible manipulator. The flexible
joint of the manipulator was modeled as a linear spring by Spong [14]. Wei et al.
[15] applied the optimal trajectory planning for eliminating the residual vibrations
of a rigid–flexible two-link manipulator with a flexible joint.
The main objective of the present work is to carry out the dynamic analysis
of different shapes of flexible links with or without flexible joints using the finite
element method. Here, five types of manipulators have been considered, viz., (1)
rigid-joint and flexible-link manipulator with a rectangular cross-section, (2) rigid-
joint and flexible-link manipulator with rectangular holes, (3) rigid-joint and tapered
flexible-link manipulator, (4) single flexible-link flexible-joint manipulator, and (5)
curved flexible-fink and rigid-joint manipulator with circular holes. In these works,
the modal frequencies of the manipulators have been found out using the ANSYS
software package.

2 Modeling

Figure 1 shows the model of a rigid-joint and flexible-link manipulator. Initially,

this 3D model is developed by using SOLIDWORKS software. Here, the parameters
considered for the flexible link are (L × H × B) = (0.5 × 0.05 × 0.003 m) and
the initial curvature angle is θ = 300 , mass per unit length is 0.403 kg/m, bending
stiffness (EI = 7.85 N.m2 ), and tip payload mass Mp = 0.5 kg. Figure 2 shows the
SOLIDWORKS model of the rigid-joint and flexible-link manipulator with circular
holes. Here, the hole diameter is d = 10 mm, and the remaining parameters are the
same as in Fig. 1. Figure 3 shows the rigid joint and tapered flexible-link manipulator
having a = 0.05 m at the left end and b = 0.025 m at the right end, and the remaining
parameters are the same as in the first case. Figure 4 shows an ANSYS model of
a curved flexible-link and rigid-joint manipulator with circular holes, and physical
parameters for this manipulator are given in Table 1. Figure 5 shows a schematic
diagram of a flexible-joint and flexible-link manipulator. Here parameters considered
are flexible link length (L = 0.5 m), width of the planar link (B = 0.03 m), thickness
of the link (H = 0.0015 m), density of the material of the link (7800 kg/m3 ), Young’s
modulus (E = 200 GPa), and joint stiffness (torsional spring constant K s = 750 N.m/
rad). It may be noted that the flexible joint is modeled as a torsional spring with
stiffness K s as shown in Fig. 5.
Finite element modeling has been carried out in ANSYS by exporting the SOLID-
WORKS model (Figs. 1, 2 and 3) for the first 3 cases and in the last 2 cases, the
models have been developed in ANSYS itself. Three-dimensional 4-noded tetra-
hedron solid structural elements have been considered in these models. The total
number of elements and nodes used in the 5 different cases is given in Table 2. In
492 R. Ranjan et al.

Fig. 1 SOLIDWORKS
model of rigid-joint and
flexible-link manipulator
with a rectangular
cross-section

Fig. 2 Rigid-joint and

flexible-link manipulator
with circular holes

Fig. 3 Rigid-joint and

tapered flexible-link
manipulator

b
a

Fig. 4 ANSYS model of

curved flexible-link and
rigid-joint manipulator with
circular holes
Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 493

Table 1 Physical parameters of curved flexible-link and rigid-joint manipulator with circular holes
S. No. Description Curved manipulator
1 Young’s modulus of elasticity E = 193 GPa
2 Density ρ = 7950 kg/m3
3 Length L = 520 mm
4 Distance of the payload from the end point d payload = 20 mm
5 Distance of the accelerometer center from the center of the d sensor = 70 mm
payload
6 Thickness h = 4 mm
7 Number of finite elements ne = 2229
8 Rayleigh damping coefficient β = 3.8 × 10–4
9 Weight of the sensor M s = 0.054 kg
10 Weight of the payload M L = 0.62 kg
11 Motor rotational spring constants K m2 = 16,000 Nm/rad
12 Time step Δt = 0.005 s

Fig. 5 Schematic diagram Y Payload

of flexible-joint and
flexible-link manipulator
X1

Y1

Ks
X

the first 4 cases, the left end of the manipulator is considered to be clamped and the
other end is considered to have tip mass. In the fifth case, the left end is provided
with a torsional spring and the other end is similar to the previous cases.
494 R. Ranjan et al.

Table 2 Mesh-generated total elements and total nodes

Case Modal name Elements Nodes
1 Rigid-joint and flexible-link manipulator with a rectangular 658 1569
cross-section
2 Rigid-joint and flexible-link with holes manipulator system 877 2058
3 Rigid-joint and tapered flexible-link manipulator system 558 2032
4 Flexible-joint and flexible-link manipulator system 1378 5292
5 Curved flexible-link and rigid-joint manipulator with circular 528 2008
holes

3 Results and Discussion

3.1 Dynamic Analysis of Flexible-Link and Rigid-Joint

Manipulator

Modal analysis has been carried out in ANSYS and the mode shapes for the flexible
link with rigid-joint manipulators are shown in Figs. 6 and 7 for 0 kg (no payload
condition) and 0.5 kg payload, respectively. Here, the first 2 modes have been consid-
ered. The first 3 modal frequencies are given in Table 3 and are compared with those
values available in the literature (KhalilIbrahim et al. [7]). The difference in the
modal frequencies is found to be very less. It is clearly observed that by increasing
the payload the modal frequencies are decreasing.
Now by making holes in the flexible links, Fig. 8 shows the first two modes of
the manipulators with rigid joints. The modal frequencies are found to be 9.72 and
58.38 Hz which are more than those found in the first case (i.e., without holes).
So by appropriately incorporating holes in the manipulator, one may control the tip
deflection.

Fig. 6 Rectangular flexible-link and rigid-joint mode shapes of Mp = 0 kg

Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 495

Fig. 7 Mode shapes of rectangular flexible-link and rigid-joint manipulator with Mp = 0.5 kg

Table 3 Modal frequencies of manipulator of rectangular flexible-link and rigid-joint with various
payloads
Mode Shape Mode frequencies (Hz)
no Mp = 0 kg Mp = 0.25 kg Mp = 0.5 kg
[7] Present work [7] Present work [7] Present work
1 7.97 5.95 6.92 5.29 6.31 5.15
2 23.973 22.38 18.35 18.44 15.17 17.716
3 64.501 63.201 49.37 51.86 44.76 45.1

Fig. 8 Mode shapes of rectangular flexible-link and rigid-joint manipulator with holes and Mp =
0 kg

Figure 9 shows the first 2 mode shapes of the system shown in Fig. 3 where the
link is considered to be tapered. Here, the modal frequencies are found to be 12.635
and 67.337 Hz which are more than those found in the first two cases. These values
are found to be more than those of the first 2 cases.
496 R. Ranjan et al.

Fig. 9 Mode shapes of tapered flexible-link manipulator with Mp = 0 kg

Table 4 Comparison of modal frequencies of flexible-link and rigid-joint manipulator with other
different shapes of flexible links
Model no. Modal frequency (Hz)
Mp = 0 kg Mp = 0.5 kg
Case 1 Case 2 Case 3 Case 1 Case 2 Case 3
1 5.95 9.728 12.635 5.15 5.3538 5.8175
2 22.38 58.384 67.337 17.716 29.624 29.816
3 63.201 71.432 91.793 45.1 47.596 49.351

Table 4 gives the comparison of the first 3 modal frequencies of the above-
mentioned 3 cases of flexible-link and rigid-joint manipulators. One may choose
the manipulators based on applications and kinematic constraints. The manipulators
must operate away from the modal frequencies to avoid excessive vibration due to
resonance.

3.2 Dynamic Analysis of Flexible-Joint and Flexible-Link

Manipulator

The manipulator shown in Fig. 5 is analyzed in this subsection. FEA simulation

has been carried out on a single flexible-link and flexible-joint manipulator with the
joint torsional stiffness K s = 750 Nm/rad. In Fig. 10, the first mode of vibrations
of the flexible-link flexible-joint manipulator has been shown while in Fig. 11, the
second mode of vibrations of the manipulator has been shown. In each of these cases,
frequencies of the flexible-link and flexible-joint manipulators are calculated from
their equilibrium position.
The frequencies for the first three modes of vibrations of the flexible-link and
flexible-joint manipulator have been given in Table 5.
Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 497

Fig. 10 First mode of the flexible-link and flexible-joint manipulator

Fig. 11 Second mode of the flexible-link and flexible-joint manipulator

Table 5 Modal frequencies

Mode of vibration Frequency (Hz)
of vibrations of a single
flexible-link flexible-joint First mode 4.4905
manipulator Second mode 19.955
Third mode 28.484
498 R. Ranjan et al.

3.3 Dynamic Analysis of Curved Flexible-Link

and Rigid-Joint Manipulator with Circular Holes

Simulation has been carried out using curved flexible links and rigid joints with
circular holes. Different mode shapes are shown in the figures below. In each of
these cases, the natural frequencies of the flexible link tip are calculated from its
equilibrium position. Figures 12 and 13 are mode shapes of curved flexible links and
rigid joints with circular holes of steel material and aluminum alloy.
The first three modal frequencies with a variety of materials are given in Table 6.
It is observed that aluminum alloy material gives higher natural frequencies than
steel material.

Fig. 12 First and second modes of curved flexible-link and rigid-joint manipulator with circular
holes

Fig. 13 First and second modes of curved flexible-link and flexible-joint manipulator with circular
holes

Table 6 Mode frequencies of curved flexible link and rigid joint with circular holes
Mode shape no. Mode frequencies (Hz)
Experimental [11] Present work Simulation of AL Alloy
1 6.251 6.34 6.5131
2 54.590 55.083 56.598
3 87.390 89.155 91.185
Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 499

3.4 Dynamic Analysis of Curved Flexible-Link

and Rigid-Joint Manipulator with Different Types
of Holes

In this subsection, the curved manipulator shown in Fig. 4 is analyzed using ANSYS.
Four different types of holes, viz., circular (Fig. 13), hexagonal (Fig. 14), pentagonal
(Fig. 15), and rectangular (Fig. 16) have been considered, and the first two modes
have been shown in the figures. Table 7 gives the modal frequencies of the first 4
modes. The results of the circular holes have been compared with those available in
the literature [11] and are found to be in good agreement.
Simulation of dynamic analysis of flexible-link manipulator has been carried out
using curved flexible-link and rigid-joint manipulator with hexagonal holes. Different
mode shapes are shown in the figures below. In each of these cases, the natural
frequencies of the flexible link are calculated from its equilibrium position. Figure 14
shows the mode shapes of a curved flexible-link and rigid-joints manipulator with
hexagonal holes of steel material. Figure 15 shows the mode shapes of a curved
flexible-link and rigid-joints manipulator with pentagonal holes of steel material.

Fig. 14 First and second modes of curved flexible-link and rigid-joint manipulator with hexagonal
holes

Fig. 15 First and second modes of curved flexible-link and rigid-joint manipulator with pentagonal
holes
500 R. Ranjan et al.

Fig. 16 First and second modes of curved flexible-link and rigid-joint manipulator with rectangular
holes

Table 7 Comparison of modal natural frequencies of steel manipulator with different types of holes
Mode shape no. Mode frequencies (Hz)
Circular holes [11] Hexagonal holes Pentagonal holes Rectangular holes
1 6.34 6.2778 6.2106 6.5004
2 55.083 54.153 54.245 56.281
3 89.155 87.719 87.128 90.744
4 142.9 141.68 140.7 146.06

Figure 16 shows the mode shapes of a curved flexible-link and rigid-joint manipulator
with rectangular holes of steel material.
The first four modal frequencies with the variation of a type of holes are given in
Table 7 of steel material and aluminum alloy. It has been observed that rectangular
holes give higher natural frequencies than other types of holes.

4 Conclusion

In the present paper, the modal analysis of different types of flexible-link manipulators
mentioned in Sect. 3 with or without joint flexibilities has been performed in the
commercial finite element software package ANSYS. The following conclusions
can be drawn from the above analyses:
. The natural frequencies of the first three modes are greater in tapered flexible-link
manipulators with rigid joints than in rectangular flexible-link manipulators with
holes and rigid joints.
. It is noticed that the manipulator of aluminum alloy material gives greater natural
frequencies than that of steel material in the first three modal frequencies of curved
flexible-link and rigid-joint manipulators with circular holes.
Dynamic Analysis of Flexible-Joint and Single Flexible-Link … 501

. In the first four modal frequencies of the curved flexible-link manipulators with
a variety of types of holes, the one with rectangular holes has been found to have
higher natural frequencies than with other types of holes.

References

1. Dwivedy SK, Eberhard P. Dynamic analysis of flexible manipulators, a literature review. Mech
Mach Theory. 2006;41(7):749–77.
2. Kiang CT, Spowage A, Yoong CK. Review of control and sensor system of flexible manipulator.
J Intell Rob Syst. 2015;77(1):187–213.
3. Ankarali A, Diken H. Vibration control of an elastic manipulator link. J Sound Vib.
1997;204(1):162–70.
4. Wang FY, Guan G. Influences of rotatory inertia, shear and loading on vibrations of flexible
manipulators. J Sound Vib. 1994;171(4):433–52.
5. Wang FY, Russell JL. Optimum shape construction of flexible manipulators with total weight
constraint. IEEE Trans Syst Man Cybern. 1995;25(4):605–14.
6. Wang FY, Russell JL. A new approach to optimum flexible link design. In: Proceedings of
1995 IEEE international conference on robotics and automation, vol 1. IEEE, p. 931–6
7. KhalilIbrahim AAAB, Ismail AA. Mode shape analysis of a flexible robot arm. Int J Control
Autom Syst 1(2)
8. Shin K, Brennan MJ. Two simple methods to suppress the residual vibrations of a translating
or rotating flexible cantilever beam. J Sound Vib. 2008;312(1–2):140–50.
9. Korayem MH, Nikoobin A. Maximum payload for flexible joint manipulator in point-to-point
task using optimal control approach. Int J Adv Manuf Technol. 2008;38(9–10):1045–60.
10. Malgaca L, Karagülle H. Numerical and experimental study on integration of control actions
into the finite element solutions in smart structures. Shock Vib. 2009;16(4):401–15.
11. Malgaca L, Yavuz Ş, Akdağ M, Karagülle H. Residual vibration control of a single-link flexible
curved manipulator. Simul Model Pract Theory. 2016;67:155–70.
12. De Luca A, Siciliano B. Inversion-based nonlinear control of robot arms with flexible links. J
Guid Control Dyn. 1993;16(6):1169–76.
13. Abe A. Trajectory planning for residual vibration suppression of a two-link rigid-flexible
manipulator considering large deformation. Mech Mach Theory. 2009;44(9):1627–39.
14. Spong MW. Modeling and control of elastic joint robots. J Dyn Syst Meas Contr. 1987;109:310–
8. https://doi.org/10.1115/1.3143860.
15. Wei J, Cao D, Liu L, Huang W. Global mode method for dynamic modeling of a flexible-link
flexible-joint manipulator with tip mass. Appl Math Model. 2017;48:787–805.
Prediction of Fatigue Crack Growth
Behavior Under Cold-Turbistan
Spectrum Loads Using Variable Crack
Closure Approach in GTM 720 Nickel
Base Superalloy

Sharanagouda G. Malipatil, Sandeep Kumar, Anuradha N. Majila,

D. Chandru Fernando, M. Manjuprasad, and C. M. Manjunatha

Abstract In this investigation, the fatigue crack growth, FCG behavior of a

nickel base superalloy GTM720 under a standard cold-TURBISTAN spectrum load
sequence was predicted and compared with existing experimental data. For the
purpose of prediction, fatigue crack growth law of the alloy GTM720 and the crack
closure level at various stress ratios determined in an earlier investigation were
employed. Empirical equations were derived for closure estimation at any other
given stress ratios and load amplitudes. Using the derived crack growth law and crack
closure values, the crack propagation behavior under spectrum load was predicted.
The crack closure value for each load cycle was considered to vary unlike the use
of constant crack closure value for entire load block. Predicted result was conser-
vative and reasonably good but accuracy of prediction could further be enhanced

S. G. Malipatil (B) · C. M. Manjunatha

Structural Integrity Division, CSIR-National Aerospace Laboratories, Bangalore 560 017, India
e-mail: [email protected]
C. M. Manjunatha
e-mail: [email protected]
S. G. Malipatil · S. Kumar
Academy of Scientific and Innovative Research (AcSIR), Ghaziabad 201 002, India
e-mail: [email protected]
S. Kumar · M. Manjuprasad
Structural Technologies Division, CSIR-National Aerospace Laboratories, Bangalore 560 017,
India
e-mail: [email protected]
A. N. Majila · D. C. Fernando
Materials Group, Gas Turbine Research Establishment (GTRE), DRDO, Bangalore 560 093, India
e-mail: [email protected]
D. C. Fernando
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 503
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_34
504 S. G. Malipatil et al.

by considering load-interaction effects on crack closure level values for each cycle
during prediction.

Keywords Fatigue crack growth · Crack closure · Cold-TURBISTAN · GTM720

1 Introduction

In gas turbine aero-engine components like blades and disks, nickel-based superal-
loys are frequently employed. Such components are subjected to variable amplitude
fatigue loads in service. Initiation and propagation of cracks under spectrum fatigue
loads may cause premature fractures. Hence, it is necessary to evaluate the damage
tolerance capability of materials. Such requirement calls for an accurate prediction
of fatigue crack propagation in these materials under spectrum loads.
The crack driving force for the fatigue crack growth rate (FCGR), da/dN of a mate-
rial is the applied stress intensity factor range, ∆K and the stress ratio, R = σmin /σmax .
The crack growth law, da/dN = f(∆K, R) [1, 2] derived from the constant amplitude
FCGR behavior of the material is employed to estimate the growth behavior under
spectrum loads. The general methodology for prediction thus involves (i) fatigue
load cycle counting in the spectrum, and (ii) determine crack length extension for all
counted load cycles using crack growth law.
The phenomenon of crack closure [3] has been shown to alter the applied SIF
range, ∆K = (Kmax −Kmin ) to ∆Keff = (Kmax −Kcl ), where Kcl is closure level,
Kcl = f(∆K, R). The experimental determination of closure level is quite difficult in
service. Crack closure level is expected to vary for every cycle in the spectrum [4]. In
order to simplify prediction procedures, a constant crack closure value for the entire
spectrum is assumed in the past [5, 6]. Although it has yielded fairly good results,
it is observed that it cannot be used as a generalized methodology for all types of
spectrum loads. Eliminating/nullifying the effect of stress ratios, merging all curves
into a single narrow band, and estimating a specific fatigue crack growth law, that
is appropriate for any stress ratio are all important for calculating the fatigue crack
growth law for the material. The fatigue crack closure concept [3] has been employed
frequently in the past to eliminate the impacts of stress ratios on a variety of mate-
rials. Instead of the basic fatigue crack driving parameter ∆K, the crack growth rate,
da/dN, is displayed as a function of ∆Keff , and the FCGR curves at different stress
ratios seem to fall into a small band. This curve, which is then utilized to calculate
the crack advance for each rainflow counted fatigue cycle in variable amplitude, VA
loads, is approximated in the form da/dN = f (∆Keff ).
In this investigation, the crack closure for each individual load cycle estimated
from the constant amplitude data was used to determine ∆Keff and then used to
determine crack extension through crack growth law. As a result, the cycle-by-cycle
approach was used to predict the spectrum FCGR behavior under cold-TURBISTAN
and compare it to the existing experiment values.
Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 505

2 Spectrum Load Sequence

The standard cold-TURBISTAN load sequence [7] was used in this investigation for
prediction of fatigue crack growth in a nickel base superalloy. Figure 1a displays the
entire normalized cold-TURBISTAN load spectrum which has 15,452 reversals for
100 flights (is equal to one block). Figure 1b illustrates the magnified image of flight
no. 10 have 105 reversals. The maximum load for prediction was set at 15 kN. To
get the load sequence block, all the load reversal points were multiplied by 15. The
predicted fatigue crack growth behavior under this cold-TURBISTAN load sequence
was compared with the existing experiment results.

1.00
one block
Normalized load

0.75

0.50

0.25

0.00
0 2000 4000 6000 8000 10000 12000 14000 16000

Reversal number

(a) Full Spectrum

1.00
Normalized load

0.75

0.50

0.25

0.00
0 10 20 30 40 50 60 70 80 90 100 110

Reversal number
(b) Magnified image of flight number 10

Fig. 1 Standard cold-TURBISTAN spectrum load sequence [7]

506 S. G. Malipatil et al.

3 Material and Specimen

In this investigation, a nickel base superalloy GTM720 was used. The chemical
composition of major alloying elements in superalloy is shown in Table 1. The 0.2%
of yield strength, ultimate tensile
√ strength, and fracture toughness of this material
were 1115, 1501 & 110 MPa m respectively. The microstructure of this material is
shown in Fig. 2. Standard compact-tension (CT) specimens were cut and machined
from the forged disk in loading in circumferential, C and crack growth in radial, R,
i.e., in C-R direction. For prediction and experimentation, a nickel-based superalloy,
GTM720 was considered. ASTM Standard [8] compact tension (CT) specimen as
shown in Fig. 3 with an initial crack length is 10 mm, thickness is 10 mm, and width
is 38 mm.

Table 1 Chemical composition (weight %) of a GTM720 nickel-based superalloy

Cr Co Ti Mo Al W Fe Si B Ni
16.21 14.53 5.08 3.00 2.55 1.21 0.24 0.03 0.019 57.06

Fig. 2 Initial microstructure

of a GTM720 nickel-based
superalloy
Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 507

Fig. 3 Compact tension specimen considered for prediction [8]

4 Fatigue Crack Growth Law

The constant amplitude, CA fatigue crack growth rate, FCGR test results obtained at
various stress ratios, R for GTM720 nickel-based superalloy [10] is shown in Fig. 4.
As expected, fatigue crack growth rate was observed to increase with stress ratio. The
threshold stress intensity factor range, ∆Kth values determined by extrapolation of
the data to 10–7 mm/cycle for all R ratios, is shown in Fig. 4. As increasing the stress
ratio, R was observed that threshold stress intensity factor range decreases, ∆Kth
(Fig. 4). Similar behavior in FCGR of structural metallic alloys has been observed
by many authors [5, 9].
An effective stress intensity factor range ∆Keff is determined using Eqs. 2, 3, and
4 at various stress ratio, R (0.1–0.7). Then, fatigue crack growth rate curves at various
stress ratios for GTM720 alloy (Fig. 4) are replotted as a function effective stress
intensity factor range, ∆Keff, is shown in Fig. 5. Also, the narrow banded FCGR
curve (Fig. 5) was fitted Newman’s FCG law [2] Eq. (1), solid line (Eq. 1) is shown
in same Fig. 5.
[ ( ∆K th )2 ]
1− ef f
∆K e f f
da/d N = C1 (∆K e f f )(C2 ) [ ( )2 ] (1)
∆K e f f
1− Cs

∆K e f f = K max − K cl (2)
508 S. G. Malipatil et al.

Fig. 4 CA FCGR curves at 1.E-1

various R of a GTM720
nickel base superalloy [10]
GTM 720
RT, Lab air
1.E-2 ν = 10 Hz
sine wave

log da/dN (mm/cycle)

1.E-3
R = 0.1
R = 0.3
1.E-4 R = 0.5
R = 0.7

1.E-5

1.E-6

1.E-7
1 10 100
log ∆K (MPa√m)

∆K e f f = (1 − γ )∆K (3)

K cl Pcl
γ = = (4)
K max Pmax

where ∆K = stress
√ intensity C1 = 3.473 × 10–8 mm/cycle, C2 = 2.57,
factor range,√
C3 = 67.0 MPa m, ∆K e f f = 7.5 MPa m, γ = crackclosure. Pcl and Pmax are
th

crack closure and maximum applied load in N, respectively. Similarly,

√ Kcl and Kmax
are crack closure and maximum stress intensity factor in MPa m, respectively.

5 Fatigue Crack Closure

The crack closure values were measured from load-COD data obtained at various
∆K. Typical load-COD data obtained at various ∆K and stress ratio, R is shown
in Fig. 6. From each load-COD curve at constant R (ranging from 0.1 to 0.7) to
determine Pmax, Pmin , and Pcl, and using Eqs. 5, 6, and 7 to determine corresponding
K max, K min , and K cl.

Pmax × f (x)
K max = √ (5)
B w
Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 509

Fig. 5 CA FCGR curve of a 1.E-1

GTM720 nickel base
superalloy was replotted as
function of ∆Keff
1.E-2
R = 0.1
R = 0.3

da/dN (mm/cycle)
1.E-3
R = 0.5
R = 0.7
Eqn. (1)
1.E-4

1.E-5

1.E-6

1.E-7
1 10 100
∆Keff (MPa√m)

Pmin × f (x)
K min = √ (6)
B w
( )
Pcl × f wa
K cl = √ (7)
B w

where x = wa , f (x) = (2 + x) × (0.886 + 4.64 × x − 13.32 × x 2 + 14.72 × x 3 −

5.6 × x 4 )/(1 − x)1.5 .
The variation of crack closure level is plotted as a function of ∆K at R = 0.1,
0.3, 0.5, and 0.7 is shown in Fig. 7. It was observed that the crack closure decreases
with either increasing ∆K at any given R or increasing R at any given ∆K. The
crack closure was observed to be zero for any value of ∆K at high stress ratio, R =
0.7. The variation of crack closure, Kcl at four different stress ratios for GTM720 is
shown in Fig. 7. While estimating crack growth using crack growth law (Eq. 1), it
is necessary to compute crack closure level for specific R ratio of the load cycle. In
the cold-TURBISTAN load sequence shown in Fig. 1, R is ranging from 0 to 1, so
it is necessary to estimate crack closure value at any R ratio between 0 and 1. The
data in Fig. 7 was used to derive Eqs. (8, 9, 10, and 11) and using cubic and linear
interpolation method in MATLAB, the crack closure was estimated at other R ratios
510 S. G. Malipatil et al.

12000 20000
R = 0.1 R = 0.3
∆K (MPa√m) Pcl (N) γ ∆K (MPa√m) Pcl (N) γ
b a 13.27 2452 0.43
10000 a 16.67 2529 0.56 d
b 26 4239 0.37 b 18 3527 0.25
c 37 1209 0.16 15000 c 23 4092 0.08
8000 d 27 0 0
Load, ∆P (N)

d 55 0 0

Load, ∆P (N)
c
6000 Pcl 10000
c d

a b
4000
5000
2000 a
Pcl
0 0
0 0.2 0.4 0.6 0 0.05 0.1 0.15
COD (mm) COD (mm)
12000 30000
R = 0.5 R = 0.7
∆K (MPa√m) Pcl (N) γ ∆K (MPa√m) Pcl (N) γ
10000 a 9.72 2165 0.16 25000 a 8.7 0 0
b 11 2632 0.12 d b 10 0 0
c 13 3384 0.04 c 13 0 0 d
d 16 0 0 20000 d 18 0 0
8000
Load, ∆P (N)
Load, ∆P (N)

6000 c 15000 c
Pcl
b 10000
4000
a b

2000 5000 a

0 0
0 0.05 0.1 0.15 0 0.1 0.2 0.3
COD (mm) COD (mm)

Fig. 6 Determination of crack closure values from CA FCGR load-COD data in a GTM720 nickel
base superalloy at various R = 0.1 to 0.7 and corresponding ∆K

as shown in Fig. 8. These formulations and methods were used to estimate closure
level for any given R ratio during prediction of fatigue crack growth under spectrum
loads.

R = 0.1, γ = 0.00029∆K 2 − 0.03602∆K + 1.11316 (8)

R = 0.3, γ = 0.00064∆K 2 − 0.05896∆K + 1.10355 (9)

R = 0.5, γ = 0.00356∆K 2 − 0.11810∆K + 0.97828 (10)

R ≥ 0.7, γ = 0 (11)
Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 511

Fig. 7 Fatigue crack closure 0.6

variation with ∆K & R in
GTM720 alloy R = 0.1
0.5
R = 0.3
R = 0.5
0.4

Crack closure,
R = 0.7

0.3

0.2

0.1

0
0 20 40 60 80
ΔK (MPa√m)

Fig. 8 Crack closure value

at various R by interpolation
method for GTM720 alloy

6 Fatigue Crack Growth Prediction

The spectrum fatigue crack growth behavior of GTM720 under cold-TURBISTAN

load sequence was predicted in a standard compact-tension, C(T) specimen. The
fatigue crack growth prediction considered the following factors: (a) correlating the
stress ratio, R, effects on the CA FCGR of GTM720 nickel base superalloy by crack
closure based effective stress intensity factor range, ∆Keff ; (b) deriving the unified
Newman’s crack growth law [2] using the material’s CA FCGR; (c) counting each
512 S. G. Malipatil et al.

cycle’s stress ratio, R, load range, ∆P, cycle count, n from spectrum load using
rainflow counting method; and (d) In order to calculate the crack extension for each
load cycle using the unified Newman’s fatigue crack growth law, the crack extension,
“a,” is calculated in terms of number blocks, “NB ,” and the relationship between “a”
and “NB ” is plotted. Also, Flowchart depicting the procedure followed for crack
growth prediction is shown in Fig. 9. An in-house MATLAB code was developed
for these calculations.
In Fig. 9, w, B, and a are width, thickness, and crack length of the specimen
respectively, NB is number of blocks, Nh is half cycles, m and C are Paris constants,

Fig. 9 Schematic flow chart for fatigue crack growth prediction

Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 513

Kmax and Kmin are maximum and minimum stress intensity factor, respectively, ∆K
is stress intensity factor range, K1C fracture toughness of the material, γ fatigue crack
closure, ∆Keff is effective stress intensity factor range, C and m are Paris constants.

7 Results and Discussion

The predicted spectrum fatigue crack growth behavior in a C(T) specimen of

GTM720 nickel-based superalloy under cold-TURBISTAN load sequence, the
procedure shown in Fig. 9 is shown in Fig. 10. As expected, the growth rate was
low at short crack lengths and increased as the crack length is increased. Figure 10
also shows an experimental result of fatigue crack growth behavior determined under
cold-TURBISTAN load sequence [10]. It can be observed in this Fig. 10 that the
predicted results are conservative but reasonably good as the total spectrum FCG
life predicted. Npred is about 40 blocks as against experimental life, Nexpt of about 46
blocks. The spectrum FCG life ratio Npred /Nexpt was about 0.87.
It may be noted that the load sequence effects are not considered in this inves-
tigation and inclusion of such effects may further improve the prediction accuracy.
The fatigue tensile overloads and compression underloads have been shown to result
in retardation and acceleration of crack speed, respectively [11–13]. It is expected
that load sequence effects in spectrum loads alter the ∆Keff by altering the Kcl and
this factor should be considered further in modeling crack growth behavior under
spectrum loads.

30

25 Experimental [10]
Crcack length (mm)

Predicted
20

15

10

0
0 10 20 30 40 50
Number of block, Nb

Fig. 10 Predicted and experimental [10] FCG behavior under cold-TURBISTAN load sequence in
GTM720 alloy
514 S. G. Malipatil et al.

8 Concluding Remarks

The predicted spectrum FCG behavior of GTM720 nickel-based superalloy under

cold-TURBISTAN load sequence was compared with existing experimental results.
The prediction was made using variable crack closure approach. The predicted results
were conservative but fairly good when compared to experimental observations. The
predicted spectrum fatigue crack growth, FCG life, Npred is about 40 blocks against
experimental FCG life, Nexpt was about 46 blocks, FCG life ratio, Npred /Nexpt is
0.87. Incorporation of load-interaction effects in the model may further improve the
accuracy of prediction. Also, observed that fatigue crack closure decreases with either
increasing stress ratio, R at a given stress intensity factor range, ∆K or increasing
∆K at a given R.

Acknowledgements The authors are thankful to Mr. JJ Jadhav, Director, NAL and Mr. MZ
Siddique, Director, GTRE for facilitating the conduct of these experiments and for providing the
permission to publish this work. The technical staff members of Materials Evaluation Laboratory,
Structural Integrity Division, CSIR-NAL are thanked for their help in conducting the tests.

References

1. Paris PC, Erdogan F. A critical analysis of crack propagation laws, transactions of ASME. J
Basic Eng. 1963;85:528–33.
2. Newman Jr JC. A crack-closure model for predicting fatigue crack growth under aircraft
spectrum loading. In: Methods and models for predicting fatigue crack growth under random
loading. ASTM STP 748. Philadelphia, PA American Society for Testing and Materials;1981.
p. 53–84.
3. Elber W. The significance of fatigue crack closure, in damage tolerance in aircraft structures,
ASTM STP 486. Philadelphia PA: American Society for Testing and Materials; 1971. p. 230–42.
4. Newman JC. Application of a closure model to predict crack growth in three engine disc
materials. Int J Fract. 1996;80(2–3):193–218.
5. Manjunatha CM. Fatigue crack growth prediction under spectrum load sequence in an
aluminum alloy by K*-RMS Approach. Int J Damage Mech. 2008;17(6):477–92.
6. Zhuang WZ. Prediction of crack growth from bolt holes in a disc. Int J Fatigue. 2000;22:241–50.
7. Ten Have AA. Cold turbistan–final definition of a standardized fatigue test loading sequence
for tactical aircraft cold section engine disc. Nationaal Lucht-en Ruimtevaartlaboratorium
1987;NLR TR 87054 L.
8. ASTM. Standard test method for measurement of fatigue crack growth rates, ASTM E 647–95.
Am Soc Test Mater PA. 1997;3(01):578–614.
9. Sree PC, Kujawski D. A two-parameter fatigue crack growth correlation using ∆K and Kmax
parameters. In ASME International Mechanical Engineering Congress and Exposition, 2014,
(Vol. 46583, p. V009T12A033). American Society of Mechanical Engineers.
10. Malipatil SG, Majila AN, Fernando DC, Manjuprasad M, Manjunatha CM. Fatigue crack
propagation behavior of a nickel base super alloy GTM720 under cold-TURBISTAN spectrum
load sequence. Theoretical and applied fracture mechanics, vol. 112. 2021. https://doi.org/10.
1016/j.tafmec.2021.102913
11. Manjunatha CM, Parida BK. Prediction of fatigue crack growth after single overload in an
aluminum alloy. AIAA J. 2004;42(8):1536–42.
Prediction of Fatigue Crack Growth Behavior Under Cold-Turbistan … 515

12. Tvergaard V. Effect of underloads or overloads in fatigue crack growth by crack-tip blunting.
Eng Fract Mech. 2006;73(7):869–79.
13. Zhang W, Simpson CA, Lopez-Crespo P, Mokhtarishirazabad M, Buslaps T, Pippan R, Withers
PJ. The effect of grain size on the fatigue overload behaviour of nickel. Mater Des. 2020;189:
108526.
Optimization of Active Vibration
Absorber by Acceleration Feedback
Using Fixed Point Theory and Genetic
Algorithm

S. Mohanty and Santosha Kumar Dwivedy

Abstract In the present work, analyses on two different optimization techniques,

viz., fixed point theory (FPT) and genetic algorithm (GA) have been used for finding
the optimum tuning ratio and damping ratio for an active vibration absorber (AVA).
Here, the AVA is used to suppress the response amplitude of an SDOF (single degree
of freedom) spring-mass damper primary system under external harmonic excitation.
The AVA consisting of a PZT (lead zirconate titanate) stack actuator, spring, damper
and a mass, which is attached to the SDOF primary system. The active force by
AVA is produced by considering acceleration feedback from the primary system.
The transfer function of the system is obtained to study the responses of the system.
The comparison is also carried out between the optimized damping ratio obtained
by the FPT and GA techniques. The optimal active force is also evaluated at which
the total vibrational energy of the primary system is minimum. From the analyses,
it is observed that a significant reduction (60–90%) in the response amplitude of
the primary system is achieved by the active force, obtained by using the optimal
parameters than the passive vibration absorber.

Keywords Tuned vibration absorber · Feedback · Optimization · Fixed point

theory · Genetic algorithm

S. Mohanty (B) · S. K. Dwivedy

Department of Mechanical Engineering, Indian Institute of Technology Guwahati, Assam 781039,
India
e-mail: [email protected]
S. K. Dwivedy
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 517
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_35
518 S. Mohanty and S. K. Dwivedy

1 Introduction

The use of tuned vibration absorber (TVA) to decrease the vibration in the structures
have been studied for over a century. One may find the first patented work in this field
by Frahm in 1911 [1]. However, the undamped TVA used in [1] produces two high-
amplitude peaks near the resonating frequency and to reduce these peaks, initially,
dampers are used which are known as tuned mass dampers (TMDs). Ormondroyd
and Den Hartog [2] first obtained the optimal parameters for the spring stiffness and
damping ratio of the TMD in the year 1928, by using fixed point theory (FPT) or
H∞ optimization, which is later famously known as Den Hartog’s equal peaks in the
frequency response [3, 4]. These FPT techniques are later used in many recent articles
[4–7] considering the use of TMD for various applications taking different mass
ratios and modifying the design of the TMD by considering active/smart materials
[8–14] which provide actuation to suppress the vibration of the primary system. The
integration of actuators and sensors with the TMDs is often known as active tuned
mass damper (ATMD) or active vibration absorber (AVA) which generally attenuates
vibration of the primary system more efficiently than the passive vibration absorber
(PVA) with less absorber mass [4, 8–10, 14, 15] and for a wider range of operating
frequencies. Cheung and Wong [5–7] contributed a significant amount of work on
the optimization of passive and active TMDs to suppress vibration of the primary
system effectively for a wider range of operating frequencies using FPT and H2
optimization techniques which show Den Hartog’s equal peaks in the frequency
response. The optimal parameters of the TMD are still an active area of research as
observed from many of the recently published articles from 2019 to 2021 [11, 12,
16–19] incorporating recent techniques such as genetic algorithm (GA), artificial
intelligence and machine learning [16–19]. The consideration of nonlinearity in the
TMD is an important aspect of study as mentioned in many literature [5–11, 13,
20–27] to study the effectiveness of the TMD which is essential in many practical
applications. The optimization techniques for vibration reduction of the nonlinear
primary system by the nonlinear passive or active TMDs are a new area of research
as very limited works are available in literature [28]. Esen and Koç [18] suppress
the nonlinear vibration of the gun barrel tip which is modelled as an SDOF primary
system by the use of optimal TMD using the combination of GA and finite element
method. Batou and Adhikari [29] obtained optimal parameters of viscoelastic TMDs
and showed a slightly better vibration suppression than conventional TMD. Fallah
and Ebrahimnejad [30] studied the performance of PZT stack actuators in the TMDs
for vibration suppression of a 10-storey tall building due to seismic response. They
used various control schemes to command the PZT actuators and obtained optimum
arrangements of actuators by GA.
From the literature, it can be observed that the new optimization techniques
slightly improve the vibration suppression capabilities of both the passive and active
TMDs. Also, from previous literature, it is observed that there were no works done
in obtaining the optimal parameters for the AVA with acceleration feedback. As the
actuation by acceleration feedback is reasonably straightforward via accelerometer
Optimization of Active Vibration Absorber by Acceleration Feedback … 519

it is used in this study with a novel PZT stack actuator and a spring in series combi-
nation. So, in the present work, formulae were obtained for the optimum tuning ratio
and damping ratio for the AVA with acceleration feedback using FPT technique. The
range of active force for which the system remains stable is evaluated considering the
Routh stability criterion. Further, the optimal active force is evaluated. The compar-
ison analyses were also carried out between the optimal parameters obtained by FPT
and GA techniques. The analyses are focused on the efficacy of the proposed AVA
with acceleration feedback and the obtained optimal parameters by FPT and GA
techniques to suppress maximum vibration of the primary system than the PVA. In
the next section, the formulation of the system with the proposed AVA is discussed.

2 Formulation of the System

In this section, the dynamics of the proposed AVA is studied by considering the
mathematical model as shown in Fig. 1, where m i , ki , ci and xi , denotes the mass,
spring stiffness, damping coefficient and displacements of the primary system and
the secondary system for i = 1 and i = 2, respectively. The primary system is
subjected to an external excitation with respect to the time t of the form F(t) and the
active force f a is produced by the combination of a spring having stiffness k3 and a
PZT stack actuator in series connection. The negative acceleration control feedback
of the primary system is used for the actuation by the AVA. The stiffness of the
PZT stack actuator is denoted as k PE The governing equation of motion of the system
shown in Fig. 1 can be written as follows:

m 1 ẍ1 + k1 x1 + k1 (x1 − x2 ) + c1 ẋ1 + c2 (ẋ1 − ẋ2 ) = F(t) − f a (1)

m 2 ẍ2 + k2 (x2 − x1 ) + c2 (ẋ2 − ẋ1 ) = f a (2)

The active force produced by the AVA with the PZT stack actuator and k 3 in
series combination is expressed as f a = k PE (x1 + δ0 − δ), where δo is the nominal
displacement of the actuator and δ is the displacement of the spring at the junction

Fig. 1 PZT-based active

vibration absorber with
acceleration feedback
520 S. Mohanty and S. K. Dwivedy

of the PZT actuator and secondary absorber mass m 2 . The expression of the active
force f a can also be modified by considering f a = k3 (δ − x2 ). From these rela-
k x +k PE (x1 +δ0 )
tions, the expression of δ can be obtained as δ = 3 2 k +k E . Substituting this δ
3 P
value in the/(above f a) expression, one can express f a = K (x 1 + δ0 − x 2 ), where
K = k PE k3 k PE + k3 . The δo for the PZT stack actuator can also be expressed as
δo = nd33 V , where n is the number of wafers used in the PZT stack actuator, V
is the applied voltage and d33 is dielectric charge constant [31]. From the control
law with the acceleration feedback, V can be expressed as V = −kc ẍ1 , where kc
is the controller gain for the negative acceleration feedback. Considering the non-
dimensional time τ = ω1 t, x 0 and V0 as reference displacement and voltage quantity,
respectively, the governing Eqs. (1) and (2) are non-dimensionalized as given below:

Z̈ 1 + Z 1 + 2ζ1 Ż 1 − μΩ2 Z − 2ζ2 μΩ Ż = F1 (τ ) + Fa (τ ) Z̈ 1 (3)

Fa (τ )
Z̈ + Ω2 Z + 2ζ2 Ω Ż = − Z̈ 1 − (4)
μ

The non-dimensional parameters used in Eqs. (3) and (4) are defined as follows:
x1 x2 m2 ω2 c1
Z1 = , Z2 = , Z = Z2 − Z1, μ = ,Ω= , ζ1 = ,
x0 xo m1 ω1 2m 1 ω1
/ /
c2 k1 k 2 + kr nd33 Vo V
ζ2 = , ωo = , ω2 = ,λ= , V̄ = ,
2m 2 ω2 m1 m2 xo Vo
kr λV̄ F
Fa = , F1 =
k1 m 1 ω1 2 x 0

In the next section, the mathematical analysis of the above equations of motions is
carried out considering the transfer function approach and the optimized parameters
of the proposed AVA are obtained.

3 Mathematical Analysis

In this section, the mathematical analysis of the system is carried out by taking
Laplace transformations of Eqs. (3) and (4), which yield the following equations:
( ) ( )
s 2 + 2ζ1 s + 1 Z 1 (s) − μΩ2 + 2ζ2 μΩs Z (s) = F1 (s) + Fa (s)s 2 Z 1 (s) (5)

( 2 ) Fa (s)s 2 Z 1 (s)
s + 2ζ2 Ωs + Ω2 Z (s) + s 2 Z 1 (s) = − (6)
μ
Optimization of Active Vibration Absorber by Acceleration Feedback … 521

Solving Eqs. (3) and (4) the transfer function of the primary system and the AVA
are obtained as follows:

Z 1 (s) s 2 + 2ξ2 Ωs + Ω2
G 1 (s) = = (7)
F1 (s) N14 s + N13 s 3 + N12 s 2 + N11 s + N10
4

Z (s) (1 + Fa /μ)s 2
G 2 (s) = =− (8)
F1 (s) N14 s 4 + N13 s 3 + N12 s 2 + N11 s + N10

where the coefficients N10 , N11 , N12 , N13 and N14 are expressed as follows:

N10 = Ω2 , N11 = 2ζ1 Ω2 + 2ζ2 Ω, N12 = Ω2 + 4ζ1 ζ2 Ω + μΩ2 + 1,

N13 = 2ζ2 Ω + 2ζ1 + 2ζ2 Ωμ, N14 = 1 − Fa

Following Routh’s stability criterion to obtain the range of the active force (Fa )
by the absorber at which the system responses are stable, the following array of the
coefficients is obtained from Eq. (7) for the primary system:
|
s 4 | N14 N12 N10
|
s 3 || N13 N11
2 | N13 N12 −N11 N14 N (9)
s | N13 10
s 1 || N13 N12NN11N−N−N
14 N11 −N13 N10
2 2

s 0 | N10
13 12 11 N14

The stable responses are ensured when the real parts of all poles are negative.
From Eq. 9, it can be derived from the first column of the Routh table that all the
coefficients of the array are positive for −μ < Fa < 1. This means that the proposed
system is stable within the above range of feedback gain (Fa ).
In the next sub-section, optimization procedures such as FPT and GA techniques
are undertaken to obtain the optimal tuning ratio, damping ratio and the optimal
controlling force required by the AVA to attenuate the response amplitude of the
primary system.

3.1 Optimization of the AVA

In this section, the tuning ratio and the damping ratio of the AVA are optimized
by following the FPT and the GA optimization techniques using the mass ratio
μ = 0.05 in the entire analysis. The tuning ratio (Ω) is the ratio between the natural
522 S. Mohanty and S. K. Dwivedy

Fig. 2 Responses of the system with μ = 0.05 a undamped primary system b passive vibration
absorber (PVA) c primary system with optimized PVA d PVA with optimum parameters

Fig. 3 Response amplitude of the primary system with AVA at the fixed points a Fa = 0 (passive)
b Fa = 0.5 (AVA)

frequency of the absorber (ω2 ) and the primary system (ω1 ). To obtain the optimal
parameters of the proposed AVA, initially, the response of the system is studied
with various tuning ratios and damping ratios, which are shown in Figs. 2 and 3.
It may be noted that the frequency response of the system is obtained by solving
Optimization of Active Vibration Absorber by Acceleration Feedback … 523
√
Eqs. (7) and (8) by considering s = j ω, where ω = ωω111 , j = −1. Here, ω11
is the frequency of external harmonic excitation acting on the primary system. The
reference displacement (x 0 ) and reference voltage (v0 ) are considered equal to 1,
and the damping ratio of the primary system (ζ1 ) is considered equal to 0 throughout
the analysis. In Fig. 2, the frequency response of the primary system is shown for
four different values of damping ratios considering zero active force (Fa = 0). From
Fig. 2a, it can be observed that, at ωa and ωb , the response amplitude of the primary
system remains same irrespective of the various values of damping ratio ζ2 . These two
points (A and B) in the frequency responses are known as fixed points [4, 9]. There
are many studies carried out on these fixed points starting with the first work from
Den Hartog in the year 1942 to the many recent works [4–12] to obtain the optimum
tuning and damping ratio of the absorber so that these peaks are minimum. One of
the widely used methods is the fixed point theory of optimization technique which
was first proposed by Den Hartog [2–9] and famously known as Den Hartog equal
peaks in the frequency response. The response of the absorber is shown in Fig. 2b,
from which one can observe that the vibrational energy of the primary system is
transferred to the absorber due to which its response is high. Also, from Fig. 2a, one
can observe that, with an increase in the ζ2 values, the response amplitude of the
primary system increases rather than decreases. Further, Fig. 2b shows the common
phenomenon that is the response amplitude of the absorber decreases with increases
in ζ2 values, which means the vibrational energy from the primary system is not
effectively transferred to the absorber. From these figures, one can notice that there
should be some optimal parameters for the tuning ratio and damping ratio at which
the response of the primary system is suppressed effectively by the passive vibration
absorber (PVA). From the previous literature [4], it is found that, for the passive
TMD, the optimum parameters for damping ratio (ζ2 ) and tuning√ ratio (Ω) in terms
of the mass ratio μ are given as Ω = 1/(1 + μ) and ζ2 = 3μ/(8(1 + μ)). Using
these optimal parameters, the response of the system with the passive absorber are
shown in Fig. 2c and d. From these figures, one can notice that primary system
response is considerably decreased and the peaks are at same amplitude which are
known as Den Hartog equal peaks in the response. However, it can be observed that,
with the optimal parameters, the system response is very high. Due to this, here active
vibration absorber is considered and the corresponding optimal parameters for the
proposed AVA are carried out in the next section.
524 S. Mohanty and S. K. Dwivedy

3.2 Fixed Point Theory and GA Techniques for the Proposed

AVA

To obtain the optimized Ω and ζ2 , the frequency response function of the primary
system and the absorber which are obtained√in Eqs. (7) and (8) are modified as given
below by considering s = j ω, where j = −1.

P + jQ
G 1 ( j ω) = (10)
R + jS

(1 + Fa /μ)ω2
G 2 ( j ω) = (11)
R + jS

where

P = Ω2 − ω2 , Q = 2ζ2 Ωω, R = N14 ω4 − N12 ω2 + N10 , S = N11 ω − N13 ω3

Now considering the FPT optimization techniques, one can observe from Fig. 2a
that the response amplitude of the primary system at the fixed points (A and B)
remains the same whether ζ2 = 0 or ζ2 = ∞. Substituting these two values of
ζ2 in Eq. (10), and equating the resulting equations one may obtain the following
expression:

2Ω2 (1 + μ) + 2 2 2Ω2
ω4 − ω + =0 (12)
2 + μ − Fa 2 + μ − Fa

The roots of Eq. (12) and the corresponding response amplitude of the primary
system at A and B with the proposed AVA are obtained as follows:

| [ ] )
1 + (1 + μ)Ω2 || 1 + (1 + μ)Ω2 2 2Ω2
ωa2 , ωb2 = ± − (13)
2 + μ − Fa 2 + μ − Fa 2 + μ − Fa
| |
| 1 |
|G 1 (ωa )| = || |
2|
(14)
1 − (1 + μ)ωa
| |
| 1 |
|G 1 (ωb )| = || |
2|
(15)
1 − (1 + μ)ω b

Following the observation in Fig. 2, one can notice that the optimum damping
ratio of the AVA can be evaluated when the response amplitude of the primary
system at ωa and ωb is same and minimum, which can be mathematically expressed
as |G 1 (ωa )| = |G 1 (ωb )|. From this condition, the optimum√tuning ratio
/ (Ω) of
the system for the proposed AVA can be written as Ωopt = (1 − Fa ) (1 + μ).
This optimum value of Ωopt giving rise to the condition that the active force by the
Optimization of Active Vibration Absorber by Acceleration Feedback … 525

actuator must be Fa 1. This obtained optimal Ωopt is also verified in Fig. 3 by

substituting Eq. (13) into Eqs. (14) and (15) and plotting the curves for the passive
and active vibration absorber. From Fig. 3a, one can observe that the optimal tuning
ratio is the intersection point of curves that are marked C and D for the passive and
AVA, respectively. The optimal Ωopt obtained in Fig. 3 are also satisfied with the
above-obtained formula for the passive and active vibration absorber for μ = 0.05.
Substituting the optimal Ω value in Eq. 10, one may obtain the following expression
for ωa and
/( ))
1 μ + Fa
ωa , ωb
2 2
= 1± (16)
(1 + μ) (2 + μ − Fa )

ωb . Substituting Eq. (16) into Eqs. (14) and (15), the

/ corresponding response
/
amplitude at the fixed points ωa and ωb are obtained as (2 + μ − Fa ) (Fa + μ).
To obtain the optimal damping ratio ζ2 for the AVA, |G 1 ( j ω)| is obtained from
Eq. (7), which is given below:
/
P 2 + Q2
|G 1 ( j ω)| = (17)
R1 2 + S1 2

The optimum damping ratio of the AVA can be obtained at which the fixed points
become the maximum response amplitude of the primary system, which gives rise
to the following expression [4]:

∂ | ∂ |
|G 1 (ω)|2 |ω=ω = |G 1 (ω)|2 |ω=ω = 0 (18)
∂ω 2 a ∂ω 2 b

Substituting the value of optimal Ω, ωa from Eqs. (16) and (17) into Eq. (18), one
may obtain the following expression for obtaining the optimal damping ratio (ζ2 ) for
the proposed AVA:
√
−C12 ± C12 2 − 4C11 C13
ζ2 (ωa ) =
2
(19)
2C11

Similarly, ζ2 ! considering ωb can be written as follows:

/
−C̃12 ± 2
C̃12 − 4C̃11 C̃13
ζ2 2 (ωb ) = (20)
2C̃11

The formula for the optimal ζ2 for the proposed AVA with the acceleration
feedback can be given by the following equation in terms of mass ratio %μ:
526 S. Mohanty and S. K. Dwivedy
/
ζ2 2 (ωa ) + ζ2 2 (ωb )
ζ2 (optimal) = (21)
2

where

C11 = C2 C4 − C6 C9 , C12 = C2 C3 + C1 C4 − C6 C7 C8 − C5 C9 ,
C13 = C1 C3 − C5 C7 C8 , C̃11 = C̃2 C̃4 − C̃6 C̃9 ,
C̃12 = C̃2 C̃3 + C̃1 C̃4 − C̃6 C̃7 C̃8 − C̃5 C̃9 ,
C̃13 = C̃1 C̃3 − C̃5 C̃7 C̃8 , C1 = ((1 − Fa )ωa4 − (1 + (1 + μ)ωopt
2
)ωa2 + Ω2opt )2 ,
C2 = 4Ω2opt ωa2 + 4Ω2opt (1 + μ)2 ωa6 − 8Ω2opt (1 + μ)ωa4 ,
C3 = −2(Ω2opt − ωa2 ), C4 = 4Ω2opt , C5 = (ωopt
2
− ωa2 )2 , C6 = 4ωa2 Ω2opt ,
C7 = 2((1 − Fa )ωa4 − (1 + (1 + μ)ωr2 )ωa2 + Ω2opt ),
C8 = 2(1 − Fa )ωa2 − (1 + (1 + μ)Ω2opt ),
C9 = 4Ω2opt + 12Ω2opt (1 + μ)2 ωa4 − 16Ω2opt (1 + μ)ωa2 ,
C̃1 = ((1 − Fa )ωb4 − (1 + (1 + μ)Ω2opt )ωb2 + Ω2opt )2 ,
C̃2 = 4Ω2opt ωb2 + 4Ω2opt (1 + μ)2 ωb6 − 8Ω2opt (1 + μ)ωb4 ,
C̃3 = −2(Ω2opt − ωb2 ), C̃4 = 4Ω2opt , C̃5 = (ωopt
2
− ωb2 )2 , C̃6 = 4ωb2 Ω2opt ,
C̃7 = 2((1 − Fa )ωb4 − (1 + (1 + μ)Ω2opt )ωb2 + Ω2opt ),
C̃8 = 2(1 − Fa )ωb2 − (1 + (1 + μ)Ω2opt ),
C̃9 = 4Ω2opt + 12Ω2opt (1 + μ)2 ωb4 − 16Ω2opt (1 + μ)ωb2 ,
√
Ωopt = (1 − Fa )/(1 + μ),
/( ))
1 μ + Fa
ωa =
2
1+ ,
(1 + μ) (2 + μ − Fa )
/( ))
1 μ + F a
ωb2 = 1−
(1 + μ) (2 + μ − Fa )

The damping ratio of the proposed AVA is also optimized by using the GA toolbox
in MATLAB. The optimized values of the proposed AVA by GA are obtained by the
following steps. First, the maximum peak of the frequency response curve is obtained
from Eq. 17 by considering |G 1 ( jω)| as the fitness function and selecting the upper
and lower bounds of ζ2 between 0 and 1. In the next section, the performance of the
proposed AVA is studied with the obtained optimized tuning ratio (Ω) and damping
ratio (ζ2 ) by the FPT and GA for various values of active force Fa .
Optimization of Active Vibration Absorber by Acceleration Feedback … 527

4 Results and Discussion

In this section, the responses of the system is studied with the obtained optimal
parameters for the AVA by acceleration feedback considering μ = 0.05 in the entire
analysis. In Fig. 4, the response of the system is shown for various values of Fa with
optimized parameters of tuning ratio Ω and damping ratio ζ2 obtained by using FPT
and GA techniques by solving Eqs. (7) and (8). From Fig. 4a, it can be observed
that, with increasing value of Fa , there is a corresponding decrease in the response
amplitude of the primary system. For Fa = 0.19, there is an almost 60% decrease in
the response amplitude of the primary system. Similarly, for Fa = 0.5 and 0.8, the
response amplitude of the primary system reduces significantly close to the static
deflection of the primary system. The responses obtained by using the obtained
optimal ζ2 show similar response, which are shown in Fig. 4a and b. However, the
zoom of Fig. 4a shows distinct observations which are shown in Fig. 4c to h for the
primary system. From Fig. 4c and d with Fa = 0.19, one can observe that, with GA
optimized ζ2 value, the Den Hartog’s equal peaks in the response are observed with
minimum peaks unlike FPT optimal ζ2 . But the response amplitude of the primary
system with Fa = 0.19, at the resonance, is less compared to the GA optimized ζ2 .
Similar, observations are also noticed in Fig. 4e to h where with the optimal ζ2 by GA
shows almost equal minimum Den Hartog’s equal peaks in the response amplitude
compared to the FPT optimal ζ2 but, at resonating frequencies, the FPT optimized
ζ2 performs marginally better. In Fig. 4i to k, the zoom version of Fig. 4b is shown
for studying the comparison between the FPT and GA optimized ζ2 for various
values of Fa From Fig. 4i to k, it is observed that with FPT the response amplitude
of the absorber is more than the GA optimized ζ2 The damping ratio obtained for
various values of the Fa are given in Table 1. From this table, one can observe that
the optimized value of ζ2 by FPT technique is slightly less than that of the GA
optimized ζ2 Due to this, it can be observed that, with GA, the response amplitude
of the absorber is comparatively less than with FPT optimized ζ2 . From Fig. 4, it
can be noticed that, with the proposed AVA, if simultaneous vibration attenuation of
the primary system and the absorber is required, then one can use GA optimization
or if the vibration attenuation of the primary system is the sole purpose than one
can use FPT technique. Also, with the GA optimization technique, it is easier to
evaluate the optimal damping ratio for the AVA and is less cumbersome than in the
FPT technique. The results shown in Fig. 4a for the PVA (Fa = 0) are also in good
agreement with the work of Batou and Adhikari [29] (Fig. 8 classical TMD). Further,
from Table 1, one can observe that the optimal ζ2 value for the PVA (Fa = 0) is also
in good agreement with the published work [3–9]. From Fig. 4a, one can also observe
that the response amplitude of the primary system significantly reduces from Fa = 0
to Fa = 0.19 but this large reduction in the response amplitude is not observed when
Fa is increased from 0.5 to 0.8. These observations suggest that one can obtain the
optimal active force Fa for which the total vibrational energy of the primary system
is minimum. The optimal active force Fa at which the total vibration amplitude of
the primary system is minimum is shown in Fig. 5a and b. In Fig. 5a, the / optimal
active force Fa is obtained by using the passive optimal parameters Ω = 1 (1 + μ)
528 S. Mohanty and S. K. Dwivedy

Fig. 4 Frequency response of the system with the optimized AVA a primary system b AVA c–h
zoom of a, i–k zoom of b
Optimization of Active Vibration Absorber by Acceleration Feedback … 529

Fig. 4 (continued)

Table 1 Optimized damping ratio (ζ2 ) values for different active forces (Fa ) by using FPT and
GA optimizations
Active control force (Fa ) ζ2 (optimal) by FPT ζ2 (optimal) by GA
0 0.1336 0.133
0.19 0.2928 0.295
0.5 0.4432 0.453
0.8 0.5510 0.565

/ /
and ζ2 = 3μ 8(1 + μ). In Fig. 5a, the total vibrational energy is calculated from
Eq. (7), which is nothing but the area and under the frequency response curve. From
this figure, it is found that, for Fa = 0.19, the area under the frequency response curve
is minimum, which corresponds to the optimal active force. ( In√the same way/ for the)
proposed AVA using the obtained optimal tuning ratio Ω = (1 − Fa ) (1 + μ)
and damping ratio ζ2 (Eq. (21)), the corresponding optimal active force Fa is obtained
from Fig. 5b, which is equal to 0.38. In Fig. 6a and b, a comparison between the PVA
and AVA with the obtained optimized parameters from GA is studied considering
how the mean fitness value converses to best fitness and also the corresponding fitness
value. In Fig. 6a, the best fitness value is shown for the PVA from which one can
530 S. Mohanty and S. K. Dwivedy

Fig. 5 The vibrational energy of the primary system with a passive optimal parameters b with
present optimal parameters

observe that the best fitness value is around 6.46 and the mean fitness value starts from
28 which requires more time to converge to the best fitness value than the proposed
AVA, which is shown in Fig. 6b. From Fig. 6b, the best fitness value is around 1.66
and the mean fitness value is 2.9 which is converged faster than PVA. The fitness
value shown in Fig. 6 can be termed as the maximum response amplitude of the
primary system. From Fig. 6b, it can be noticed that, with the AVA, the response
of the primary system is nearly one-fourth than with PVA (Fig. 6a). In Fig. 7a and
Fig. 7b, the step response of the system is shown with PVA and AVA, respectively.
From these figures, one can observe the settling time to reach steady-state response
and the steady-state error is more in the case of PVA (Fig. 7a) than with AVA (Fig. 6b).
The response of the primary system improves tellingly in all aspects with active force
as illustrated in Fig. 7(b) than with PVA (Fig. 7a). From Fig. 7b, it can be observed
that with GA (black color) the primary system response is marginally better than the
FPT optimization technique (blue color). In Fig. 8 the root locus plot of the system
is shown for PVA and AVA. From Fig. 8a, it can be noticed that all four poles of
the system have negative real parts but with PVA the roots are marginally stable as
the solution approaches very closely to the zero than the AVA (Fig. 8b). The two
complex conjugate poles of the primary system transfer function (Eq. 7) with PVA
are crossing the real axis at −0.02, whereas, with AVA, these poles are at −0.36.
From these figures, one can observe that, with active force by the proposed AVA, the
primary system is more stable than with PVA, and it reduces the response amplitude
of the primary system significantly.
Optimization of Active Vibration Absorber by Acceleration Feedback … 531

Fig. 6 Fitness value of the primary system with optimized a PVA b AVA

Fig. 7 Step response of the primary system with the optimal parameter using a PVA b AVA

Fig. 8 Root locus plot of the primary system a PVA b AVA

532 S. Mohanty and S. K. Dwivedy

5 Conclusion

In this paper, the fixed point theory (FPT) and genetic algorithm (GA) optimization
techniques are used to obtain the optimum tuning and damping ratio of a PZT stack
actuator-based AVA with acceleration feedback to suppress resonant vibration of
an SDOF primary system. The formulae are obtained by using the FPT technique
for the optimized tuning ratio and damping ratio in terms of the mass ratio (μ)
between the absorber mass and the primary system mass for the proposed AVA
with acceleration feedback. A reduction of 60–90% in the response of the primary
system is achieved with the obtained optimized parameters by the proposed AVA.
The optimum active force is also evaluated and significant vibration suppression
in terms of total energy is achieved in both the primary system and the absorber.
From the previous literature, the optimized damping ratio and tuning ratio for the
PZT-based AVA with acceleration feedback were not found. So, the proposed AVA
uses acceleration feedback which is easier to implement in many applications in the
form of an accelerometer compared to other feedbacks. The active force produced
by the novel design AVA, where the PZT stack actuator is connected in series with a
spring, also produces more blocking force on the primary system and can withstand
a relatively high amplitude of oscillations than directly connecting the PZT stack
actuator. The proposed PZT stack actuator can also simultaneously act as an actuator
and sensor by converse and diverse piezoelectric effects. Further, the active force by
the combination of the PZT stack actuator and the spring makes the system more
stable as it reduces the response amplitude of the system for a wider range of operating
frequencies. It is observed that one may obtain better vibration suppression by using
the GA-optimized damping parameters than by using the FPT-optimized damping
parameter.

References

1. Frahm H. Device for damping vibrations of bodies. U.S. Patent, 989958, 1911
2. Ormondroyd J, Den Hartog JP. The theory of the dynamic vibration absorber. ASME J Appl
Mech. 1928;50(7):9–22.
3. Den Hartog JP. Mechanical vibrations. Dover Publications; 1985.
4. Cheung YL, Wong WO, Cheng L. Design optimization of a damped hybrid vibration absorber.
J Sound Vib. 2012;331:750–66. https://doi.org/10.1016/j.jsv.2011.10.011.
5. Cheung YL, Wong WO. H∞ and H2 optimizations of a dynamic vibration absorber for
suppressing vibrations in plates. J Sound Vib. 2009;320(1–2):29–42. https://doi.org/10.1016/
j.jsv.2008.07.024.
6. Cheung YL, Wong WO. Isolation of bending vibration in a beam structure with a translational
vibration absorber and a rotational vibration absorber. J Vib Control. 2008;14(8):1231–46.
https://doi.org/10.1177/1077546307083174.
7. Wong WO, Cheung YL. Optimal design of a damped dynamic vibration absorber for vibration
control of structure excited by ground motion. Eng Struct. 2008;30(1):282–6. https://doi.org/
10.1016/j.engstruct.2007.03.007.
Optimization of Active Vibration Absorber by Acceleration Feedback … 533

8. Mohanty S, Dwivedy SK. Linear and nonlinear analysis of traditional and nontraditional piezo-
electric vibration absorber with time-delay feedback for simultaneous resonance conditions.
Mech Syst Signal Proc. 2021;161:107980. https://doi.org/10.1016/j.ymssp.2021.107980.
9. Mohanty S, Dwivedy SK. Nonlinear dynamics of piezoelectric-based active nonlinear vibration
absorber using time delay acceleration feedback. Nonlinear Dyn. 2019;98(2):1465–90. https://
doi.org/10.1007/s11071-019-05271-4.
10. Mohanty S, Dwivedy SK. Traditional and non-traditional active nonlinear vibration absorber
with time delay combination feedbacks for hard excitation. Commun Nonlinear Sci Numer
Simul. 2023;117: 106919. https://doi.org/10.1016/j.cnsns.2022.106919.
11. Liu K, Coppola G. Optimal design of damped dynamic vibration absorber for damped primary
systems. Trans Can Soc Mech Eng. 2010;34(1):119–35. https://doi.org/10.1139/tcsme-2010-
0008.
12. Farzam MF, Jalali HH, Gavgani SAM, Kayabekir AE, Bekdaş G. Current trends in the optimiza-
tion approaches for optimal structural control. Adv Struct Eng Optimizat. 2021;326:133–79.
https://doi.org/10.1007/978-3-030-61848-3_5
13. Mohanty S, Dwivedy SK. Nonlinear analysis of active vibration absorber as a wearable rest
tremor suppression in Parkinson’s disease. In: Recent advances in computational and exper-
imental mechanics. Singapore, Springer;2022. p. 661–9. https://doi.org/10.1007/978-981-16-
6738-1_53
14. Hossein Lavassani SH, Alizadeh H, Homami P. Optimizing tuned mass damper parameters
to mitigate the torsional vibration of a suspension bridge under pulse-type ground motion: a
sensitivity analysis. J Vib Control. 2020;26:1054–67. https://doi.org/10.1177/107754631989
1591.
15. Liu Y, Wang K, Mercan O, Chen H, Tan P. Experimental and numerical studies on the optimal
design of tuned mass dampers for vibration control of high-rise structures. Eng Struct. 2020;211:
110486. https://doi.org/10.1016/j.engstruct.2020.110486.
16. Yucel M, Bekdaş G, Nigdeli SM, Sevgen S. Estimation of optimum tuned mass damper param-
eters via machine learning. J Build Eng. 2019;26: 100847. https://doi.org/10.1016/j.jobe.2019.
100847.
17. Cetin H, Aydin E. A new tuned mass damper design method based on transfer functions. KSCE
J Civ Eng. 2019;23(10):4463–80. https://doi.org/10.1007/s12205-019-0305-x.
18. Esen İ, Koç MA. Optimization of a passive vibration absorber for a barrel using the
genetic algorithm. Expert Syst Appl. 2015;42(2):894–905. https://doi.org/10.1016/j.eswa.
2014.08.038.
19. Salvi J, Rizzi E, Rustighi E. Ferguson NS. Optimum tuning of passive tuned mass dampers
for the mitigation of pulse-like responses. J Vibrat Acoust. 2018;140(6), p. 061014. https://doi.
org/10.1115/1.4040475
20. Mohanty S, Dwivedy SK. Active nonlinear vibration absorber for a nonlinear system with
a time delay acceleration feedback under the internal resonance, subharmonic, superhar-
monic and principal parametric resonance conditions simultaneously. J Aerospace Syst Eng.
2019;13(5):9–15. https://doi.org/10.20910/JASE.2019.13.5.9
21. Mohanty S, Dwivedy SK. Active nonlinear vibration absorber for a harmonically excited beam
system. In: Nonlinear dynamics and control. Cham: Springer;2020. p 3–11. https://doi.org/10.
1007/978-3-030-34747-5_1
22. Mohanty S, Dwivedy SK. Dynamic analysis of active vibration absorber by time delay accel-
eration feedback using higher order method of multiple scales. In: ASME, Gas Turbine India
Conference. Bangalore, India;2017. https://doi.org/10.1115/GTINDIA2017-4850
23. Mohanty S, Dwivedy SK. Active vibration absorber for a nonlinear system with time-delay
acceleration feedback for superharmonic and subharmonic resonance conditions. In: Machines,
Mechanism and Robotics. Singapore, Springer;2019. p. 681–90. https://doi.org/10.1007/978-
981-10-8597-0_58
24. Mohanty S, Dwivedy, SK. Nonlinear active vibration absorber using HBM method.
In: Advances in rotor dynamics, control and structural health monitoring. Singapore,
Springer;2020. p. 571–8. https://doi.org/10.1007/978-981-15-5693-7_39
534 S. Mohanty and S. K. Dwivedy

25. Mohanty S, Sikder, S, Dwivedy, SK. Nonlinear analysis of rotational inertial double-tuned mass
damper by harmonic balance method. In: Advances in mechanical engineering. Singapore:
Springer;2020. p. 1183–90. https://doi.org/10.1007/978-981-15-0124-1_105
26. Mohanty S, Dwivedy SK. Active vibration absorber for super-harmonic resonance condition.
J Phys: Conf Ser. 2020;1706: 012102. https://doi.org/10.1088/1742-6596/1706/1/012102.
27. Mohanty S, Dwivedy SK. Linear and nonlinear analysis of piezoelectric based vibration
absorber with acceleration feedback. Procedia Eng. 2016;144:584–91. https://doi.org/10.1016/
j.proeng.2016.05.045.
28. Habib G, Detroux T, Viguié R, Kerschen G. Nonlinear generalization of Den Hartog’s equal-
peak method. Mech Syst Signal Proc. 2015;52:17–28. https://doi.org/10.1016/j.ymssp.2014.
08.009.
29. Batou A, Adhikari S. Optimal parameters of viscoelastic tuned-mass dampers. J Sound Vib.
2019;445:17–28. https://doi.org/10.1016/j.jsv.2019.01.010.
30. Fallah N, Ebrahimnejad M. Active control of building structures using piezoelectric actuators.
Appl Soft Comput. 2013;13(1):449–61. https://doi.org/10.1016/j.asoc.2012.08.010.
31. Mallik AK. Chatterjee. Principles of passive and active vibration control: East West Press;
2014.
Study of Time-Varying Oil Film Damping
and Contact Stiffness of Ball Bearings

Sameera Mufazzal, S. M. Muzakkir, and Sidra Khanam

Abstract In current scenarios, where the thrust is kept on the incipient fault detection
and its size estimation, the dependency on the vibration signals and processing tech-
niques is growing day by day. The correct understanding of the mechanism behind
the sources of this signal is therefore called upon. Like other sources of damping,
the effect of squeezed oil film damping on the vibration response of rolling element
bearings cannot be overlooked. The variation in the squeezed film damping and
contact stiffness, as a result of orbital motion of cage, and the resulting load distri-
bution pattern play important role in determining the time-varying system behavior.
The values of contact stiffness and damping are governed by multiple factors which
necessitate their quantitative investigations to gain deep insight. In this work, the vari-
ation in squeezed film damping and contact stiffness in bearing owing to its orbital
motion and that due to localized defect on outer race have been studied. The influence
of operating parameters and defect geometry factors on dynamic contact character-
istics have also been numerically investigated. A considerably large variation in the
values of the contact parameters was witnessed with rotation, especially in the pres-
ence of defect. The variations were found as high as 89% in squeezed film damping
and 38% in contact stiffness for the selected test bearing rotating at 1000 RPM under
100 N radial load. The consideration of such huge variations will help in correct
simulation of the time localized responses of both healthy and defective bearings.

Keywords Point defect · Outer race defect · Lubricated contact · Load

redistribution · Additional deflection · Varying compliance

S. Mufazzal (B) · S. M. Muzakkir

Department of Mechanical Engineering, Jamia Millia Islamia, New Delhi, India
e-mail: [email protected]
S. Khanam
Department of Mechanical Engineering, ZHCET, Aligarh Muslim University, Aligarh, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 535
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_36
536 S. Mufazzal et al.

1 Introduction

Since decades, there has been consistent effort to support smooth and uninterrupted
services of industrial machines. The generation of vibration produced by machines
themselves can have poor impact on the product quality and operational perfor-
mance. This requires proper system design and installation as per the operational
requirements, before bringing the system in use. Among other indispensable machine
components, rolling element bearings are one of the major sources of compliance
and damping, influencing the overall systems’ response. Sunnersjo [1] brought into
light the notion of varying compliance vibration in rolling element bearings which
exists in all the bearings irrespective of its health condition and installment issues.
This is caused by periodical load redistribution between the finite number of rolling
elements in the load zone, thereby causing periodic variation in the stiffness of the
bearing assembly. Later to this discovery, several researchers attempted to model
this time-varying stiffness in terms of different operational and geometric parame-
ters. Liew and Lim [2] formulated generalized model for time-varying stiffness as a
function of orbital rotation, based on Hertzian theory. Liu and Shao [3] evaluated the
contact forces between the roller and localized surface defect with edge topographies
using Finite Element Method and found that the number of impacts at the exit edge is
more than that at the entry edge. Liu et al. [4] investigated the effects of defect edge
topographies on the time-varying contact stiffnesses of ball bearing using analytical
formulations. Liu and Zhang [5] computed time-varying stiffness by finding time-
varying load distribution, and hence, the number of balls in the load zone of angular
contact bearing.
The damping characteristic of the vibratory system is essential for suppressing
the vibration, as soon as it is produced, and prevents structural transmission. This
helps in achieving better operational performance and eventually improved system
life. However, any improvement in the damping characteristics requires comprehen-
sive knowledge of system vibration mechanism, identification of the major sources
of damping and different influencing factors. In rolling bearings, the damping can
arise either of the following reasons: structural deformation at ball raceway contact,
viscous effect in the presence of lubricant, friction between moving surfaces and at
the joints, or hysteresis in the structure. Based on their experimental study, Walford
and Stone [6] discovered that the joints in bearings are a good source of damping,
which requires proper consideration for good correlation between the theoretical and
the experimental values. Next, Walford and Stone [7] proposed a vibration model
of contact zone of angular contact ball bearing. The analyses revealed that the oil
film in the contact zone becomes extremely stiff and it is many orders higher than
the contact stiffness of the solid body. Thus, the damping associated with oil film
compression is negligible compared to the squeezed film damping. It is for this reason
that in the later models for contact zone, the compression oil film damping and the
structural stiffness are ignored yet achieving reasonably accurate results. Hagiu and
Gafitanu [8] analyzed the dynamic characteristics of elasto-hydrodynamic contact
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 537

in high-speed angular contact ball bearing. The stiffness and oil film damping at the
inlet of lubricated contact zone were evaluated using Reynold’s theory.
Kraus et al. [9] examined the effect of inner race speed and pre-load on the stiffness
and damping characteristics of ball bearing with free outer race using modal anal-
ysis and mathematical relations. Their study revealed that the transverse damping
decreases while axial damping improves with the amount of preload. Moreover,
damping was found to decrease with speed, as reported in Ref. [10]. According to
experimental findings, Mitsuya et al. [10] reported that the damping improves with
clearance and number of balls. Further, application of grease has insignificant influ-
ence on the damping characteristics during rotation. Opposed to this, Dietl et al. [11]
established from analytical and experimental findings that the lubrication appreciably
enhances the damping capability of rolling bearing.
Jacobs et al. [12] investigated the bearing stiffness and damping by measuring the
electrical resistance. The influence of temperature and speed was also studied. The
damping increased during run-up while it decreased dramatically during warm-up,
due to loss in viscosity. Lambert et al. [13] used steady-state equations to find elasto-
hydrodynamic oil film damping and concluded that the oil film damping is nearly
independent of viscosity.
Sarangi et al. [14] derived empirical relations for stiffness and damping of asperity
contact, considering viscosity-pressure relationship by solving classical Reynold’s
equation. Sarangi et al. [15] extended the work of [14] to mixed lubrication condi-
tion. The trends of equivalent stiffness and damping were found similar to past
experimental results.
Sun et al. [16] used multigrid method to find oil film pressure and thickness distri-
bution in the contact zone, and estimated bearing damping by considering external
radial load, centrifugal force, ball spin, rotational speed and oil film thickness. Like
previous observation [9, 9], the damping values were found to increase with load
and decrease with speed. Similar observations were drawn in [17, 17]. Apart from
this, Zhang et al. [17] observed improved dynamic characteristics, including stiff-
ness and damping coefficients of rolling element bearing, for lubricants with high
ambient viscosity and low pressure-viscosity coefficient. Moreover, Zhang et al. [18]
found that the surface waviness amplitude had a little effect on lubricant stiffness and
damping. Zhang et al. [19] performed parametrical studies to investigate the influ-
ence of EHL on the stiffness matrix and found that EHL has a negligible effect on
the radial stiffness whereas it is more influential for rotational and coupling stiffness
coefficients.
Hooke and Espejel [20] used Maxwell model to obtain the EHL damping at
contact under low amplitude, sinusoidally varying load. The alterations in pressure
and clearance at inlet under varying load were seen many times greater than those
obtained for slow load variation.
Peterson et al. [21] examined the varying stiffness considering surface roughness
in a radially loaded defective double row bearing, containing raceway defect of
varying depth and size. It was found that the variation in stiffness was large in the
presence of defect, and this variation is mainly a function of defect geometry and
also the applied load and bearing macro-geometry.
538 S. Mufazzal et al.

The varying compliance vibration resulting from continuous load redistribution

has been established well in the literature. However, the investigation on time-varying
oil film damping has not been reported yet. Since the damping value can considerably
affect the overall system response, its evaluation seeks equal attention as does the
stiffness.
Further, the presence of defect can dramatically alter the oil film damping and
contact stiffness. Hence, it becomes essential to model these behaviors to precise
find the system’s dynamic response and diagnosing the faults. Considering these
aspects, efforts have been put in the present work to evaluate the varying nature of
damping and stiffness in healthy and defective bearing, with single point defect on
the outer raceway. Moreover, a quantitative study has been presented on the effect of
controlling factors, to obtain a detailed view of dynamic characteristic of lubricated
ball-raceway contact.

2 Mathematical Model

A. Healthy Bearing

Linearized contact stiffness

Under the application of static radial load (Fr ), the ball-raceway contact gets elas-
tically deformed and takes the shape of elliptical patch. The elastic deformation
(δ) due to non-conformal bodies in dry contact can be obtained using Hertzian non-
linear theory. Even in the presence of lubricating oil, this deformation is substantially
the same as for unlubricated case [13], which can be obtained using the following
Hertzian non-linear relationship:
3/2
Fψ = K n δψ (1)

where, δψ denotes the surface deflection of the contact located at an angle ψ w.r.t
the load line, under the effect of normal load Fψ . K n is the load–deflection constant
and can be obtained using the following expression:
√
K n = π κ E ' 2ε R/9 3 (2)

where, κ is the ellipticity parameter of the contact patch, E ' is the effective elas-
ticity modulus, R is the effective radius of curvature of the contacting bodies and
ε and represent the elliptic integrals of first and second kind, respectively. The
linearized stiffness can be calculated by substituting K n in the generalized expression
of stiffness, and it is given by:

3 2/3 1/3
K cψ = K F (3)
2 n ψ
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 539

The values of contact load at different angular location can be approximated with
reasonable accuracy using the load distribution formulation illustrated in Ref. [22].
The net value of contact stiffness (K cT ) along the load line will be obtained by
summing the directional contribution (K cl ) due to each ball-raceway contact in the
load zone, i.e.:

ψl
∑
K cT = K cl cosψ; where[−ψl , ψl ] is the angular limit of load zone (4)
−ψl

Due to the involvement of finite number of rolling elements taking up the load, the
load distribution pattern changes continuously during rotation in a periodic manner.
As a result, the stiffness, being a load dependent parameter, also follows a similar
periodic trend.
Squeezed film damping
When the ball rolls over the surface, the oil film at the entry region tends to squeeze
due to which it owns considerable magnitude of stiffness and damping. The viscous
damping due to squeezing action of the oil can have a prominent role in governing the
vibration behavior of the entire bearing and it can be estimated using the following
relationship given by Hagiu and Gafitanu [8]:
( )
3π Rx 1.5
Ce = √ η0 (2a) (5)
2 h min

where, Rx is the effective radius of curvature in the direction of ball roll, h min is
the minimum film thickness of lubricating oil, η0 is the oil dynamic viscosity at
atmospheric condition and a is the semimajor axis of the elliptical contact patch.
The minimum film thickness (h min ) is calculated using the dimensionless film thick-
ness (Hmin = h min /Rx ) formula, derived by Hamrock and Dowson [23] as a func-
tion of three other dimensionless parameters (U = η0 u x /E ' Rx , G = ξ E ' and,
W = Fψ /E ' Rx2 ) related to material and operating parameters considering isothermal
conditions (i.e. neglecting the effect of temperature on lubricant’s viscosity), as given
below:
( )
3.63U 0.68 G 0.49 1 − e−0.68κ
Hmin = (6)
W 0.073
where U represents dimensionless speed parameter, G represents dimensionless
material parameter, while W denotes dimensionless load parameter. Further, κ is the
ellipticity parameter of the ball-raceway contact patch. The net value of squeezed
film damping (CeT ) along the load line will be obtained by summing the directional
contribution (Ceψ ) due to the contact of each ball in the load zone, i.e.:
540 S. Mufazzal et al.

ψl
∑
CeT = Ceψ cosψ; where [−ψl , ψl ] is the angular limit of load zone (7)
−ψl

It is evident that the variables defining the oil film damping, including the contact
patch geometry, film thickness and load parameter, have small or large dependency
on normal load acting at the contacting elements, the damping also follows contin-
uous but periodic time varying trend, like the effective load on element and contact
stiffness.
B. Defective bearing
In the presence of defect on the outer raceway, the ball undergoes additional deflection
(h s ) while traversing that defect, which is given by:

h s (ψ) = (1 − cosβ)(Rb + rd ); for small rd (8)

where, Rb is the ball radius and rd is the defect edge radius, β is the instantaneous
angular location of the ball w.r.t. the entry location the defect zone, and it can be
calculated from kinematics and geometric constraints of the bearing. The total radial
deflection of the balls at defect location becomes:

δT = δψ + h s (ψ); (9)

The additional deflection, h s is zero for healthy bearings, or for ball outside
the defect zone. Using the equation of overall deflection, the varying stiffness and
the squeezed damping can be evaluated iteratively for healthy as well as defective
bearings.
As will be evident from the numerical simulation results in the later section,
the contact stiffness is a speed-independent parameter, whereas the damping has
a strong relationship with speed. Moreover, both the parameters are a function of
applied radial load.

3 Result and Discussion

In order to simulate the dynamic contact characteristics, a single row, deep groove
ball bearing was chosen whose specifications are provided in Table 1. The values of
contact stiffness and EHL damping were computed by solving Eqs. (4) and (7) in
MATLAB 2019b environment, under the desired operating conditions. The simula-
tions were carried out both for the healthy bearing and the bearings with localized
defect on the outer race. The defect parameters considered in this study are also listed
in Table 1.
Figure 1 shows the results of squeezed film damping (Ce ) and the linearized
contact stiffness (K cl ) for a perfectly healthy bearing.
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 541

Table 1 Bearing geometry and defect parameters

Bearing geometry Defect parameters
Inner race diameter, di (mm) 24.24 Defect length, ad (mm): 0.1; 1; 1.5
Outer race diameter, do (mm) 41.64
Groove radii, r g (mm) 5
Diametral clearance, Pd (µm) 10 Angular location of defect, αd (degree): 0; 10;
Ball diameter, db (mm) 8.70 20
Number of balls, Z 7 Defect Edge radius, rd (µm): 1; 10; 50
Inner and outer ball-race conformities, 0.5747
f (mm)

Fig. 1 Varying compliance parameters for healthy bearing at Ns = 2000RPM and Fr = 1kN

It can be observed that as the bearing rotates, the values of these parameters vary
periodically at angular ball spacing (θs ) because after each ball reaches the position
of its next ball, the bearing configuration becomes the same. As a result, there is a
continuous change in the load distribution between the loaded balls, which causes
repetitive widening and shrinking of the load zone. Consequently, the number of
balls in the load zone (N L Z ) changes periodically with rotation, as observed from
Fig. 2.
It is clearly revealed jointly from Figs. 1 and 2 that when the geometric configu-
ration attains symmetry-I condition about the load line, there are maximum number
of balls sharing the load, thereby contributing large to the values of Ce and K cl .
As the bearing starts rotating and the configuration changes, the values of these
contact characteristics reduce, attaining minima at symmetry-II configuration. For
the bearing with specifications, given in Table 1, with IR rotating at 2000RPM under
the application of 100N radial load, the values of N L Z changes from 2 to 3 while
moving from symmetry-I to symmetry-II configuration.
542 S. Mufazzal et al.

Fig. 2 a Type I symmetry (with maximum number of balls in the load zone), and b Type II symmetry
(with minimum number of balls in the load zone)

The plots for Ce and K cl for the case of bearing with defective OR have been
depicted in Fig. 3.
The varying sinking depth of the ball inside the defect while moving from entry to
exit of the spall can also be seen from Fig. 3. During this interval, there are dramatic
reductions both in damping and stiffness values. It is mainly due to the change in
the contact geometry after the bearing witnesses a defect. On arrival of defect in
the bearing, the effective radius of curvature along the direction of ball roll (Rox ),
minimum film thickness (h min,o ) and semimajor axis of elliptical contact (ao ) at ball-
OR contact become smaller. For the present case, the reduction of about 44.881%,
17.968% and 61.255% was seen, respectively in Rox , h min,o and ao , for a single

Fig. 3 Varying compliance parameters for defective bearing at Ns = 2000RPM, Fr = 1kN,

ad = 1 mm, rd = 10μm and αd = 0°
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 543

contact. The cumulative effect of these geometric parameters for the contacts due to
all the balls in the load zone will give a smaller value of Ce , according to Eq. (7).
Similarly, the increments of about 46.088% and 13.028% were seen, respectively,
in the ellipticity parameter (κ) and the elliptic integral of first kind ( ), while the
elliptic integral of second kind (ε) and the effective radius of curvature (R) reduced
by 3.839% and 41.139%, respectively. This yields smaller value of load–deflection
constant (K n ) and the linearized contact stiffness (K cl ), as per Eq. (4). The sudden
reductions in the values of stiffness and damping at the defect entry and exit locations
are attributed to the sharp edge of the newly formed defect, assumed in the present
work. As the ball repetitively hits the defect edges with time, the edges become dull,
and the gradient in the stiffness and damping values will become gradual.
Since the dynamic contact characteristics are strongly dependent on operational
conditions and defect geometry, the effects of these variables have been studied in
the following sub-sections.
A. Effect of operational parameters:

Effect of shaft speed:

Figure 4 shows the variation in the values of Ce and K cl at three values of shaft speed
for 0.1 s duration. It reveals that as the shaft speed increases, the damping is reduced,
while there is no effect on the magnitude of K cl . As the speed rises, the reductions in
both Ce and K cl become more frequent due to increasing outer race defect frequency.
It is evident from here that the vibration in the bearing will become stronger at higher
speed, both due to lowering of speed dependent damping coefficient and the frequent
reduction in stiffness and damping due to additional deflection and load redistribution
while passing the defect.

Fig. 4 Plots of Ce and K cl at different values of shaft speed with Fr = 1kN, ad = 1 mm,
rd = 10μm and αd = 0°
544 S. Mufazzal et al.

Further, the variation in contact stiffness is stronger than that in damping. However,
similar strength of variation is observed at different shaft rpms for both the plots.
Effect of static radial load:
Figure 5 shows the variation in the values of Ce and K cl at three values of radial load.
It reveals that as the load increases, both the damping and the stiffness improve.
This is due to the involvement of higher number of load sharing balls at larger loads,
thereby creating more contacts, each contributing to the total value of damping and
stiffness. The trend of these curves and the location of maxima and minima vary on
changing the load because of different load redistribution pattern. At higher values
of load, the bearing vibration gets suppressed due to increase in load dependent
damping coefficient and stiffness.
The study of non-linear contact parameters indicates their strong dependence on
the operational load and speed. Hence, these parameters should be appropriately
considered during simulation of the system’s response.
B. Effect of Defect Geometry:

Effect of defect length:

Figure 6 shows the plots of damping and stiffness for three different values of defect
size. These defect sizes are sufficiently small to avoid hitting of the ball with the defect
base and thus they represent initial stage of spall generation. The figure indicates
that the defect size has no influence on the strength and the trend of damping and

Fig. 5 Plots of Ce and K cl at different values of radial load with Ns = 1000RPM, ad = 1mm,
rd = 10μm and αd = 0°
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 545

Fig. 6 Plots of Ce and K cl at different values of defect length with Ns = 1000 RPM, Fr = 1kN,
rd = 10μm and αd = 0 ◦

stiffness coefficients. However, the interval during which the damping and stiffness
get reduced while passing across the defect, changes according to the defect length.
Effect of defect angular location:
Figure 7 shows the variation in the values of Ce and K cl at three values of defect
angular locations. It may be observed that as the angular location of the defect is
varied, the regions of fall in the damping and stiffness values shift correspondingly.
The angular location of defect, however, does not affect the magnitude of decay in
damping and stiffness values as long as the ball resides inside the load zone. This is
because the total load supported by balls inside the load zone is same as the applied
radial load, irrespective of the defect location. Although it may be expected that the
effect of the presence of defect should be low for defects in the region away from the
load line because the balls contribute to small load sharing in that region. But when
the ball loses its contact while negotiating the defect, the load born by that ball is
now distributed to the remaining balls in the load zone, and hence their contributions
to the overall values of contact characteristics are raised.
Effect of defect edge radius:
Figure 8 shows the variation in the values of Ce and K cl at three values of defect
edge radius. Like the defect size, the edge radii have non-considerable impact on
the magnitude of damping and stiffness characteristics. The reduction in damping is
more in new defects with smaller edge radius while this reduction becomes a little
lesser as the defect grows and the edge radius becomes larger. For the present case,
the minimum damping values in Ns/m are 1111.1, 1141.1 and 1216.3, at edge radius
546 S. Mufazzal et al.

Fig. 7 Plots of Ce and K cl at different values of defect location with Ns = 1000 RPM, Fr = 1
kN, ad = 1mm and rd = 10μm

of 0.001 mm, 0.01 mm and 0.05 mm, respectively. On the other hand, the reduction
in stiffness is extremely minute and it does not occur one way with respect to the
edge radius.
Finally, to quantify the impact of each parameter on the time varying values of
stiffness and damping, the trends corresponding to different conditions have been

Fig. 8 Plots of Ce and K cl at different values of defect edge radius with Ns = 1000 RPM, Fr = 1
kN, ad = 1 mm and αd = 0 ◦
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 547

evaluated, and results listed in Table 2. In nearly all the plots, the maximum values
are same for healthy and defective cases, however, the minimum values depend on
the reduction caused due to the presence of the defect. Hence, the maximum values
provided in Table 2 belong to both healthy and defective bearings.
The variations in K c,d are smaller than in Ce,d in all the cases. A variation of about
56.24% in Ce,d and 20% in K c,d are observed in defective bearing for all the three
values of shaft rpms, whereas a maximum variation of 88.91% in Ce,d and 38.43%
in K c,d is observed at the smallest static load of 100 N. Moreover, the maxima and
the minima remain the same irrespective of the defect length. Among all the five
factors considered, the influence of load on the strength of stiffness and damping is
the highest, followed by the shaft speed, while the influence of defect parameters are
the least.

4 Conclusion

The time-varying oil film damping and contact stiffness have been evaluated under
different sets of operational parameters and defect geometry. It was found that as the
shaft rpm increases, the damping is reduced, while there is no effect on the magnitude
of contact stiffness. On the other hand, the damping as well as the stiffness were found
to increase with increase in load. At higher values of load, the bearing vibration gets
suppressed due to increase in the values of load-dependent damping coefficient and
stiffness.
In case of defective bearing, the damping and stiffness fall sharply when the ball
encounters a defect on the outer raceway. The reduction in damping is more in newer
defects having smaller edge radii while this reduction becomes relatively smaller
as the defect grows older and the edge radius becomes larger due to repetitive ball
impacts.
A considerably large variation in the damping and stiffness was witnessed during
the rotation. The variations are as high as 89% in squeezed film damping and 38%
in contact stiffness for the selected test bearing rotating at 1000RPM under 100N
radial load. The percentage variations in K c,d are smaller than that in Ce,d in all
the cases. A variation of about 56.24% in Ce,d and 20% in K c,d occur in defective
bearing for all three values of shaft rpms, whereas a maximum of 88.91% change in
Ce,d and 38.43% in K c,d are observed at the smallest static load of 100N. Moreover,
the maxima and minima remain same irrespective of the defect length.
Among all the five factors considered, the effect of load on the strength of stiffness
and damping is the highest, followed by the shaft speed, while the effects of defect
parameters are the least.
The consideration of these time-varying contact parameters will help in correct
modeling of the time-localized bearing vibration response of healthy and defective
bearing, useful for both design and diagnostic applications.
Table 2 Summary of the effect of controlling parameters on the time-varying squeezed film damping and contact stiffness
548

Shaft speed
Ns (RPM) Ce,H (Ns/m) Ce,D (Ns/m) K cl,H or K cl,D (N/m)
Max Min % Variation Min % Variation Max Min % Variation
1000 2.61E + 03 2.45E + 03 6.086 1.14E + 03 56.24 1.41E + 08 1.12E + 08 20.743
2000 1.28 E + 03 1.21E + 03 562.68
3000 850.33 798.58 372.09
Applied radial load
Fr (kN) Ce,H (Ns/m) Ce,D (Ns/m) K cl,H (N/m) K cl,D (N/m)
Max Min % Variation Min % Variation Max Min % Variation Min % Variation
0.1 886.60 728.9 17.78 98.36 88.91 5.19E + 07 4.05E + 07 22.07 3.75E + 07 38.43
1.5 3.13E + 03 2.94E + 03 5.99 1.41E + 03 54.99 1.63E + 08 1.28E + 08 21.25 1.28E + 08 26.98
2.5 3.93E + 03 3.75E + 03 4.5 1.81E + 03 53.83 1.94E + 08 1.69E + 08 12.71 1.69E + 08 14.56
Defect length (for all the three sizes)
Ce,H (Ns/m) Ce,D (Ns/m) K cl,H or K cl,D (N/m)
Max Min % Variation Min % Variation Max Min % Variation
2.61E + 03 2.45E + 03 6.09 1.14E + 03 56.23 1.41E + 08 1.12E + 08 20.74
(continued)
S. Mufazzal et al.
Table 2 (continued)
Defect edge radius
rd (mm) Ce,H (Ns/m) Ce,D (Ns/m) K cl,H or K cl,D (N/m)
Max Min % Variation Min % Variation Max Min % Variation
0.001 2.61E + 03 2.45E + 03 6.09 1.11E + 03 54.65 1.41E + 08 1.12E + 08 20.74
0.01 1.14E + 03 53.42
0.05 1.22E + 03 50.36
Angular location of defect
αd (Deg) Ce,D (Ns/m) K cl,D (N/m)
Max Min % Variation Max Min % Variation
0 2.61E + 03 1.14E + 03 53.42 1.41E + 08 1.12E + 08 26.17
10 1.15E + 03 55.90
20 1.15E + 03 56.05 1.04E + 08 35.47
Note
Subscripts H and D imply characteristics associated with healthy and defective bearings, respectively
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball …
549
550 S. Mufazzal et al.

Acknowledgements The first author of this paper acknowledges the financial support of Depart-
ment of Science & Technology (DST), India, for supporting this research in the form of Junior
Research Fellowship (JRF)-INSPIRE.

References

1. Sunnersjö CS. Varying compliance vibrations of rolling bearings. J Sound Vib. 1978;58(3):363–
73.
2. Liew H, Lim TC. Analysis of time-varying rolling element bearing characteristics. J Sound
Vib. 2005;283:1163–79.
3. Liu J, Shao Y. A numerical investigation of effects of defect edge discontinuities on contact
forces and vibrations for a defective roller bearing. 2016; 230(4):387–400.
4. Liu J, Shao Y, Zhu WD. A new model for the relationship between vibration characteristics
caused by the time-varying contact stiffness of a deep groove ball bearing and defect sizes.
2015;137(3).
5. Liu Y, Zhang Y. A research on the time-varying stiffness of the ball bearing considering the
time-varying number of laden balls and load distribution. J Mech Eng Sci. 2018;233(12):4381–
96.
6. Walford TLH, Stone BJ. The measurement of the radial stiffness of Rolling Element Bearings
under oscillating conditions. Proc Inst Mech Eng C J Mech Eng Sci. 1980;22(4):175–81.
7. Walford TLH, Stone BJ. The sources of damping in rolling element bearings under oscillating
conditions. Proc Inst Mech Eng C J Mech Eng Sci. 1983;197(4):225–32.
8. Hagiu GD, Gafitanu MD. Dynamic characteristics of high speed angular contact ball bearings.
Wear. 1997;211:22–9.
9. Kraus J, Blech JJ, Braun SG. In situ determination of rolling bearing stiffness and damping by
modal analysis. J Vibr Acoustic Stress Reliab Design. 1987;109(7):235–40.
10. Mitsuya Y, Sawai H, Shimizu M, Aono Y. Damping in vibration transfer through deep groove
ball bearings. J Tribol. 1998;120:413–20.
11. Dietl P, Wensing J, Nijen GCV. Rolling bearing damping for dynamic analysis of multi-body
systems—experimental and theoretical results. Proc Instit Mech Eng Part K J Multi-body Dyn.
2000;214(1):33–43.
12. Jacobs W, Boonen R, Sas P, Moens D. The influence of the lubricant film on the stiffness and
damping characteristics of a deep groove ball bearing. Mech Syst Signal Process. 2014;42:335–
50.
13. Lambert RJ, Pollard A, Stone BJ. Some characteristics of rolling-element bearings under oscil-
lating conditions: part 1: theory and rig design. Proc Instit Mech Eng Part K J Multi-Body Dyn.
2006;220:157–170.
14. Sarangi M, Majumdar BC, Sekhar AS. Stiffness and damping characteristics of lubricated ball
bearings considering the surface roughness effect: part 2: numerical results and application.
Proc Instit Mechanical Engineers, Part J: Journal of Eng Tribol. 2004;218(6):539–48.
15. Sarangi M, Majumdar BC, Sekhar AS. On the dynamics of elastohydrodynamic mixed lubri-
cated ball bearings part i formulation stiffness and damping coefficients. Proc Instit Mech Eng
Part J: J Eng Tribol. 2005;219(6):411–21.
16. Sun P, Chen W, Shen Y, Wang D. Calculation method for comprehensive damping of ball
bearings based on multigrid method. Ind Lubr Tribol (Accepted). 2020.
17. Zhang YY, Wang X, Yan X. Dynamic behaviors of the elastohydrodynamic lubricated contact
for rolling bearings. J Tribol. 2013;135(2):0215011–8.
18. Zhang Y, Liu H, Zhu C, Liu M, Song C. Oil film stiffness and damping in an elastohydrodynamic
lubrication line contact-vibration. J Mech Sci Technol. 2016;30(7):3031–9.
Study of Time-Varying Oil Film Damping and Contact Stiffness of Ball … 551

19. Zhang Y, Xie L, Lv X, Sun G. The influence of elastohydrodynamic lubrication on the stiffness
of deep groove ball bearing. J Vibroeng. 2016;18(7):4178–92.
20. Hooke CJ, Espejel GEM. The effects of small sinusoidal load variations in elastohydrodynamic
line contacts. J Tribol. 2016;138(3):0315011–9.
21. Peterson D, Howard C, Sawalhi N, Ahmadi AM, Singh S. Analysis of bearing stiffness varia-
tions, contact forces and vibrations in radially loaded double row rolling element bearings with
raceway defects. Mech Syst Signal Process. 2015;50–51:139–60.
22. Harris TA, Kotzalas MN. Essential concepts of bearing technology. 5th ed. New York: Taylor &
Francis; 2016.
23. Hamrock BJ, Dowson D. Isothermal elastohydrodynamic lubrication of point contacts: part III
fully flooded results. J Tribol. 1977;99:264–75.
Static and Vibration Response Analysis
of Pzt-5A/PT Based Smart Functionally
Graded (SFG) Plate Subjected
to Electromechanical Loading

Pawan Kumar and S. P. Harsha

Abstract The present research concerns the static and free vibration study of a smart
functionally graded (FG) plate subjected to electromechanical loading at various
boundary conditions (BC’s). The distribution of material properties along the thick-
ness direction is estimated using the sigmoid law (S-Law). Hamilton’s principle,
displacement fields, and first order shear deformation theories (FSDT) are used to
find the governing equation for the motion of the SFG plate. Then, the governing
equation is solved using the finite element method (FEM) and 9-noded Lagrange
elements with 6 degrees of freedom (DOFs) per node. Convergence and compar-
ison tests were performed for the suggested approach’s efficacy. Numerical results
are presented comprehensively for the different side to thickness ratios, material
gradient exponent, and applied electric voltage with different BC’s.

Keywords Smart functionally graded (FG) plate · Finite element method (FEM) ·
Hamilton’s principle · Electromechanical load

1 Introduction

Smart FG plate is a metal and piezoelectric ceramic-based plate broadly used in smart
composite structures. The variation of material properties according to the power law,
exponential law, and S-Law. Several shear deformation theories, including CLPT,
FSDT, and HSDT, are used to formulate the governing equation of motion. For
solving the governing equation of motion using various methods such as analytical
and numerical techniques [1–3]. Because of its exceptional smart properties, piezo-
electric materials are widely used as functionally graded forms in modern times
[4–7]. Huang XL et al. investigated the dynamic behavior of an FG plate integrated

P. Kumar (B) · S. P. Harsha

Department of Mechanical and Industrial Engineering, IIT Roorkee, Roorkee, Uttrakhand, India
e-mail: [email protected]
S. P. Harsha
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 553
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_37
554 P. Kumar and S. P. Harsha

with piezoelectric material layers in a thermal environment. Batra et al. [8–10] used
analytical and meshless techniques to present a FG plate’s dynamic analysis using
third-order shear deformation theory. Zenkour [11] uses 3-D trigonometric shear
deformation theory to describe the bending behavior of a Functionally graded (FG)
piezoelectric plate. The results were obtained under various conditions, such as elec-
tromechanical loading. Lee et al. [12] investigate the elastic behavior of the FG
material plate under a thermal environment. Various researchers [13–20] examined
the dynamic investigation of the FG plate, sandwich plate and the FGP material plate
under different thermoelectromechanical loads [21–23]. The governing equation was
solved using a different technique, and the results were found under various BC’s.
Xue et al. [24] investigate the vibration analysis of an FGP plate subjected to ther-
moelectric loads. The thermomechanical properties are varied using the power law.
Lieu B. Nguyen et al. [25] investigate the vibration behaviour of the FG piezoelec-
tric porous plate using a Bezier isogeometric FEM analysis. Kumar et al. [26–30]
examine a piezoelectric FG plate’s static and vibration analysis. The material distri-
bution law is used to attain the properties variation. A finite element method is used
to solve the obtained governing equation motion.
Considering the facts above, the present study investigates the numerical behavior
of a smart FG plate exposed electromechanical loading with various BC’s. In the
thickness direction, all electromechanical properties vary according to the S-Law.
The FSDT displacements are used to get the final equation of the motion, which
is then solved by FEM. To establish the usefulness and precision of the current
technique, an assessment of the present study with literature outcomes is performed.
In addition to detailing the influence of the applied electric voltage, material gradient
exponent, and a/h ratios with different BC’s.

2 Mathematical Formulation

The smart FG plate (Fig. 1) comprises PZT-5A (top-side) and Pt (bottom-side). The
properties of the smart FG plate vary according to S-Law. The volume fraction stated
as (Fig. 2):
( ) ) ⎫
/ ( / )( K ( / )⎪
Vc1 (z) = − h 2 − z h +1 ; 0≤z≤ h 2 ⎪
⎬
)/ (1)
( / )( K ( / ) ⎪
⎪
Vc2 (z) = h 2 + z h ; −h 2 ≤ z ≤ 0 ⎭

The generalized S-Law for the FG plate material properties is stated in Eq. (2),
and the character K signifies the material gradient exponent. It can be stated as:

(z) = m +( c − m )Vc1
(2)
(z) = m +( c − m )Vc2
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 555

Fig. 1 Smart FG square plate configurations

Fig. 2 S-law distributions of Young’s modulus E(z)

where (z) is the generalized properties of FG plate’s.

The FSDT displacement field is as:
}
u 0 = u 01 + zϕx ; v0 = v01 + zϕ y ; w0 = w01 (3)

where u 01 , v01 and w01 are the displacement, ϕx and ϕ y are the bending rotation. The
strain–displacement relation is:
556 P. Kumar and S. P. Harsha
( ) ( ) ⎫
∂u 01 ∂ϕx 1 ∂w01 2 ∂v01 ∂ϕ y 1 ∂w01 2 ⎪
⎪
εx = +z + , εx = +z + ⎪
⎪
∂x ∂x 2 ∂x ∂y ∂y 2 ∂y ⎪
⎪
( ) ( ) ⎪
⎬
∂u 01 ∂v01 ∂w01 ∂w01 ∂ϕx ∂ϕ y
γx y = + + +z + , ⎪
(4)
∂y ∂x ∂x ∂y ∂y ∂x ⎪
⎪
⎪
⎪
∂w01 ∂w01 ⎪
⎪
γx z =ϕx + , γ yz = ϕ y + ⎭
∂x ∂y

Also, the Eq. (4) are written in terms of linear (L) and nonlinear (NL) vectors,
which are denoted as:

ε = (ε L + ε N L ); (5)

The fundamental equations of the FG smart plate can be specified as [34]:

{ } }
σi j = {ε}[C] − {E}[e]; {Di } = [κ]{E} + [e]{ε} (6)

where [C], [e] and [κ] are the elastic constants, piezoelectric and dielectric coefficient
matrices, respectively. A shear correction (SC) term has been used to reduce the
discrepancy between the imaginary stress and the actual stress. Each matrix are
described as below:
⎡ ⎤ ⎡ ⎤ ⎫
C 11 C 12 0 0 0 0 0 e31 ⎪ ⎪
⎢ ⎥ ⎢ ⎥ ⎪ ⎪
⎢ C 12 C 22 0 0 0 ⎥
⎢ ⎥ ⎢ 0 0 e32 ⎥ ⎪ ⎪
⎪
⎢ ⎥ ⎢ ⎥ ⎪
⎪
[C] =⎢ 0 0 C 44 0 0 ⎥ , [e] = ⎢ 0 e 0 ⎥ , ⎪
⎪
⎢ ⎥ ⎪
⎥ ⎪
24
⎢ ⎥ ⎢ ⎪
⎢0 0 0 C 0 ⎥ ⎣ e 0 0 ⎦ ⎬
⎣ 55 ⎦ 15
(7)
0 0 0 ⎪
⎪
0 0 0 0 C 66 ⎪
⎪
⎡ ⎤ ⎪
⎪
κ 11 0 0 ⎪
⎪
⎪
⎪
⎢ ⎥ ⎪
⎪
[κ] =⎣ 0 κ 22 0 ⎦ ⎪
⎪
⎭
0 0 κ 33

where,
{ }T { }T { }T }
{σ } = σx x σ yy τx z τ yz τx y , {D} = Dx x D yy Dzz , {E} = E x x E yy E zz
⎫
C132
C13 C23 C232
⎪
C 11 = C11 − , C 12 = C12 − , C 22 = C22 − , κ 11 = κ11 ,⎪
⎪
⎪
C33 C33 C33 ⎪
⎪
⎪
⎬
C13 e33
Also, C 44 = C44 , C 55 = C55 , C 66 = C66 , e31 = e31 − , e15 = e15 ,
C33 ⎪
⎪
⎪
⎪
C23 e33 2
e13 ⎪
⎪
e32 = e32 − , e24 = e24 , κ 22 = κ22 , κ 33 = κ33 + ⎪
⎭
C33 C33
(8)
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 557

Let’s us consider an electric θ (x, y, z, t) function with applied electric voltages

V, and it is stated as:
/ /
θ = 2V (z h ) − cos(π z h )θ (x, y, t) (9)

And the electric field can be specified as:

⎧ ⎫ ⎧ ⎫
⎪
⎨ Ex x ⎪
⎬ ⎨ θ,x ⎪
⎪ ⎬
{E} = E yy = − θ,y (10)
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪ ⎭
E zz θ,z

3 Energy Expressions

The SE includes the electromechanical loading of the plate is specified as:

⎫
˚ ⎬
1 ( T )
US E = ε L Cε + ε TN L Cε − ε LT eε − ε TN L eε − E D d V (11)
2 ⎭
V

The PE expression of the smart FG plate due to electric loading is specified by:
⎫
h/2 a a ( )2 ( )2 ⎪
⎪
1 2V ∂w0 2V ∂w0 ⎪
UP E = e31 + e32 d xd ydz ⎪
⎪
⎪
⎪
2 h ∂x h ∂y ⎪
⎬
−h/2 0 0
(12)
h/2 ⎪
⎪
2V 2V ⎪
⎪
where N0x = e31 dz, N0y = e32 dz ⎪
⎪
⎪
⎪
h h ⎭
−h/2

The KE is stated as:

˚ [( )2 ( )2 ( )2 ]
1 ∂u 0 ∂v0 ∂w0
UK E = ρ(z) + + dV (13)
2 ∂t ∂t ∂t
V
558 P. Kumar and S. P. Harsha

Integrating Eqs. (3) into (13), the kinetic energy can be specified as:
⎧ [( ) ( ) ( ) ] ( )⎫ ⎫
⎪
⎪ ∂u 0 2 ∂v0 2 ∂w0 2 ∂u 0 ∂ϕx ∂v0 ∂ϕ y ⎪
⎪ ⎪
⎪
a a⎪
⎪ I0 + + + 2I1 + ⎪ ⎪
⎪
1 ⎨ ∂t ∂t ∂t ∂t ∂t ∂t ∂t ⎪ ⎬ ⎪
⎪
⎪
UK E = [( ] d xd y ⎪
⎪
⎪ ) ( ) ⎪ ⎪
⎪
2 ⎪
⎪ ∂ϕ 2 ∂ϕ y
2 ⎪
⎪ ⎪
⎬
0 0 ⎩ ⎪ +I2
x
+ ⎪
⎭
∂t ∂t
⎪
⎪
⎪
⎪
h/2 ⎪
⎪
) ( ⎪
⎪
with (I0 , I1 , I2 ) = 2
1, z, z ρ(z)dz ⎪
⎪
⎪
⎪
⎭
−h/2
(14)

The external workdone on the smart FG plate can be specified as:

¨
UW D = qw0 d A (15)
A

Thus, the overall energy is:

U = U K E − U S E − U P E + UW D (16)

Substituting the Eqs. (11), (12), (14), (15) into (16). Then the attain eqs are:
⎫
∂ Nx x ∂ Nx y 2 ∂ Nx y ∂ N yy ⎪
⎪
+ = I0 ü 01 + I1 ϕ̈x ;
2
+ = I0 v̈01 + I1 ϕ̈ y
2 2
⎪
⎪
∂x ∂y ∂x ∂y ⎪
⎪
⎪
⎪
∂ Sx x ∂ S yy ∂ 2
w ∂ 2
w ⎪
⎪
+ + q + Nx 0 0
+ N0 0
= I ẅ 2 ⎪
⎪
∂x ∂y ∂x 2 y
∂y 2 0 01 ⎬
∂ Mx x ∂ Mx y ∂ Mx y ∂ M yy ⎪
+ − Sx x = I1 ü 201 + I2 ϕ̈x2 ; + − S yy = I1 v̈01
2
+ I2 ϕ̈ y2 ⎪
⎪
⎪
∂x ∂y ∂x ∂y ⎪
⎪
⎪
⎪
∂ Dx x ∂ Dx y ⎪
⎪
⎪
⎪
+ − D zz = 0 ⎭
∂x ∂y
(17)

where
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 559
⎫
h/ h/ ⎪
2 2 ⎪
[ ] [ ] [ ] [ ] ⎪ ⎪
⎪
N x x N yy N x y = σx x σ yy τx y dz, Sx x S yy = k τx z τ yz zdz ⎪ ⎪
⎪
⎪
⎪
⎪
⎪
h
− /2 h
− /2 ⎪
⎪
⎪
⎪
⎪
⎪
h/ ⎪
⎪
2 ⎪
⎪
[ ] [ ] ⎪
⎪
Mx x M yy Mx y = σx x σ yy τx y zdz, ⎪
⎪
⎪
⎪
⎪
⎪
h ⎪
⎬
− /2
(18)
h/ ⎪
⎪
2 ⎪
⎪
( / ) ⎪
⎪
Dx x = − cos(π z h ) Dx x dz ⎪
⎪
⎪
⎪
⎪
⎪
h ⎪
⎪
− /2 ⎪
⎪
⎪
⎪
⎪
⎪
h/ h/
2 ( / ) ⎪
⎪
2 π z ⎪
⎪
( π / ) d − cos( h ) ⎪
⎪
D yy = − cos( z ) D dz, D = D dz ⎪
⎪
h yy zz
dz
zz ⎪
⎪
⎪
⎭
h
− /2 h
− /2

4 Solution Methodology

The mathematical approach of the governing eqs for the static and vibration behaviour
of the smart FG plate is investigated to gratify the different combinations of BC’s
(Fig. 3a). The presumed applied boundary conditions of the smart FG plate are
CCCC, CFCF, SCSF, CFFF, SFSF, SSSS, and SCSC. It represented as:

a. X-axis along BC’s:

[ ( )]
Clamped case: (u 01 = v01 = w01 = 0); ϕx = ϕ y = 0 ;
[ ( )]
Simply supported case: (u 01 = w01 = 0); ϕx = M y = 0

b. Y-axis along BC’s:

[ ( )]
Clamped case: (u 01 = v01 = w01 = 0); ϕx = ϕ y = 0 ;
[ ( )]
Simply supported case: (v01 = w01 = 0); ϕ y = Mx = 0 ;

A finite element (FE) formulation with a continuous C0 9-noded isoparametric

FE (Fig. 3b) with 6-DOFs per node is used.
 560

Fig. 3 The arrangement of the BC’s (a) and nine noded isoparametric finite elements with 6-DOFs (b) Plate discretized at 14 × 14 mesh element (c)
P. Kumar and S. P. Harsha
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 561
⎫
∑
e
N
⎪
⎪
{ }= i{
}i ,
e ⎪
⎪
⎪
⎪
i=1 ⎪
⎪
⎪
⎪
or ⎪
⎪
⎪
⎪
⎪
⎪
∑9 ∑
9 ⎪
⎪
{u 01 } = e
{u } ; {v } = e
{v } ; ⎪
⎬
i 01 i 01 i 01 i
(19)
i=1 i=1
⎪
⎪
∑
9 ∑
9 ⎪
⎪
⎪
⎪
{w01 } = e
{w } , {θ } = e
{θ } ⎪
⎪
i 01 i i i ⎪
⎪
⎪
⎪
i=1 i=1
⎪
⎪
∑9 ∑
9 ⎪
⎪
⎪
⎪
{ϕx } = e
i {ϕ x }i ; {ϕ x } = e
i {ϕ x }i ; ⎪
⎭
i=1 i=1

where { }i are wide-ranging vector which contains 6-DOFs. N e is the number of

nodes per element. The 9-noded shape ( i ) function is given as [31]:
⎛ ⎞
1 ( )( )
⎜
e
1 = (1 − ξ )(−ξ − η − 1)(1 − η) + 1 − ξ 2 1 − η2 ⎟
⎜ 4 ⎟
⎜ 1 ( )( ) ⎟
⎜ e
= (1 + ξ )(ξ − η − 1)(1 − η) + 1 − ξ 2 1 − η2 ⎟
⎜ 2
4 ⎟
⎜ ⎟
⎜ 1 ( )( ) ⎟
⎜ e
= (1 + ξ )(ξ + η − 1)(1 + η) + 1 − ξ 1 − η
2 2 ⎟
⎜ 3 ⎟
⎜ 4 ⎟
⎜ 1 ( )( )⎟
⎜ e
= (1 − ξ )(−ξ + η − 1)(1 + η) + 1 − ξ 1 − η ⎟
2 2 ⎟
⎜ 4
⎜ 4 ⎟
⎜ ( )( ) ⎟
e
=⎜ e 1
= (1 − ξ )(1 + ξ )(1 − η) − 1 − ξ 1 − η
2 2 ⎟ (20)
i
⎜ 5 ⎟
⎜ 2 ⎟
⎜ ( )( ) ⎟
⎜ 1 ⎟
⎜ e
= (1 + ξ )(1 − η)(1 + η) − 1 − η 1 − ξ
2 2
⎟
⎜ 6
2 ⎟
⎜ ( )⎟
⎜ 1 2 ⎟
⎜ e
= (1 − ξ )(1 + ξ )(1 + η) − (1 − ξ )(1 + ξ ) 1 − η ⎟
⎜ 7
2 ⎟
⎜ ( ) ( ) ⎟
⎜ 1 ⎟
⎜ e
= (1 − ξ ) 1 − η − (1 − η)(1 + η) 1 − ξ
2 2 ⎟
⎝ 8
2( ⎠
)( )
e
9 = 4 1 − ξ 2 1 − η2

Inserting the Eqs. (17) and (18) into Eqs. (19) and (20), and reordering the terms.
Hence, the attained eqs is:
[ ] { }
[K nl ]{U } − K p {U } + [M] Ü + [K l ]{U } + [K uθ ]{θ } = {Fme } (21)

[K θu ]{U } − [K θθ ]{θ } = {Fel } (22)

By inserting the Eqs. (22) into (21) and obtain as:

562 P. Kumar and S. P. Harsha

[ ] [ ]−1 )[ ] [ ] [ ] [ ][ ]−1 ( { }
{Fme } + K uθ {Fme } K θθ = K nl + K l − K p + K θu K θθ {U } + [M] Ü (23)

The equilibrium generalized illustration of the Eq. (23) is:

{ }
[K ]{U } + [M] Ü = {F} (24)

The above-generalized eqs of the static and free vibration are expressed as follows:

[K ]{U } = {F} (25)

{ }
[K ]{U } + [M] Ü = 0 (26)

and the corresponding eigenvalue eqs is:

|( )|
| K − Ω2 M | = 0 (27)

where Ω is the frequency of the plate. Also, Fme and Fel are the externally applied
mechanical and electrical load, respectively. The modified iterative technique [32,
33] is used to solve the Eqs. (25) and (26).

5 Numerical Results and Discussions

In this segment, the present method’s convergence, usefulness and constancy are
confirmed by different examples and comparisons. The present investigation is used
to analyze the smart FG plate’s static and free vibration behaviour. The arrangement
of the BC’s is displayed in Fig. 3. The presumed nondimensional parameters are
(Table 1). ) 2(
Nondimensional axial stress σ ∗ = σ qa h
2 .

∗
( )
Nondimensional center deflection w = w h1 .
) 2 ( )/ (
ρ
Nondimensional frequency Ω∗ = Ω ah C11
.
upper

5.1 Convergence Assessment

The convergence assessment of the Smart FG plate has been shown for the steadiness
of the present technique. The presumed dimension is K = 2, a = b, and a = 10 h.
Table 2 shows that the convergence amount of the stress, deflection and frequency
of the Smart FG clamped plate and 14 × 14 is the optimal mesh element (Fig. 3c)
for the investigations.
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 563

Table 1 Smart FG plate’s material properties [18, 21]

Properties PZT-5A Pt
E(G Pa) – 147
ρ (Kg/m3 ) 7750 2150
v 0.35 0.39
Elastic constant (GPa)
C11 = C22 99.2010 –
C33 86.8560 –
C12 54.0160 –
C13 = C23 50.7780 –
C44 = C55 21.10 –
C66 22.60 –
/
Piezoelectric constant (C m2)
e15 = e24 12.3220 –
e31 = e32 −7.2090 –
e33 15.1180 –
/
Dielectric constant (×10−9 F m)
κ11 = κ22 1.530 –
κ33 1.500 –

Table 2 Convergence assessment of smart plate

Parameters 9*9 10*10 11*11 12*12 13*13 14*14
Ω∗ 0.5943 0.59439 0.59461 0.59465 0.59469 0.59471
w∗ 0.2527 0.25279 0.25285 0.25289 0.02541 0.02542
σ∗ 0.3324 0.33245 0.33252 0.33255 0.3326 0.33262

5.2 Comparison Assessment

The effectiveness and precision of the present technique are established by comparing
it with existing works. Four different problems are assumed for comparison of the
current technique.

Problem 1 The illustration depicts the nondimensional deflection of a simply

supported FGM plate made of Al/ZrO2 . The material and geometrical properties
are assumed to be the same
/ ( as Ref. The attained nondimensional center deflec-
( [15]. ))
tion w∗ = 100wE m h 3 12qa 4 1 − v2 is shown (Table 3) and compared with
existing works [12, 15, 16].

Problem 2 Table 4 compares the simply supported FGM plate’s nondimen-

sional axial stress with the literature values. The material and geometrical parameters
564 P. Kumar and S. P. Harsha

Table 3 Comparison of center deflection of the plate

Methods K
0.5 1 2
Lee et al. [12] 0.2403 0.2811 0.3221
Castellazzi et al. [15] 0.2349 0.2620 0.2977
Gilhooley et al. [16] 0.2505 0.2905 0.3280
Current 0.2496 0.2787 0.3189

Table 4 Comparison of axial stress of the FG plate

Methods K
2 4 6 8
Behjat et al. [18] 5.1735 5.8162 6.3429 6.8125
Zenkour [19] 5.2296 5.8917 6.4043 6.8999
Present 5.1811 5.8204 6.3482 6.8201

of the plate are the{)

same as(Refs.
} [18, 19]. The obtained nondimensional dimensional
/
axial stress σ x = h aq σx is shown in Table 4 and compared with Ref. [18, 19].

Problem 3 Table 5 compared the FGM (Al/ZrO2 ) plate’s nondimensional frequency

to the literature results. The material and geometrical parameters of the plate are
identical and taken as the same as Refs. [20, 21]. The obtained nondimensional
( 2 / )/ /
frequency ω = ω a h ρc E c are presented in Table 5 with K.

Problem 4: In this illustration, the nondimensional nonlinear to the linear frequency

of the simply supported FGM plate is obtained and equated with available results
[22]. The assumed material properties are the same as Ref. [22] and geometrical

Table 5 Comparison of the FG material plate’s frequency

BC’s Method K
1 2 5 8 10
SSSF Kumar and Harsha[20] 3.1689 3.1631 3.2297 3.2321 3.2209
Su et al. [21] 3.1695 3.1637 3.2302 3.2326 3.2214
Present 3.1705 3.1688 3.2332 3.2347 3.2254
SCSF Kumar and Harsha [20] 3.4146 3.4074 3.4766 3.4786 3.4664
Su et al. [21] 3.4153 3.4080 3.4772 3.4791 3.4669
Present 3.4168 3.4102 3.4795 3.4811 3.4698
SFSF Kumar and Harsha [20] 2.6296 2.6253 2.6820 2.6844 2.6751
Su et al. [21] 2.6297 2.6254 2.6821 2.6845 2.6752
Present 2.6325 2.6281 2.6901 2.6864 2.6788
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 565

Table 6 Comparison of the nondimensional nonlinear to the linear frequency of the Simply
supported FGM plate
( / )
Method wmax h
0.2 0.4 0.6 0.8 1.0
Abrate [22] 1.021 1.081 1.174 1.293 1.432
Present 1.023 1.083 1.176 1.292 1.436

configuration of the plate is a =( b = ) mm and h = 25 mm. Table 6 demonstrates

/ 200
the nonlinear frequency 𝛡( = ω/nl ) ωl with various values of the nondimensional
center point displacement wmax h at the power-law exponent K = 2.

5.3 Static Analysis

Under electromechanical loading, the numerical simulation of static bending and

deflection for the smart FG plate with seven BC’s is performed. Table 1 shows the
material properties of the smart FG plate and geometric dimensions
) assumed as a =
( / )( / )4 (
10 h. For the sake of simplicity, the assumed load parameter P = q E m a h
and the influence of mechanical loading ‘q’ is uniform on the smart plate. Table 7
displays the nondimensional center deflection for various BCs and material gradient
indices (K). The plate’s stiffness increases with the rise of K, and the deflection
decreases with the rise of K.
Table 8 displays the nondimensional axial stress for various BCs and material
gradient exponents. The axial stress increases as the K rises from 0 to 10. Because
stresses in the homogeneous structure were unaffected by the material and when the
K was 0 or infinity. CFFF has the maximum axial stress among all other BC’s.
Figures 4 and 5 depicts the effect of the varying K and a/h ratios on the stresses and
deflection of a smart FG plate subjected to mechanical loading. It can be supposed
that nondimensional axial stress decreases as the K rises and increases as the a/h
ratio increases because the plate stiffness is dependent on the a/h ratio and K. The

Table 7 Comparison of the center deflection with K and BC’s

K BC’s
SFSF CFCF SCSF CCCC CFFF SSSS SCSC
0 0.01721 0.00622 0.01600 0.00253 0.23172 0.00765 0.00342
2 0.01701 0.00604 0.01590 0.00253 0.23120 0.00763 0.00339
5 0.01691 0.00594 0.01577 0.00253 0.23042 0.00761 0.00337
8 0.01679 0.00584 0.01572 0.00253 0.22915 0.00758 0.00335
10 0.01674 0.00574 0.01557 0.00253 0.22787 0.00755 0.00335
566 P. Kumar and S. P. Harsha

Table 8 Comparison of the axial stress with K

K BC’s
CFFF SSSS SCSC SFSF CFCF SCSF CCCC
0 0.23172 0.00765 0.00342 0.01721 0.00622 0.01600 0.00253
2 0.23120 0.00763 0.00339 0.01701 0.00604 0.01590 0.00253
5 0.23042 0.00761 0.00337 0.01691 0.00594 0.01577 0.00253
8 0.22915 0.00758 0.00335 0.01679 0.00584 0.01572 0.00253
10 0.22787 0.00755 0.00335 0.01674 0.00574 0.01557 0.00253

Fig. 4 Effect of the K and a/h ratio on the axial stress

other BC’s have the same deflection and axial stress responses in the variations of the
a/h ratios, and K. Figures 6 and 7 depicts the effect of voltage on the deflection and
stress of a smart FG plate subjected to electromechanical conditions with a variable
a/h ratio. Stress and deflections increase as the a/h ratio increases because the smart
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 567

Fig. 5 Effect of the K and a/h ratio on the center deflection

plate becomes stiffer as the a/h ratio rises. Also, the center deflection and axial stress
decline as the applied loading rises from 0 to 200 V, increasing as the electrical
loading decreases from 0 V to −200 V. Because negative increases the axial tensile
load, which increases the stiffness, and positive loading decreases plate stiffness.

5.4 Free Vibration Study

In this section, a numerical simulation of the vibration behaviour of the smart FG

square plate exposed to electromechanical loading was performed. Table 9 shows
the plate’s first six modes of frequencies vary with various BCs at K = 2, a = 10 h.
In all the BC’s, CCCC has a higher nondimensional frequency than other BC’s.
Table 10 shows how the nondimensional frequency varies with electric voltages
under different BCs. It can be seen that rising the a/h ratio rises the nondimensional
frequencies in all BCs. This is because the geometrical nonlinearity is enhanced,
and the nonlinear stiffness matrices have an additional effect on the SFG plate’s
568 P. Kumar and S. P. Harsha

Fig. 6 Influence of the a/h ratio and V on the nondimensional center deflection

free vibration behaviour. Also, the frequency of the smart plate rises as positive
loading is applied and decreases as negative loading is applied. It implies that the
nondimensional frequency is directly affected by the nature of the sign of the voltages.
The frequency of the smart FG plate exposed to mechanical loading is depicted in
Fig. 8 at various BCs and K. The frequency appears to decrease as the K increases.
Because the K increases, a greater proportion of the metal phases. The same pattern
can be seen in all of the BC’s depicted in Fig. 8.
Figures 9 and 10 illustrates the frequency of the smart FG plate exposed to elec-
tromechanical loading with BC’s. Figure 9 depicts the nondimensional frequency
with a/h ratio when applied electrical loading. It shows that as the a/h ratio rises, the
nondimensional frequency increases in positive and declines in negative voltages.
The variation of the nondimensional frequency with the material gradient exponent is
shown in Fig. 10. It shows that increasing the K lowers the frequency under constant
electrical loading. This occurs because increasing the K enriches the metal side,
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 569

Fig. 7 Effect of the a/h ratio and voltages on axial stress

Table 9 First six frequency modes

Mode number BC’s
SCSC SFSF CFCF SCSF CCCC CFFF SSSS
1 1.91887 0.83255 1.39854 0.96204 2.24388 0.23152 1.2807
2 3.55074 1.02136 1.57664 2.09697 4.26885 0.50654 3.07245
3 4.10014 1.18075 2.56084 2.78221 4.26897 1.30574 3.07245
4 5.15455 2.26859 3.55354 3.61391 5.97396 1.41632 3.69061
5 5.48829 2.66426 3.71319 3.83975 7.0362 1.70644 3.69173
6 6.16525 2.96233 3.76541 3.88246 7.1063 1.80696 4.67548
570 P. Kumar and S. P. Harsha

Table 10 Comparison of the nondimensional frequency with voltages (K = 2)

a/h V BC’s
SCSC SFSF CFCF SCSF CCCC CFFF SSSS
10 −200 2.15687 1.07055 1.63654 1.20004 2.48188 0.46952 1.51870
−100 2.02287 0.93655 1.50254 1.06604 2.34788 0.33552 1.38470
0 1.91887 0.83255 1.39854 0.96204 2.24388 0.23152 1.28070
100 1.81487 0.72855 1.29454 0.85804 2.13988 0.12752 1.17670
200 1.78487 0.69855 1.26454 0.82804 2.10988 0.09752 1.14670
100 −200 2.36751 1.19196 1.71044 1.33112 2.67675 0.42976 1.52999
−100 2.23351 1.05796 1.57644 1.19712 2.54275 0.29576 1.39599
0 2.12951 0.95396 1.47244 1.09312 2.43875 0.19176 1.29199
100 2.02551 0.84996 1.36844 0.98912 2.33475 0.08776 1.18799
200 1.99551 0.81996 1.33844 0.95912 2.30475 0.05776 1.15799

improving the stiffness of the smart FG plate. Also, the voltages effect increases and
decreases the frequency as the voltages loading changes. Positive electric loading
produces axial compressive forces, while negative electric loading produces axial
tensile forces, causing the stiffness of the SFG plate to decrease and increase, respec-
tively. Furthermore, the expanded material gradient exponent results in a lower plate
frequency.

6 Conclusions

The present works investigate a smart FG plate’s static and vibration analysis with
various BCs subjected to electromechanical loading. The final eqs of motion are
deduced using Hamilton’s principle and displacement fields based on FSDT. The
obtained governing eqs solved by FEM using nine-noded Lagrange elements with
6-DOFs per node. This paper presents all of the results in tabular and graphical form.
The present study’s findings are as follows:
• The center deflection diminishes, and stresses rise as the K increases. Similarly,
the stresses and center deflection increase as the a/h ratio rises.
• Depending on the sign of the applied electric loading, frequency, stresses and
deflection are significantly affected.
• At constant K and a/h ratios, the first six frequencies increase as the modes
increase.
• The nondimensional frequency increases and decreases when the a/h ratio and K
rise.
• The nondimensional frequency decreases with rises of K at a fixed applied voltage.
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 571

Fig. 8 Nondimensional frequency response of the smart FG plate with K and a/h ratio
572 P. Kumar and S. P. Harsha

Fig. 9 Nondimensional frequency response of the Smart plate with voltage and a/h ratio
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 573

Fig. 10 Comparison of the nondimensional frequency of the smart plate with V and K
574 P. Kumar and S. P. Harsha

• The current method is also used for other different combinations of the BC’s under
electromechanical conditions for a smart FG structure, irrespective of the BC’s
and loading environments.

References

1. Koizumi M, Niino M. Overview of FGM research in Japan. MRS Bull. 1995;20:19–21.

2. Wang CM, Reddy JN, Lee KH. Shear deformable beams and plates. Shear Deform Beams
Plates. 2000.
3. Reddy JN. Analysis of functionally graded plates. Int J Numer Methods Eng. 2000;47:663–84.
4. Praveen GN, Reddy JN. Nonlinear transient thermoelastic analysis of functionally graded
ceramic-metal plates. Int J Solids Struct. 1998;35:4457–76.
5. Batra RC, Vidoli S. Higher order piezoelectric plate theory derived from a three-dimensional
variational principle. AIAA J. 2002;40(1):91–104.
6. Huang XL, Shen HS. Nonlinear vibration and dynamic response of functionally graded plates
in thermal environments. Int J Solids Struct. 2005;41:2403–27.
7. Huang XL, Shen HS. Vibration and dynamic response of functionally graded plates with
piezoelectric actuators in thermal environments. J Sound Vib. 2006;289:25–53.
8. Ferreira AJM, Batra RC, Roque CMC, Qian LF, Martins PALS. Static analysis of functionally
graded plates using third-order shear deformation theory and a meshless method. Compos
Struct. 2005;69:449–57.
9. Qian LF, Batra RC. Design of bidirectional functionally graded plate for optimal natural
frequencies. J Sound Vibr. 2005;280:415–24.
10. Ferreira AJM, Fasshauer GE, Batra RC, Rodrigues JD. Static deformations and vibration
analysis of composite and sandwich plates using a layerwise theory and RBF-PS discretizations
with optimal shape parameter. Compos Struct. 2008;86:328–43.
11. Zenkour AM, Hafed ZS. Bending analysis of functionally graded piezoelectric plates via quasi-
3D trigonometric theory. Mech Adv Mater Struct. 2020;27(18):1551–62.
12. Lee YY, Zhao X, Liew KM. Thermoelastic analysis of functionally graded plates using the
element-free kp-Ritz method. Smart Mater Struct. 2009;18(3): 035007.
13. Batra RC, Jin J. Natural frequencies of a functionally graded rectangular plate. J Sound Vibr.
2005;282:509–16.
14. Harsha A, Kumar P. Thermoelectric elastic analysis of bi-directional three-layer function-
ally graded porous piezoelectric (FGPP) plate resting on elastic foundation. Forces Mech.
2022;8:100112.
15. Castellazzi G, Gentilini C, Krysl P, Elishako I. Static analysis of functionally graded plates
using a nodal integrated finite element approach. Compos Struct. 2013;103:197–200.
16. Gilhooley DF, Batra RC, Xiao JR, McCarthy MA, Gillespie JW. Analysis of thick functionally
graded plates by using higher-order shear and normal deformable plate theory and MLPG
method with radial basis functions. Compos Struct. 2007;80(4):539–52.
17. Kumar P, Harsha SP. Static, buckling and vibration response analysis of three-layered func-
tionally graded piezoelectric plate under thermo-electric mechanical environment. J Vibr Eng
Technol. 2022;10(4):1561–1598.
18. Behjat B, Salehi M, Armin A, Sadighi M, Abbasi M. Static and dynamic analysis of functionally
graded piezoelectric plates under mechanical and electrical loading. Sci Iran. 2011;18(4):986–
94.
19. Zenkour AM. Generalized shear deformation theory for bending analysis of functionally graded
plates. Appl Math Model. 2006;30(1):67–84.
20. Kumar P, Harsha A. Vibration response analysis of the bi-directional porous functionally graded
piezoelectric (BD-FGP) plate. Mech Based Des Struct Mach. 2022. https://doi.org/10.1080/
15397734.2022.2099418.
Static and Vibration Response Analysis of Pzt-5A/PT Based Smart … 575

21. Su Z,Jin G, Ye T. Electromechanical vibration characteristics of functionally graded piezoelec-

tric plates with general boundary conditions. Int J Mech Sci. 2018:138–139, 42–53.
22. Abrate S. Free vibration, buckling, and static deflection of functionally graded plates. Compos
Sci Technol. 2005;66:2383–94.
23. Vel SS, Mewer RC, Batra RC. Analytical solution for the cylindrical bending vibration of
piezoelectric composite plates. Int J Solids Struct. 2004;41:1625–43.
24. Xue Y, Li J, Li F et al. Active control of plates made of functionally graded piezoelectric material
subjected to thermo-electro-mechanical loads. Int J Struct Stab Dyn. 2019;19:1950107.
25. Nguyen LB, Thai CH, Zenkour AM, Nguyen-Xuan H. An isogeometric Bézier finite element
method for vibration analysis of functionally graded piezoelectric material porous plates. Int J
Mech Sci. 2019:157–158, 165–183.
26. Kumar P, Harsha SP. Vibration response analysis of exponential functionally graded piezoelec-
tric (EFGP) plate subjected to thermo-electro-mechanical load. Compos Struct. 2021:113901.
https://doi.org/10.1016/j.compstruct.2021.113901
27. Kumar P, Harsha SP. Static analysis of porous core functionally graded piezoelectric (PCFGP)
sandwich plate resting on the Winkler/Pasternak/Kerr foundation under thermo-electric effect.
Mater Today Commun. 2022;32:103929.
28. Kumar P, Harsha S. Response analysis of hybrid functionally graded material plate subjected
to thermo-electro-mechanical loading. Proc Instit Mech Eng Part L J Mater Des Appl. 2020.
https://doi.org/10.1177/1464420720980031S.
29. Kumar P, Harsha SP. Vibration response analysis of PZT-4/PZT-5H based functionally graded
tapered plate subjected to electromechanical loading. Mech Res Commun. 2021;116:103765.
https://doi.org/10.1016/j.mechrescom.2021.103765
30. Kumar P, Harsha SP. Dynamic analysis of porosity dependent functionally graded sigmoid
piezoelectric (FGSP) plate. In: Structures, vol. 46. Elsevier; 2022. p. 1737–52. https://doi.org/
10.1016/j.istruc.2022.11.021
31. Jiashi Y. An introduction to the theory of piezoelectricity. USA: Springer Science; 2005.
32. Bergan PG, Clough RW. Convergence criteria of iteration process. AIAA J. 1972;10:1107–8.
33. Bathe KJ. Finite element procedures. New Jersey: Prentic-Hall; 1996.
34. Reddy JN. An introduction to finite element method. 3rd ed. New York: TataMcGraw-Hill
Edition; 2009.
Influence of Squeeze Film Damper
on the Rub-Impact Response
of a Dual-Rotor Model

K. Prabith and Praveen Krishna

Abstract The main aim of this paper is to study the influence of squeeze film damper
on the dynamic characteristics of a dual-rotor model undergoing rub-impact. Gener-
ally, the two-spool rotor system in an aero-engine is mounted on nonlinear supports,
including rolling contact bearings, squirrel-cage elastic support, and a squeeze film
damper. Even though the squeeze film damper is an arrangement to reduce the vibra-
tions, it acts as a source of nonlinearity. A dynamic model is developed using one-
dimensional finite elements to study its effect on the nonlinear dynamics of a dual-
rotor system. In this model, the low-pressure rotor consists of a fan, a compressor,
and a turbine, while the high-pressure rotor consists of a compressor and a turbine.
Both the rotors are supported on bearings in which the bearing near the fan disk is
mounted inside a squeeze film damper. The equation of motion of the squeeze film
damper is derived using Reynold’s equation based on short bearing approximation.
The results are obtained using the numerical integration technique and are analyzed
using frequency response, FFT spectrum, and Poincaré maps. The effects of different
squeeze film parameters on the response behavior are also investigated. The results
show that the amplitudes of vibrations are significantly reduced due to the presence of
a squeeze film damper. The parameters such as squeeze film clearance and coefficient
of viscosity have a remarkable effect on the dynamics of the dual-rotor system.

Keywords Aero-engine · Dual-rotor · Rub-impact · Nonlinear support ·

Squeeze-film damper

K. Prabith (B) · Praveen Krishna

Department of Aerospace Engineering, IIST, Thiruvananthapuram, India
e-mail: [email protected]
Praveen Krishna
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 577
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8_38
578 K. Prabith and Praveen Krishna

Nomenclature

η Co-efficient of viscosity
δ Rotor-stator clearance
μ Co-efficient of friction
κ Speed ratio
c0 Ball bearing clearance
c SFD radial clearance
e SFD journal eccentricity
Cb Hertz contact stiffness
kc Rotor-stator contact stiffness
L Width of the damper
Nb Number of balls in the raceway
R SFD journal radius
DOF Degrees of Freedom
FFT Fast Fourier Transform
HP High Pressure
LP Low Pressure
SFD Squeeze Film Damper

1 Introduction

In aero-engines, there are several reasons for occurring engine vibrations, such as
the presence of unbalanced components, looseness in the assemblies, external excita-
tions including wind blasts, and rudder flutter [1]. The excessive vibration can cause
disturbance to the aircraft crew and damage the critical engine parts. Sometimes, it
may lead to the destruction of the aeroengine as well. In modern aircraft engines,
the rotor structure is supported inside a squeeze film damper (SFD) to minimize
vibration amplitudes. However, a poor design of the SFD will induce nonlineari-
ties in the machine, which in turn threatens its safe working. As of now, significant
numbers of scientific researches have been carried out to investigate the vibration
reduction mechanism of the SFD arrangement. In 1975, Gunter [2] examined the
effect of the SFDs on the unbalance response of the aero-engine rotors. It is observed
that the faulty design of the SFD can increase the hydrodynamic stiffness of the
bearing, which will reduce the damping rate and increase the force transmission
rate. By preventing the cavitation inside the fluid film, it is possible to decrease the
hydrodynamic stiffness. In such cases, a retainer spring is employed to provide load
carrying capacity to the SFD. Inayat-Hussain [3] studied the bifurcation character-
istics of a flexible rotor mounted inside an SFD, connected with retainer springs. As
the value of retainer spring stiffness increases, the speed at which the bifurcation
has happened increases. Saddle-node bifurcation and secondary Hopf bifurcation
are mainly noticed for different SFD parameters. Chu [4] analyzed the influence of
Influence of Squeeze Film Damper on the Rub-Impact Response … 579

various parameters and observed that the onset of instability was delayed due to the
presence of SFD.
In turbomachines, the SFD arrangement consists of a ball bearing, an oil-film
damper, and a squirrel-cage elastic support. It is a combined nonlinear support in
which the ball bearing and the rotor shaft whirl inside the oil-film damper during its
operation. Luo [5] performed research to understand the vibration characteristics of
a flexible rotor mounted inside a combined nonlinear support. The force produced
by the nonlinear support is analyzed, and its effects on the rotor motion are studied.
Various combinations of excitation frequencies are noticed in the response due to
nonlinearity [6]. The presence of a ball bearing can introduce a jump phenomenon in
the frequency response because of its hard spring characteristics. However, the SFD
can effectively reduce amplitude jump by providing excessive damping in the model
[7]. The studies also prove that the sub- and super-harmonic vibrations generated due
to the flight maneuvring [8] and base excitations [9] are also significantly reduced
due to the presence of SFDs. Moreover, there are some modified versions of the SFDs
as well, called floating-ring squeeze film dampers (FSFD), having two layers of oil
films, separated by a floating ring [10]. Whatever the design, incorporating SFDs is
beneficial during the fan blade loss for reducing the transient vibrations generated
due to the sudden unbalance force. It is verified by Huang [11] by performing a
transient analysis in a dual-rotor system supported by SFD. Sometimes, the blade
loss will result in a rubbing phenomenon when the disk deflection exceeds the blade-
stator clearance. It creates contact nonlinearity in the model, leading to motions such
as full annular rub, backward whirl, sub & super-harmonic vibrations, and partial
rubs [12]. Moreover, different types of bifurcations also appear in the system while
analyzing the stability [13]. A very few researchers [14–16] have investigated the
influence of SFD on the rubbing response of a rotor system. However, most of them
have employed a simple Jeffcott rotor model to understand the vibration behavior.
But it will be helpful for the qualitative analysis only.
In this work, the influence of SFD on the rub-impact nature of a dual-rotor model
is studied. The model consists of an overhung fan, and it is subjected to unbalance
forces during its operation. The rotors are mounted inside the combined nonlinear
support, including ball bearing, oil-film damper, and squirrel-cage elastic support.
The dynamic model is constructed using finite elements, and the solution procedure
is carried out using the Runge–Kutta method. The Hertz contact theory is used to
determine the ball bearing forces, while Reynold’s equation is employed to derive the
oil-film forces. The responses of the model for various values of SFD parameters are
investigated and compared. The frequency response curves, time signals, Poincaré
maps, and orbit plots are utilized to study the response characteristics.
The current article is categorized into five sections. Section 1 provides an introduc-
tion and review of the different studies concerned with SFDs. Later, Sect. 2 describes
the dynamic modeling of the two-spool aero-engine system undergoing rotor–stator
rub. The different nonlinear forces in the model, such as ball bearing force, oil-film
force, and rub-impact force, are also expressed in detail. Section 3 explains the mode
shapes and critical speeds of the model and are verified using the ANSYS results.
In Sect. 4, the response characteristics of the model undergoing rotor–stator rub are
580 K. Prabith and Praveen Krishna

analyzed. The influences of various parameters such as SFD oil viscosity and SFD
clearance are also studied. Finally, Sect. 5 presents the conclusion.

2 Dynamic Model of the Two-Spool Aero-Engine

The simplified form of a two-spool aero-engine comprises two co-axial rotors

connected by an inter-shaft bearing. The arrangement is also termed as a dual-rotor
model, including high-pressure (HP) & low-pressure (LP) rotors. Each rotor contains
single-disk turbine and compressor units, as shown in Fig. 1. An overhung disk is
attached to the LP rotor representing the aero-engine fan assembly. The fan, turbine,
and compressor units are modeled using rigid disks with the inertial properties equiva-
lent to them. Both the rotors are supported on ball bearings except inter-shaft bearing,
which is assumed to be a linear spring-damper model. It should be noted that the
bearing near the fan disk is mounted inside an SFD arrangement while all other bear-
ings are attached with elastic supports. The rotating speeds of two rotors are different
with a ratio that is a fractional number “κ.” The physical properties and dimensions
of the two-spool rotor are given in Appendix A.
The finite element model of an aero-engine two-spool rotor is developed using
one-dimensional beam elements based on Nelson and McVaugh [17] finite element
formulation. The shafts are modeled using the Timoshenko beam elements, including
their rotational inertia and gyroscopic moments in the formulation. The LP rotor
is divided into eight elements with nine nodes, whereas the HP rotor is divided
into four elements with five nodes. The nodal degrees of freedom (DOF) are four,
representing two translational and rotational motions. As a result, the complete
finite element model has a DOF of 56. The model has five rigid disks representing
the fan, compressors, and turbines. The inertial properties of disks are applied to
corresponding nodes.
The governing equation of the LP rotor is expressed as

Fig. 1 Schematic drawing of a simplified two-spool aero-engine model consisting of a single-stage

turbine and compressor units mounted inside ball bearings. Bearing 1 is supported inside an SFD
arrangement to reduce the amplitude of vibration generated due to unbalance force in the fan
Influence of Squeeze Film Damper on the Rub-Impact Response … 581

ML q̈L + (CL + ωL GL )q̇L + KL qL + FnL = FL (1)

where ML , CL , KL, and GL are the mass, damping, stiffness, and gyroscopic matrices
respectively. The subscript “L” denotes the LP rotor. The matrix CL is obtained using
the relation CL = α ML + β KL , where α and β are the Rayleigh damping coefficients.
The angular velocity is denoted using ωL , while the displacement, velocity, and
acceleration vectors are represented using qL , q̇L , and q̈L respectively. FL is the
unbalance force vector and Fnl is the summation of ball bearing force, oil-film force
and rub-impact force vectors (FnL = Fball + Fsfd + Frub ). More details about these
nonlinear forces are given in Sect. 2.1. Similarly, the governing equations of the HP
rotor can be expressed as

MH q̈H + (CH + ωH GH )q̇H + KH qH + FnH = FH (2)

where the subscript “H” represents the HP rotor. It should be noted that FnH doesn’t
include rub-impact force and oil-film force. The inter-shaft bearing couples the
dynamics of both rotors; hence, the governing equations of the complete model
are derived by assembling Eqs. 1 and 2 as given

Mq̈ + Cq̇ + Kq + Fn = F (3)

where M, C, K, and G are the assembled mass, damping, stiffness, and gyroscopic
matrices respectively. The size of these global matrices is 56 × 56. F and Fn are the
assembled unbalance force and nonlinear force vectors of size 56 × 1.

2.1 Nonlinear Forces in the Dual-Rotor Model

As mentioned earlier, the nonlinear forces in the dual-rotor model include ball bearing
forces, oil-film forces due to SFD, and rub-impact forces. The expressions for these
forces are explained in detail as follows.

2.1.1 Ball Bearing Model

Ball bearings are widely used in many applications since they can operate at high
speeds and carry loads in axial and radial directions. In a two-spool rotor model,
the outer race of the ball bearing is connected to the housing/SFD while the inner
race is jointed to the rotor shaft. The schematic diagram of the ball bearing is shown
in Fig. 2. During the operation, point contacts happen between the balls and the
raceway, which reduces the friction between them. The nonlinear restoring forces
[18, 19] due to the contacts can be obtained using the Hertz contact theory as given
below
582 K. Prabith and Praveen Krishna

Fig. 2 A schematic drawing

of the ball bearing model

∑
Nb
( )3 ( )
f bx = Cb x cos β j + y sin β j − c0 2 H x cos β j + y sin β j − c0 cos β j
j=1

∑
Nb
( )3 ( )
f by = Cb x cos β j + y sin β j − c0 2 H x cos β j + y sin β j − c0 sin β j (4)
j=1

where f bx & f by are the nonlinear forces due to ball bearing in horizontal and vertical
directions. H () is the heavy side function whose value may be 0 or 1, depending
on the sign of the argument. The relative displacements between bearing rings are
denoted using x and y in horizontal and vertical directions. The angular position of
the jth ball is denoted using β j and is described as follows:

2π
β j =ωcage t + ( j − 1), j = 1, 2, . . . ., Nb
Nb
rb
ωcage = ωi (5)
Rb + r b

where ωi represents the inner ring angular velocity. Generally, it is equal to the rotor’s
angular velocity to which the bearing is connected. The inner and outer radii of the
ball bearing are denoted by rb and Rb , respectively. The ball bearing vector Fball
is created by placing f bx & f by in the corresponding positions related to each ball
bearing.
Influence of Squeeze Film Damper on the Rub-Impact Response … 583

2.1.2 Squeeze Film Damper Model

The introduction of the SFD in bearing-1 is mainly intended to reduce the rotor
vibrations produced due to unbalanced forces. A schematic diagram of the SFD
arrangement is given in Fig. 3. The rotating shaft and the ball bearing constitute the
damper’s journal, and it whirls inside the squeeze film damper during its motion.
Meantime, the rotation of the ball bearing’s outer ring is prevented by connecting it
to the squirrel-cage elastic support. The oil is squeezed inside the annular clearance
during the whirling action to produce radial and tangential oil-film forces. They
are derived from Reynold’s equation based on short bearing approximation. It is
explained as follows.
The pressure distribution inside a cavitated oil film is written in rotating
coordinates [9] as
( )
6η 2 L 2 (εω sin θ + ε̇ cos θ )
P(θ, z) = z − (6)
c2 4 (1 + ε cos θ )3

where ε is the eccentricity ratio (ε = e/c) and ε̇ is its first derivative. Now, the force
components in radial and tangential directions can be written as

Fig. 3 A schematic diagram of the SFD arrangement including journal, squirrel-cage elastic
support, and housing
584 K. Prabith and Praveen Krishna

f sr θ1 +π L cos θ
= − ∫ ∫ P(θ, z)R dzdθ (7)
f sθ θ1 0 sin θ

By expanding Eq. 7, the radial and tangential force components can be rewritten
as

ηR L3 ηR L3
f sr = − [I 1 ε̇ + I 2 εω] & f sθ = − [I2 ε̇ + I3 εω] (8)
c2 c2
where
θ1 +π cos2 θ θ1 +π sin θ cos θ
I1 = ∫ dθ I2 = ∫ dθ
θ1 (1 + ε cos θ ) θ1 (1 + ε cos θ )
3 3
( )
θ1 +π sin2 θ −1 ε̇
I3 = ∫ dθ θ 1 = tan − (9)
θ1 (1 + ε cos θ )
3 εω

The integrals I 1 , I 2 , and I 3 are calculated analytically by taking the integral interval
as [π, 2π ]. Finally, the nonlinear oil-film forces in radial and tangential directions
can be expanded as,
[ ( ) ]
η R L 3 π ε̇ 1 + 2ε2 2ωε2
f sr = ( )5/2 + ( )2
c2 2 1 − ε2 1 − ε2
[ ]
ηR L3 2εε̇ π ωε
f sθ = ( )2 + ( )3/2 (10)
c2 1 − ε2 2 1 − ε2

r the numerical analysis, the radial and tangential squeeze film forces are transformed
into the Cartesian coordinate system
as,
[ ]
f sx cos θ − sin θ f sr
= (11)
f sy sin θ cos θ f sθ

The oil-film force vector Fsfd is created by inserting f sx and f sy in the

corresponding positions related to the SFD location.

2.1.3 Rub-Impact Model

The rub-impact in a rotor system happens when the rotor deflection exceeds the rotor–
stator clearance. Hence, a piecewise contact modeling strategy is utilized for repre-
senting the rub-impact phenomenon. The schematic drawing of a disk contacting the
stator is given in Fig. 4. The normal and tangential forces are determined using the
penalty method and are expressed as follows:
Influence of Squeeze Film Damper on the Rub-Impact Response … 585

Fig. 4 The schematic

diagram of a cylindrical disk
rubbing the stator

0 for r < δ
f n (xd , yd ) = (12)
kc (r − δ) for r ≥ δ

/
The variable r is the deflection of the disk in radial direction which is determined
using the expression r = xd2 + yd2 , where xd and yd are the horizontal and vertical
deflections of the disk. Now, the components of the rub-impact forces in horizontal
and vertical directions can be obtained as
( )[ ]
fr x δ 1 −μ xd
= kc 1 − (13)
fr y r μ 1 yd

where is a heavy side function. The rub-impact force vector Frub is created by
inserting fr x and fr y in the corresponding positions wherever the rubbing happens.

3 Mode Shapes and Critical Speeds

Initially, the mode shapes and critical speeds of the rotor model are determined by
performing the modal analysis. The speed ratio κ is assumed as 1.2. The first four
natural frequencies of the model during non-rotating conditions are listed in Table 1.
In addition, the first four mode shapes are also obtained and are validated with a
rotor model developed in the ANSYS software as shown in Fig. 5. Exact matching
is obtained when comparing both the results. The critical speeds of the dual-rotor
model are calculated from the Campbell diagram as given in Fig. 6. The intersection
586 K. Prabith and Praveen Krishna

Table 1 Natural frequencies (in Hz) of two-spool rotor during non-rotation

Model Order
First Second Third Fourth
Current model frequency (Hz) 47.18 123.94 210.39 233.86
ANSYS model frequency (Hz) 47.14 123.83 210.24 233.61
Percentage of error (%) 0.084 0.088 0.071 0.107

points of the natural frequency curves and the synchronous excitation line give the
critical speeds. They are recorded in Table 2.

4 Steady State Response of Two-Spool Rotor During

Rub-Impact

In the current work, the numerical analysis is performed using the Runge–Kutta
method. The unbalance forces are applied at the fan disk (Disk 1) and the HP
compressor disk (Disk 4) positions. The corresponding mass unbalance values are
listed in Appendix A. Initially, the unbalance response of the model is obtained
without considering rub-impact. The effects of SFD clearance and oil viscosity
are analyzed and compared as shown in Fig. 7. The percentage reduction in the
amplitudes of rotor responses for different values of SFD parameters is listed in
Table 3.
Due to the presence of ball bearings, a jump-down phenomenon is noticed in
the frequency response. It mainly happens because of the hard spring nature of
ball bearings. When an SFD has been introduced in the model significant reduc-
tion in the amplitude is happened. As the value of the oil-film clearance decreases,
the amplitude of the model response also reduces. In contrast, a decrease in the
oil viscosity increases the vibration amplitude. It implies that an SFD with low
clearance and moderately high oil viscosity can significantly reduce the vibration
amplitude. However, reducing the clearance below a particular value may result in
contact nonlinearity, and increasing the viscosity above a specific value may result
in locking. Hence, the designer must calculate the optimum SFD parameters for
effective vibration reduction.
To study the influence of SFD on the rub response, a stator is introduced at the
fan disk location. The clearance between the fan and the stator is fixed to be 0.1 mm.
When the fan deflection exceeds rotor–stator clearance, it contacts the stator. As
described in Sect. 2.1.3, normal and tangential forces are developed at the contact
point. The values of contact stiffness and friction coefficient are taken as k c = 1.025
× 108 N/m and μ = 0.01 respectively. The graphs showing the comparison of rotor
responses with respect to the SFD clearance and oil viscosity are given in Fig. 8.
The rub response is the same as the unbalance response until rotor–stator contact. By
checking the Poincaré map at 11,858 rpm, it is seen that a period-5 rotor motion is
 Influence of Squeeze Film Damper on the Rub-Impact Response …

Fig. 5 Mode shapes of two-spool rotor a at 47.18 Hz (first mode) b at 123.94 Hz (second mode) c at 210.39 Hz (third mode) and d at 233.86 Hz (fourth mode)
587
 588

(a) with respect to inner rotor speed (b) with respect to outer rotor speed

Fig. 6 Campbell diagrams of two-spool rotor model

K. Prabith and Praveen Krishna
Influence of Squeeze Film Damper on the Rub-Impact Response … 589

Table 2 Critical speeds of two-spool rotor model

w.r.t. inner rotor speed (rpm) w.r.t. outer rotor speed (rpm)
Order Backward Forward Backward Forward
1 1830 9202 1924 8202
2 5702 11,180 5912 10,730
3 10,740 14,810 11,120 14,330

Fig. 7 Comparison of the rotor responses during no-rub condition for different values of a SFD
clearance and b oil viscosity

Table 3 Percentage reduction in the amplitudes of rotor response compared to no SFD response
Parameter SFD Clearance SFD oil viscosity (Pa.s)
(mm)
c = 0.10 c = 0.05 η = 2.56 × 10−3 η = 5.66 × 10−3 η = 8.86 × 10−3
% of reduction 31.12 73.57 22.37 31.12 37.46

observed in the unbalance response. In addition, the frequency components, including

(2, −1), (1, −2), (3, −2), (2, −3), and (4, −3) are noticed in the FFT spectrum besides
the excitation frequency components (1, 0) & (0, 1). This happens mainly because of
nonlinearity in the ball bearings. When rub-impact occurs, a forward-leaning of the
response curve is noticed due to the increase in model stiffness. The Poincaré map
and the FFT spectrum are utilized to study the nature of rubbing responses. They are
given in Figs. 9 and 10 for certain values of rotating speed (without considering the
SFD). Once the contact has happened, a chaotic nature is observed in the response
that can be identified by looking at the Poincaré map and the FFT spectrum at
11,952 rpm, 12,057 rpm, and 12,510 rpm. After 13,400 rpm, a period-5 rotor motion
is again observed, which can be verified from the Poincaré map at 14,378 rpm. At
14,940 rpm, a jump-down phenomenon has happened, and the motion is transferred
to the no-rub region. While checking the FFT spectrum at 15,165 rpm, it is noticed
590 K. Prabith and Praveen Krishna

Fig. 8 Comparison of two-spool rotor responses during rub-impact for different values of a SFD
clearance and b oil viscosity

that only (1, 0) & (0, 1) are coming in the response. It indicates that the nonlinear
effects of ball bearings are negligible at this speed.
The introduction of SFD in the model made significant variations in the response.
From Fig. 8, it is seen that by decreasing the SFD clearance and by increasing the
oil viscosity, the speed range of rub-impact can be reduced remarkably. For low
clearance and high viscosity, the jump-down phenomenon happens so early that the
model quickly shifts to the no-rub region. Moreover, for very low oil-film clearances,
the percentage of amplitude reduction is so large that the rotor never touches the stator
during its operation. It is shown in Fig. 8 for a clearance value of c = 0.05 mm. This
is the main advantage of SFD; it avoids rotor–stator contact. Figures 11 and 12 show
the Poincaré map and the FFT spectrum of the response when an SFD is incorporated
in the model. The parameters of the SFD are taken in such a way that it doesn’t avoid
rub-impact. A chaotic motion is observed during the rub-impact response, and a
period-5 rotor motion is noticed during the no-rub period. It indicates that the nature
of rubbing responses isn’t altered due to SFD, although the vibration amplitudes are
significantly reduced.

5 Conclusions

In the current paper, the influences of SFD on the rub-impact response of a two-
spool rotor model are studied in detail. The rotor model comprises a fan, turbines,
and compressors in both rotors and is supported on ball bearings. The fan disk is
subjected to an unbalanced force that leads to the whirling of rotors. To minimize
the deflection of the fan, an SFD is incorporated in the bearing near the fan disk.
The dynamic model is created using the one-dimensional finite element method in
which the rotor shaft is discretized using the Timoshenko beam elements. The ball
bearing forces are determined using the Hertz contact theory, while the SFD forces
 Influence of Squeeze Film Damper on the Rub-Impact Response …

Fig. 9 The Poincaré map and FFT spectrum during rub-impact without considering the SFD at a 11,858 rpm, b 11,952 rpm, and c 12,057 rpm
591
 592

Fig. 10 The Poincaré map and FFT spectrum during rub-impact without considering the SFD at a 12,510 rpm, b 14,378 rpm, and c 15,165 rpm
K. Prabith and Praveen Krishna
 Influence of Squeeze Film Damper on the Rub-Impact Response …

Fig. 11 The Poincaré map and FFT spectrum during rub-impact with the SFD (c = 0.087 mm, η = 5.66 × 10−3 ) at a 11,858 rpm, b 12,510 rpm, and c 14,378 rpm
593
 594

Fig. 12 The Poincaré map and FFT spectrum during rub-impact with the SFD (c = 0.1 mm, η = 8.86 × 10−3 ) at a 11,858 rpm, b 12,510 rpm, and c 14,378 rpm
K. Prabith and Praveen Krishna
Influence of Squeeze Film Damper on the Rub-Impact Response … 595

are derived using Reynold’s equation based on short bearing approximation. The
governing equations of the rotor model are solved using numerical integration, and
the results are presented as frequency response curve, FFT spectrum, and the Poincaré
map.
Initially, the no-rub response is analyzed. Due to the presence of ball bearings, a
rightward bending of the response curve is noticed. Besides the excitation frequency
components, the components including (2, −1), (1, −2), (3, −2), (2, −3), and (4, −
3) are observed in the response. A jump-down phenomenon has happened because
of the hard spring characteristics of ball bearings. When an SFD is introduced, the
deflection of the fan disk is significantly reduced. The amplitude of vibration is
reduced by 73.57% when an SFD with a clearance of 0.05 mm is incorporated. In
addition, when an oil having viscosity of 8.66 × 10−3 is utilized, the amplitude of
whirling is reduced by 37.46%. Hence, it can be concluded that an SFD with low
clearance and high oil viscosity will effectively suppress the amplitude of vibration.
During the no-rub response, a period-5 rotor motion is observed. Once the rub-impact
happens, a forward-leaning of the response curve is noticed. The response contains
chaotic motions for a certain range of rotating speeds, then changes to the period-5
motion. Finally, a jump-down phenomenon happens to the no-rub region, which only
contains the excitation frequency components as its constituents.

Appendix: Physical Dimensions and Properties of the Model

See Tables 4, 5, 6, and 7.

Table 4 Physical dimensions of dual-rotor shaft elements

Node Axial location (m) Bearing/disk Outer dia. (m) Inner dia. (m)
1 0.000 Disk 1 0.03 0.00
2 0.076 Bearing 1 0.03 0.00
3 0.152 Disk 2 0.03 0.00
4 0.235 0.03 0.00
5 0.330 0.03 0.00
6 0.400 0.03 0.00
7 0.482 Bearing 4 0.03 0.00
8 0.533 Disk 3 0.03 0.00
9 0.584 Bearing 2 0.03 0.00
10 0.228 Bearing 3 0.06 0.05
11 0.279 Disk 4 0.06 0.05
12 0.355 0.06 0.05
13 0.432 Disk 5 0.06 0.05
14 0.482 Bearing 4 0.06 0.05
596 K. Prabith and Praveen Krishna

Table 5 Properties of dual-rotor model

Property Value Property Value
Mass of Disk 1 (Kg) 16.2 Diametral moment of inertia of Disk 1 0.255
(Kg.m2 )
Mass of Disk 2 (Kg) 10.5 Diametral moment of inertia of Disk 2 0.043
(Kg.m2 )
Mass of Disk 3 (Kg) 7.0 Diametral moment of inertia of Disk 3 0.034
(Kg.m2 )
Mass of Disk 4 (Kg) 7.0 Diametral moment of inertia of Disk 4 0.021
(Kg.m2 )
Mass of Disk 5 (Kg) 3.5 Diametral moment of inertia of Disk 5 0.013
(Kg.m2 )
Polar moment of inertia of Disk 1 0.509 Elastic modulus (GPa) 207
(Kg.m2 )
Polar moment of inertia of Disk 2 0.086 Poisson ratio 0.3
(Kg.m2 )
Polar moment of inertia of Disk 3 0.068 Mass density (Kg/m3 ) 8300
(Kg.m2 )
Polar moment of inertia of Disk 4 0.042 Rayleigh Damping Coeff. (Mass) 11.63
(Kg.m2 )
Polar moment of inertia of Disk 5 0.026 Rayleigh Damping Coeff. (Stiffness) 2.55e-6
(Kg.m2 )
Elastic support stiffness k s1 (N/m) 36e6 Elastic support damping cs1 (Ns/m) 100
Elastic support stiffness k s2 (N/m) 36e6 Elastic support damping cs2 (Ns/m) 100
Inter-shaft bearing stiffness (N/m) 9e6 Inter-shaft bearing damping (Ns/m) 100
Mass unbalance in Disk 1 (Kgm) 4e-5 Mass unbalance in Disk 4 (Kgm) 6e-5

Table 6 Parameters of ball bearings

Order Rb (mm) r b (mm) Nb C b (N/m3/2 ) c0 (µm)
Bearing 1 22.746 15.246 18 6.58e8 10
Bearing 2 22.746 15.246 33 13.25e8 12
Bearing 3 31.196 30.126 24 11.58e8 16

Table 7 Parameters of the

Parameter Description Value
SFD used in the dual-rotor
model R (mm) Journal radius 22.746
L (mm) Width of damper 11.000
c (mm) Clearance 0.1
μ (Pa.s) Viscosity 5.66e-3
K a (N/m) Squirrel-cage stiffness 52e6
Influence of Squeeze Film Damper on the Rub-Impact Response … 597

References

1. Fabry S, Ceskovic M. Aircraft gas turbine engine vibration diagnostics. MAD-Mag Aviation
Develop. 2017;5(4):24–8.
2. Gunter E, Barrett L, Allaire P. Design of nonlinear squeeze-film dampers for aircraft engines.
J Lubr Technol. 1977;99(1):57–64.
3. Inayat-Hussain JI. Bifurcations in the response of a flexible rotor in squeeze-film dampers with
retainer springs. Chaos, Solitons Fractals. 2009;39(2):519–32.
4. Chu F, Holmes R. The effect of squeeze film damper parameters on the unbalance response
and stability of a flexible rotor. J Eng Gas Turbines Power. 1998;120(1):140–8.
5. Luo Z, Wang J, Tang R, Wang D. Research on vibration performance of the nonlinear combined
support-flexible rotor system. Nonlinear Dyn. 2019;98(1):113–28.
6. Wang F, Luo GH, Yan S, Cui HT. A comparison study on co-and counterrotating dual-rotor
system with squeeze film dampers and intermediate bearing. Shock Vibr. 2017;2017.
7. Xinxing M, Hui M, Haiqin Q, Xumin G, Chenguang Z, Mingyue Y. Nonlinear vibration
response characteristics of a dual-rotor-bearing system with squeeze film damper. Chin J
Aeronaut. 2021.
8. Han B, Ding Q. Forced responses analysis of a rotor system with squeeze film damper during
flight maneuvers using finite element method. Mech Mach Theory. 2018;122:233–51.
9. Chen X, Gan X, Ren G. Nonlinear responses and bifurcations of a rotor-bearing system
supported by squeeze-film damper with retainer spring subjected to base excitations. Nonlinear
Dyn. 2020;102(4):2143–77.
10. Zhou Hl, Feng Gq, Luo Gh, Ai Yt, Sun D. The dynamic characteristics of a rotor supported
on ball bearings with different floating ring squeeze film dampers. Mech Mach Theory.
2014;80:200–213.
11. Huang J, Wang Y. Transient Analysis for a twin spool rotor with squeeze-film dampers
considering blade loss. J Vibr Eng Technol. 2020;8(1):95–104.
12. Prabith K, Krishna IP. The numerical modeling of rotor–stator rubbing in rotating machinery:
a comprehensive review. Nonlinear Dyn. 2020;101:1317–63.
13. Prabith K, Krishna IP. The stability analysis of a two-spool rotor system undergoing rub-impact.
Nonlinear Dyn. 2021;104:941–69.
14. Ding Q. Backward whirl and its suppression of a squeeze film damper mounted rotor/casing
system in passage through critical speed with rubs. J Vib Control. 2004;10(4):561–73.
15. Chen G, Li C, Wang D. Nonlinear dynamic analysis and experiment verification of rotor-ball
bearings-support-stator coupling system for aeroengine with rubbing coupling faults. J Eng
Gas Turbines Power. 2010;132(2).
16. Chang-Jian CW. Dynamic analysis of a hybrid squeeze film damper mounted rub-impact rotor-
stator system. J Appl Math. 2012;2012.
17. Nelson H, McVaugh J. The dynamics of rotor-bearing systems using finite elements. J Eng Ind.
1976;98(2):593–600.
18. Lu Z, Zhong S, Chen H, Wang X, Han J, Wang C. Nonlinear response analysis for a dual-rotor
system supported by ball bearing. Int J Non-Linear Mech. 2021;128: 103627.
19. Haslam AH, Schwingshackl CW, Rix AI. A parametric study of an unbalanced Jeffcott rotor
supported by a rolling element bearing. Nonlinear Dyn. 2020;99:2571–604.
Index

A Base excitation, 59, 60, 61, 63, 192, 196,

Acceleration, 66, 71–75, 82, 83, 86, 88, 94, 579
99, 106, 108–111, 113, 194, 209, Basins of attraction, 191, 193, 200, 204
210, 212, 217, 285–290, 292, 294, Bearing, 1, 3–5, 22, 71–80, 82, 83, 116,
376–380, 382, 384, 386, 387, 389, 131–134, 136–138, 140, 144, 145,
395, 513, 517–520, 525, 527, 532, 149–152, 154, 155, 159, 237–241,
581 245, 246, 252, 253, 255, 257–261,
Accelerometer, 59, 65, 66, 117, 210–212, 263, 265–267, 282–285, 288, 289,
257, 317, 325, 375–387, 389, 291, 301–305, 308, 310, 313, 317,
391–393, 493, 518, 532 323–329, 395–397, 453–460,
Active Magnetic Bearing (AMB), 1–11, 463–471, 535–544, 547, 549,
13–22, 131, 133–138, 147, 149–161, 577–583, 586, 590, 595, 596
237–250, 252, 253, 301–305, Bearing Faults, 116, 291, 313, 317,
307–310 323–326
Active Noise Control (ANC), 163–167, Bending, 25–28, 30–32, 34–38, 42, 53–55,
169–171 63, 272–275, 279, 338, 362, 407,
Additional deflection, 540, 543 474, 491, 554, 555, 565, 595
Aero-engine, 504, 577–580 Bistable Energy Harvester, 191, 193, 204
Ambulance, 207–212, 216, 217 Bluff body, 271–279
Analytical method, 407, 474 Boundary conditions, 25, 34, 36, 47, 62, 65,
Angular momentum, 133 91, 182, 186, 187, 208, 210, 221,
Arrival time, 100–103, 113 225, 227, 228, 344, 346, 384, 405,
Artificial intelligence, 116, 283, 284, 323, 413–416, 420, 456, 457, 473–475,
397, 518 481–486, 553, 559
Asymmetric potential function, 191, 193, Bow, 150, 159, 240, 245–247, 249, 250,
204 252
Area Under the Curve (AUC), 401 Braided Composite, 405–408, 414, 415,
Autocorrelation, 168 473–475, 478, 479, 481–484
Buckling, 25–28, 34, 36–42, 227, 407, 440,
474
B Bump Foil, 454, 455, 463, 464, 466–468
Backward whirl, 240, 579
Balanced draft gear, 423, 425, 430
Balancing, 4, 5, 151, 153, 154, 157, 158, C
160–162, 281, 292, 293, 302 Campbell diagram, 45–48, 54–57,
Bandwidth, 204, 380, 381 138–140, 143, 585, 588
© The Editor(s) (if applicable) and The Author(s), under exclusive license 599
to Springer Nature Singapore Pte Ltd. 2023
R. Tiwari et al. (eds.), Vibration Engineering and Technology of Machinery, Volume I,
Mechanisms and Machine Science 137, https://doi.org/10.1007/978-981-99-4721-8
600 Index

Centre buffer coupler, 423, 425 239–241, 257, 259, 263, 267, 303,
Chaotic noise, 172–176 305, 308, 360, 363, 376, 379, 380,
Cold-TURBISTAN, 503–505, 509, 511, 389, 396, 397, 413, 423, 425, 426,
513, 514 430, 431, 441, 449, 454, 465, 480,
Composite beam, 26, 439–443, 449 493, 517, 518, 521–527, 529, 532,
Composite laminate, 220, 227, 229, 443 535–540, 542–549, 581
Composites, 26, 219–222, 224–229, 232, Data driven model, 314
233, 235, 405–408, 410, 420, Decomposition, 184, 359, 361, 364–366
439–444, 446, 449, 473–475, 478, Deep learning, 115–119, 122, 128, 284,
479, 481–484, 553 285, 293, 295, 296, 397
Compression-after-impact, 219, 221, 227, Delamination, 26, 219–221, 223, 224,, 229,
228, 233, 235 230, 232, 233, 235, 474
Condition indicators, 323, 325, 327, 329, Dual-rotor, 577, 579–581, 585, 595, 596
331, 332 Dynamic analysis, 45–49, 52–57, 106, 193,
Condition monitoring, x, 71–74, 77, 78, 83, 246, 387, 439, 490, 491, 494, 496,
116, 150, 239, 240, 256, 257, 302, 498, 499, 554
323–325 Dynamic characterization, 60
Conoidal shells, 405–412, 414–421 Dynamic modeling, 489
Contact stiffness, 535, 536, 538–540, 543, Dynamic modelling of the reaction wheel,
547, 548, 586 131
Continuum approach, 99 Dynamic response, 48, 85, 88, 95, 101,
Coupled, 3–5, 117, 133, 164, 171, 173, 106, 131, 376, 377, 380, 432, 538
181–184, 186–189, 216, 263, 272, Dynamic vibration absorber, 208
434, 463, 475
Coupling, 3–5, 32, 64, 117, 154, 164, 181,
183, 184, 189, 194, 208, 216, 241, E
245, 272, 302, 303, 305, 308, 310, Elastic magnifier, 191, 203
317, 338, 423–425, 434, 436, 537 Electromechanical load, 553, 554, 557,
Coventorware, 375, 380, 384–389, 391, 392 565, 567, 568, 570
Crack, 27, 46–49, 57, 84–90, 94–97, 150,
Electromechanical system, 313, 315, 317,
224, 228–229, 237–243, 245, 247,
319, 321
252, 253, 283, 337–339, 342, 344,
Energy harvesting, 59–61, 63, 65, 67, 69,
346–356, 359–362, 364, 366–372,
191–193, 200, 204, 271, 272, 276
503–514
Equivalent single layer theory, 439, 441,
Crack closure, 503, 504, 508–511, 513, 514
449
Crack depth, 85–87, 90, 240, 346–350, 352,
Experimental test setup, 118
355, 356, 362, 366, 371, 372
Crack detection, 240, 338, 359–361, 366,
367, 369, 371, 372
Crack inclination angle, 85, 86, 89, 95–97 F
Crack modelling, 88, 94 Fast Fourier Transform, 116, 118, 145, 246,
Crash test, 375–377, 384, 392 302, 304, 324
Critical speed, 4, 27, 45–48, 55–57, 131, Fatigue crack growth, 503–505, 507,
138, 139, 143, 157, 266, 283, 293, 509–514
301, 396, 579, 585, 586, 589 Fatigue crack, 237, 240, 245, 360, 503, 504,
Curved flexible link, 498 505, 507–514
Fault classification, 115, 118, 122, 323,
325, 326, 330, 331–333
D Fault diagnosis, 115–117, 119, 121, 123,
Damper, 60, 144, 194, 209, 255, 257, 260, 125, 127–129, 257, 268, 283, 284,
261, 265–267, 395, 397, 517, 518, 297, 313, 314, 324, 397
577, 579, 583 Feature extraction, 117, 119, 122, 128, 129,
Damping, 3, 5, 60–63, 67, 68, 94, 103, 143, 286, 289, 295–297, 314, 324, 325,
156, 157, 182, 194, 195, 209, 237, 332
Index 601

Feedback, 141, 150, 237, 517, 518, 519, Hysteresis, 423, 426, 427, 429, 433, 435,
520, 521, 523, 525, 527, 529, 531, 436, 536
532
Finite element method (FEM), 27, 46, 48,
65, 87, 220, 239, 260, 301, 376, 405, I
407, 420, 473, 474, 481, 486, 491, Identification, 3–5, 15, 21, 22, 150, 151,
518, 536, 553, 554, 590 237, 239, 240, 242–244, 246, 250,
Finite element method, 27, 46, 48, 65, 87, 252, 283, 286, 289, 292, 297, 301,
220, 239, 260, 301, 376, 405, 407, 302, 304, 308, 310, 314, 315, 318,
420, 473, 474, 481, 486, 491, 518, 320, 325, 360, 536
536, 553, 554, 590 Impact, 25, 47, 48, 61, 86–88, 95, 97, 164,
Fixed point theory, 192, 517, 518, 523, 524, 189, 192, 202, 219–221, 224–235,
532 245, 246, 273, 282, 326, 328, 375,
Flexible joint, 489, 491, 493–499, 501 424, 427, 429, 433–436, 454, 465,
Flexible rotor, 151, 239, 395, 397, 398, 504, 536, 545–547
578, 579 Impact dynamics, 25, 86, 87, 192, 245, 424,
Flow induced vibration, 271 436
Flutter, 87, 271–273, 275–279, 578 Impulsive noise, 163, 165, 166, 174–178
Foil journal bearing, 453–456, 459, 460, Induction Motor (IM), 115–117, 119,
463, 465, 466, 469, 470 122–126, 128, 129, 257, 292, 317
Forced vibration, 87 Internal damping, 239–243, 252
Free vibration, 26, 27, 34, 61, 87, 88, 138, Internal damping coefficients, 241
142, 273, 405, 407, 409, 411–413,
415, 417, 419–421, 440, 473–475, K
480, 481, 486, 553, 562, 567, 568 Kurtosis, 120, 257, 319, 320, 324, 325, 327,
Frequency analysis, 75, 76 329–332
Frequency domain data, 116–119, 122,
128, 292, 294
Frequency response function, 48, 199, 524 L
Friction, 133, 225, 285, 289, 302, 425, 426, Laminated composite beam, 440
430, 434, 536, 578, 581, 586 Laminated Composite Structure, 474
Full spectrum, 237, 240, 245–247, 249, Laser, 47, 324
250, 252, 505 Linear-quadratic regulator (LQR) Control,
Functionally Graded Material (FGM), 131, 134, 138, 139, 141, 142, 143,
25–31, 33, 34, 88, 463, 465, 470, 145–147
563–565 Load capacity, 454, 455, 460, 463, 465,
469, 470
Load redistribution, 536, 538, 543, 544
G Longitudinal train dynamics, 424
Gas turbine engine, 3 Low Velocity Impact, 219, 220, 228, 235
Geared-rotor bearing system, 317 Lubricated contact, 537
Genetic algorithm, 314, 324, 517, 518, 532
Gyroscopic moment, 22, 580
M
Machine learning, 256, 281, 283–290,
H 292–296, 314, 323, 325, 327, 332,
Hamilton’s principle, 25, 32, 412, 480, 553, 333, 397, 518
570 Machine tool, 375, 376
Hardness, 51 Magnetic bearing reaction wheels, 133, 144
Harmonic balance method, 193, 199, 204, Magnetic field, 194
397 Magnetorheological fuid, 209
High cycle fatigue, 48 Material grading patterns, 25, 28, 38
High speed rotor, 283, 284 Material non-linearity, 255
Hybrid structures, 60 Mathematical modelling, 1, 6, 22, 441
602 Index

Micro-Electro-Mechanical Systems O
(MEMS), 59, 68, 375–378, 380, On-line vibration monitoring, 440
381, 383–387, 389, 391–393 Operator based approach, 439, 444, 449
Micro-Electro-Mechanical Systems Optimization, 4, 26, 59, 68, 192, 288, 302,
(MEMS) devices, 59, 68, 376 315, 316, 319, 377, 387, 392, 490,
Micro vibration analysis, 147 491, 517, 518, 521, 523, 524, 527,
Misalignment, 1–7, 9–22, 76, 116, 150, 529, 530, 532
239, 282, 301, 302, 307–309, 311, Outer race defect, 535, 538, 540, 541, 543,
396 547
Modal parameter, 68, 337, 338, 339, 344,
346, 356, 360
Modal response, 407 P
Mode shape, 48, 53, 57, 171, 338, 347, 348, Passive vibration absorber, 517, 518, 522,
350, 352, 354, 360, 387, 474, 490, 523
494–496, 498–500, 579, 585, 587 Patient comfort, 208
Mode superposition, 59 Pendulums, 61
Modulation sidebands, 364 Perturbation, 338, 454, 455, 463, 465, 466,
Monte Carlo Simulation (MCS), 101 474
Motor fault diagnosis, 115, 116, 128, 257, PID controller, 2, 6, 10, 15, 149, 154, 156,
313, 314 157, 237, 247, 253, 301, 302, 304,
Moving mass, 85–87, 89, 95 307, 308, 310, 311
Multi-fault classification, 117 Piezoceramic (PZT), 59, 65
Multi-span beam, 339, 344, 346, 347, 353, Piezoelectric energy harvesting, 61, 191,
356 192, 193
Piezoelectric material, 60, 194, 271, 272,
276, 533, 554
N Piezoresistive, 375, 376, 377, 378, 380,
Natural frequency, 27, 35, 45, 46, 48, 53, 387, 389
55–57, 65, 68, 88, 107, 138, 157, Planetary gearbox, 117, 317
171, 210, 212, 213, 215–217, Point defect, 538
245–247, 273, 304, 338, 346, 360, Polymer hysteresis, 423, 424, 426, 427, 429
376, 380, 387, 389, 414–417, 420, Polynomial Chaos Expansion (PCE), 99,
421, 440, 447, 449, 474, 482–486, 102
498, 499–501, 585, 586 Potential energy method, 193, 197
Neural network, 115, 116, 119, 122, 128, Proportional Integral Derivative (PID)
163, 166, 169, 170, 255, 261, controller, 6, 150, 245, 301, 302
264–266, 282, 284, 288, 291–293, Pseudo-elasticity, 181, 182, 186
295, 397 Pulse loading, 157
Noise, 4, 71–75, 82, 83, 117, 119, 163–168,
171–178, 193, 289, 291, 292, 301,
302, 310, 311, 314, 360, 361, 363, R
364, 367, 372, 387, 399 Reaction wheel, 131, 133–136, 138, 139,
Nonlinear, 88, 117, 150, 166, 192, 193, 143–147
196, 199, 200, 292, 293, 296, 297, Reaction wheel assemblies, 131
314, 316, 395–397, 399, 406, 423, Receiver Operating Characteristic (ROC),
425, 426, 430, 431, 474, 475, 518, 401, 403
538, 556, 564, 565, 567, 577, 579, Regression, 238, 239, 242, 243, 255, 256,
581, 582, 584, 590 264, 285, 297, 308, 310
Nonlinear damping, 430 Response matching, 257
Nonlinear support, 577, 579 Rigid rotor, 1, 5, 7, 15–17, 21, 22, 154, 157,
Nonlinear vibrations, 150, 192, 396, 518 396
Non-stationary, 99, 101, 39 Road surface roughness, 100–103,
Nonsynchronous, 75, 395, 397 106–108, 112, 113
Index 603

Rolling element bearing, 3, 257, 260, 263, Stiffness, 2, 3, 5, 6, 8–10, 18, 26, 32, 35,
323–325, 396, 535–537 38, 39, 62, 87, 94, 100, 103, 107,
Rotary inertia, 59, 64, 440, 490 132, 151–153, 155–157, 159, 186,
Rotational spring, 337, 338, 493 191, 193–195, 197, 199, 204, 207,
Rotor, 1–17, 19–22, 26, 45, 46, 71, 76, 78, 212, 213, 219, 220, 228, 229,
80, 116, 131, 134, 138, 146, 149, 237–243, 252, 256, 259–263, 267,
150, 152, 154–157, 159, 160, 162, 293, 302, 303, 305, 308, 340, 342,
208, 237, 239–242, 244–250, 252, 344, 345, 346, 360, 362, 363, 380,
253, 255, 257–259, 261, 263–266, 396, 397, 406, 409–412, 414, 416,
281–293, 296, 301–305, 307, 308, 420, 421, 423, 425, 440–443, 447,
310, 313, 317, 360, 395–398, 403, 449, 453–455, 458–460, 463–466,
490, 577–581, 583–590, 595, 596 468–471, 473, 474, 475, 477–479,
Rotor unbalance, 4, 22, 131, 281, 302 482, 484, 490, 491, 496, 518, 519,
Rub-impact, 577, 579, 581, 584–586, 535–540, 542–548, 565, 567, 570,
589–595 578, 581, 586, 589, 596
Stochastic, 165, 265, 314
Support vector machine, 116, 256, 282,
S 285, 313, 314, 316, 325, 332–334,
397
Sensitivity analysis, 375
Sensors faults, 121
Shaft, 1, 3, 4, 7, 9, 12, 16, 17, 22, 25–34,
36–42, 76, 117, 132, 140, 151, T
153–155, 159, 237–241, 245–247, Tapered beam, 440–444
253, 257, 261, 262, 283, 302, 303, Tapered cross-section, 489
305, 317, 338, 359–364, 366–368, Testing, 119, 208, 217, 220, 239, 255, 267,
372, 396, 543, 544, 547, 548, 579, 273, 278, 283–287, 289, 291, 292,
580, 581, 583, 590, 595, 596 294, 295, 297, 305, 319, 369, 372,
Shannon energy, 395, 399, 403 376, 377, 384, 392, 395, 397, 399,
Shape Memory Alloy (SMA), 181–183, 401, 403, 423, 430
186, 187, 189 Thermo-mechanical coupling, 181, 183,
Shear Load 185, 187, 189
Shock response spectra, 490 Time histories, 198, 201, 228, 230, 273
Skewness, 120, 257, 319, 320, 324, 325, Time windows, 99, 107
327, 329–331 Timoshenko beams, 26, 34, 87
Slope discontinuity, 359, 360 Training, 115, 116, 119, 122, 255, 257,
Smart Functionally Graded (FG) Plate, 265, 266, 284, 285–287, 289, 292,
553–555, 557–559, 562–568, 570, 294–297, 316, 319, 320, 323, 327,
571 332, 334, 384, 395, 397, 399, 423
Spherical shells, 473, 475–486 Transfer matrix method, 338
Spindle Bearing, 245 Tuned vibration absorber, 518
Squeeze film damper, 256, 257, 259–261, Two spools, 577, 579–581, 586–590
263, 266, 395–397, 577, 578, 579,
583
Stability, 4, 6, 27, 87, 149, 162, 163, 165, U
166, 178, 239, 240, 272, 302, 304, Ultrasonic, 71–75, 82, 83
314, 376, 396, 454, 465, 519, 521, Unbalance mass, 76, 140, 146, 586, 596
579 Unbalance, 1–6, 8, 14, 15–18, 21, 22, 76,
Stability threshold speed, 27 116, 131–136, 140, 146, 147,
Standard deviation, 99, 101, 105–112, 120, 149–154, 157–160, 237–243, 245,
237, 240, 250–252, 257, 319, 320, 247, 249, 251–253, 255–257, 259,
324–326 260, 265, 266, 281, 283, 285,
Static analysis, 380, 386, 565 287–289, 293, 295, 297, 302, 303,
Steam turbine blade, 45–54, 56, 57 308, 310, 311, 313, 317, 396, 403,
604 Index

446, 578–581, 583, 586, 589, 590, Vibration analysis, 46, 48, 71, 73, 83, 87,
596 88, 131, 412, 440, 474, 475, 480,
Unbalanced rotor, 313, 317 481, 554, 570
Vibration and current signal, 115, 116, 119,
123, 129, 257
Vibration control, 60, 131, 209, 239, 490,
V 491
Vibration, 3–5, 6, 10, 12, 25–27, 45–48, Vibration features, 324
59–62, 71–74, 76, 82, 83, 85–90, Vibration frequency., 273, 497
94–97, 99, 108, 115–117, 119, 123, Vibration measurement, 134, 143, 210
126, 129, 131, 133, 134, 138, Vibration parameters, 356
142–147, 149–151, 154, 156, 157, Vibration signal, 116, 117, 123, 284, 286,
159, 162–164, 181, 182, 191–194, 313, 317, 327, 328, 398, 535
207–213, 215, 217, 237, 239–242, Vibro-acoustic cavity, 163, 166, 171, 172,
245–247, 252, 253, 255, 257, 259, 175–178
261, 263, 265–267, 271–276, 279, Virtual trial misalignment, 1, 2, 5, 6, 9, 13,
281–284, 286, 289, 293–295, 301, 22
302, 307, 308, 310, 311, 313, 314, Virtual trial unbalances, 149, 151, 154, 157
317, 323–328, 338, 356, 375, 376,
395–398, 401, 403, 405–407, 409,
411–413, 415, 417, 419–421, 440, W
473–475, 480, 481, 486, 489–491, Wave propagation, 88
496, 497, 517–519, 521–523, 525, Wavelet packet transform, 359, 361,
527, 529, 530–532, 535, 536, 538, 363–367, 369, 371
539, 543, 544, 547, 552, 554, 557, Wavelet transform, 314, 318, 324, 359, 360,
559, 561–563, 565, 567–570, 364, 366
577–580, 583, 586, 590, 595, Wavelet, 166, 256, 282, 313–315, 318–321,
Varying compliance, 536, 538, 541, 542 324, 359, 360, 361, 363–367, 369,
Vibration absorber, 192, 194, 207–210, 371
216, 217, 517–519, 521–523, 525, Whirling, 27, 239, 583, 590, 595
527, 529, 531 Wind tunnel experiment, 271, 273, 274