Liming Dai · Reza N.

Jazar Editors

Nonlinear
Approaches in
Engineering
Applications
Applied Mechanics, Vibration Control,
and Numerical Analysis
Nonlinear Approaches in Engineering Applications
Liming Dai • Reza N. Jazar
Editors

Nonlinear Approaches
in Engineering Applications
Applied Mechanics, Vibration Control,
and Numerical Analysis

123
Editors
Liming Dai Reza N. Jazar
Department of Industrial School of Aerospace, Mechanical
Systems Engineering and Manufacturing Engineering
University of Regina RMIT University Bundoora East Campus
Regina, SK, Canada Bundoora, VIC, Australia

ISBN 978-3-319-09461-8 ISBN 978-3-319-09462-5 (eBook)

DOI 10.1007/978-3-319-09462-5
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014953220

© Springer International Publishing Switzerland 2015

This work is subject to copyright. All rights are reserved 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. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
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.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Dedicated to Xinming and Mojgan.
Preface

This book follows the same concept of the first two volumes and is based on the
ASME 2013 and 2012 Congress in the track of Dynamics Systems and Control,
Optimal Approaches in Nonlinear Dynamics, which were organized by the editors.
The nonlinear approaches and techniques have been developed since decades ago.
The research on nonlinear science and dynamics have brought new insights into our
modeling methods of natural and engineering phenomena. Although many of these
approaches and techniques have been brought into research and engineering prac-
tices, linearization and simplification are still the dominating approaches existing in
physics and engineering. Nature is nonlinear in general as the responses of physics
and engineering systems are nonlinear. Linearization ends up with simplification,
and it damages the original characteristics of the systems. Such simplifications
usually lead to inaccuracy, misunderstanding, or even incorrect conclusions. For
example, Hooke’s law including the generalized Hooke’s law is linear and it
composes the foundation of linear elasticity and dominates numerous solutions
of physical systems and mechanical designs. However, no material is perfectly
linear. Any material used in the real world can actually be a nonlinear and complex
system, not only due to its material or structure nonlinearity but also due to the
inhomogeneousness and anisotropy of the materials.
Another challenge facing the scientists and engineers in our time is the generation
of the solutions and characterization of the nonlinear systems modeled from the
physical systems in reality. It would be greatly beneficial in accurately evaluating the
behavior of nonlinear systems and revealing the actual nature of the systems, with
utilization of the existing mathematical tools and analytical means, if the analytical
solutions of nonlinear systems could be pursued. Due to the nonlinearity and
complexity of the nonlinear systems, unfortunately, it is very difficult or impossible
to derive the analytical and closed-loop solutions for the systems. In solving or
simulating the nonlinear systems, one may have to rely on approximate or numerical
methods, which may only provide approximate results for the systems while
errors are unavoidable during the processes of generating the approximate results.

vii
viii Preface

Approximation and inaccuracy are the inescapable shadows following the current
research and engineering practices involving nonlinearity or nonlinear systems.
In the role of the editors as well as the chapter contributors of this book, we have
tried to present a collection of chapters showing the theoretically and practically
sound nonlinear approaches and their engineering applications in various areas, in
hoping that this book may provide useful tools and comprehensible examples of
solving, modeling, and simulating the nonlinear systems existing in the real world.
The carefully selected chapters contained in the present book reflect recent advances
in nonlinear approaches and their engineering applications. The book intends to
feature in particular the fundamental concepts and approaches of nonlinear science
and their applications in engineering and physics fields. It is anticipated that this
book may help to promote the development of nonlinear science and nonlinear
dynamics in engineering, as well as to stimulate research and applications of
nonlinear science and nonlinear dynamics in physics and engineering practices. It
is also expected that the book will further enhance the comprehension of nonlinear
science and stimulate interactions among scientists and engineers who are interested
in nonlinear science and who find that nonlinearity and complexity of systems play
an important role in their respective fields.
With the theme of the book, nonlinear approaches and engineering applications:
Applied Mechanics, Vibration Control and Numerical Analysis, the book covers
interdisciplinary studies on theories and methods of nonlinear science and their
applications in complex systems such as those in nonlinear dynamics, vehicle
dynamics, rigid body, solid mechanics, safety, visco-elastic mechanics, control
engineering, ocean engineering, mechatronic engineering, acoustic engineering,
and material science. Examples include: steady-state dynamic analysis of vehicles;
rigid body dynamics and Razi acceleration discussion; a deep discussion about
piecewise linear vibration isolators and challenges in discovering its time and
steady-state determination; active vibration control for axially translation cables
to limit their vibrating reactions; a new numerical solution method for ordinary
differential equations where other methods fail; visco-elastic materials analysis
and determination of their loss tangent; investigation on the frontal human-vehicle
impact and re-designing the hood of the car to save pedestrians; a new treatment and
analysis of the post-bukling of thin tube structures; study in acoustics of exhaust
system to reduce their noise; parametric segmentation of nonlinear structures in
image data; prediction and analysis of sea level.

Level of the Book

This book aims at engineers, scientists, researchers, engineering and physics

students of graduate levels, together with the interested individuals in engineering,
physics, and mathematics. This chapter-book focuses on application of the nonlinear
approaches representing a wide spectrum of disciplines of engineering and science.
Throughout the book, great emphases are placed on engineering applications,
Preface ix

physical meaning of the nonlinear systems, and methodologies of the approaches

in analyzing and solving for the systems. Topics that have been selected are of
high interest in engineering and physics. An attempt has been made to expose the
engineers and researchers to a broad range of practical topics and approaches. The
topics contained in the present book are of specific interest to engineers who are
seeking expertise in nonlinear analysis, mathematical modeling of complex systems,
optimization of nonlinear systems, non-classical engineering problems, and future
of engineering.
The primary audience of this book is the researchers, graduate students and
engineers in mechanical engineering, engineering mechanics, electrical engineer-
ing, civil engineering, aerospace engineering, ocean engineering, mathematics, and
science disciplines. In particular, the book can be used as a book for the graduate
students as well as senior undergraduate students to enhance their knowledge by
taking a graduate or advanced undergraduate course in the areas of nonlinear
science, dynamics and vibration of discrete and continuous system, structure
dynamics, and engineering applications of nonlinear science. It can also be utilized
as a guide to the readers’ fulfilment in practices. The covered topics are also of
interest to engineers who are seeking to expand their expertise in these areas.

Organization of the Book

The main structure of the book consists of two parts of analytical and practical
nonlinearity, including 12 chapters. Each of the chapters covers an independent
topic along the line of nonlinear approach and engineering applications of nonlinear
science. The main concepts in nonlinear science and engineering applications are
explained fully with necessary derivatives in detail. The book and each of the
chapters is intended to be organized as essentially self-contained. All necessary con-
cepts, proofs, mathematical background, solutions, methodologies and references
are supplied except for some fundamental knowledge well known in the general
fields of engineering and physics. The readers may therefore gain the main concepts
of each chapter with as less as possible the need to refer to the concepts of the other
chapters. Readers may hence start to read one or more chapters of the book for their
own interests.

Method of Presentation

The scope of each chapter is clearly outlined and the governing equations are derived
with an adequate explanation of the procedures. The covered topics are logically and
completely presented without unnecessary overemphasis. The topics are presented
in a book form rather than in the style of a handbook. Tables, charts, equations, and
references are used in abundance. Proofs and derivations are emphasized in such a
x Preface

way that they can be straightforwardly followed by the readers with fundamental
knowledge of engineering science and university physics. The physical model and
final results provided in the chapters are accompanied with necessary illustrations
and interpretations. Specific information that is required in carrying out the detailed
theoretical concepts and modeling processes has been stressed.

Prerequisites

The present book is primarily intended for researchers, engineers, and graduate
students, so the assumption is that the readers are familiar with the fundamentals
of dynamics, calculus, and differential equations associated with dynamics in engi-
neering and physics, as well as a basic knowledge of linear algebra and numerical
methods. The presented topics are given in a way to establish as conceptual
framework that enables the readers to pursue further advances in the field. Although
the governing equations and modeling methodologies will be derived with adequate
explanations of the procedures, it is assumed that the readers have a working
knowledge of dynamics, university mathematics and physics together with theory
of linear elasticity.

Acknowledgements

This book is made available under the close and effective collaborations of all the
enthusiastic chapter contributors who have the expertise and experience in various
disciplines of nonlinear science and engineering applications. They deserve sincere
gratitude for the motivation of creating such book, encouragement in completing the
book, scientific and professional attitude in constructing each of the chapters of the
book, and the continuous efforts toward improving the quality of the book. Without
the collaboration and consistent efforts of the chapter contributors, the completion
of this book would have been impossible. What we have at the end is a book that we
have every reason to be proud of.
It has been gratifying to work with the staff of Springer-Verlag through the
development of this book. The assistance provided by the staff members has been
valuable and efficient. We thank Springer-Verlag for their production of an elegant
book.

Regina, SK, Canada Liming Dai

Bundoora, VIC, Australia Reza N. Jazar
Contents

Part I Analytical Nonlinearity

1 Steady-State Vehicle Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Hormoz Marzbani, Reza N. Jazar, and M. Fard
2 On the Razi Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
A. Salahuddin M. Harithuddin, Pavel M. Trivailo,
and Reza N. Jazar
3 Challenges in Exact Response of Piecewise Linear
Vibration Isolator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Oleksandr Pogorilyi, Reza N. Jazar, and Pavel M. Trivailo
4 Active Vibration Control for Nonlinear Axially
Translating Cable Systems of Multi-Dimensions. . . . . . . . . . . . . . . . . . . . . . . . 81
L. Dai and L. Sun
5 Nonlinear Initial Value Ordinary Differential Equations . . . . . . . . . . . . . . 117
Mohammad M. Aghdam, Ali Fallah, and Poorya Haghi
6 The Loss Tangent of Visco-Elastic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Franz Konstantin Fuss

Part II Practical Nonlinearity

7 Optimization of Hood Design to Minimise Pedestrian

Head Injury in Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Revathi Krishnamoorthy, Monir Takla, Aleksandar Subic,
and Derek Scott
8 Nonlinear Finite Element and Post-Buckling of Large
Diameter Thin Walled Tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Toh Yen Pang, Dale Waterson, Rees Veltjens,
and Tristan Garcia
xi
xii Contents

9 Exhaust System Acoustic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

Milan Simic
10 Nonlinear Approaches in Three Dimensional Medical
Image Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Ruwan Tennakoon, Alireza Bab-Hadiashar, and Zhenwei Cao
11 Pose Estimation of a Dioptric Imaging Sensor with a
Circle-Projecting Collimated Laser Moving Inside a Pipeline . . . . . . . . 281
Abbasali Dehghan Tezerjani, Mehran Mehrandezh,
and Raman Paranjape
12 Analysis of Tide Gauge Sea Level Time Series . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Albert Parker

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
List of Figures

Fig. 1.1 Kinematics of a moving vehicle at steady-state conditions . . . . . . . 4

Fig. 1.2 The steady-state responses as functions of the velocity vx . . . . . . . . 7
Fig. 1.3 The steady-state values of R, ˇ, and r, as functions of
forward velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Fig. 1.4 The coordinate .xO ; yO / of the dynamic rotation
center of the vehicle for 0  vx  40 m = s and its
critical speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Fig. 1.5 The location of the steady-state rotation center for
0  vx  40 m =s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Fig. 1.6 A realistic step change of steer angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Fig. 1.7 The instantaneous time history of r, vy , and R for a
step steering change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Fig. 1.8 The steady-state time history of r, vy , and R for a step
steering change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Fig. 1.9 The difference between the instantaneous and
steady-state values of r during a step steering input . . . . . . . . . . . . . . . 12
Fig. 1.10 The difference between the instantaneous and
steady-state values of vy during a step steering input. . . . . . . . . . . . . . 12
Fig. 1.11 The difference between the instantaneous and
steady-state values of R during a step steering input . . . . . . . . . . . . . . 13
Fig. 1.12 The difference between the instantaneous and
steady-state values of during a step steering input . . . . . . . . . . . . . . 13
Fig. 1.13 The instantaneous path of motion of the vehicle and
the steady-state rotation center. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Fig. 1.14 The instantaneous and steady-state global velocity
components of the vehicle and their differences . . . . . . . . . . . . . . . . . . . 16
Fig. 1.15 The instantaneous and steady-state paths of motion of
the vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Fig. 1.16 Plot of X , Xs , Y , Ys , as functions of t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Fig. 1.17 Plot of X  Xs , Y  Ys , as functions of t . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Fig. 1.18 Plot of x  xs , as functions of t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
xiii
xiv List of Figures

Fig. 1.19 Plot of y  ys , as functions of t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Fig. 1.20 Cycloidal Motion lane change maneuver-moving 3 m
from one lane to the adjacent lane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Fig. 1.21 Radius of curvature for Cycloidal Motion Lane
Change Maneuver Road. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Fig. 1.22 Radius of curvature for Cycloidal Motion Lane
Change Maneuver Road. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Fig. 1.23 A front wheel steering vehicle in a right turn . . . . . . . . . . . . . . . . . . . . . . 21
Fig. 1.24 Illustration of a car which is moving on a road at a
point that Cc is the center of curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Fig. 1.25 The kinematic steering angles—Black: ıi and Red: ıo . . . . . . . . . . . . 22
Fig. 1.26 The average of the kinematic steering angles . . . . . . . . . . . . . . . . . . . . . . 24
Fig. 1.27 The side slip velocity vy of the vehicle for the
kinematic steering case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Fig. 1.28 The rate of change of the heading angle r.t / of the
vehicle for the kinematic steering case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Fig. 1.29 The x-component of the resulted motion expressed in
body frame for the kinematic case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Fig. 1.30 The y-component of the resulted motion expressed in
body frame for the kinematic case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Fig. 1.31 Resulted road as seen in the body frame for the
kinematic case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Fig. 1.32 The X -component of velocity in the global frame for
the kinematic case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Fig. 1.33 The Y -component of velocity in the global frame for
the kinematic case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Fig. 1.34 Kinematic steering road . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Fig. 1.35 The kinematic steering angles vs. steady-state steering
angles, red: kinematic—green: steady-state . . . . . . . . . . . . . . . . . . . . . . . 27
Fig. 1.36 The difference between kinematic and steady-state
steering angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Fig. 1.37 The resulted road in the global frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Fig. 1.38 Actual road (red) vs. kinematic steering road (black) . . . . . . . . . . . . . 27
Fig. 1.39 Actual road (red) vs. steady-state road (green) . . . . . . . . . . . . . . . . . . . . 28
Fig. 1.40 The difference between the steady-state and kinematic
angles—high speed maneuver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Fig. 1.41 Actual road (red) vs. steady-state road (black)—high
speed maneuver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Fig. 1.42 Actual road (red) vs. steady-state road (green)—high
speed manoeuvre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Fig. 2.1 Three coordinate frames system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Fig. 2.2 Binary coordinate frames system—a global-fixed
frame G and a rotating body frame B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Fig. 2.3 Three relatively rotating coordinate frames . . . . . . . . . . . . . . . . . . . . . . . . 38
List of Figures xv

Fig. 2.4 Multiple coordinate frames system with frame 0 as

the inertial frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fig. 2.5 Describing compound rotation motion using three
coordinate frames system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Fig. 2.6 Razi acceleration vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
P P
A !B D  D 5 rad=s; B !C D  D 1 rad=s. . . . . . . . . . . . . . . . . . . . . . . . .
B C
Fig. 2.7 51
P P
A !B D  D 4 rad=s; B !C D  D 2 rad=s . . . . . . . . . . . . . . . . . . . . . . . .
B C
Fig. 2.8 52
P P
A !B D  D 3 rad=s; B !C D  D 3 rad=s. . . . . . . . . . . . . . . . . . . . . . . . .
B C
Fig. 2.9 52
P P
A !B D  D 2 rad=s; B !C D  D 4 rad=s. . . . . . . . . . . . . . . . . . . . . . . . .
B C
Fig. 2.10 53
P P
A !B D  D 1 rad=s; B !C D  D 5 rad=s. . . . . . . . . . . . . . . . . . . . . . . . .
B C
Fig. 2.11 53
Fig. 2.12 B
! D P D 5 rad=s; C
! D P D 0 rad=s. . . . . . . . . . . . . . . . . . . . . . . . . 54
A B B C

Fig. 3.1 A dual rate bilinear vibrating system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

Fig. 3.2 A vibrating system with gap-damper engagement . . . . . . . . . . . . . . . . . 60
Fig. 3.3 Mechanical model of the piecewise linear system with
symmetric constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Fig. 3.4 A steady state periodic response of the system . . . . . . . . . . . . . . . . . . . . 63
U
Fig. 3.5 The frequency response for Z D ............................... 68

Fig. 3.6 The frequency response curve of the primary for
!1 D 1, ı D 1, 1 D 0:1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Fig. 3.7 The implicit plot of the two transcendental
equations p(3.46) and (3.47) as functions of t1 and '
for !2 D 2, 2 D 0:1, !1 D 1, w D 0:5, 1 D 0:1 . . . . . . . . . . . . . . . 70
Fig. 3.8 A sample
p of the steady response of the system for
!2 D 2, 2 D 0:1, !1 D 1, w D 0:5, 1 D 0:1. . . . . . . . . . . . . . . . . . . 70
Fig. 3.9 The implicit plot of the Eqs. (3.46) and (3.47)p
for ! D 1, w D 3 and !1 D 1, 1 D 0:1, !2 D 17; 2 D 0:1 . . . 71
Fig. 3.10 The steady response of the system for
' D 1:248353815, t1 D :9049692487 . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Fig. 3.11 The steady response of the system for
' D 1:314128509, t1 D 2:229699384 . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Fig. 3.12 The implicit plot of the Eqs. (3.46) and (3.47) p for
! D 1, w D 3 and !1 D 1, 1 D 0:1, !2 D 2, 2 D 0:1 . . . . . . . . 73
Fig. 3.13 The steady response of the system p for ! D 1, w D 3
and !1 D 1, 1 D 0:1, !2 D 2, 2 D 0:1 . . . . . . . . . . . . . . . . . . . . . . . . 73
Fig. 3.14 The implicit plot of the Eqs. (3.46) and (3.47)pfor
! D 1:4, w D 3 and !1 D 1, 1 D 0:1, !2 D 17, 2 D 0:1. . . . . 74
Fig. 3.15 The steady response of the system p for ! D 1, w D 3
and !1 D 1:4, 1 D 0:1, !2 D 2, 2 D 0:1 . . . . . . . . . . . . . . . . . . . . . . 75
Fig. 3.16 One set of solution of Eqs. (3.46) pand (3.47) for
w D 3 !1 D 1, 1 D 0:1, !2 D 17, 2 D 0:1 and
varying ! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Fig. 3.17 The mathematical model of an asymmetric piecewise
linear vibration isolator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
xvi List of Figures

Fig. 3.18 The mathematical model of a one-sided piecewise

linear vibration isolator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Fig. 4.1 The model of the axially translating cable . . . . . . . . . . . . . . . . . . . . . . . . . 85
Fig. 4.2 The wave diagram of wp without the application of
the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Fig. 4.3 (a) The contribution of the first vibration mode to
wp without the application of the control strategy.
(b) The contribution of the second vibration mode
to wp without the application of the control strategy.
(c) The wave diagram of w3 without the application of
the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Fig. 4.4 The wave diagram of wp with the application of the
control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Fig. 4.5 (a) The contribution of the first vibration mode
to wp with the application of the control strategy.
(b) The contribution of the second vibration mode to
wp with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Fig. 4.6 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 99
Fig. 4.7 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Fig. 4.8 The wave diagram of wp with the application of the
control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Fig. 4.9 (a) The contribution of the first vibration mode
to wp with the application of the control strategy.
(b) The contribution of the second vibration mode to
wp with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Fig. 4.10 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 101
Fig. 4.11 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Fig. 4.12 The wave diagram of wp with the application of the
control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Fig. 4.13 (a) The contribution of the first vibration mode
to wp with the application of the control strategy.
(b) The contribution of the second vibration mode to
wp with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Fig. 4.14 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 104
Fig. 4.15 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
List of Figures xvii

Fig. 4.16 The wave diagram of wp with the application of the

control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Fig. 4.17 (a) The contribution of the first vibration mode to wp
with the application of the control strategy. (b) The
contribution of the second vibration mode to wp
with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Fig. 4.18 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 107
Fig. 4.19 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Fig. 4.20 The wave diagram of wp with the application of the
control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Fig. 4.21 (a) The contribution of the first vibration mode to wp
with the application of the control strategy. (b) The
contribution of the second vibration mode to wp
with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Fig. 4.22 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 110
Fig. 4.23 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Fig. 4.24 The wave diagram of wp with the application of the
control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Fig. 4.25 (a) The contribution of the first vibration mode to wp
with the application of the control strategy. (b) The
contribution of the second vibration mode to wp
with the application of the control strategy. (c) The
contribution of the third vibration mode to wp with the
application of the control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Fig. 4.26 The wave diagram of wp (the blue continuous line)
and the reference signal wr (the green dash line) . . . . . . . . . . . . . . . . . . 113
Fig. 4.27 The control input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Fig. 5.1 Comparison between Euler method and second-order
Runge–Kutta method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Fig. 5.2 Domain of k-step method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Fig. 5.3 Bernstein polynomials (a) quadratic n D 2, (b) cubic n D 3 . . . . . . . 122
Fig. 5.4 Comparing Newton interpolation with Bézier curves
and Convex hull property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Fig. 5.5 Nonlinear mass–spring–damper system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Fig. 5.6 Solution of nonlinear mass–spring system with
damping for (a) h D 0.1, (b) h D 0.3, (c) h D 0.5 . . . . . . . . . . . . . . . . . . . 130
Fig. 5.7 Solution of Lotka–Volterra equation for (a) h D 0.1,
(b) h D 0.3, (c) h D 0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
xviii List of Figures

Fig. 5.8 Solution of Van der Pol equation in case of low

damping  D 0.5 for (a) h D 0.1, (b) h D 0.2, (c)
h D 0.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Fig. 5.9 Solution of Van der Pol equation in case of high
damping  D 5 for (a) h D 0.001, (b) h D 0.005, (c) h D 0.05. . . . . 134
Fig. 6.1 Sinusoidal strain curve (blue) imposed on a
visco-elastic material and resulting stress curve
(pink); ¢ 0 : stress amplitude, maximal stress; "0 : strain
amplitude, maximal strain; •: phase shift . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Fig. 6.2 Hysteresis loop of stress–strain ellipse; ¢ 0 : maximal
stress; "0 : maximal strain; •: phase angle . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Fig. 6.3 Stress–time curve (red) during the first load cycles;
blue curve: transient component; green curve: steadys
state component; the stress–time curve (red) is the
sum of transient and steady state components . . . . . . . . . . . . . . . . . . . . . 140
Fig. 6.4 Standard linear solid of Voight form; ¢: stress; ":
strain; ˜: viscosity constant; E1 : modulus of series
spring; E2 : modulus of parallel spring; "1 : strain of
series spring; "2 : strain of Kelvin–Voight model. . . . . . . . . . . . . . . . . . . 141
Fig. 6.5 Maximal stress ¢ 0 against modulus E2 of parallel
spring; f : frequency; ˜: viscosity constant; E2 : E2 at
0
which a local minimum of ¢ 0 is found . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Fig. 6.6 Phase angle • against frequency f ; ˜: viscosity
constant; E1 and E2 : moduli of series and parallel
spring; f0 : f at which a local maximum of • is found . . . . . . . . . . . . . . 146
Fig. 6.7 Peak stress ¢ 0 against frequency f ; ˜: viscosity
constant; E: velocity independent elasticity parameter . . . . . . . . . . . . 152
Fig. 6.8 Ratio of E/˜ against frequency f ; ˜: viscosity
constant; E: velocity independent elasticity parameter;
tan •: loss tangent; f0 : frequency at which tan • D /2 . . . . . . . . . . . . 154
Fig. 6.9 Phase angle • against log frequency; ˜: viscosity
constant; E: velocity independent elasticity parameter . . . . . . . . . . . . 154
Fig. 7.1 Unsafe conditions for pedestrians (World Health
Organization 2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Fig. 7.2 Kinematics of a pedestrian in PPCFC (Synaptic
analysis consulting group inc) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Fig. 7.3 Direction of impact for pedestrian fatalities due to
collision with a passenger car in the US (National
highway traffic safety administration 2009). . . . . . . . . . . . . . . . . . . . . . . . 164
Fig. 7.4 Pedestrian protection technical requirements (Euro
NCAP 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Fig. 7.5 Subsystem test method (Hardy et al. 2007) . . . . . . . . . . . . . . . . . . . . . . . . 168
Fig. 7.6 Comparison between TRL-PLI and Flex-PLI (World
Health Organization 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
List of Figures xix

Fig. 7.7 TRL-PLI lower leg impactor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

Fig. 7.8 Leg impact and head impact test machines (Centre of
automotive safety research 2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Fig. 7.9 POLAR III pedestrian dummy (Honda Worldwide 2008). . . . . . . . . 171
Fig. 7.10 Comparison of peak acceleration magnitude. . . . . . . . . . . . . . . . . . . . . . . 172
Fig. 7.11 Comparison of peak acceleration duration . . . . . . . . . . . . . . . . . . . . . . . . . 172
Fig. 7.12 Primary impact of head with hood panels with
excessive stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Fig. 7.13 Effect of inertial loading of the head form on HIC value. . . . . . . . . . 174
Fig. 7.14 Secondary impact of head with hard components
within engine bay. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Fig. 7.15 Waveform with secondary impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Fig. 7.16 Waveform efficiency (˜) of a half-sine waveform
(Lim et al. 1995) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Fig. 7.17 Realistic optimal waveform for pedestrian head
impact (Wu and Beaudet 2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Fig. 7.18 Underside view of a hood assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Fig. 7.19 Inlaid hood (A2Mac1 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Fig. 7.20 Wraparound hood (A2Mac1 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Fig. 7.21 FEA model for optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Fig. 7.22 FEA model with head impactor for optimisation . . . . . . . . . . . . . . . . . . 182
Fig. 7.23 FEA model simulation using LS-DYNA for optimisation . . . . . . . . 182
Fig. 7.24 Contour plot of hood displacement using HYPERVIEW . . . . . . . . . 183
Fig. 7.25 Plot of hood displacement vs. time using HYPERGRAPH . . . . . . . 183
Fig. 7.26 Plot of resultant acceleration vs. time using HYPERGRAPH . . . . 184
Fig. 7.27 Energy plot from the finite element model . . . . . . . . . . . . . . . . . . . . . . . . . 184
Fig. 7.28 Pedestrian head impact finite element model . . . . . . . . . . . . . . . . . . . . . . 185
Fig. 7.29 Correlation between ANCAP test and FE simulation . . . . . . . . . . . . . 186
Fig. 7.30 Correlation of hood deformation between ANCAP
test and FE simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Fig. 7.31 Coordinates for intrusion computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Fig. 7.32 Comparison of impactor intrusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Fig. 7.33 Accelerometer rotation due to impactor rebound . . . . . . . . . . . . . . . . . . 188
Fig. 7.34 ‘Intrusion T’ calculation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Fig. 7.35 Design space for hood assembly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Fig. 7.36 Scope of the project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Fig. 7.37 Hood leading edge (United Nations Economic
Commission for Europe 1998) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Fig. 7.38 Hood rear reference line (United Nations Economic
Commission for Europe 1998) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Fig. 7.39 Hood side reference line (United Nations Economic
Commission for Europe 1998) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Fig. 7.40 Head impact test area on hood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Fig. 7.41 Impact positions considered for optimisation . . . . . . . . . . . . . . . . . . . . . . 192
Fig. 7.42 Concepts for improving pedestrian protection performance . . . . . . 193
xx List of Figures

Fig. 7.43 Pugh matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

Fig. 7.44 Inner hood geometries chosen for developing an
optimal solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
Fig. 7.45 Control factors in scope for refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Fig. 7.46 Parameter diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Fig. 7.47 Predictability of response surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Fig. 7.48 ANOVA chart for mean value of HIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Fig. 7.49 Evaluation of efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Fig. 7.50 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Fig. 7.51 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Fig. 7.52 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Fig. 7.53 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Fig. 7.54 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point E. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Fig. 7.55 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Fig. 7.56 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Fig. 7.57 HIC value, effective plastic strain, displacement of
the impactor and hood with consideration to primary
impact only for point H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Fig. 7.58 Efficiency and combined hood panel thickness relationship . . . . . . 204
Fig. 7.59 Influence of deformation space on HIC value (steel) . . . . . . . . . . . . . . 205
Fig. 7.60 Head impact performance with 80 mm deformation space. . . . . . . . 206
Fig. 7.61 Influence of hood panel thickness on HIC value (aluminium) . . . . 206
Fig. 7.62 Influence of combined hood panel thickness on HIC
value (aluminium) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Fig. 7.63 Combined hood panel thickness vs. the sum of the
mean value of HIC and twice its standard deviation (aluminium) 207
Fig. 7.64 Combined thickness of aluminium outer and inner
hood panels in various vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Fig. 7.65 Influence of deformation space on HIC value (aluminium) . . . . . . . 208
Fig. 7.66 Waveform comparison-70 mm deformation space. . . . . . . . . . . . . . . . . 209
List of Figures xxi

Fig. 7.67 Waveform comparison aluminium with 85 mm

deformation space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Fig. 7.68 ANOVA chart for peak acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Fig. 7.69 Resultant acceleration against vertical intrusion . . . . . . . . . . . . . . . . . . . 211
Fig. 7.70 Comparison of rebound velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Fig. 8.1 Thin walled cylindrical tube used for the experiment . . . . . . . . . . . . . 221
Fig. 8.2 Experimental setup of test specimen in the load press. . . . . . . . . . . . . 221
Fig. 8.3 Images of experiment demonstrating mode of failure . . . . . . . . . . . . . 222
Fig. 8.4 Force vs. displacement graph of experiment . . . . . . . . . . . . . . . . . . . . . . . 222
Fig. 8.5 Stress–strain behaviour of material damage where
softening occurs, as well as degradation of material. . . . . . . . . . . . . . . 224
Fig. 8.6 Comparison of force–displacement plots for
numerical and experimental analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Fig. 8.7 Deformation of aluminium tubing using ductility and
MSFLD damage criteria for Aluminium 6060 T5 . . . . . . . . . . . . . . . . . 229
Fig. 8.8 Close-up deformation of aluminium tubing using
ductility and MSFLD damage criteria for Aluminium
6060 T5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Fig. 9.1 An exhaust system structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Fig. 9.2 Power spectrum measurement program . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Fig. 9.3 Exhaust system sound power spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Fig. 9.4 Basic acoustic subsystem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
Fig. 9.5 Gas temperature distribution along the system . . . . . . . . . . . . . . . . . . . . 244
Fig. 9.6 A segment from the acoustic ladder network . . . . . . . . . . . . . . . . . . . . . . 245
Fig. 9.7 Distributed parameters calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Fig. 9.8 Header and its equivalent physical network: Flow
sources F(t) are represented by arrows, acoustic
inductors as La and capacitors as Ca. Equivalent
circuit is given on the right where equivalent source
flow is 4F(t) and equivalent components are as given
in the text and on the figure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
Fig. 9.9 Simulink model of the cascaded acoustic filter . . . . . . . . . . . . . . . . . . . . 246
Fig. 9.10 Bode plots of the exhaust system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Fig. 10.1 In this illustration the moving image needs to be both
translated and rotated to align with the fixed image . . . . . . . . . . . . . . . 252
Fig. 10.2 (a) CT, (b), MRI, (c) PET images of the same
subject. Images are from the RIRE dataset
(http://www.insight-journal.org/rire) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Fig. 10.3 Registration flow of a typical intensity information
based registration algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Fig. 10.4 Joint histograms between the original image (top-left)
and rotated versions of it. (Top-right) No rotation,
(bottom-left) rotated by 2ı , (bottom-right) rotated by 15ı . . . . . . . . 259
xxii List of Figures

Fig. 10.5 Inverse transformation vs. forward transformation.

The figure shows that using inverse transform is
computationally advantageous. (a) Inverse transform.
(b) Forward transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Fig. 10.6 Original image (c) and the image after an affine
transform (d). The transformation of both the grid
(a, b) and the intensities are shown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
Fig. 10.7 B-Spline kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Fig. 10.8 Original image and the image after a B-Spline
transform. The transformation of both the grid and the
intensities are shown together with the control point locations . . . 269
Fig. 11.1 Robotic pipe inspection system. (a) Robotic
inspection system. (b) Imaging sensor mounted on the
robot. (c) Snapshot of the robot inside a PVC pipe . . . . . . . . . . . . . . . . 284
Fig. 11.2 Main components of the imaging unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Fig. 11.3 Proposed imaging platform and its configuration
inside the pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Fig. 11.4 Projection geometry of the imaging platform . . . . . . . . . . . . . . . . . . . . . . 286
Fig. 11.5 Projection of the laser ring on the image plane of an
equidistant fish-eye lens for a sample camera pose
and configuration f D 0:14 cm, tx D 6, ty D 4,
˛x D =6, ˛y D pi=8, d D 5 cm, and rp D 20 cm . . . . . . . . . . . . . . . 288
Fig. 11.6 Snapshot of developed GUI for generating sample data. . . . . . . . . . . 291
Fig. 11.7 Observation and analytical reference signals during
the optimization process, where the actual pose is
Qactual D< xc D 1:2988 yc D 5:6033 p D 0:2496 t D
0:2153 >: (a) after 15 iterations, (b) after 50
iterations, (c) after 100 iterations, and (d) after 140
iterations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Fig. 11.8 Performance of the pose estimation algorithm vs.
number of iterations and initial seed value. (a)
Pose estimation error vs. the number of iterations.
(b) Success rate of the pose estimation algorithm vs.
the offset value of the initial seed for the camera pose . . . . . . . . . . . . 293
Fig. 11.9 Comparison of the actual and estimated camera pose
for 60 consecutive frames of the camera moving in
a spiral path. In this experiment the last estimated
pose (last-known configuration) is used as the initial
seed for the pose estimation on the next frame.
(a) Trajectories in X-Y plane. (b) Trajectories in
pan-tilt plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
List of Figures xxiii

Fig. 11.10 Computation time and optimization error vs. number

of iterations for two initial seeds with 40 and 60 %
offset. (a) Computation time vs. number of iterations.
(b) Value of the objective function J vs. number of iterations . . . . 294
Fig. 12.1 Glacial isostatic adjustment (fromhttp://en.wikipedia.
org/wiki/Glacial_isostatic_adjustment) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
Fig. 12.2 Demography of PSMSL stations (from
http://www.psmsl.org/data/obtaining/, updated: 22
August 2012): stations with at least 40 years of
data (top) and stations from Northern and Southern
Hemisphere (bottom) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Fig. 12.3 Sea levels in Sydney (left) and Fremantle (right):
(a) monthly sea levels; (b) linear trend; (c) oscillations
of monthly sea levels about the linear trend; (d) land
motion in nearby GPS stations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
Fig. 12.4 Sea levels in Sydney (left) and Fremantle (right):
(a) time history of SLR computed with 20, 30 and 60
years windows; (b) time history of SLR computed
by using all the data available at any time—this is
the conventional sea level velocity; (c) time history
of SLA computed as the time rate of change of the
above SLR; (d) apparent present velocity that could
have been computed with the linear model in case the
measuring record could have started in different times . . . . . . . . . . . 309
Fig. 12.5 Sea levels in Sydney (left) and Fremantle (right):
(a) time history of SLA computed with 20, 30 and 60
years windows; (b) time history of SLA computed
by using all the data available at any time—this is
the average sea level acceleration since the start of
the record; (c) apparent present acceleration that
could have been computed as twice the second
order coefficient of a polynomial fitting in case the
measuring record could have started in different times . . . . . . . . . . . 310
Fig. 12.6 Sea levels in Sydney (left) and Fremantle (right):
(a) periodogram of the monthly oscillations about the
linear trend; (b), (c) and (d) measured and computed
monthly sea levels by the proposed multiple nonlinear
fitting over different time windows—better accuracy
may be obtained by using other periodic functions . . . . . . . . . . . . . . . 312
 Part I
Analytical Nonlinearity
Chapter 1
Steady-State Vehicle Dynamics

Hormoz Marzbani, Reza N. Jazar, and M. Fard

Keywords Vehicle dynamics • Autodriver algorithm • Autonomous vehicle

Steering dynamics

1.1 Dynamic Equations of Motion

The equations of motion of the planar bicycle car model, expressed in the principal
body coordinate frame B, are governed by the following set of nonlinear coupled
ordinary differential equations. In the equations, the steering angle ı acts as the input
and the mass center forward velocity vx , lateral velocity vy , and yaw rate r act as
the outputs (Jazar 2014).
1
vP x D Fx C r vy (1.1)
m
 
Cˇ Cr Cı
vP y D vy  vx  rC ı .t / (1.2)
mvx m m
Dˇ Dr Dı
rP D vy C rC ı .t / (1.3)
Iz vx Iz Iz
where
@Fy a1 a2
Cr D D  C˛f C C˛r (1.4)
@r vx vx

H. Marzbani () • R.N. Jazar • M. Fard

School of Aerospace, Mechanical, and Manufacturing Engineering
RMIT University, Melbourne, VIC, Australia
e-mail: [email protected]; [email protected];
[email protected]

© Springer International Publishing Switzerland 2015 3

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__1
4 H. Marzbani et al.

Fig. 1.1 Kinematics of a

moving vehicle at
steady-state conditions

@Fy
Cˇ D D C˛f  C˛r (1.5)
@ˇ
@Fy
Cı D D C˛f (1.6)
@ı
@Mz a2 a2
Dr D D  1 C˛f  2 C˛r (1.7)
@r vx vx
@Mz
Dˇ D D a2 C˛r  a1 C˛f (1.8)
@ˇ
@Mz
Dı D D a1 C˛f (1.9)
@ı
vy
ˇD (1.10)
vx

The orthogonal body coordinate frame B is fixed to the vehicle at its mass center C
and is set such that the x-axis is longitudinal, y-axis is lateral, and z-axis is vertical
upward. The body coordinate and kinematics of the bicycle car model in a forward
motion on a positive turn is illustrated in Fig. 1.1.
The coefficients Cr , Cˇ , Cı , Dr , Dˇ , Dı in the equations of motion are slopes
of the curves for lateral force Fy and yaw moment Mz as a function of r, ˇ, and
ı, respectively. The coefficients Cr and Dr are functions of vx , and the coefficients
Cˇ , Cı , Dˇ , Dı are constant for a given vehicle. The solution of the equations to a
given time dependent steer angle ı .t / would be

vx D vx .t / (1.11)
1 Steady-State Vehicle Dynamics 5

vy D vy .t / (1.12)
r D r .t / (1.13)

The equations of motion are not analytically solvable, however, researchers

traditionally assume that the forward velocity vx is constant to reduce the equation
to the following linear set.
Generally, this happens when the forward velocity is known and there is no need
for calculating vx .

Fx D mr vy (1.14)
2 C C 3 2 3
Cı
  ˇ r
 vx  
vP y 6 7 y v 6 m 7
D 4 mv
D
x m
D 5 C6
4
7 ı .t /
5 (1.15)
rP ˇ r r D ı
Iz vx Iz Iz

When vx remains constant or in more general cases is known, Eq. (1.15) become
independent from (1.14). The set of Eq. (1.15) can then be written in the form

qP D ŒA q C u (1.16)

in which ŒA is a constant coefficient matrix, q is the vector of control variables, and

u is the vector of inputs.
2 C C 3
ˇ r
 vx
6 7
ŒA D 4 mv
Dˇ
x m
Dr 5
Iz vx Iz
2 3
C˛f C C˛r a1 C˛f C a2 C˛r
6   vx7
mvx mvx
D64 a1 C˛f  a2 C˛r a12 C˛f C a22 C˛r 5
7 (1.17)
 
Iz vx Iz vx


vy
qD (1.18)
r
2 3 2 3
C˛f Cı
6 m 7 6 7
uD6 7 ı .t / D 6 m 7 ı .t / (1.19)
4 a1 C˛f 5 4 Dı 5
Iz Iz
6 H. Marzbani et al.

1.2 Steady-State Responses

When the vehicle is turning at a steady-state condition, its behavior is governed by

the following equations:
m
Fx D  vx vy (1.20)
R
  1
Cˇ ˇ C Cr vx  m v2x D Cı ı (1.21)
R
1
Dˇ ˇ C Dr vx D Dı ı (1.22)
R

from which, we can define a set of steady-state responses (Jazar 2013):

1. Curvature response, S

1 Cı D ˇ  Cˇ D ı
S D D D   (1.23)
ı Rı vx Dr Cˇ  Cr Dˇ C mvx Dˇ

2. Sideslip response, Sˇ

ˇ Dı .Cr  mvx /  Dr Cı
Sˇ D D (1.24)
ı Dr Cˇ  Cr Dˇ C mvx Dˇ
3. Yaw rate response, Sr
r Cı D ˇ  C ˇ D ı
Sr D D vx D S vx D (1.25)
ı ı Dr Cˇ  Cr Dˇ C mvx Dˇ
4. Centripetal acceleration response, Sa
 
v2x =R Cı Dˇ  Cˇ Dı vx
Sa D D vx D S vx D
2 2
(1.26)
ı ı Dr Cˇ  Cr Dˇ C mvx Dˇ
5. Lateral velocity response, Sy

vy Dı .Cr  mvx /  Dr Cı
Sy D D Sˇ vx D vx (1.27)
ı Dr Cˇ  Cr Dˇ C mvx Dˇ

Figure 1.2 illustrates a sample of steady-state responses with respect to vx for a

car with the following characteristics.

C˛f D 50;000 N = rad C˛r D 50;000 N = rad

m D 1;000 kg Iz D 1;650 kg m2 (1.28)
a1 D 1:0 m a2 D 1:5 m
1 Steady-State Vehicle Dynamics 7

Fig. 1.2 The steady-state responses as functions of the velocity vx

Fig. 1.3 The steady-state values of R, ˇ, and r, as functions of forward velocities

The vehicle is understeer because of a positive stability parameter K.

 
m a2 a1
KD  D 0:0016 > 0 (1.29)
l2 C˛f C˛r

Employing the steady-state responses, we can determine the variation of the

kinematics of motion as functions of forward velocity vx . Figure 1.3 depicts the
steady-state radius of rotation R of the mass center of the car, vehicle side slip angle
ˇ, and yaw rate r, at different forward velocities.
8 H. Marzbani et al.

1.3 Rotation Center

Having the steady-state responses Sk D 1=R=ı and Sˇ D ˇ=ı, we are able to

determine the steady-state position of the rotation center .xO ; yO / of a vehicle in
the vehicle body coordinate frame as illustrated in Fig. 1.1.

1  
xO D R sin ˇ D  sin Sˇ ı (1.30)
Sk ı
1  
yO D R cos ˇ D cos Sˇ ı (1.31)
Sk ı

At steady-state conditions the radius of rotation R can be found from the curvature
response S , and the vehicle sideslip angle ˇ from the sideslip response Sˇ .
 
1 vx Dr Cˇ  Cr Dˇ C mvx Dˇ
RD D   (1.32)
ıS Cı D ˇ  C ˇ D ı ı
Dı .Cr  mvx /  Dr Cı
ˇ D ıSˇ D ı (1.33)
Dr Cˇ  Cr Dˇ C mvx Dˇ

Therefore, the steady-state position of the dynamic center of rotation O in the body
coordinate frame B, about which the vehicle will actually turn, is at
   
vx Dr Cˇ  Cr Dˇ C mvx Dˇ Dı .Cr  mvx /  Dr Cı
xO D    sin ı
Cı D ˇ  C ˇ D ı ı Dr Cˇ  Cr Dˇ C mvx Dˇ
(1.34)
   
vx Dr Cˇ  Cr Dˇ C mvx Dˇ Dı .Cr  mvx /  Dr Cı
yO D   cos ı
Cı D ˇ  C ˇ D ı ı Dr Cˇ  Cr Dˇ C mvx Dˇ
(1.35)

Figure 1.4 illustrates the coordinates xO and yO of the vehicle for 0  vx 

40 m=s at a constant steer angle ı D 0:1 rad . At a constant steer angle, the dynamic
rotation center of the understeer vehicle moves away and forward by increasing the
forward velocity. The rate of the displacement of the rotation center increases by
speed. The figure also depicts the critical speed vx D .Cr Dı  Dr Cı / =m=Dı D
13:693 m=s at which the velocity vector of the vehicle at its mass center is in the
x-direction v D vx {O. At the critical speed, we have ˇ D 0 and the dynamic rotation
center is on the y-axis. Figure 1.5 illustrates the location of the steady-state rotation
center for 0  vx  40 m=s, relative to the vehicle.
1 Steady-State Vehicle Dynamics 9

Fig. 1.4 The coordinate .xO ; yO / of the dynamic rotation center of the vehicle for 0  vx 
40 m = s and its critical speed

Fig. 1.5 The location of the steady-state rotation center for 0  vx  40 m =s

1.4 Effect of Changing Steering Angle on Transient

and Steady-State Responses

In the following, we calculate the transition response of an understeer passenger

car that is moving at a constant forward velocity, under variable steering angle. The
transition behavior of the car will be determined by solving the equations of motion.
The same transition behavior will be determined using steady-state responses to
examine the proximity of the two analyses. It will be shown that steady-state
response equations are good enough to predict the transition behavior of the car.
10 H. Marzbani et al.

Fig. 1.6 A realistic step change of steer angle

The results of this investigation will be useful in speeding the online calculation
of autodriver algorithm, as well as obtaining data to determine the steady-state
behavior of vehicles (Tahami et al. 2003). The maneuvers can be considered
as the steady state circular path which is a classical vehicle dynamic parameter
determination test. The steady-state circular test is an open-loop test in which either
the circular track radius or the steering angle or the vehicle’s forward velocity must
be kept constant (ISO4138 2012).
A realistic step change of steer angle can be expressed by
   
t
ı D ı0 H .t  t0 / C sin2 H .t0  t / (1.36)
2t0 =

as illustrated in Fig. 1.6 for t0 D 1 s, where H .t  t0 / is the Heaviside function


0 t < t0
H .t  t0 / D (1.37)
1 t  t0

and t0 is the response time and ı0 is the maximum constant steer angle.
We determine the reaction of the vehicle for

ı0 D 0:1 rad t0 D 1 s vx D 20m=s (1.38)

and determine the solution of the equations of motion.

r D H .t  t0 / e 4:962.1t/ .0:0588 sin 3:7687.t  1/  0:01048 cos 3:7687.t  1//

C H .t  t0 / .0:233 cos. t / C 0:0869 sin. t / C 0:2439/
C e 4:962t .0:0588 sin. t /  0:01048 cos. t //
 0:233 cos. t /  0:0869 sin. t / C 0:2439 (1.39)
1 Steady-State Vehicle Dynamics 11

Fig. 1.7 The instantaneous time history of r, vy , and R for a step steering change

vy D H .t  t0 / e 4:962.1t/ .0:0492 sin 3:7687.t  1/ C 0:2924 cos 3:7687.t  1//

 H .t  t0 / .0:1222 cos. t / C 0:4028 sin. t / C 0:4146341465/
C e 4:962t .0:2924 cos.3:7687t / C 0:0492 sin.3:7687t //
C 0:1222 cos. t / C 0:4028 sin. t /  0:414 (1.40)
vx
RD (1.41)
r
Let us call the dynamic variables that are calculated by solving the equations
of motion (1.1)–(1.3) instantaneous variable and instantaneous vehicle, and the
variables that are calculated based on responses (1.14)–(1.15) the steady-state
variable and steady-state vehicle. Figure 1.7 depicts the instantaneous time history
of r, vy , and R, during and after steering change. The steer angle is variable for
0 < t  1 s and is fixed at ı0 D 0:1 rad for t > 0. It takes less than one second for
the vehicle to achieve the final steady-state conditions, while their states are not too
far from the steady-states.
The steady-state expressions of r, vy , and R as functions of time for the step
change of steer angle are:
 
t
rs D 0:4878 sin 2
H.1  t / C 0:4878 H.t  1/ (1.42)
2=
 
t
vys D 0:829 sin 2
H.1  t /  0:8292 H.t  1/ (1.43)
2=
4:1
Rs D   (1.44)
t
0:1 sin2 H.1  t / C 0:1H.t  1/
2=

which are shown in Fig. 1.8.

The difference between the instantaneous and steady-state values of r, vy , and R
is shown in Figs. 1.9, 1.10, 1.11. The maximum difference of r  rs happens around
12 H. Marzbani et al.

Fig. 1.8 The steady-state time history of r, vy , and R for a step steering change

Fig. 1.9 The difference between the instantaneous and steady-state values of r during a step
steering input

Fig. 1.10 The difference between the instantaneous and steady-state values of vy during a step
steering input

middle of the steering change at t  0:56 s and is r  rs  0:083913 rad=s which

is 17:2 % of the steady-state value of 0:4878 rad=s. As r < rs during t < t0 the
instantaneous yaw rate of the vehicle is higher than the steady-state, and therefore,
the instantaneous vehicle is turning faster than the steady-state vehicle. This yields
that the angular position of the steady-state vehicle has a constant lag.
1 Steady-State Vehicle Dynamics 13

Fig. 1.11 The difference between the instantaneous and steady-state values of R during a step
steering input

Fig. 1.12 The difference between the instantaneous and steady-state values of during a step
steering input

The angular orientation of the body coordinate frame B .xyz/ with respect to the
global frame G .X Y Z/ is indicated by the angle between axes X and x measured
about the Z-axis. Figure 1.12 depicts , s , and  s indicating that steady-state
vehicle is a few degrees behind the instantaneous vehicle.
Z t
D r .t / dt (1.45)
0

D H .t  t0 / .e 4:962.1t/ .0:00435 cos 3:769.t  1/  0:0085 sin 3:769.t  1//

C H .t  t0 / .0:0277 cos. t / C 0:074 sin. t /  0:267// C 0:244t
 e 4:962.1t/ .0:004356 cos 3:769.t  1/  0:0085 sin 3:769.t  1//
C 0:0277 cos. t /  0:0743 sin. t / C 0:244t  0:02332 (1.46)
14 H. Marzbani et al.

The global frame G is fixed on the ground, and B coincides with G at the start of
maneuvers. The B-frame moves with the vehicle while the axes z and Z are always
parallel. Therefore, the velocity vector of the vehicle in the global frame is
  
 sin
cos vx
G
vD G
RB B v D
sin
cos vy
   
vx cos  vy sin v
D D X (1.47)
vy cos C vx sin vY

where G RB is the transformation between B and G and the velocity vector of the
vehicle in the body frame is
 
v
B
vD x (1.48)
vy

Therefore, the global coordinates of the mass center of the vehicle would be
Z Z
t t  
XD vX dt D vx cos  vy sin dt (1.49)
0 0
Z Z
t t  
Y D vY dt D vy cos C vx sin dt (1.50)
0 0

When the steer angle is kept constant, the vehicle will eventually be turning on a
constant circular path. The position of the steady-state rotation center of the vehicle
in the B-frame is at

xO D R sin ˇ D 1:6994 m (1.51)

yO D R cos ˇ D 40:965 m (1.52)

because

lim R D 41m (1.53)

t!1

lim ˇ D 0:04146 rad (1.54)

t!1

The global coordinates of the steady-state rotation center is

     
XO xO 13:617 m
D G G rB C G
R B D (1.55)
YO yO 41:388 m

where GG rB is the G-expression of the position vector of the origin of the B-frame
with respect to the origin of the G-frame at any point on the steady-state conditions
(Jazar 2011). In this example, we used the calculated data at t D 2 s when we have
1 Steady-State Vehicle Dynamics 15

Fig. 1.13 The instantaneous path of motion of the vehicle and the steady-state rotation center

   
xO 1:6994 m
D (1.56)
yO 40:965 m
D 0:6851 rad (1.57)
 
38:2226 m
G rB D
G
(1.58)
8:5916 m

Figure 1.13 illustrates the instantaneous path of motion of the vehicle in global
frame and its final steady-state rotation center. The magnification of the transient
stage depicts how the vehicle approaches its steady-state circular path.
To compare the instantaneous and steady-state, or in other words, the exact and
approximate maneuvers, we substitute vys from (1.19) in (1.47) and calculate the
steady-state velocity components of the vehicle in G-frame.
16 H. Marzbani et al.

Fig. 1.14 The instantaneous and steady-state global velocity components of the vehicle and their
differences

   
v cos  vys sin v
G
vs D G
RB B
vs D x s s
D Xs (1.59)
vys cos s C vx sin s vYs

Figure 1.14 depicts the instantaneous and steady-state global velocity components
of the vehicle and their differences.
Integration of vXs and vYs provides us with the path of motion based on steady-
state calculation. Figure 1.15 illustrates the instantaneous and steady-state paths
of motion of the vehicle in the global frame. The magnification of the transient
stage depicts how different the paths are. Figure 1.16 depicts X , Xs , Y , Ys , as
functions of t , and Fig. 1.17 depicts X  Xs , Y  Ys . The maximum error of
X  Xs is 3:808 m which is less than 6:89 % of the maximum X of 54:522 m. The
maximum error of Y Ys is also 2:116 m which is less than 2:59 % of the maximum
Y of 82:468 m. The errors in the local coordinate are much less, as are shown in
Figs. 1.18 and 1.19.
The difference between instantaneous and steady-state calculation decreases
exponentially by increasing t0 . When t0 D 2 s the maximum error of X  Xs is
3:796 m which is less than 5:89 % of the maximum X of 64:45 m and the maximum
error of Y  Ys is also 2:114 m which is less than 2:55 % of the maximum Y of
83:154 m. When t0 D 10 s the maximum error of X  Xs is 3:188 m which is less
than 2:33 % of the maximum X of 136:886 m and the maximum error of Y  Ys is
also 0:0002 m which is less than 0:00025 % of the maximum Y of 90:161 m.

1.5 Steady-State Dynamic Steering

According to the findings of the previous section the steady-state responses of a

vehicle are in a good approximation range compared with the results of solving
the dynamic equations of motion of the vehicle. This concept, initiated the idea of
finding the steering angles required to keep a vehicle on a specific road by using
the steady-state responses instead of the dynamic equations. The mentioned idea is
tested in this section and is called the Steady-State Dynamic Steering (S SDS ).
1 Steady-State Vehicle Dynamics 17

Fig. 1.15 The instantaneous and steady-state paths of motion of the vehicle

Fig. 1.16 Plot of X, Xs , Y , Ys , as functions of t

After examining the output–input relationships of the steady-state responses it

was found that the curvature response of the vehicle could be used to find the
steering angles of the vehicle.
18 H. Marzbani et al.

Fig. 1.17 Plot of X  Xs , Y  Ys , as functions of t

Fig. 1.18 Plot of x  xs , as functions of t

Fig. 1.19 Plot of y  ys , as functions of t

k 1
sk D D (1.60)
ı Rı

where R is the radius of curvature of the road, k D R1 is the curvature, and ı is the
steady-state steering angle of the vehicle.
To determine the road curvature center, it is necessary to find the radius of
curvature of the road at any time during the travel. Now, consider having the
1 Steady-State Vehicle Dynamics 19

Fig. 1.20 Cycloidal Motion lane change maneuver-moving 3 m from one lane to the adjacent lane

equation of a road which could be used to find the radius of curvature of the road. By
replacing the radius of curvature found from the equation of the road in Eq. (1.60)
we can find the steady-state steering angle of the vehicle and validate the findings
by solving the equations of motion of the vehicle using the resulted steering angles.
The equation for lane change maneuvering which will be examined is the following:

y.t / D d. ˛ x.t /  1
2
sin.2˛x.t ///
(1.61)
x.t / D vx t

where d is the scaling factor and ˛ is

r
2am
˛D (1.62)
d

where am is the maximum acceleration.

Equation (1.61) indicates a Cycloid motion. This produces a sine acceleration
curve. The only problem with such a curve is the existence of jerk in the start of the
motion, but the acceleration and velocity start and end at zero. Figure 1.20 shows
the mentioned curve drawn using the following values:

am D 0:02m=s2 d D 3m vx D 1m=s (1.63)

As mentioned above the equation for the steering angles can be easily calculated by
the use of the curvature response, but in order to check the differences we’ll solve
the equations for both kinematic and steady-state situations.
The forward velocity has been adjusted on 1m=s to start with. This will be
changed to higher velocities to check the effect of the speed change as well.
The first step is to find the center of rotation of the road, and radius of curvature.
r q
p .v2x C 0:45d v2x am .1  cos.2:82vx t am 2 3
// /
.xP 2 C yP 2 /3 d
RD D 0:78 q
yR xP  xR yP sin.2:82 adm t vx /
(1.64)
Radius of curvature from Eq. (1.64) is shown in Fig. 1.21.
Center of curvature of the road can be found using the radius of curvature and
the unit normal vectors. The center of curvature of the road is shown in Fig. 1.22.
20 H. Marzbani et al.

Fig. 1.21 Radius of curvature for Cycloidal Motion Lane Change Maneuver Road

Fig. 1.22 Radius of

curvature for Cycloidal
Motion Lane Change
Maneuver Road

From Fig. 1.22 it can be seen that for the start of the motion the car is on a straight
line which has the center of rotation in the positive side towards infinity. Then the
radius of curvature gets closer to the road and starts getting far from the road again
as the car gets closer to the center of the road. Right at the center the radius of
curvature moves toward infinity once again and as soon as the side of the road curve
changes it repeats the same motion backwards but this time on the negative side.
The steering angles of a front-wheel-steering vehicle can be found using the
following two equations:

l
ıi D arctan. / (1.65)
R1  w
2
1 Steady-State Vehicle Dynamics 21

Fig. 1.23 A front wheel

steering vehicle in a right turn

Fig. 1.24 Illustration of a car

which is moving on a road at
a point that Cc is the center of
curvature

l
ıo D arctan. / (1.66)
R1 C w
2

where ıi is the angle for the wheel in the inside of the turn, ıo is the angle for the
wheel in the outside of the turn, w is the width of the front of the car, and R1 is the
vertical distance of the center of mass of the vehicle from the center of rotation of
the car. These are illustrated in Fig. 1.23.
There is a global coordinate frame G attached to the ground, and a vehicle
coordinate frame B attached to the car at its mass center C , as shown in Fig. 1.24.
The z and Z axes are assumed to be parallel and the angle indicates the orientation
of B in G. is the angle between x and X axes, and is called the heading angle
of the car. If (Xc ; Yc ) indicates the coordinates of the road curvature center cc in the
global coordinate frame G, then the coordinate of cc in B would be
22 H. Marzbani et al.

Fig. 1.25 The kinematic

steering angles—Black: ıi
and Red: ıo

B
rC D Rz; .G rC  G
d/ (1.67)

2 3 2 3 02 3 2 31
xc cos sin 0 Xc X
4yc 5 D 4 sin cos 05 @ 4 Yc 5  4 Y 5A
0 0 0 1 0 0
2 3 (1.68)
.Xc  X / cos C .Yc  Y / sin
D 4 .Yc  Y / cos  .Xc  X / sin 5
0

When the equation of the road is given, Eq. (1.68) determines the point that
the car should turn about. We use Eq. (1.68) to calculate the position of the
curvature center and set the steer angles to have the same point as the turning center
of the car. Having the coordinates of the road curvature center cc in the vehicle
coordinate frame is enough to determine the kinematic characteristic R1 .

R1 D yc D .Yc  Y / cos  .Xc  X / sin (1.69)

Figure 1.25 illustrates the outer and inner wheel’s kinematic steering angle in
order for the car to be kept on the road shown in Fig. 1.20. These are calculated
using nominal values 1:5 m and 3 m for the width and total length of the vehicle,
respectively.
Recalling from introduction, we can use the following equations of motion of the
vehicle, and the kinematic steering angles found above to find out the actual road
resulted from applying these steering angles, which we will call the kinematic road
from now on.
1
vP x D Fx C r vy (1.70)
m
 
Cˇ Cr Cı
vP y D vy  vx  rC ı .t / (1.71)
mvx m m
Dˇ Dr Dı
rP D vy C rC ı .t / (1.72)
Iz vx Iz Iz
1 Steady-State Vehicle Dynamics 23

where

@Fy a1 a2
Cr D D  C˛f C C˛r (1.73)
@r vx vx
@Fy
Cˇ D D C˛f  C˛r (1.74)
@ˇ
@Fy
Cı D D C˛f (1.75)
@ı
@Mz a2 a2
Dr D D  1 C˛f  2 C˛r (1.76)
@r vx vx
@Mz
Dˇ D D a2 C˛r  a1 C˛f (1.77)
@ˇ
@Mz
Dı D D a1 C˛f (1.78)
@ı
vy
ˇD (1.79)
vx

The solution of the equations to a given time dependent steer angle ı .t / would be

vx D vx .t / (1.80)
vy D vy .t / (1.81)
r D r .t / (1.82)

Equations of motion of the vehicle are calculated using the following nominal
values:

C˛f D 100;000 C˛r D 100;000 m D 1;000 kg

a1 D 1:2 m a2 D 1:8 m Iz D 1;650 kg m2 (1.83)

and ı.t / can be calculated by getting the average of ıi and ıo according to the
following equation:
 
2 tan ıi tan ıo
ı.t / D arctan (1.84)
tan ıi C tan ıo

ı.t / is illustrated in Fig. 1.26.

The equations of motion will be solved for ı.t / from Eq. (1.84) to find out what
will be the final path taken by the vehicle having these steering angles. By solving
the equations we will get to see the kinematic answers for vy which indicates the
lateral velocity, and r which is the rate of change of heading angle of the vehicle.
These can be seen in Figs. 1.27, 1.28, 1.29, 1.30, 1.31, 1.32, 1.33.
24 H. Marzbani et al.

Fig. 1.26 The average of the kinematic steering angles

Fig. 1.27 The side slip velocity vy of the vehicle for the kinematic steering case

The resulted road from using the kinematic steering angles is shown in Fig. 1.34.
Now the problem will be solved using the new method based on the steady-state
responses of the vehicle as mentioned above. More specifically the response used
here is the curvature response of the vehicle.
The curvature response sk of the vehicle may be calculated using Eq. (1.85) and
nominal values (1.83)

Cd D b  Cb D d 1
sk D D D 0:3331112592 (1.85)
vx .Dr Cb  Cr Db C mVx Db / Rsteady ı.t /

where Rsteady is the radius of curvature using steady-state conditions, and ı.t / is
the new steering angle which will be found by putting the steady-state curvature
radius equal to the radius of curvature of the road. This will result in the following
equations:
1 Steady-State Vehicle Dynamics 25

Fig. 1.28 The rate of change of the heading angle r.t / of the vehicle for the kinematic steering
case

Fig. 1.29 The x-component of the resulted motion expressed in body frame for the kinematic case

Fig. 1.30 The y-component of the resulted motion expressed in body frame for the kinematic case
26 H. Marzbani et al.

Fig. 1.31 Resulted road as seen in the body frame for the kinematic case

Fig. 1.32 The X-component of velocity in the global frame for the kinematic case

Fig. 1.33 The Y -component of velocity in the global frame for the kinematic case

Fig. 1.34 Kinematic steering road

1 Steady-State Vehicle Dynamics 27

Fig. 1.35 The kinematic steering angles vs. steady-state steering angles, red: kinematic—green:
steady-state

Fig. 1.36 The difference between kinematic and steady-state steering angles

Fig. 1.37 The resulted road in the global frame

Fig. 1.38 Actual road (red) vs. kinematic steering road (black)

3:002
Rsteady D (1.86)
ı.t /

0:7644530225 sin.0:2309401076t /
ı.t / D p (1.87)
.1 C .0:1102657791  0:110265779 cos.0:2309401076t //2 /3

The equation for the steering angles has been found and illustrated in comparison
with the kinematic steering angles in Fig. 1.35.
As it can be seen in both Figs. 1.35 and 1.36, the difference between the two
is very small in case of a slow forward velocity (vx D 1), and it happens mostly
while returning to the straight section of the road. In order to validate the new
method and the steering angles found using this method, the equation of motion
of the vehicle was solved once again and this time by substituting the new steering
angles found using the steady-state responses of the vehicle. The road resulted using
the steady-state dynamic angles is shown in Fig. 1.37. The figure proves to be very
similar to the actual road but new plots will show how accurate the two methods are.
It can be seen from Figs. 1.38 and 1.39 that the road resulted from using the
steady-state steering angles is generating less error and matches the actual road
28 H. Marzbani et al.

Fig. 1.39 Actual road (red) vs. steady-state road (green)

Fig. 1.40 The difference between the steady-state and kinematic angles—high speed maneuver

Fig. 1.41 Actual road (red) vs. steady-state road (black)—high speed maneuver

Fig. 1.42 Actual road (red) vs. steady-state road (green)—high speed manoeuvre

much better than the kinematic steering case. In order to take the testing of the new
method further, same maneuvers will be repeated with a higher forward velocity
equal to 10 m/s this time. The results are shown in Figs. 1.40 and 1.41.
Comparing the difference between the angles in this case with the previous
case proves that the deviation is much more than before in the case of high speed
maneuvers.
Figures 1.40 and 1.41 show that the major source of the bigger difference in the
steering angles as a result of increasing the speed is the kinematic steering angles
calculations. As seen in Fig. 1.42 the steering angles calculated using the steady-
state dynamics are consistent after the change of speed but the deviation from the
road gets much bigger for the kinematic steering case.
To have a closer look, the final position of the actual road will be compared
with the final position of the vehicles traveling on the road using the two different
methods which are being investigated. Although this might not be the place on
which the maximum deviation happens at, it gives a good understanding of the
results. On the other hand, from the figures it is obvious that all of these maneuvers
will take the car in the desired direction but the main difference happens at the end
of the road. The values of the end point of the actual road can be easily found by
substituting the time at which the traveling ends which is equal to T D 2:72 s in the
case with the higher forward velocity case.
1 Steady-State Vehicle Dynamics 29

The position of the end point of the road is at .x; y/ D .27:2; 3/ m. This is
shown in figures of the road. The car travels 27:2 m horizontally and this results in
a 27:36 m total travel.
The final point of travel for the case of the kinematic steering angles is .x; y/ D
.26:98; 2:66/ m which is a total travel of about 27:1 m. This point for the case of
steady state steering angles is at .x; y/ D .26:95; 2:94/ m which is again 27:1 m of
travel.
Looking back at the main purpose of the maneuver, which was changing the lane
that the vehicle is moving in to the adjacent lane with 3 m of vertical displacement,
one can come to the conclusion that the steady-state steering angles is providing a
much more accurate and faster response. The displacement in the vertical direction
in the steady-state case is only deviating about 1:6 % from the actual road which is
very accurate. The amount of deviation in the other case, however, is about 11:3 %
from the actual road.

1.6 Conclusion

The proximity of the transient and steady-state responses has been checked and
validated by applying a step change input to the steering angle of a vehicle with
constant velocity. This proved that if using the steady-state responses could help
to solve any problem, they are close enough to be used instead of the transient
responses of the vehicle. A unique method for finding steering angles for the use of
the autodriver algorithm was introduced using the curvature response of the vehicle.
This method which is called Steady-State Dynamic Steering (S SDS ) helps to find
the angles which are accurate enough compared to the dynamic steering angles, with
short calculations and for every kind of road equations. The angles found using the
S SDS method were applied on different scenarios, which proved its high accuracy,
less complication, and consistency in the resulted road. The use of this method
will result in much shorter calculation times, less complicated and smaller steering
angles equations, and a much smaller work load on the control feedback system
as a result. Besides, the S SDS method proves to be consistent even in the case of
completed road edition which made it impossible to solve the problem dynamically
(Marzbani 2012; Jazar 2010; Pacejka 1973).

1.7 Key Symbols

d Translation scalar
d Translation vector
di Element i of d
D Homogeneous translation matrix
A
DB Homogeneous translation matrix from B-frame to A-frame
E Total illumination
30 H. Marzbani et al.

Er Distance-dependent illuminance
E' Incident angle-dependent illuminance
I; Ii Light intensity
nO Normal unit vector to the collector
O Origin point of a coordinate frame
r Distance between target and source of light, position scalar
r Position vector
ri Element i of r
rij Element of row i and column j of matrix R
A
r Vector r expressed in A-frame
rO Direction of light ray with respect to the collector
r,, Local spherical coordinates
R Rotation transformation matrix
A
RB Rotation transformation matrix from B-frame to A-frame
Rx;˛ Rotation transformation about x-axis with ˛ angle
S Sun vector
T Homogeneous transformation matrix
A
TB Homogeneous transformation matrix from B-frame to A -frame
x,y,z Local coordinate axes
X ,Y ,Z Global coordinate axes

˛ Yaw angle
ˇ Pitch angle
Orientation of the vehicle body-frame in the global frame
 Azimuth angle
 Elevation angle
' Angle of incidence

References

Jazar RN (2013) Vehicle dynamics: theory and applications, 2nd edn. Springer, New York
Jazar RN (2014) Vehicle dynamics: theory and applications, 2nd edn. Springer, New York
Jazar RN (2011) Advanced dynamics: rigid body, multibody, and aerospace application. Wiley,
New York
Jazar RN (2010) Mathematical theory of autodriver for autonomous vehicles. J Vib Control
16(2):253–279.
Marzbani H (2012) Flat ride; problems and solutions in vehicle. Nonlinear Eng 1(3–4):101–108
Pacejka, HB (1973) Simplified analysis of steady-state turning behaviour of motor vehicles. Part 1.
Handling diagrams of simple systems. Vehicle Syst Dyn 2:161–172
Tahami F, Kazemi R, Farhanghi, S (2003) A novel driver assist stability system for all-wheeldrive
electric vehicles. IEEE Trans. Vehicular Technol 52(3):683–692
ISO4138 (2012) Passenger Cars Steady-state circular driving behaviour, Open Loop test procedure,
ISO 4138
Chapter 2
On the Razi Acceleration

A. Salahuddin M. Harithuddin, Pavel M. Trivailo, and Reza N. Jazar

Keywords Razi acceleration • Rigid body kinematics • Multiple coordinate

system kinematics • Coordinate transformation

2.1 Introduction

Jazar presented a revised method of calculating vector derivatives in multiple

coordinate frames using the extended variation of the Euler derivative transfor-
mation formula (Jazar 2012). The paper derives the general expression for vector
derivatives for a system of three relatively rotating coordinate frames called the
mixed derivative transformation formula. The formula is demonstrated as follows:
Let A, B, and C be three arbitrary, relatively rotating coordinate frames sharing an
origin as shown in Fig. 2.1. Let B  be a generic vector expressed in the frame B,
A !B be the angular velocity of the frame B with respect to the frame A, and C !B be
the angular velocity of the frame B with respect to the frame C . The time-derivative
of the B-vector as seen from the A frame can be expressed as
A
dB C
dB  
D  C BA !B  BC !B  B  (2.1)
dt dt
where the left superscript indicates the frame in which the vector is expressed.
The formula relates the A-derivative of a B-vector to the C -derivative of the
same B-vector by means of the angular velocity differences between the three

A.S.M. Harithuddin • P.M. Trivailo () • R.N. Jazar

School of Aerospace, Mechanical, and Manufacturing Engineering,
RMIT University, Melbourne, VIC, Australia
e-mail: [email protected]; [email protected];
[email protected]

© Springer International Publishing Switzerland 2015 31

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__2
32 A.S.M. Harithuddin et al.

Fig. 2.1 Three coordinate

frames system A

zA B
zC zB

C

B

yC

yA

xB yB

xC
xA

coordinate frames. The formula presents a more general expression of the classical
derivative transformation formula between two relatively rotating frames
A B
dB dB
D  C BA !B  B  (2.2)
dt dt

where the frame C is assumed to coincide with B (C  B), thus BC !B D 0. One

advantage of the general mixed derivative transformation formula is that it leaves
the need to compute the derivative of B d B =dt in its local coordinate frame.
An interesting result appears from the application of a reference frame system
with three coordinate frames. A new acceleration term, given the name the Razi
acceleration (Jazar 2011), is shown to appear when the derivatives of an acceleration
vector are taken from two different coordinate frames. Such differentiation, which
resulting expression is called mixed acceleration, contains the Razi term, .A !C 
B !C /  r. To demonstrate the Razi acceleration term, let there be three relatively
rotating coordinate frames A, B, and C , shown in Fig. 2.1. Assume a position vector
r is expressed in its local frame which is C . Taking the first derivative of r in C from
B using the vector derivative transformation formula Eq. (2.2) yields
B C
dC dC
rD r C CB !C  C r (2.3)
dt dt
Bd
C
Taking the derivative of dt
r again, but from the frame A yields:

A
d Bd C
rD C
AB a
dt dt
2 On the Razi Acceleration 33

C 
D C CC a C CA !C  CC v C CBC ˛C  C r C A !C  CB !C  C r
 
CCB !C  CC v C CB !C  CA !C  C r (2.4)

where
C
d2 C
C
CC a  r
dt
C
dC
C
Cv  r (2.5)
dt
C
dC
C
CB ˛C  !C
dt B
and the left superscripts in (2.4) show that every vector is expressed in the frame C .
The term CA !C  CB !C  C r is a new term in the acceleration expression called
the Razi acceleration.
For comparison, let us assume that the frame B is not rotating and coincides
with the frame A, making B  A. Equation (2.4) is now reduced to the classical
expression of acceleration between two relatively rotating frames B and C
B
d2 C
r D BBC a D C CC a C 2CB !C  CC v C CBC ˛C  C r
dt 2
 
CCB !C  CB !C  C r (2.6)

Equation (2.6) shows that there are three supplementary terms to the absolute
acceleration of a point in local frame C : The Coriolis acceleration 2CB !C  CC v,
 acceleration CB ˛C  r, and the centripetal acceleration B !C 
C C C
the tangential
B !C  r . The difference between the expressions in Eqs. (2.4) and (2.6) is the
C C

Razi term which has the form of a vector product between a product of two angular
velocities and a position. This shows that adding another rotating coordinate frame
and differentiating a displacement vector from two different frames reveals a new
acceleration term.
It is shown in the published work of Jazar that the Razi acceleration is a novel
finding in classical mechanics. In its current status, however, the Razi acceleration
has only found a limited application in relative kinematics. This paper seeks to
reveal the aspect of the Razi acceleration in mechanics of relative rigid body motion
specifically in the multiple coordinate system kinematics. We show the inclusion
of the new Razi term in the equation of motion of a rigid body described in
non-inertial rotating coordinate frames. To do this, we redefine and extend the
Euler derivative transformation formula to be applied in a system with arbitrary
number of coordinate frames. We provide an illustrative example of a rigid body in
compound rotation motion about a fixed point to show the appearance of the Razi
acceleration. We compare the magnitudes of the Razi acceleration with the classical
centripetal acceleration to show that in the case of complex rotating motion, the Razi
acceleration may appear as a result of rotational inertia.
34 A.S.M. Harithuddin et al.

2.2 Time-Derivative and Coordinate Frame

The time-derivative of a vector depends on the coordinate frame in which it is

differentiated. In a multiple coordinate frames environment, it is important to know
from which coordinate the derivative vectors, e.g. velocity and acceleration, are
calculated and to which coordinate frame it is related. In this section we introduce
the derivative transformation formula and expand it to include more than two
reference frames.

2.2.1 Euler Derivative Transformation Formula

Any vector r in a coordinate frame can be represented by its components using the
O be an arbitrary coordinate frame in which
basis vectors of the frame. Let B.O {O|Ok/
r is expressed. The vector is written using the basis vectors of B as

r D xO{ C y |O C zkO (2.7)

For r, a vector function of time, its components x, y, and z are scalar functions of
time. The time-derivative of the vector r is

d P
P { C yP |O C zPkO C x {PO C y |PO C zkO
r D xO (2.8)
dt
Since the basis vector is constant with respect to its own frame, the time-derivative
of a vector expressed in B differentiated from its own frame B is
B
dB
B
BrP D P { C yP |O C zPkO
r D xO (2.9)
dt
This is called a local or simple derivative since the vector derivative is taken from
the coordinate frame in which it is also expressed. The result is a simple derivative
of its scalar components.
Now consider global-fixed coordinate frame G.O IOJO K/ O and a rotating body-
O
fixed coordinate frame B.O {O|Ok/ as shown in Fig. 2.2. Due to the relative motion of
the coordinate frames, the basis vectors of one frame are not constant in relation to
the other frame. The time-derivative of a vector expressed in B differentiated from
frame G is expressed as
G G G G
dB d d d O
B
GrP D P { C yP |O C zPkO C x
r D xO {O C y |O C z k (2.10)
dt dt dt dt
Similarly, the time-derivative of a vector expressed in G differentiated from frame
B is expressed as
2 On the Razi Acceleration 35

Fig. 2.2 Binary coordinate

frames system—a G
global-fixed frame G and a
JK B
rotating body frame B
Jk

r (t)

B B B B
dG d O d O d O
G
BrP D r D XP IO C YP JO C ZP KO C X I CY J CZ K (2.11)
dt dt dt dt
An angular velocity vector can be defined in terms of a time-derivative of basis
vectors coordinate frame in relation to another coordinate frame. The angular
velocity of a rotating frame B in coordinate frame G, G !B , expressed using the
B-coordinates, is written as
 G  !  G 
G O
O d |O d k d {O O
G !B D k  {O C {O  |O C |O 
B
k (2.12)
dt dt dt

And the angular velocity of a rotating frame G in coordinate frame B, B !G ,

expressed using the G-coordinates, is written as
! ! !
B
d JO B
d KO B O
dI
G
B !G D KO  IO C IO  JO C JO  KO (2.13)
dt dt dt

Equations (2.10) and (2.11) can be represented as

G B
dB dB
B
GrP D rD r C BG !B  B r (2.14)
dt dt
B G
dG dG
G
BrP D rD rCG
B !G  r
G
(2.15)
dt dt
An alternate proof of Eqs. (2.14) and (2.15) is given in Appendix 1.
36 A.S.M. Harithuddin et al.

The resulting equations show the method of calculating and transforming the
derivative of a vector expressed in one frame to another. This results in the most
fundamental result of kinematics: Let A and B be two arbitrary frames which
motion is related by the angular velocity !, for every vector  expressed in B,
its time derivative with respect to A is given by
A B
dB dB
D  C BA !B  B  (2.16)
dt dt
This formula is called the Euler derivative transformation formula. It relates the
first time-derivative of a vector with respect to another coordinate frame by means
of angular velocity.
Despite its importance and ubiquitous application in classical mechanics, there
is no universally accepted name for the Euler derivative transformation formula in
Eq. (2.16). Several terminologies are used: kinematic theorem (Kane et al. 1983),
transport theorem (Rao 2006), and transport equation (Kasdin and Paley 2011).
These terms, although terminologically correct, are more prevalent in the subject
of fluid mechanics to refer to entirely different concepts. We use Zipfel’s (Zipfel
2000) and Jazar’s (Jazar 2011) terminology, which we opine as more descriptive
and unique to the subject of derivative kinematics. The name also is a homage
to Leonard Euler who first used the technique of coordinate transformation in his
works on rigid body dynamics (Koetsier 2007; Tenenbaum 2004; Baruh 1999; Van
Der Ha and Shuster 2009).

2.2.2 Second Derivative

One of the most important applications resulting from the vector derivative transfor-
mation method is the development of inertial accelerations. Inertial accelerations,
or sometimes called d’Alembert accelerations or, misnomerly, “fictitious” accel-
erations, are the acceleration terms that appear when the double-derivatives of a
position vector are transformed from a rotating frame to an inertial frame or vice
versa.
To demonstrate this, let G be an inertial coordinate frame and B a non-inertial
coordinate frame rotating with respect to G with an angular velocityG !B . Using the
Euler derivative transformation formula, the first derivative of a vector B r as seen
from G is
G B
dB dB
rDG
B P
r D r C BG !B  B r (2.17)
dt dt

Using the formula again to get the second derivative of the vector B r from G yields
2 On the Razi Acceleration 37

G B 
d Gd B G
d dB
rD r C G !B  r
B B
dt dt dt dt
B   B 
d Bd B dB
D r C G !B  r C G !B 
B B B
r C G !B  r
B B
dt dt dt
B
d2 B B
dB  
D r C B
G ˛ B  B
r C 2B
G ! B  r C BG !B  BG !B  B r
dt 2 dt
D B
BB a C BG ˛B  B r C 2BG !B  BB v
 
CBG !B  BG !B  B r (2.18)

Equation (2.18) is the classical expression for the acceleration of a point in a

rotating frame as seen from an inertial frame. The term BBB a is the local acceleration
of a point in B with respect to the frame itself regardless of the rotation of B in G.
The second term BG ˛B  B r is called the tangential acceleration as its direction is
always tangential to the rotation path. Tangential acceleration term is represented
by a vector product of an angular acceleration and a position vector. The third term
2BG !B  BB v is an acceleration term traditionally named after Coriolis. The Coriolis
acceleration term appears from the two different sources: half of the term comes
from a local differentiation of BG !B  B r, and the other half is from the product of
angular velocity and the local linear velocity vector. The Coriolis acceleration in
the classical second derivative between two frames is written as twice the vector
product of an angular velocity and a linear velocity. The final term in the expression
B
G ! B  B
G ! B  B
r is called the centripetal acceleration as its direction is towards
the center of rotation. The centripetal acceleration has the form of a vector product
of an angular velocity with another vector product of an angular velocity and a
position vector.

2.2.3 Derivative Transformation Formula in Three

Coordinate Frames

The Euler derivative transformation formula is now extended to include a third

coordinate frame. Let A, B, and C be three relatively-rotating coordinate frames
sharing a same origin as shown in Fig. 2.3.
The first time-derivative of a B-vector in relative to the A-frame is written as
B
Av D BB v C BA !B  B r (2.19)

And the first time derivative of the same B-vector in relative to the C -frame is
written as
B
Cv D BB v C BC !B  B r (2.20)
38 A.S.M. Harithuddin et al.

Fig. 2.3 Three relatively

rotating coordinate frames A
ZA B
ZC ZB

C Br

yC
yA

yB
xB

xA xC

Combining Eqs. (2.19) and (2.20) yields

B 
B
Av D BC v C A !B  BC !B  B r (2.21)

The resulting equation relates the A-derivative of a B-vector (2.19) and the C -
derivative of the same B-vector (2.20) by means of the angular velocity differences
between the three coordinate frames. This result also shows that if the relative
velocities and angular velocities with respect to one of the three coordinate frames
are known, the simple derivative in local frame can be neglected. In another
perspective, by introducing an auxiliary coordinate frame, one can forgo altogether
the calculation of the vector derivatives in the local coordinate frame.
The extended Euler derivative transformation in three coordinate frames can be
generalized as
B P
B 
A
P C
D BC  A !B  BC !B  B  (2.22)

where  can be only generic vector.

This formula is called the mixed derivative transformation formula (Jazar 2012).
It can be used to relate vector derivatives in three frames by means of angular
velocities difference between the three coordinate frames.

2.2.4 Kinematic Chain Rule

To extend the Euler derivative transformation to account for arbitrary number of

moving coordinate frames, we derive the general formula for the composition of
2 On the Razi Acceleration 39

3ω4

3
2ω3
1ω2

0ω1

0 1

Fig. 2.4 Multiple coordinate frames system with frame 0 as the inertial frame

multiple angular velocity vectors. This is useful for, but not limited to, a system with
a chain of coordinate frames where the motion of one frame is subject to another.
One example of such system is depicted in Fig. 2.4.
Consider now n angular velocities representing a set of n .n D 1; 2; : : : ; n/ non-
inertial rotating coordinate frames in an inertial frame indicated by n D 0. Their
relative angular velocities satisfy the relation

X
n
0
0 !n D 00 !1 C 01 !2 C 02 !3 C    C n10 !n D 0
i1 !i (2.23)
iD1

if and only if they all are expressed in the same coordinate frame, which in this
case frame 0. Using the rule, the composition of the angular velocity0 !n can be
represented in any coordinate frame f in the set.
40 A.S.M. Harithuddin et al.

f
X
n
f
0 !n D f R0 00 !n D i1 !i (2.24)
iD1

Equation (2.24) represents the additional rule of the composition of multiple angular
velocity vectors. Any equation of angular velocities that has no indication of the
coordinate frame in which they are expressed is incomplete and generally incorrect.

0 !n ¤ 0 !1 C 1 !2 C 2 !3 C    C n1 !n (2.25)

Angular acceleration is the first time-derivative of an angular velocity vector. For

an angular velocity vector of B with respect to G, for example,

B
G !B D !x {O C !y |O C !z kO (2.26)

the angular acceleration is defined as

B
dB
B
G ˛B D !B D !P x {O C !Py |O C !Pz kO (2.27)
dt G
Suppose there are n angular velocity vectors representing a set of n coordinate
frames expressed in an arbitrary frame f .

f f f f f
0 !n D 0 !1 C 1 !2 C 2 !3 C    C n1 !n (2.28)

To differentiate the composition of angular velocities, which are expressed in f ,

from an arbitrary frame g in the set, one has to apply the Euler derivative
transformation formula in Eq. (2.16).
!
g
d X
n
f
g
df
i1 !i D !n
dt iD1
dt 0
g
d f f f
D 0 !1 C 1 !2 C    C n1 !n
dt
f f f f
D 0 ˛1 C fg !1  0 !1 C 1 ˛2 C fg !2  1 !2 C   
f f
 C n1 ˛n C fg !n  n1 !n

X
n
f f
D i1 ˛i C fg !i  i1 !i (2.29)
iD1

where ˛ is defined as the differentiation of ! in its own frame f d f !=dt . The

formula provides a method to express the first derivative of a composition of angular
velocity vectors.
2 On the Razi Acceleration 41

As an example, let the number of rotating coordinate frames be n D 3. The

angular velocity of frame 3 with respect to the inertial frame 0 is written as a
composition of three angular velocities, expressed in frame 3.
3
0 !3 D 30 !1 C 31 !2 C 32 !3 (2.30)

Applying differentiation from frame 0, using the formula in Eq. (2.29), one obtains
0
d3 3  3 
0 !3 D 0 ˛1 C 0 !1  0 !1 C 1 ˛2 C 0 !2  1 !2
3 3 3 3
dt
 
C 32 ˛3 C 30 !3  32 !3
   
D 30 ˛1 C 31 ˛2 C 30 !2  31 !2 C 32 ˛3 C 30 !3  32 !3 (2.31)

Looking at the third term in the right-hand side, 32 ˛3 is the rate of magnitudinal
change of 32 !3 in frame 3, and 30 !3  32 !3 is the convective rate of change of 32 !3
due to its angular velocity of frame 3 with respect to the frame it is differentiated,
3
0 !3 . The second right-hand side term can be interpreted similarly. The total angular
acceleration of frame 1 with respect to frame 0 is only 30 ˛1 as the direction of the
angular velocity of frame 1 with respect to frame 0 30 !1 does not change in frame
from which it is differentiated as 30 !1  30 !1 D 0
Therefore, any equation of angular acceleration that has no indication of in which
coordinate frame they are expressed, or without the convective rate of change term
P n
f f
g !i  i1 !i is incomplete and generally incorrect.
iD1

0 ˛n ¤ 0 ˛1 C 1 ˛2 C 2 ˛3 C    C n1 ˛n (2.32)
!
g
d X
n
f f f f f
i1 !i ¤ 0 ˛1 C 1 ˛2 C 2 ˛3 C    C n1 ˛n (2.33)
dt iD1

2.3 The Razi Acceleration

Jazar (2012) showed that the Razi term appears as a relative acceleration due to
transforming vector derivative in multiple reference frames. Here, we show that the
Razi term can be exhibited by the same method that the inertial accelerations in a
fixed-axis rotation motion are calculated, such as the Coriolis and the centripetal
acceleration. We examine the system of three nested coordinate frames—a body-
fixed coordinate frame C is rotating in another coordinate frame B, which in
turn is rotating in a globally fixed, inertial frame A, with all coordinate frames
sharing a common origin. The position vector of a body point C r is expressed in
the body frame C and its double derivative, hence the acceleration of the body
42 A.S.M. Harithuddin et al.

point, is taken from the inertial frame A. Any resulting acceleration term, apart
from the local absolute acceleration C CC a, that appears from the transformation
of the vector derivatives is defined as an inertial acceleration. The derivative
transformation between two coordinate frames produces inertial accelerations such
as centripetal, Coriolis, and tangential acceleration as shown in Eq. (2.18). We
show that nested rotations involving more than two coordinate frames produce more
inertial accelerations terms, in which one of them is the Razi acceleration.

2.3.1 Razi Acceleration in Mixed Acceleration Expression

We reproduce the derivation of the Razi acceleration as shown in Jazar (2011, 2012)
as a review. Consider three relatively rotating coordinate frames A, B, and C , where
the frame A is considered as a global-fixed frame, while the frame B is rotating
in A, and a body frame C is rotating in B. This motion is called compound or
nested rotation since the rotation of C is inside the frame B, which in turn is also
rotating inside another frame, A. Assume a position vector C r is expressed in its
local frame C . Taking the first derivative of C r from the B-frame using the vector
derivative transformation formula yields
B C
dC dC
rD r C CB !C  C r (2.34)
dt dt
Bd
C
Taking the derivative of dt
r from the A-frame now yields

A
d Bd C
C
AB a D r
dt dt
A C 
d dC
D r C CB !C  C r
dt dt
C   C 
d Cd C dC
D r C CB !C  C r C CA !C  r C CB !C  C r
dt dt dt
D C
CC aC CB ˛C  C r C CA !C  CC v C CB !C  CC v
   
C CA !C  CB !C  C r C CB !C  CA !C  C r (2.35)

From the expression, the mixed acceleration expression contains the local
acceleration term C CC a; two acceleration terms that have the Coriolis form of a
product of an angular velocity and a linear velocity, which are CA !C  CC v and
B !C  C v; the tangential  term B ˛C  r; and a mixed centripetal
C C C C
Cacceleration
 B !C  A!C  r .
C C
acceleration term
The term CA !C  CB !C  C r is the Razi acceleration. Note the difference
between the Razi acceleration and the mixed centripetal acceleration is the order
2 On the Razi Acceleration 43

of cross product. This indicates that when a coordinate body frame is rotating
relative to other two rotating coordinate frame, the mixed double derivative of C r
contains these inertial acceleration terms: mixed Coriolis acceleration, tangential
acceleration, mixed centripetal acceleration, and the Razi acceleration. An applica-
tion example of mixed derivative transformation formula with this Razi acceleration
is illustrated on Jazar (2012, pp. 858–864).
However, the dynamical interpretation of the mixed acceleration is not clear.
In practice, the double derivative of a position vector is usually directly trans-
formed from the local frame to the inertial frame without undergoing a mid-frame
differentiation. The reason is that a double-differentiation of a point position in
a rotating frame from the inertial frame gives the additional acceleration terms,
which contribute to inertial effect such as centrifugal and Coriolis forces. In mixed
acceleration, in contrast, the differentiation is taken from two different frames. The
A
mixed acceleration ABC a D dtd CB v can be used to complement the acceleration of a
point in C from B using the mixed derivative transformation formula in Eq. (2.22).
B
dC A
dC C  C
Bv D B v C B !C  A !C  B v
C
(2.36)
dt dt
Next, we show that if there are more than one nested rotation, the Razi
acceleration would appear even when the vector derivative is transformed directly
from the local frame to the inertial frame without using the mixed acceleration
technique.

2.3.2 Razi Acceleration as an Inertial Acceleration

The derivation of the Razi acceleration here uses the Euler derivative transformation
formula in conjunction with the kinematic chain rule. It differs from the mixed
acceleration method in such a way that the both derivatives of a body point in a
local coordinate frame are taken from the inertial frame, the same way the Coriolis,
centripetal, and tangential accelerations are derived. To demonstrate, consider a two
coordinate frames system with G as the global-fixed inertial frame and B as the
rotating body-fixed frame. Since the relative rotation between the two frames can
be represented by one angular velocity G !B , we can use Eq. (2.16) twice to find
the double derivative of a body-fixed position vector B r from the inertial frame G.
Looking at the resulting expression of the acceleration in two frames, shown in
Eq. (2.18), one can separate the acceleration terms as follows
44 A.S.M. Harithuddin et al.

B
BB a  local acceleration in B
B
G ˛B  B r  tangential acceleration
(2.37)
2BG !B  BB v  Coriolis acceleration
B
G !B  .BG !B  B r/  centripetal acceleration

Now, consider a slightly complex case of nested rotations. Consider three

relatively rotating frames A, B, and C where the A-frame is treated as the inertial
frame, the B-frame is rotating in A, and the C -frame, which is the body-fixed
frame, is rotating in B. The motion of a rigid body which is spinning in one or
more rotating coordinate frames is called compound rotation or nested rotation.
Compound rotation is usually seen in the gyroscopic motion, or the precession and
nutation of a rigid body. To express the velocity of a point in C as seen from the
inertial frame, the Euler derivative transformation formula in (2.16) is invoked to
transform the velocity from the local frame C to the inertial frame A
A
dC
r D CA rP D CC rP C CA !C  C r (2.38)
dt
Since, the frame C is in B, and the frame B is in A, the angular velocity CA !C can
be expanded using the angular velocity addition rule from Eq. (2.24).
C
A !C D CA !B C CB !C (2.39)

The expression of the acceleration of a point in C from the inertial frame can be
expressed as
A
d Ad C A
dC
r D Ca C !C  C r C 2CA !C  C v C CA !C  .CA !C  C r/ (2.40)
dt dt dt A
A
The angular velocity derivative dtd CA !C can be expanded using the rule of addition
for angular acceleration in (2.29) with n D 2
A A  
dC d C
A !C D !B C CB !C
dt dt A
D CA ˛B C CA !B  CA !B C CB ˛C C CA !C  CB !C
D CA ˛B C CB ˛C C CA !C  CB !C (2.41)

The expanded expression of the acceleration of a point in C from the inertial frame
now becomes
A
d Ad C  
r D C CC a C CA ˛B  C r C CA !C  CB !C  C r C CB ˛C  C r
dt dt
C2CA !C  C v C CA !C  .CA !C  C r/ (2.42)
2 On the Razi Acceleration 45

One can now list the acceleration terms expressed in local frame C and include the
Razi term
C
CC a  local acceleration in C
C
A ˛C  C r  tangential acceleration
2CA !C  CC v  Coriolis acceleration (2.43)
C
A !C  .CA !C  C r/  centripetal acceleration
C  C
A !C  B !C  r  Razi acceleration
C

 
The Razi acceleration CA !C  CA !B  C r comes from differentiating the composi-
tion of angular velocity vectors from a different frame as shown in Eq. (2.41). Next,
we show that for the case of n > 2 nested rotating frames, Razi acceleration terms
appear in the equation of motion.
It is important to note that the Razi acceleration can only appear in compound
rotation motion. A rigid body in a rotation about a fixed axis does not experience
an inertial force caused by the Razi acceleration. This is because a rotation about
a fixed axis can always be simplified to a two coordinate frames system. In such
case the Razi acceleration vanishes as there is only one angular velocity, making
A !B  A !B D 0.

2.3.3 Acceleration Transformation in Multiple Coordinate

Frames with a Common Origin

Given a set of n non-inertial rotating coordinate frames in nested rotations with

i1 !i representing the angular velocity of i -th coordinate frame with respect to the
preceding coordinate frame, we present the general expression for the acceleration
(the double derivative of a vector) of the n-th coordinate frame as seen from the
inertial coordinate frame 0.
0
d2 n Xn Xn
n  n
2
r D 00
n
a D n
nn a C n
i1 ˛ i  n
r C 0 !i  i1 !i  r
n
dt iD1 iD1
!
X n Xn Xn

0 !i  n v C 0 !i  0 !i  r
n n n n n
2 (2.44)
iD1 iD1 iD1
46 A.S.M. Harithuddin et al.

where
n
nn a  local acceleration
X
n
n
i1 ˛i  n r  tangential acceleration
iD1

X
n
2 n
0 !i  nn v  Coriolis acceleration
iD1 (2.45)
!
X
n X
n
n
0 !i  n
0 !i  n r  centripetal acceleration
iD1 iD1

X
n
 
n
0 !i  i1n !i  n r  Razi acceleration
iD1

The general expressions for acceleration when a position vector in a local frame
is differentiated twice from the inertial frame are shown in Table 2.1. Examples are
given for up to a five coordinate frames system. The frame A is dedicated to the
inertial frame and the successive relatively rotating frames are named B, C , and so
on, with the last letter indicates the local coordinate frame. For example, in a system
of three coordinate frames, A is the inertial frame, B is the intermediate frame, and
C is the local frame. All vectors are expressed in the local coordinate frame as
indicated by the left superscript on each vector. Addition of angular acceleration
P n
vectors is simplified as n0 ˛n  n
i0 ˛i .
iD1
The resulting acceleration expression in Table 2.1 shows that if a rigid body
moves linearly with respect to an inertial space, its motion can be sufficiently
described by a linear acceleration vector A a. When a rotating reference frame is
introduced and the motion is described in this frame, the additional acceleration
terms—centripetal, Coriolis, and tangential accelerations—act as “supplementary”
accelerations to describe the motion in a rotating reference frame. When another
rotating frame is introduced in the existing rotating frame making a three coordinate
frames system, the expression adds another acceleration term: the Razi acceler-
ations. As more coordinate frames are introduced in the nested rotating frames
system, the acceleration expressions becomes more complex to accommodate the
relative motion between the coordinate frames. In general, Eq. (2.44) includes all
of the relative accelerations acting on a rigid body in compound motion which
are otherwise unobservable with the classical derivative kinematics method and not
intuitive to the analyst.
2 On the Razi Acceleration 47

Table 2.1 Inertial acceleration terms for describing rigid body in multiple
frames system
No. of frames Acceleration terms Enclosed system
A

Ar

Aa

A
B

Br

A
B
C

Cr

A
B
C
D

Dr

A
B
C
D

Er
48 A.S.M. Harithuddin et al.

2.4 Mechanical Interpretation of the Razi Acceleration

Classical dynamics asserts that there exists a rigid frame of reference in which
Newton’s second law of motion is valid. Assuming the mass is constant at all times,
the law of motion is written as

F D ma (2.46)

where m indicates a mass scalar and a represents the acceleration vector. The
reference frame in which this relationship is valid is called an inertial frame. An
inertial frame is fixed in space, or at most, moves in relation to fixed points in space
with a constant velocity. On the other hand, a non-inertial reference frame can be
defined as a frame which accelerates with respect to the inertially fixed points. The
calculation of motion observed from a non-inertial reference frame does not follow
Newton’s second law due to the relative acceleration between the observer’s frame
and the inertial frame.
Starting from an inertial frame, the method of derivative kinematics is used
to calculate and transform the vector derivatives to the non-inertial observer.
Such transformation produces additional acceleration terms to the Newton’s law
in Eq. (2.46). For the classical case of one stationary inertial frame and a body
frame which rotates about a fixed axis, these additional acceleration terms are
generally divided into tangential acceleration, Coriolis acceleration, and centripetal
acceleration, as shown in the expression in Eq. (2.18).
 
F D m a C atangential C aCoriolis C acentripetal (2.47)

Multiplying the acceleration terms with mass gives the expression of the forces that
is acting on a point mass in the rotating coordinate frame.
In the case of multiple nested coordinate frames, the classical method of
transforming derivative between two frames is limiting. This method is sufficient for
any motion consideration in two-dimensional space, due to the assumption that the
direction of a rigid body’s angular velocity is fixed with respect to the inertial frame
at all time. However, a general formula for a three-dimensional space is needed to
account for the subtle accelerations due to a non-constant angular velocity.
The derivative transformation method can be extended to include the effects of
moving angular velocities. To do this, the complex rotation caused by the non-
constant angular velocity vector is decomposed into several simpler coordinate
frame representation such that each coordinate frame then will have a fixed-axis
angular velocity vector. Such technique results on multiple coordinate frames in
nested rotations. To analyze such system, we incorporate the extended derivative
transformation formula and the kinematic chain rule, which method reveals new
results and application.
For the case of three coordinate frames system (two nested rotating frames
in an inertial frame), we have seen that the acceleration expression includes
2 On the Razi Acceleration 49

a new rotational inertia called which is the Razi acceleration. We also have
shown the general expression for all four inertial acceleration—tangential, Coriolis,
centripetal, and Razi—for an arbitrary number of nested rotating frames.
 
F D m a C atangential C aCoriolis C acentripetal C aRazi (2.48)

Extending the derivative transformation formula and incorporating the kinematic

chain rule reveal several new results and applications. The Razi acceleration is given
a focus due to its appearance as one of the additional acceleration terms in the system
of more than two coordinate frames.
We take the simplest case of the Razi acceleration where there are three nested
coordinate frames—A, B, and C —sharing an origin with the body coordinate frame
C being the “innermost” frame. Such system can be better illustrated by a flat disc
in compound rotation motion, as shown in Fig. 2.5. We fix the C coordinate frame
in the disc, which is spinning on its own axis with an angular speed P and tilted on
a turning shaft by degrees. We fix the B coordinate frame on the shaft, which
is turning with respect to the global-fixed frame with an angular speed . P The two
P P
angular velocities can be written as B !C D  {O and A !B D  |O.
The total acceleration acting on a point r on the disc, as seen from the inertial
frame, can be expressed by Eq. (2.42). Decomposing the expression into multiple
terms, the Razi acceleration, if expressed in the local body coordinates of C , is
written as
C 
aRazi D A !C  CB !C  C r (2.49)

The cross product of two angular velocities CA !C  CB !C is the convective rate

of change of B !C due to the angular velocity of C in A, A !C . Therefore, the Razi
acceleration can be described as an inertial effect caused by the change of the local
angular velocity vector direction. This is separable from the tangential direction
which is due to the change of an angular velocity magnitude, hence producing
acceleration in the direction which is tangent to the rotation curve.
To examine the direction of action of the Razi acceleration, we visualize the
resulting vectors in Fig. 2.6. Assuming the position vector of the point of interest r
is constant in C . Since the angular velocity of C in A can be decomposed using the
addition rule, i.e. CA !C D CA !B C CB !C , the Razi acceleration is written as
C 
aRazi D A !B  CB !C  C r (2.50)

The total acceleration, thus, can be expressed as

A
d Ad C  
r D CA !B  CB !C  C r C CA !C  .CA !C  C r/ (2.51)
dt dt
In Fig. 2.6, the total acceleration vector of a body point in C is decomposed into the
centripetal acceleration vector CA !C  .CA !C  C r/ and the Razi acceleration vector
50 A.S.M. Harithuddin et al.

a yA

yB
ψ
A ωB xC

yC B ωC

O xB
θ
C
xA
zA r(t)
zB

zC

Body coordinate frame C spinning in shaft

coordinate frame B. The B-frame, in turn, is
rotating in inertial frame A.

yA
c
b
C yB
r(t)

zC

θ O
O zB xB
ϕ
zA zB xA
C
r(t)
yC

C-frame rotating in B-frame with B-frame rotating in A-frame with

.^ .^
B ωC = θi
C
A ωB = ϕi
B

Fig. 2.5 Describing compound rotation motion using three coordinate frames system

C 
A !B  CB !C  C r. Using the disc to represent the plane of rotation of the C -frame,
we can see the resultant vector of the Razi acceleration is always in the out-of-plane
direction. This can be tested by finding the direction of the cross products in the
Razi term using the right-hand rule.
Apart from the magnitudes of CA !B , CB !C , and C r, the magnitude of the
Razi acceleration vector depends on the angle between the two angular velocity
vectors, . Figures 2.7, 2.8, 2.9, 2.10, 2.11, 2.12 show the magnitudes of Razi
2 On the Razi Acceleration 51

instantaneous
axis of rotation
A ωC

A ωB

B ωC

C
r

a centripetal = Aω C ( A ωC × r)
a Razi = ( A ωB × B ωC ) × r

Fig. 2.6 Razi acceleration vector

30

acentripetal
20
m/s2

10

aRazi

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.7 B
A !B D P D 5 rad=s; C
B !C D P D 1 rad=s

and centripetal acceleration experienced by the body point C r in the example. The
plots show the effect of to the magnitude of Razi acceleration for different
combinations of angular velocities. Since most readers are more familiar with
the centripetal acceleration effect and its influence on bodies under rotation, we
compare its magnitude with the magnitude of Razi acceleration.
52 A.S.M. Harithuddin et al.

30

acentripetal
20
m/s2

10

aRazi

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.8 B
A !B D P D 4 rad=s; C
B !C D P D 2 rad=s

30

20 acentripetal
m/s2

10

aRazi

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.9 B
A !B D P D 3 rad=s; C
B !C D P D 3 rad=s

2.5 Conclusion

By revising and extending the classical Euler vector derivative transformation

formula, we update the interpretation of the Razi acceleration and reveal its
application in rigid body motion. The Razi acceleration is revealed as one of
the inertial effects, apart from the classical Coriolis, centripetal, and tangential
accelerations, affecting rigid bodies in compound rotation motion. It is shown to
appear as a result of the convective rate of change of a body-frame’s angular velocity
Pn   n
0 !i  i1 !i  r, can only
n n
in nested rotation. Razi’s mathematical expression,
iD1
2 On the Razi Acceleration 53

30

acentripetal
20
m/s2

10

aRazi

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.10 B
A !B D P D 2 rad=s; C
B !C D P D 4 rad=s

30

acentripetal

20
m/s2

10

aRazi

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.11 B
A !B D P D 1 rad=s; C
B !C D P D 5 rad=s

be discovered by using multiple coordinate frames in analyzing the kinematics of

complexly rotating bodies. A practical example is illustrated to show the typical
direction of action of Razi acceleration. Its magnitude is compared to the more
widely acknowledged centripetal acceleration to show that the Razi acceleration
can have a significant impact to the structures of a body in compound rotation.
54 A.S.M. Harithuddin et al.

30

acentripetal
20
m/s2

10

0
0 30 60 90 120 150 180
ψ (deg)

Fig. 2.12 B
A !B D P D 5 rad=s; C
B !C D P D 0 rad=s

Appendix 1: Alternative Proof of Derivative Transformation

Formula

Angular velocity can be defined in terms of time-derivative of rotation matrix.

Consider the two coordinate frames system as shown in Fig. 2.2. The angular
velocity of B in relative to G, G !B is written as follows
G
G !B D G RP B G RBT if expressed in G
(2.52)
B
G !B D G RBT G RP B if expressed in B

To prove Eq. (2.52), we start with the relation between a vector expressed in two
arbitrary coordinate frames G and B.
G
r D G RB B r (2.53)

Finding the derivative of the G-vector from G, we get

G
GrP D G RP B B r D G RP B G RBT G r D G
G !B  r
G
(2.54)

Here, it is assumed that the vector B r is constant in B. If it is not constant, we have

to include the simple derivative of the vector in B and rotate it to G
G
GrP D G RP B B r C G RB BB rP D G G G
P
G !B  r C B r (2.55)

The angular velocities between two arbitrary frames are equal and oppositely
directed given that both are expressed in the same frame

G !B D B !G (2.56)
2 On the Razi Acceleration 55

Using this angular velocity property, rearranging Eq. (2.54) gives

G
BrP DG
GrPCG
B !G  r
G
(2.57)

Appendix 2: Proof of the Kinematic Chain Rule

for Angular Acceleration Vectors

To prove Eq. (2.29), we start with a composition of n angular velocity vectors which
are all expressed in an arbitrary frame g.

g
X
n
g g g g
0 !n D i1 !i D 0 !1 C 1 !2 C    C n1 !n
iD1

D R1 10 !1 C g R2 21 !2 C    C g Rn .n1/
g n
!n (2.58)

We use the technique of differentiating from another frame in Appendix 1.

Differentiating the angular velocities from an arbitrary frame g gives

d X g
g n g
d
i1 !i D R1 10 !1 C g R2 21 !2 C    C g Rn .n1/
g n
!n
dt iD1 dt

D .g R1 10 ˛1 C g RP 1 10 !1 / C .g R2 21 ˛2 C g RP 2 21 !2 / C   
   C .g Rn .n1/
n
˛n C g RP n .n1/
n
!n /

D .g R1 10 ˛1 C g RP 1 g R1T 0 !1 / C .g R2 21 ˛2 C g RP 2 g R2T 1 !2 / C   
g g

˛n C g RP n g RnT
g
   C .g Rn .n1/
n
.n1/ !n /
g g g g
D .0 ˛1 C gg !1  0 !1 / C .1 ˛2 / C gg !2  1 !2 / C   
g g
   C ..n1/ ˛n C gg !n  .n1/ !n / (2.59)

The whole expression can be transformed to any arbitrary frame f by applying the
rotation matrix from g to f , f Rg .
! !
d X f g X
g n n
d g
!i D f
Rg !i
dt iD1 i1 dt iD1 i1
!
g
d f Xn
g
D Rg i1 !i
dt iD1

X
n
f f
D i1 ˛i C fg !i  i1 !i (2.60)
iD1
56 A.S.M. Harithuddin et al.

Notations

In this article, the following symbols and notations are being used. Lowercase bold
letters are used to indicate a vector. Dot accents on a vector indicate that is a time-
derivative of the vector.
• r, v (or rP ), and a (or rR ) are the position, velocity, and acceleration vectors
• ! and ˛ (or !) P are angular velocity and angular acceleration vectors.
Capital letter A, B, C , and G are used to denote a reference frame. In examples
where only B and G are used, the former indicates a rotating, body coordinate frame
and the latter indicates the global-fixed, inertial body frame. When a reference frame
is introduced, its origin point and three basis vectors are indicated. For example:

O
A.Oa1 a2 a3 / G.OX Y Z/ B.oO{ |Ok/

Capital letters R is reserved for rotation matrix and transformation matrix. The
right subscript on a rotation matrix indicates its departure frame, and the left
superscript indicates its destination frame. For example:
B
RA D rotation matrix from frame A to frame B

Left superscript is used to denote the coordinate frame in which the vector is
expressed. For example, if a vector r is expressed in an arbitrary coordinate frame
O it is written as
A which has the basis vector .O{ ; |O; k/,

A
r D r1 {O C r2 |O C r3 kO
D vector r expressed in frame A

Left subscript is used to denote the coordinate frame from which the vector is
differentiated. For example,
V
dA
A
BrP D r
dt
D vector r expressed in frame A and differentiated in frame B
B
d Bd A
BB
A
rR D r
dt dt
D vector r expressed in frame A and differentiated twice in frame B
2 On the Razi Acceleration 57

C
d Bd A
CB
A
rR D r
dt dt

D vector r expressed in frame A and differentiated in frame B

and differentiated again in frame C

Right subscript is used to denote the coordinate frame to which the vector is referred,
and left subscript is the coordinate frame to which it is related. For example, the
angular velocity of frame A with respect to frame B is written as

B !A D angular velocity vector of A with respect to B

If the coordinate frame in which the angular velocity is expressed is specified

C
B !A D angular velocity vector of A with respect to B
expressed in C

2.6 Key Symbols

a DRr Acceleration vector

v DPr Velocity vector
r Displacement vector
˛ Angular acceleration vector
! Angular velocity vector
F Force vector
m Mass scalar
n Number of elements in the set of noninertial coordinate frames
i i th element in a set
f; g Elements in the set of noninertial coordinate frames
IO; JO ; KO Orthonormal unit vectors of an inertial frame
{O; |O; kO Orthonormal unit vectors of a noninertial frame
O Origin point of a coordinate frame
P Angular speed of body (spin)
P Angular speed of body (precession)
Nutation angle, angle between spin vector and precession vector
R Rotation transformation matrix
 Arbitrary vector
A
r Vector r expressed in A-frame
A
BrP Vector rP expressed in A-frame and differentiated in B-frame
58 A.S.M. Harithuddin et al.

A
P
CB r Vector rR expressed in A-frame, first-differentiated in B-frame, and
second-differentiated from C -frame
A
RB Rotation transformation matrix from B-frame to A-frame

References

Baruh H (1999) Analytical dynamics. WCB/McGraw-Hill, Boston

Jazar RN (2012) Derivative and coordinate frames Nonlinear Eng 1(1):25–34
Jazar RN (2011) Advanced dynamics: rigid body, multibody, and aerospace applications. Wiley,
Hoboken, pp 858–864
Kane TR, Likins PW, Levinson DA (1983) Spacecraft dynamics. McGraw-Hill, New York,
pp 1–90
Kasdin NJ, Paley DA (2011) Engineering dynamics: a comprehensive introduction. Princeton
University Press, Princeton
Koetsier T (2007) Euler and kinematics. Stud Hist Philos Math 5:167–194
Rao A (2006) Dynamics of particles and rigid bodies: a systematic approach. Cambridge
University Press, Cambridge
Tenenbaum RA (2004) Fundamentals of applied dynamics. Springer, New York
Van Der Ha JC, Shuster MD (2009) A tutorial on vectors and attitude (Focus on Education).
Control Syst IEEE 29(2):94–107
Zipfel PH (2000) Modeling and simulation of aerospace vehicle dynamics. AIAA, Virginia
Chapter 3
Challenges in Exact Response of Piecewise
Linear Vibration Isolator

Oleksandr Pogorilyi, Reza N. Jazar, and Pavel M. Trivailo

Keywords Piecewise linear systems • Nonlinear vibration isolator • Frequency

response • Exact methods • Numerical methods

3.1 Introduction

Piecewise linear vibration isolator is a vibration isolator whose system has nonlinear
geometric characteristics of stiffness and damping, where nonlinearity is a result of
moving among finite number of linear segments.
The scientific story of piecewise linear vibration isolator starts with the work
of Den Hartog who analyzed an undamped bilinear stiffness system as is shown
in Fig. 3.1 (Den Hartog and Mikina 1932). He showed that the exact solution
for the frequency response of the system can be obtained by solving two tran-
scendental equations. Solving those equations, Den Hartog presented a set of
frequency response plots for a few different values of stiffness. His calculations
were confirmed by experimental results of Jacobsen and Jaspersen (1935) who
explored the bilinear stiffness system and received results close to the Den Hartog
mathematical results.
The next step in study of piecewise linear vibration isolator was analysis of a
single degree of freedom system with clearance in the viscous damping as shown in
Fig. 3.2. Gurtin presented an approximate method for solving of the system (Gurtin
1961). The method is based on finding an equivalent viscous damper.
The mathematical modeling and analysis of piecewise linear systems have been
grown in parallel to their application. Aizerman tried to ground the mathematical

O. Pogorilyi • R.N. Jazar () • P.M. Trivailo

School of Aerospace, Mechanical, and Manufacturing Engineering,
RMIT University, Melbourne, VIC, Australia
e-mail: [email protected]; [email protected]; [email protected]

© Springer International Publishing Switzerland 2015 59

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__3
60 O. Pogorilyi et al.

m stiffness
∆

k1 + k2
k2 k1 k1
u
y –∆ ∆

Fig. 3.1 A dual rate bilinear vibrating system

m damping
∆

c k1
u
y –∆ ∆

Fig. 3.2 A vibrating system with gap-damper engagement

foundation for the piecewise linear systems (Aizerman and Gantmakher 1957, 1958;
Aizerman and Lur’e 1963) and developed a number of methods to deal with these
systems. Also, Fleishman (1965) analyzed a system of coupled piecewise linear
equations from a control viewpoint and described the required conditions for a
periodic response.
The simplicity of the piecewise linear vibrating systems helped scientists to
perform many experiments. They not only discovered validation of the mathematical
predictions, but they also found several unexpected behaviors that were not pre-
dictable by the simple mathematical models. Hayashi (1964) did a notable work in
application direction. His results show that great complication is to be expected even
in simple modeling cases. Masri (1965, 1978), Masri and Stott (1978) used a general
one DOF vibrating system to explain the difficulties that introduce in formulation of
the steady state motion. Marsi used an experimental device to analyze the sensitivity
of the physical system for some of the parameters involved. Watabene (1978) used
the method of Den Hartog to plot the frequency response of an equivalent bilinear
mass-spring vibrator to a base excited cantilever beam with clearance. He examined
the effect of the amplitude of harmonic excitation in his plots.
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 61

Iwan (1973) tried to use approximation methods using averaged periodic

solutions. Nguyen et al. (1986a,b) used numerical integration methods and
described the time domain behavior of the system.
Rosenberg (1966) developed the mathematical foundations for the existence of
steady state solution for equation xR C f .x/ D p  g.t / D p  g.t C T /, where
g.t / is a cosine-like periodic function. He showed that the necessary condition
for the existence of steady state periodic solution of the equation is that f .x/ be
everywhere analytic, odd, monotonically increasing function xf .x/ > 0 for x D 0.
The piecewise linear system without damping is of this type.
Natsiavas (1989), Natsiavas and Gonzalez (1992) developed a stability analysis
for asymmetric piecewise linear force excited system using methodology that was
firstly found by Den Hartog and received system of transcendental equations that
can be used to calculate amplitudes of relative displacement analytically. Narimani
et al. (2004a) predicted the frequency response of piecewise linear system using
perturbation methods. They experimentally verified the perturbation results and
discovered a new phenomenon called the frequency island (Narimani et al. 2004b).
Deshpande et al. (2006) optimized the secondary isolator employing the RMS
method (Jazar et al. 2003; Alkhatib et al. 2004; Christopherson and Jazar 2005; Jazar
et al. 2006a). Jump avoidance conditions of the system were investigated by Jazar
et al. (2006b, 2007). The importance and application of the system were reviewed
by several investigators (Jazar and Golnaraghi 2002; Marzbani et al. 2012, 2013).
Schulman (1983) is the first scientist who discovered chaos in piecewise linear
systems. Later this phenomenon was analyzed in vast systems of this type.
The simplest practical model of a piecewise linear vibration isolator is a system
with bilinear stiffness and dumping characteristics as is shown in Fig. 3.3. The first
spring and damper that are directly attached to m is called “Primary suspension”
and the second stage, which is effective beyond the clearance amplitude , is
called “Secondary suspension.” The clearance  generates a switch to engage the
secondary suspension. This represents a sudden change in the system properties

x damping
c1 + c2
m c1
∆ u
c2 –∆ ∆
c1

k2 k1
stiffness
y k1 + k2

k1
u
–∆ ∆

Fig. 3.3 Mechanical model of the piecewise linear system with symmetric constraints
62 O. Pogorilyi et al.

which provides a hard nonlinearity of the piecewise linear system. In this article the
examined system is subjected to a periodic base excitation y.t / with period T > 0.
The equations governing the motion of the system could be written as

mxR C g1 .x; x/
P D f1 .y; y/
P (3.1)

P and f1 .y; y/
where g1 .x; x/ P are piecewise linear functions presenting sudden
changing characteristics of the system and sudden changing excitation, respectively:
8
ˆ
ˆ.c1 C c2 /xP C .k1 C k2 /x  k2  xy >
<
P D
g1 .x; x/ c xP C k1 x jx  yj <  (3.2)
ˆ 1
:̂.c C c /xP C .k C k /x C k  x  y < 
1 2 1 2 2

8
ˆ
<.c1 C c2 /yP C .k1 C k2 /y
ˆ xy >
P D c1 yP C k1 y
f1 .y; y/ jx  yj <  (3.3)
ˆ
:̂.c C c /yP C .k C k /y x  y < 
1 2 1 2

The equation of motion for the system shown in Fig. 3.3 may also be written in a
nondimensional form:
8
ˆ
<zR C 22 !2 zP C !2 z D w! sin .!t  '/ C !3 z > 1
ˆ 2 2 2

zR C 21 !1 zP C !12 z D w! 2 sin .!t  '/ jzj < 1 (3.4)

ˆ
:̂zR C 2 ! zP C ! 2 z D w! 2 sin .!t  '/  ! 2 z < 1
2 2 2 3

where
u Y
zD wD uDxy
 
k1 k 1 C k2
!12 D !22 D (3.5)
m m
c1 c 1 C c2
1 D p 2 D p
2 k1 m 2 .k1 C k2 /m

We seek the frequency response of the system (3.4) by developing and detecting its
exact steady state time response.
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 63

3.2 Exact Solution

As long as the relative displacement of m is less than , the system is a one degree-
of-freedom base exciting system with well-known time and frequency responses
(Jazar 2013, 2014). However, there is no general closed form solution for the system
when the relative displacement exceeds . In order to determine the frequency
response of the system, we search for possible steady state time response of the
system and detect its maximum amplitude. Repeating this method ends up with a
series of amplitudes for different values of excitation frequencies.
A steady state periodic response of the system should look like Fig. 3.4. To
ensure that the secondary suspension is engaged, we seek a periodic solution of
the system (3.4) with the following initial conditions:

z.0/ D 1 zP.0/ > 0 (3.6)

To set the time axis to begin at this condition, we introduce a phase lag ' in the
excitation function y.

y D Y sin .!t  '/ (3.7)

Furthermore, we assume that there is a steady state periodic response with exactly
the same frequency ! as the excitation frequency. We also assume that in the first
half of the period, the response passes through z D 1 with zP.0/ < 0 at a time t D t1 ,
0 6 t1 6 2=!.
Due to these assumptions, the solution will be jz.t /j D 1 at t D 0, t D t1 ,
t D =!, t D =! C t1 , and t D 2=! in the period starting at t D 0. Let us call
the solution in domain z > 1 as z2 .t / and the solution in domain jzj < 1 as z1 .t /
(see Fig. 3.4). The continuity and compatibility of z1 .t / and z2 .t / would be satisfied
by imposing the following conditions:

z2
1

z1
π
ω
0 t
0 t1 2π
ω
z1

–1
–z2

Fig. 3.4 A steady state periodic response of the system

64 O. Pogorilyi et al.

z2 .0/ D 1 zP2 .0/ D v2 z2 .t1 / D 1 zP2 .t1 / D v1

 
z1 .t1 / D 1 zP1 .t1 / D v1 z1 . / D 1 zP1 . / D v2 (3.8)
! !
There is no however a one-to-one relationship between the input amplitude Y and
the output response z. For some values of Y there may be no such solutions, while
for the other values of Y there may be several.
The underdamped solutions z2 .t / and z1 .t / are:

z1 .t / D e 1 !1 t .A1 sin.!d1 t / C B1 cos.!d1 t //

CC1 sin.!t / C D1 cos.!t / (3.9)

z2 .t / D e 2 !2 t .A2 sin.!d2 t / C B2 cos.!d2 t //

!32
CC2 sin.!t / C D2 cos.!t / C (3.10)
!22

and therefore,

zP1 .t / D 1 !1 e 1 !1 t .A1 sin.!d1 t / C B1 cos.!d1 t //

C!d1 e 1 !1 t .A1 cos.!d1 t /  B1 sin.!d1 t //
C!.C1 cos.!t /  D1 sin.!t // (3.11)

zP2 .t / D 2 !2 e 2 !2 t .A2 sin.!d2 t / C B2 cos.!d2 t //

C!d2 e 2 !2 t .A2 cos.!d2 t /  B2 sin.!d2 t //
C!.C2 cos.!t /  D2 sin.!t // (3.12)

where
q q
!d2 D !2 1  22 !d1 D !1 1  12 (3.13)

The coefficients A1 , B1 , A2 , B2 will be found by imposing the initial and

compatibility conditions (3.8). The coefficients C1 , D1 , C2 , D2 depend on the
forcing function and will be found by collecting the coefficients of sin.!t / and
cos.!t / in z1 .t / and z2 .t /

C1 D Q1 cos '  Q2 sin ' (3.14)

D1 D Q2 cos '  Q1 sin ' (3.15)
C2 D Q3 cos '  Q4 sin ' (3.16)
D2 D Q4 cos '  Q3 sin ' (3.17)
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 65

where

.!12  ! 2 /! 2
Q1 D w (3.18)
.!12  ! 2 /2 C .21 !!1 /2
21 ! 3 !1
Q2 D w (3.19)
.!12  ! 2 /2 C .21 !!1 /2
.!22  ! 2 /! 2
Q3 D w (3.20)
.!22  ! 2 /2 C .22 !!2 /2
22 ! 3 !2
Q4 D w (3.21)
.!22  ! 2 /2 C .22 !!2 /2

Imposing the eight boundary conditions (3.8) on Eqs. (3.9)–(3.12), produces eight
transcendental equations with eight unknowns A2 , B2 , A1 , B1 , t1 , v1 , v2 , and ':
!32
B2 C D2 C D1 (3.22)
!22
2 !2 B2 C !d2 A2 C C2 ! D v2 (3.23)

e 2 !2 t1 .A2 sin.!d2 t1 / C B2 cos.!d2 t1 //

!32
CC2 sin.!t1 / C D2 cos.!t1 / C D1 (3.24)
!22

2 !2 e 2 !2 t1 .A2 sin.!d2 t1 / C B2 sin.!d2 t1 //

C!d2 e 2 !2 t1 .A2 cos.!d2 t1 /  B2 sin.!d2 t1 //
C!.C2 cos.!t1 /  D2 sin.!t1 // D v1 (3.25)

e 1 !1 t1 .A1 sin.!d1 t1 / C B1 cos.!d1 t1 //

CC1 sin.!t1 / C D1 cos.!t1 / D 1 (3.26)

1 !1 e 1 !1 t1 .A1 sin.!d1 t1 / C B1 cos.!d1 t1 //

C!d1 e 1 !1 t1 .A1 cos.!d1 t1 /  B1 sin.!d1 t1 //
C!.C1 cos.!t1 /  D1 sin.!t1 // D v1 (3.27)
  
e 1 !1 ! .A1 sin.!d1 / C B1 cos.!d1 //  D1 D 1 (3.28)
! !
  
 1 !1 e 1 !1 ! .A1 sin.!d1 / C B1 cos.!d1 //
! !
  
C!d1 e 1 !1 ! .A1 cos.!d1 /  B1 sin.!d1 //
! !
C1 ! D v2 (3.29)
66 O. Pogorilyi et al.

The parameters B2 and A2 may be found from the Eqs. (3.22) and (3.23),

!32
B2 D 1  D2  (3.30)
!22

v2 C 2 !2 B2  C2 ! 2 !32
A2 D D Q5  (3.31)
!d2 !2 !d2

and the parameters B1 and A1 from the Eqs. (3.28) and (3.29)
1 !1  !d1  !d1 
B1 D e ! Q6 sin C .D1  1/ cos (3.32)
! !

1 !1  !d1  !d1 
A1 D e ! Q6 cos C .D1  1/ sin (3.33)
! !
where

2 !2 .D2  1/ C C2 !  v2
Q5 D (3.34)
!d2
1 !1 .D1  1/ C C1 !  v2
Q6 D (3.35)
!d1

Now, substitution of A1 , B1 , A2 , and B2 in (3.24)–(3.27) produces four equations

for v1 , v2 , t1 , and '

Q7 sin.!d2 t1 / C Q8 cos.!d2 t1 /
!32
CC2 sin.!t1 / C D2 cos.!t1 / C D1 (3.36)
!22

Q9 sin.!d2 t1 / C Q10 cos.!d2 t1 /

C! .C2 cos.!t1 /  D2 sin.!t1 // D v1 (3.37)

Q11 sin .!d1 t1 / C Q12 cos .!d1 t1 /

CC1 sin.!t1 / C D1 cos.!t1 / D 1 (3.38)

Q13 sin.!d1 t1 / C Q14 cos.!d1 t1 /

C! .C1 cos.!t1 /  D1 sin.!t1 // D v1 (3.39)
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 67

where

Q7 D A2 e 2 !2 t1 Q8 D B2 e 2 !2 t1 (3.40)

Q9 D e 2 !2 t1 .2 !2 A2 C !d2 B2 / (3.41)

2 !2 t1
Q10 D e .!d2 A2  2 !2 B2 / (3.42)
Q11 D A1 e 1 !1 t1 Q12 D B1 e 1 !1 t1 (3.43)
1 !1 t1
Q13 D e .1 !1 A1 C !d1 B1 / (3.44)
Q14 D e 1 !1 t1 .!d1 A1 C 1 !1 B1 / (3.45)

Eliminating v1 between Eqs. (3.37) and (3.39) reduces the number of equations to
three to find v2 , t1 , and '.

Q7 sin.!d2 t1 / C Q8 cos.!d2 t1 /
!32
CC2 sin.!t1 / C D2 cos.!t1 / C D1 (3.46)
!22

 Q9 sin.!d2 t1 / C Q10 cos.!d2 t1 / C Q13 sin.!d1 t1 /  Q14 cos.!d1 t1 /

C! ..C2  C1 / cos.!t1 /  .D2  D1 / sin.!t1 // D 0 (3.47)

Q11 sin .!d1 t1 / C Q12 cos .!d1 t1 /

CC1 sin.!t1 / C D1 cos.!t1 / D 1 (3.48)

The parameter v2 is embedded in A2 , B1 , and A1 . Therefore, Q5 to Q7 and Q9 to

Q14 are functions of v2 . Substituting Q11 and Q12 in Eq. (3.48) produces a linear
equation of Q6

Q15 Q6 C Q16 D Q17 (3.49)

where
 
1 !1

t   !t1
Q15 D e ! 1 sin !d 1 (3.50)
!
Q16 D e 1 !1 t1 .C1 sin.!t1 / C D1 cos.!t1 //
 
 !

t   !t1
Ce 1 1 ! 1 .D1  1/ cos !d1 (3.51)
!

1 !1
Q17 D D1 e ! (3.52)
68 O. Pogorilyi et al.

Q6 is solvable for v2

v2 D !C1 C .D1  1/ 1 !1  Q6 !d1 (3.53)

and therefore, eliminating v2 in Eqs. (3.46) and (3.47) reduces the number of
transcendental equations to two to determine t1 and ' for a set of given system
and excitation.

3.3 Time Response

If we were able to solve Eqs. (3.46)–(3.48) analytically, then we could complete the
solution z2 .t / and z1 .t /. In that case the maximum value of z2 .t / would be the steady
state amplitude. To determine the steady state response of the system, we need to
numerically determine the parameters.
The primary suspension’s frequency response is
q
!2
Z1 D C12 C D12 D q w (3.54)
.!12  ! 2 /2 C .21 !!1 /2

and shows in Fig. 3.5 for w D 1. The area above the line Z D 1 indicates the
engagement of the secondary suspension. The curves will approach the line Z D
w asymptotically. If w < 1, then there would be a frequency span in which the
secondary will engage, and there would be no engagement outside the span. If w >
1, then the engagement will begin at a frequency and never let go. As a result, the
curves above Z D 1 will change because of the sudden change of stuffiness and
damping.

Fig. 3.5 The frequency

U
response for Z D

3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 69

Fig. 3.6 The frequency response curve of the primary for !1 D 1, ı D 1, 1 D 0:1

Without losing generality we may assume

!1 D 1 (3.55)

and set

w D 0:5 < 1 1 D 0:1 (3.56)

and plot the associated frequency response curve of the primary in Fig. 3.6. The
amplitude of the primary will be greater than Z1 D 1 when the excitation frequency
is in the following range.

0:8337115427 < ! < 1:385012057 (3.57)

To express the method and determine a sample of steady state time response, let us
set the secondary suspension parameters as
p
!2 D 2 2 D 0:1 (3.58)

Substituting the system characteristics we should solve the two equations of (3.46)
and (3.47) t1 and '. Figure 3.7 illustrates the implicit plot of the equations, with a
unique solution at

' D 1:472642330 t1 D 1:688156972 (3.59)

Employing t1 and ' we calculate

v1 D 1:154853554 v2 D 1:218400685

and determine z2 .t / and z1 .t / to plot the steady state solution as shown in Fig. 3.8.
70 O. Pogorilyi et al.

Fig. 3.7 The implicit

p plot of the two transcendental equations (3.46) and (3.47) as functions of t1
and ' for !2 D 2, 2 D 0:1, !1 D 1, w D 0:5, 1 D 0:1

p
Fig. 3.8 A sample of the steady response of the system for !2 D 2, 2 D 0:1, !1 D 1, w D 0:5,
1 D 0:1

3.4 Challenges and Unsolved Problems

Regardless of the simple configuration and arrangement of the piecewise linear

vibration isolator, determination of the frequency and steady state time responses
of the system is not completely covered to date. There are several challenges and
unsolved problems that we review in this section.

3.4.1 Multiple Solutions

The eight conditions (3.8) are not enough to guarantee a single solution. Depending
on the parameters, we may get multiple possible steady state responses. This
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 71

multiplicity is different than the existence of multiple solutions for some values of
frequency in the frequency response of nonlinear vibrating systems. As an example,
let us consider a system with the following characteristics
p
!1 D 1 1 D 0:1 !2 D 17 2 D 0:1 (3.60)

and the excitation with

!D1 wD3 (3.61)

Figure 3.9 illustrates the intersection of Eqs. (3.46) and (3.47) for the possible
solutions of t1 and '.
The intersections provide us with two sets of solutions:

' D 1:248353815 t1 D 0:9049692487 (3.62)

' D 1:314128509 t1 D 2:229699384 (3.63)

The steady response of the system for solution (3.62) is illustrated in Fig. 3.10, and
for solution (3.63) illustrated in Fig. 3.11.
Considering
p
!1 D 1 1 D 0:1 !2 D 2 2 D 0:1 (3.64)

we find the illustration of Eqs. (3.46) and (3.47) as shown in Fig. 3.12, with a unique
solution for the t1 and '.

' D 0:6226982402 t1 D 2:620975511

p The implicit plot of the Eqs. (3.46) and (3.47) for ! D 1, w D 3 and !1 D 1, 1 D 0:1,
Fig. 3.9
!2 D 17; 2 D 0:1
72 O. Pogorilyi et al.

Fig. 3.10 The steady response of the system for ' D 1:248353815, t1 D :9049692487

Fig. 3.11 The steady response of the system for ' D 1:314128509, t1 D 2:229699384

The associated steady state response of the system is shown in Fig. 3.13.
Comparison of Figs. 3.13 and 3.10, 3.11 suggests some open problems to
be solved in the future.
p It indicates p that there would be a periodic bifurcation
point in the interval 2 < !2 < 17 for !1 D 1, 1 D 0:1, 2 D 0:1,
w D 3, ! D 1. Any other combination may have the same feature as well.
Double hitting or multiple hitting may also happen without violating the imposed
boundary conditions. Searching only for the responses with single hitting needs
more conditions to be introduced. In case of multiple hitting, such as the one in
Fig. 3.11, it must be discussed to clarify if the first or the highest amplitude should
be recorded.
Furthermore, when plot of Eqs. (3.46) and (3.47) show two intersections, one of
them must be associated with a stable and the other one associated with an unstable
solution. Based on the nonlinear vibration understanding, the solution with higher
amplitude must be the stable solution, and the lower amplitude be associated with
unstable solution. In case of three possible amplitudes, the middle one is unstable.
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 73

p The implicit plot of the Eqs. (3.46) and (3.47) for ! D 1, w D 3 and !1 D 1, 1 D 0:1,
Fig. 3.12
!2 D 2, 2 D 0:1

p
Fig. 3.13 The steady response of the system for ! D 1, w D 3 and !1 D 1, 1 D 0:1, !2 D 2,
2 D 0:1

However, the stability of the solutions cannot be determined by analyzing one period
of the steady state solutions. It must be done by a stability analysis.

3.4.2 Range of Nonlinearity

Although the system is piecewise linear, it shows the behavior of nonlinear systems
because of the sudden change in stiffness and damping. One of the common features
of nonlinear vibrating systems is the jump phenomenon. The jump is in general a
harmful phenomenon in real systems. It is the reason why jump avoidance design
must be taken (Jazar et al. 2006a). In the piecewise linear vibrating system, the jump
may appear when the secondary suspension is still engaged while the excitation
74 O. Pogorilyi et al.

frequency is beyond the upper limit of the frequency interval of the line Z1 D
1 intersecting the frequency response of the primary. As an example, consider a
system with
p
!1 D 1 1 D 0:1 !2 D 17 2 D 0:1 (3.65)

and the excitation with

! D 1:4 wD3 (3.66)
The excitation frequency is higher than upper limit of (3.57). Therefore, the system
should not ever engage with the secondary suspension. We expect that if such
an engagement happens, the system let the secondary go and settle down on a
steady state vibration supported only by the primary suspension. However, Fig. 3.14
indicates that there is a possible solution for system to vibrate on a steady state
condition while the secondary suspension engages. The response is depicted in
Fig. 3.15.
Theoretically, variation of ! with some small step and solution of Eqs. (3.46)–
(3.48) allow us to calculate the steady state amplitude of the system. In the case
of nonlinear systems the highest value of the frequency for which a steady state
response happens cannot be predicted exactly. The current practical way is to
increase the excitation frequency ! little by little until no solution can be achieved
from Eqs. (3.46) and (3.47). Additionally, we should monitor the tangent slope of

dZ2 Z2n  Z2n1

' (3.67)
d! !n  !n1

The point where the slope of the tangent to the response curve is almost vertical the
jump will occur. However, when the slope gets close to vertical, the sensitivity of
the solution to the variation of ! becomes higher. Therefore, the step size of ! must
become finer. One set of solution of Eqs. (3.46) and (3.47) for different values of

Fig. 3.14 The implicit

p plot of the Eqs. (3.46) and (3.47) for ! D 1:4, w D 3 and !1 D 1,
1 D 0:1, !2 D 17, 2 D 0:1
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 75

p The steady response of the system for ! D 1, w D 3 and !1 D 1:4, 1 D 0:1,

Fig. 3.15
!2 D 2, 2 D 0:1

p
Fig. 3.16 One set of solution of Eqs. (3.46) and (3.47) for w D 3 !1 D 1, 1 D 0:1, !2 D 17,
2 D 0:1 and varying !

! is shown in Fig. 3.16. No solution for ! < 0:8 or ! > 4:8 could be found. At
! D 4:8 the slope of the frequency response curve is close to vertical.

dZ2 6:21014567  7:928497661

' D 17:184 (3.68)
d! 0:1
arctan .17:184/ ' 1:512 7 rad ' 86:673 deg (3.69)

3.4.3 Asymmetric Systems

The symmetric piecewise linear vibration isolator is a limited portion of the

piecewise linear isolators. The system may easily become asymmetric in practice.
Any change of mass will make the positive and negative gap sizes unequal. The
76 O. Pogorilyi et al.

damping
x
c + cn
Δp c + cp
m c
Δn u
Δn Δp

kp k kn stiffness
cp c cn k + kn
y
k + kp
k
u
Δn Δp

Fig. 3.17 The mathematical model of an asymmetric piecewise linear vibration isolator

spring may behave different in compression and extension. Similarly, real dampers
resist differently in bound and rebound.
Figure 3.17 illustrates the mathematical model of an asymmetric piecewise linear
vibration isolator. Determination of the frequency and steady state time responses
of the system is more complicated, as the number of boundary conditions and
transcendental equations increases.
The equations governing the motion of the system could be written as

mxR C g1 .x; x/
P D f1 .y; y/
P (3.70)

P and f1 .y; y/
where g1 .x; x/ P are piecewise linear functions presenting nonlinear
characteristics of the system and nonlinear excitation, respectively:
8
ˆ
<.c C cp /xP C .k C kp /x  kp p
ˆ x  y > p
P D c xP C kx
g1 .x; x/ n < jx  yj < p (3.71)
ˆ
:̂.c C c /xP C .k C k /x C k  x  y < n
n n n n

8
ˆ
<.c C cp /yP C .k C kp /y
ˆ x  y > p
P D c yP C ky
f1 .y; y/ n < jx  yj < p (3.72)
ˆ
:̂.c C c /yP C .k C k /y x  y < n
n n

In an extreme case of the asymmetric system, the secondary suspension might

be engaged only in one direction. Figure 3.18 illustrates the mathematical model of
such system.
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 77

damping
x

Δp c + cp
m c
u
Δp

kp k stiffness
cp c y
k + kp
k
u
Δp

Fig. 3.18 The mathematical model of a one-sided piecewise linear vibration isolator

3.5 Conclusion

In this chapter we have shown that obtaining exact frequency response of piecewise
linear vibration isolator associates with a few numerical problems. It has been
shown in the past that the exact solution of the system depends on a set of
transcendental equations. For the first time, in this study the number of independent
transcendental equations has been reduced to two. This reduction simplifies many
of the numerical problems. However, not all of the problems can be resolved at
the moment. The main remained challenges were discussed in this chapter. Among
them are: difficulties in determining range of nonlinearity, multiple solutions and
transition from one amplitude/frequency curve to the other (from stable solution to
unstable and vice versa). It is still unclear how to designate the nonlinearity range,
to calculate the maximum value of frequency for double engagement. The other
challenges are solved and discussed in detail in this paper.

3.6 Key Symbols

Ai sin coefficient in zi .t /
Bi cos coefficient in zi .t /
c Damping coefficient [N s = m]
c1 Damping coefficient of the primary suspension [N s = m]
c2 Damping of the system when secondary engages [N s = m]
Ci Damping sine coefficient in zi .t /
Di Damping cosine coefficient in zi .t /
k Stiffness [N = m]
k1 Stiffness of the primary suspension [N = m]
k2 Stiffness of the system when secondary engages [N = m]
78 O. Pogorilyi et al.

m Mass [kg]
Qi Short notation of mathematical expression
t Time [s]
t1 Time when z2 D z1 D C1 [s]
u Relative displacement [m]
w Dimensionless base excitation amplitude
x Absolute displacement [m]
y Base excitation function [m]
Y Base excitation amplitude [m]
z Dimensionless relative displacement
Z Amplitude of relative displacement
Z1 Amplitude of relative displacement of primary suspension
Z2 Amplitude of relative displacement of secondary suspension
ı Gap distance
v1 Velocity in t1
2 Velocity in z2 .t / D 0
 Damping ratio
1 Damping ratio of the primary suspension
2 Damping ratio of the system when secondary engages
 Phase lag
! Excitation frequency [1= s]
!1 Natural frequency of the primary suspension
!2 Natural frequency of the system when secondary engages
!di Damped natural frequency

References

Aizerman M, Gantmakher F (1957) Determination of periodic modes in systems with

piecewise-linear characteristics, composed of segments parallel to two specified straight lines.
Autom Remote Control 18:105–120
Aizerman M, Gantmakher F (1958) Determination of stability by linear approximation of a
periodic solution of a system of differential equations with discontinuous right-hand sides. Q J
Mech Appl Math 11(4):385–398
Aizerman M, Lur’e A (1963) Methods for construction of periodic motions in piecewise-linear
systems. In Symposium on Non-Linear Vibrations, vol. 1, pp. 27–50. International Union for
Theoretical and Applied Mechanics, Kiev
Alkhatib R, Jazar, RN, Golnaraghi MF (2004) Optimal design of passive linear mounts with genetic
algorithm method. J Sound Vib 275(3–5):665–691
Christopherson J, Jazar, RN (2005) Optimization of classical hydraulic engine mounts based on
RMS method. J Shock Vib 12(2):119–147
Den Hartog J, Mikina S (1932) Forced vibrations with nonlinear spring constraints. Trans ASME
Appl Mech 54-APM-15:157
Deshpande S, Mehta S, Jazar RN (2006) Optimization of secondary suspension of piecewise linear
vibration isolation systems. Int J Mech Sci 48(4):341–377
Fleishman B (1965) Forced oscillations and convex superposition in piecewise-linear systems.
SIAM Rev 7(2):205
3 Challenges in Exact Response of Piecewise Linear Vibration Isolator 79

Gurtin M (1961) On the use of clearance in viscous dampers to limit high frequency force
transmission. J Eng Ind 83(1):50–52
Hayashi C (1964) Nonlinear oscillations in physical systems. McGraw-Hill, New York,
McGraw-Hill
Iwan WD (1973) A generalization of the concept of equivalent linearization. Int J Nonlinear Mech
8(3):279–287
Jacobsen L, Jaspersen H (1935) Steady forced vibration of a single mass system with a nonlinear
restoring force. J Franklin Inst 220:467–496
Jazar RN (2013) Advanced vibrations: a modern approach. Springer, New York
Jazar RN (2014) Vehicle dynamics: theory and application, 2nd edn. Springer, New York
Jazar RN, Alkhatib R, Golnaraghi MF (2006a) Root mean square optimization criterion for
vibration behavior of linear quarter car using analytical methods. J Vehicle Syst Dyn 44(6):
477–512
Jazar RN, Golnaraghi MF (2002) Engine mounts for automotive applications: a survey. Shock Vib
Digest 34(5):363–379
Jazar RN, Houim R, Narimani A, Golnaraghi F (2006b) Nonlinear passive engine mount,
frequency response and jump avoidance. J Vib Control 12(11):1205–1237
Jazar RN, Mahinfalah M, Deshpande S (2007) Design of a piecewise linear vibration isolator for
jump avoidance. IMechE Part K J Multi Body Dyn 221(K3):441–450
Jazar RN, Narimani A, Golnaraghi MF, Swanson DA (2003) Practical frequency and time optimal
design of passive linear vibration isolation mounts. J Vehicle Syst Dyn 39(6):437–466
Marzbani H, Jazar RN, Khazaei A (2012) Smart passive vibration isolation: requirements and
unsolved problems. J Appl Nonlinear Dyn 1(4):341–386
Marzbani H, Jazar RN, Fard M (2013) Hydraulic engine mounts: a survey. J Vib Control
19(16):253–279
Masri S (1965) Analytical and experimental studies of impact dampers. California Institute of
Technology
Masri S (1978) Analytical and experimental studies of a dynamic system with a gap. J Mech Des
100(3):480–486
Masri S, Stott S (1978) Random excitation of a nonlinear vibration neutralizer. J Mech Des
100(4):681–689
Narimani A, Golnaraghi MF, Jazar RN (2004a) Frequency response of a piecewise linear system.
J Vib Control 10(12):1775–1894
Narimani A, Jazar RN, Golnaraghi MF (2004b) Sensitivity analysis of frequency response of a
piecewise linear system in frequency island. J Vib Control 10(2):175–198
Natsiavas S (1989) Periodic response and stability of oscillators with symmetric trilinear restoring
force. J Sound Vib 134(2):315–331
Natsiavas S, Gonzalez H (1992) Vibration of harmonically excited oscillators with asymmetric
constraints. J Appl Mech Trans ASME 59(2):S284–S290
Nguyen D, Noah S, Kettleborough C (1986a) Impact behaviour of an oscillator with limiting stops,
part I: a parametric study. J Sound Vib 109(2):293–307
Nguyen D, Noah S, Kettleborough C (1986b) Impact behaviour of an oscillator with limiting stops,
part II: Dimensionless design parameters. J Sound Vib 109(2):309–325
Rosenberg R (1996) Steady-state forced vibrations. Int J Nonlinear Mech 1(2):95–108
Schulman J (1983) Chaos in piecewise-linear systems. Phys Rev A 28(1):477
Watabene T (1978) Forced vibration of a continues system with nonlinear boundary conditions.
Trans ASME J Mech Des 100:487–491
Chapter 4
Active Vibration Control for Nonlinear Axially
Translating Cable Systems of Multi-Dimensions

L. Dai and L. Sun

Keywords Active vibration control • Nonlinear behaviors • Chaos • Axially

translating cable • Dynamic systems of multi-dimensions

4.1 Introduction

Axially translating systems, such as cables, strings, belts, blades, and beams, have
been wildly studied for their important applications in the fields of civil, mechanical,
architecture, and industrial engineering. The research on the dynamical responses
and vibration controls of the systems, usually highly flexible and nonlinear, has
attracted the attentions of the researchers and engineers in the fields and significant
contributions have been made toward theoretical development and engineering
applications. Regardless the research contributions in the fields, systematical and
thorough investigations on the behavior of the axially translating systems and the
control of the nonlinear behavior of the systems are still in demanding. It is the
effective and reliable control of such nonlinear systems that has great significances
in engineering applications.
Among the research works on the axially translating structures, a significant
majority of the research is on axially translating beams, plates, and blades to which
the modeling and analysis approaches of axially translating cables are similar. An
axially traveling plate, with a constant length along the traveling direction, was
studied (Luo and Hamidzadeh 2004) in designing the aerospace and aeronautical
structures. In the study, the analytical solutions of high-speed traveling plates, as
well as the buckling stability, were derived. In analyzing the axially translating

L. Dai () • L. Sun

Industrial Systems Engineering, University of Regina, Regina, SK, S4S 0A2, Canada
e-mail: [email protected]

© Springer International Publishing Switzerland 2015 81

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__4
82 L. Dai and L. Sun

beams, the shear deformation of the beam plays an important rule (Carrera and
Giunta 2011). An axially moving Timoshenko beam was introduced in (Ghayesh
and Amabili 2013) with the considerations of the shear deformations. In this study,
a 20-dimension nonlinear dynamic system of Timoshenko beams with invariant
length was modeled to study the bifurcations as well as chaos of the system.
The class of the axially moving structures with varying length has also been
investigated wildly for its various applications in the areas of engineering. Coming
from the dynamics of spacecraft antenna, of which the length is varying with time,
the equation of motion of a cantilevered beam was established (Tabarrok et al. 1974).
In the study, a 16-dimension system of the beam has been derived, and a decreasing
frequency has been discovered with respect to the increasing length of the beam.
In (Fung et al. 1998), the four nonlinear axially moving cantilevered beam models,
including Timoshenko, Euler, simple-flexible, and rigid-body beam models were
established in considering a tip mass. In analyzing the hydraulic and motor driven
systems, the authors pointed out that the rigid-body motion and the flexible vibration
of the beam could be nonlinearly coupled and there existed Coriolis forces in the
systems.
When cables and highly flexible structures are considered, the dynamic system
modeling is an important part for studying the systems’ dynamic behavior and
stabilities which are critical to the control of the systems. With the interests in
self-spinning tethered satellites, an Euler–Bernoulli beam was adopted to represent
a tether with varying length (Tang et al. 2011). The deployment process of two-
self-spinning tethered satellite systems was successfully simulated through a newly
proposed hybrid Eulerian and Lagrangian frame work. A beam model together with
a cable model was established by Zhu and Ni (2000) in order to investigate the
dynamics of the flexible robot arms and elevator hoist cables considered in their
study. The energetics and stability of translating media of the systems considered
were investigated by the authors. The rates of energy changes of the established
models were studies in both the extension and retraction processes of the systems
considered. It is interesting to notice that a linear five-dimensional dynamic system
was derived to ensure the reliability of the conducted study, as shown in their
numerical simulation of a specified elevator hoist cable.
In the dynamic analysis of some mechanical axially traveling systems, such as
saw blades and magnetics, the nonlinear vibration of a traveling tensioned beam
was modeled and studied by Wickert (1992). The reported results showed that
the inclusion of the nonlinearity in the proposed system was most important at
near-critical speeds, where the modal stiffness was small and dominated by a
nonlinear extensional stiffness. Based on the results regarding the supercritical
stability characteristics, it was pointed out by the authors that the investigation
needed to be improved in the case of higher-order equilibria of the translating beam.
With the models established for the axially traveling systems, the control of the
dynamic motion of the systems can be analyzed, simulated, and applied. In the
study by Zhu (2002), the equations of motion of translating media were derived
via the Eulerian frame of reference, and then the rate of energy change of translating
media was distinguished with control volume. Therefore, translating media could be
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 83

divided into two classes: the translating media with invariant control volume, and
those with variable control volume. The rate of energy change was emphasized in
characterizing the dynamic stability of translating media and stationary media with
moving boundaries.
The control volume of an axially translating beam, which is located in between
two spatially fixed supports, was discussed in the study of Humer (2013). As
demonstrated in the study, this type of problems, in which kinematic constraints and
loads were inherently non-material, could be found in many industrial processes
such as belt drives for power transmission. In this study, the free vibration of an
extended beam was investigated, and the length of the protruding part showed
significant influence on the behavior of the structure.
An energy-based control scheme was proposed by Yang et al. (2005). In this
study, the applications of an axially translating beam were restricted in high-speed
and precision systems, due to the unwanted vibrations of moving continua. In
the application of the control scheme, as shown in the study, the vibration of the
translating beam could be suppressed.
For theoretically and numerically analyzing the vibrations of the axially trans-
lating systems and simulating the control of the vibrations, almost all the studies
available in the archived literature are based on discretizing the governing equations
of the systems modeled via the Galerkin method. On the basis of the discretized
system, some of the control strategies were developed. A typical such approach
can be seen in Yang et al. (2005), in which large-amplitude of the vibrations of
axially translating beams was considered and an energy-based control strategy was
developed. It should be noticed that, as indicated in the works (Zhu and Ni 2000;
Wickert 1992), a multi-dimensional dynamic system will be favored in order to
derive reliable results.
In engineering applications, corresponding to the nonlinear vibrations discovered
in cables, strings, beams, and plates, several control schemes have been developed
in the last several decades. Utkin (1992) proposed a control scheme in 1992,
named sliding mode control (SMC) strategy, the SMC strategy has been applied
in controlling the vibrations of nonlinear engineering structures, and this control
strategy has gradually shown its applicability and efficiency in the works following
Utkin’s work.
Corresponding to the nonlinear dynamic systems with uncertain external distur-
bance, the fuzzy sliding mode control (FSMC) strategy was developed, in which the
fuzzy logic rule was introduced into the original SMC strategy. The FSMC strategy
showed its effectiveness in controlling nonlinear and chaotic motions discovered
in the nonlinear dynamics systems (Yau and Kuo 2006; Kuo and Shieh 2007; Yau
et al. 2010). In the study by Haghighi and Markazi (2009), for example, both the
geometric nonlinearity and the nonlinear electrostatic force are considered, and a
chaotic motion was found and controlled by applying the FSMC control strategy.
The authors showed the elimination of the chaotic motion and stabilization of the
nonlinear system with the control strategy.
In the studies (Yau and Kuo 2006; Kuo and Shieh 2007; Yau et al. 2010;
Haghighi and Markazi 2009), although the effectiveness of the FSMC strategy
84 L. Dai and L. Sun

has been demonstrated in controlling the nonlinear vibrations, it can be found

that the application of FSMC strategy is restricted on one-dimensional nonlin-
ear dynamics systems. Corresponding to the vibrations of the multi-dimensional
dynamic system, there is currently no control strategy available in the literature.
Therefore, for controlling the nonlinear vibrations of multi-dimensional systems, the
FSMC control strategy currently used needs to be modified. The multi-dimensional
systems, developed through the Galerkin method, can be used to provide more
accurate and reliable results through numerical simulations in comparing with that
of single-dimensional system. The higher the order of the multi-dimensional system,
theoretically, the more accurate will be the simulations of the nonlinear axially
translating systems. This has been recognized by the researchers in this field (Zhu
and Ni 2000; Wickert 1992).
With the considerations described above, this chapter intends to introduce a
newly developed approach for quantitatively controlling the vibrations of nonlinear
axially translating cable systems. This includes a development of an active control
strategy and the application of the control strategy in controlling the nonlinear
cable systems. For developing such a control strategy, a cable system model
consisting of the equations of motion is to be established based on von Karman-
type equations. In developing for the solutions of the cable system and for the sake
of applying the strategy in the nonlinear vibration control, the governing equations
in the forms of partial differential equations will be non-dimensionalized and then
transformed into three ordinary differential equations via a third-order Galerkin
method. Corresponding to the derived multi-dimensional dynamic system, an active
control strategy is developed based on the FSMC strategy. The applicability of the
control strategy developed will be demonstrated in some numerical simulations
based on the model established. A few cases of chaotic vibrations of the cable
system are considered for applying the control strategy. The suppression and
stabilization of the chaotic vibrations are to be demonstrated graphically to show the
application and efficiency of the active control strategy in controlling the nonlinear
axially translating cable systems.

4.2 Equations of Motion

The axially translating cable considered in this research is sketched in Fig. 4.1. The
equations of motion of the cable are to be derived based on Hamilton’s principle
and von Karman-type equations. As can be seen from Fig. 4.1, the translating cable
with pinned–pinned boundaries is allowed to move axially at a constant rate v0 , and
the length of the cable is given as l. The displacement of any point of the cable along
the x- and z- axes is designated as u and w.
Starting from the origin at the left support of the cable, a position vector, r, of
any point x(t) of the translating beam without deformation is given as
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 85

Fig. 4.1 The model of the z

axially translating cable
v0 → x
o

l0

r D x.t /i C 0k; (4.1)

where i and k are the unit vectors of the fixed Cartesian coordinate shown in the
figure.
Thus, the displacement field of the axially translating cable can be derived as

R D .x.t / C u0 .x.t /; t // i C w0 .x.t /; t / k; (4.2)

where u0 (x(t), t) and w0 (x(t), t) are the displacement components along the x- and z-
directions respectively, of a point of the cable.
Taking full differentiation of R with respect to time t, one may obtain
 
dR dx.t / d u0 .x.t /; t / d w0 .x.t /; t /
D C iC k; (4.3)
dt dt dt dt

where the derivative of x(t) with respect to time is equal to the translating rate of the
cable, and the full derivative of w0 is

d w0 .x.t /; t / @w0 .x.t /; t / @w0 .x.t /; t /

D C v0 ; (4.4-a)
dt @t @x.t /

d 2 w0 .x.t /; t / @2 w0 .x.t /; t / @w0 .x.t /; t / 2

2 @ w0 .x.t /; t /
D C 2v0 C v0 : (4.4-b)
dt 2 @t 2 @x.t / @x 2 .t /

Hence, the kinetic energy of the translating cable over a volume V of the cable is
expressible as
Z l  
1 dR dR
T D dx; (4.5)
0 2 dt dt

where denotes the mass per unit length.

The von Karman-type equations of strains of large deflection associated with the
displacement field, normal to the cross section of the cable along the x direction, in
Eq. (4.2) can thus be given by
 
@u0 .x.t /; t / 1 @w0 .x.t /; t / 2
"11 D C : (4.6)
@x.t / 2 @x.t /
86 L. Dai and L. Sun

Therefore, the total strain energy of the cable can be given by

Z l
1
U D Q11 ."11 "11 / dx; (4.7)
0 2

where Q11 represents the elastic coefficient in the same direction with "11 .
The virtual work done by the force is 0,

W D 0; (4.8)

In the following analysis, the Hamilton’s principle will be employed to obtain the
nonlinear equations of motion for the axially translating cable. The mathematical
statement of the Hamilton’s principle is given by
Z t2
.ıL C ıW / dt D 0; (4.9)
t1

where the total Lagrangian function L is given by

L D T  E: (4.10)

For the sake of clarity, hereafter, use x, u0 , w0 to replace x(t), u0 (x(t), t), and
w0 (x(t), t) respectively. Substitute Eqs. (4.5)–(4.8) into Eq. (4.10), and the first term
in Eq. (4.9) can be developed as below,
Z t2 Z t2
ıLdt D ı .T  U / dt
t1 t1
Z (Z    )
t2 l
dR dR
D ı  Q11 ."11 "11 / dt
t1 0 dt dt
Z t2 Z l    Z t2 Z l
d R d ıR
D dt  ŒQ11 ."11 ı"11 / dt
t1 0 dt dt t1 0
Z Z t2  2  Z t2 Z
d R
D0 ıRdt d V  Q11 "11 ı"11 d Vdt
V t1 dt 2 t1 V
Z t2 Z l  2 
d x d 2 u0
D ı .x C u0 / C dxdt
t1 0 dt 2 dt 2
Z t2 Z l
d 2 w0
 ıw0 dxdt
t1 0 dt 2
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 87

Z Z  !
t2 l
@u0 1 @w0 2
 Q11 ı C "11 dxdt
t1 0 @x 2 @x

D L1 C L 2 C L 3

where
Z t2 Z l  
d 2x d 2 u0
L1 D  ı .x C u0 / C dxdt ; (4.11-a)
t1 0 dt 2 dt 2
Z t2 Z l
d 2 w0
L2 D  ıw0 dxdt ; (4.11-b)
t1 0 dt 2
Z Z   !
t2 l
@u0 1 @w0 2
L3 D  Q11 ı C "11 dxdt : (4.11-c)
t1 0 @x 2 @x

2
Since the cable is moving axially at a constant velocity ( ddtv20 D 0), from
Eq. (4.11-a), it can be derived
Z t2 Z l  
d 2x d 2 u0
L1 D  ı .x C u0 / C dxdt
t1 0 dt 2 dt 2
Z t2 Z l  
d 2 u0
D .0 C ıu0 / 0 C dxdt
t1 0 dt 2
Z t2 Z l
d 2 u0
D ıu0 dxdt (4.12-a)
t1 0 dt 2

From Eq. (4.11-c), it can be derived in as follows

Z Z   !
t2 l
@u0 1 @w0 2
L3 D  Q11 ı C "11 dxdt
t1 0 @x 2 @x
Z t2 Z l   !
@ıu0 1 @w0 2
D Q11 C ı "11 dxdt
t1 0 @x 2 @x
Z t2 Z l Z t2 Z l
@ıu0 @w0 @ıw0
D Q11 "11 dxdt  Q11 "11 dxdt
t1 0 @x t1 0 @x @x
Z t2 " Z l #
D .Q11 "11 ıu0 /jl0  Q11 ıu0 d "11 dt
t1 0
88 L. Dai and L. Sun

Z " ˇl Z l #
t2
@w0 ˇ @w
ıw0 ˇˇ 
0
 Q11 "11 Q11 "11 ıw0 d dt
t1 @x 0 0 @x
Z Z !
t2 l
D 0 Q11 ıu0 d "11 dt
t1 0

Z t2 Z l  !
@w0
 0 Q11 ıw0 d "11 dt
t1 0 @x
Z t2 Z l ! Z t2 Z l !
@"11 @2 w0
D  Q11 ıu0 d x dt   Q11 "11 ıw0 dx dt
t1 0 @x t1 0 @x 2
Z t2 Z l !
@"11 @w0
  Q11 ıw0 dx dt
t1 0 @x @x
Z t2 Z l   ! !
@ @u0 1 @w0 2
D  Q11 C ıu0 d x dt
t1 0 @x @x 2 @x
Z t2 Z l   ! !
@u0 1 @w0 2 @2 w0
  Q11 C ıw0 dx dt
t1 0 @x 2 @x @x 2
Z t2 Z l   ! !
@ @u0 1 @w0 2 @w0
  Q11 C ıw0 dx dt
t1 0 @x @x 2 @x @x
Z t2 Z l  2  !
@ u0 @w0 @2 w0
D Q11 C ıu0 d x dt
t1 0 @x 2 @x @x 2
Z t2 Z l   ! !
@u0 @2 w0 1 @w0 2 @2 w0
C Q11 C ıw0 dx dt
t1 0 @x @x 2 2 @x @x 2
Z t2 Z l   ! !
@2 u0 @w0 @2 w0 @w0 2
C Q11 C ıw0 dx dt (4.12-b)
t1 0 @x 2 @x @x 2 @x

From Eqs. (4.12-a), (4.12-b), and (4.11-b), the nonlinear governing equation of
an axially translating cable can be derived in the following,
 
d 2 u0 @2 u0 @w0 @2 w0
 C Q11 C D 0; (4.13-a)
dt 2 @x 2 @x @x 2

 
d 2 w0 @2 u0 @w0 @u0 @2 w0 3 @w0 2 @2 w0
 C Q11 2 C Q11 C Q11 D 0; (4.13-b)
dt 2 @x @x @x @x 2 2 @x @x 2
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 89

Associated with the nonlinear dynamic equations and the boundary conditions
(Nayfeh and Mook 1979; Abou-Rayan et al. 1993), the strain can be obtained as
  Z  
@u0 1 @w0 2 1 l @w0 2
D C dx: (4.14)
@x 2 @x 2l 0 @x

Then, substitute Eq. (4.14) into Eq. (4.13-b), and the nonlinear differential
governing equation of the axially translating cable in z direction is derived as
Z  
d 2 w0 l
1 @w0 2 @2 w0
 C Q11 dx D 0: (4.15)
dt 2 0 2l @x @x 2

4.3 Non-dimensionalization

To validate the governing equation Eq. (4.15) and facilitate the numerical sim-
ulations in the consequent sections, the following non-dimensional variables are
introduced,
s
Q11 x
tD t D  t; x D ; (4.16)
l

and

w0 d w0 1 d w0 d 2 w0 1 d 2 w0
w0 D ; D ; 2
D : (4.17)
l dt  l dt dt  2 l dt 2

With the non-dimensional variables shown in Eqs. (4.16) and (4.17) introduced
into Eq. (4.14), the non-dimensional governing equation of the axially translating
cable can be expressed as
Z 1  2
d 2 w0 @w0 @2 w0
 2
C dx D 0; (4.18)
dt 0 @x @x 2

where
11 1
D Q11 :
2l l 2
90 L. Dai and L. Sun

4.4 Series Solutions

Based on the Galerkin method of discretization, the transverse displacement w0 is

expanded in a series form, in terms of a set of comparison functions as,
1
X 
w0 D 'n .x/ wn t : (4.18)
nD1

Corresponding to the pinned–pinned boundaries of the axially translating string,

'n .x/ can be given as follows,

'n .x/ D sin .nx/ : (4.19)

Substitute the series solution of Eq. (4.19) into Eq. (4.17), and to assist
 2
presentation, replace ®n , wn , wP n , wR n , v, and t for 'n .x/, wn t , ddtwn , ddtw2n , v0, and t
respectively. With the application of the Galerkin method at n D 3, and

w1 D w1;1 ; w2 D w2;1 ; w3 D w3;1 ;

wP 1;1 D w1;2 wP 2;1 D w2;2 ; wP 3;1 D w3;2 :

The discretized governing equations of the axially translating cable with the
pinned–pinned boundaries can be obtained as the following.
8
ˆ
ˆ wP 1;1 D w1;2
ˆ
ˆ
ˆ
ˆ 1;2 2 .1 C 1 /
ˆ
<
wP D
wP 2;1 D w2;2
(4.20)
ˆ
ˆ wP D 2 .2 C 2 /
ˆ
ˆ
2;2
ˆ
ˆ wP 3;1 D w3;2
:̂
wP 3;2 D 2 .3 C 3 /

where,

8 1
1 D vw2;1 C v2  2 w1;1 ;
3 2
24 8
2 D vw3;1  vw1;1 C 2v2  2 w2;1 ;
5 3
24 9
3 D  vw3;1 C v2  2 w3;1 ;
5 2
 
1 4 3 9 4
1 D   w1;1   w1;1 w3;1   w1;1 w2;1 ;
2 4 2
4 4
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 91

 
2 D  4 w2;1 w21;1  9 4 w2;1 w23;1  4 4 w32;1 ;
 
9 4 81 4 3
3 D   w3;1 w1;1   w3;1  9 w3;1 w2;1 ;
2 4 2
4 4

4.5 Control Strategy Development

With the developed governing equations, boundary conditions, and the solutions
of the governing equations, an active vibration control strategy can be developed.
Based on the previous works (Utkin 1992; Haghighi and Markazi 2009; Kuo and
Shieh 2007), the proposed active control strategy is developed for the vibration
control of a multi-dimensional nonlinear dynamic system shown in Eq. (4.20).
For a nonlinear governing equation in the following general form

wR D ˆ .w; w;
P t/ ; (4.21)

denote U as the control input and P .w; w/ P the unknown external disturbance
applying on the beam, the governing equation Eq. (4.21) for the beam with the
control input and the external disturbance can be given by

wR D ˆ .w; w;
P t / C U C P .w; w/
P : (4.22)

With application of the control, it is expected that the vibration of the beam can
be controlled
If the nth Galerkin method is applied in the discretization of the governing
equation given in Eq. (4.22), a series of second order ordinary differential equations
considering the control input U and the unknown external disturbance will be
derived as follows,
8
ˆ
ˆ wP 1;1 D w1;2
ˆ
ˆ
ˆ
ˆ wP 1;2 D  1 .W; t / C u1 C f1 .W; t /
ˆ
ˆ
ˆ
ˆ wP 2;1 D w2;2
ˆ
ˆ
ˆ
ˆ wP 2;2 D 2 .W; t / C u2 C f2 .W; t /
ˆ
ˆ
ˆ
< ::
:
; (4.23)
ˆ
ˆ P
w D wi;2
ˆ
ˆ
i;1
ˆ
ˆ wP i;2 D i .W; t / C ui C fi .W; t /
ˆ
ˆ
ˆ
ˆ ::
ˆ
ˆ :
ˆ
ˆ
ˆ
ˆ wP n;1 D wn;2
:̂
wP n;2 D n .W; t / C un C fi .W; t /
92 L. Dai and L. Sun

P t /, U, and
where  i (W, t), ui , and fi (W, t) represent the expressions of ˆ .w; w;
P .w; w/P after the application of the Galerkin discretization. With the Galerkin
discretization, the column vector W in Eq. (4.23) is given below,

W D Œw1;1 w1;2 w2;1 w2;2    wi;1 wi;2    wn;1 wn;2 T :

Considering the response of a point of the axially translating cable along the x
direction, based on Eq. (4.23) and the expression in Eq. (4.18), the non-dimensional
response of the selected point wp can be given as

1
X  
wp D 'n xp wn .t /; (4.24)
nD1

where xp denotes the location of the selected point.

For a desired vibration or reference vibration expressed as

wr D Ar sin .!r t / ; (4.25)

where, Ar and ! r are the amplitude and frequency of the reference signal.
The control input U can be given as,

U D Ueq  Ur ; (4.26)

where Ueq and Ur are expressed as below,

   
P C wp  ‰ ; Ur D kf s Uf s :
Ueq D  wP p  ‰ (4.27)

In Eq. (4.27), designates the control parameters governing the sliding surface,
P < kf s 2 RC , and the value of Ufs depends on the fuzzy
kfs is given as jF .w; w/j
rule shown in the table below.

The fuzzy rule of Ufs

Ufs Ueq
1 2/3 1/3 0 1/3 2/3 1
d Ueq
dt
1 1 1 1 1 2/3 1 0
2/3 1 1 1 2/3 1 0 1/3
1/3 1 1 2/3 1 0 1/3 2/3
0 1 2/3 1 0 1/3 2/3 1
1/3 2/3 1 0 1/3 2/3 1 1
2/3 1 0 1/3 2/3 1 1 1
1 0 1/3 2/3 1 1 1 1
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 93

With the control strategy demonstrated in Eqs. (4.22)–(4.27), the control of an

axially translating cable governed with the general governing equation Eq. (4.21)
can be realized.
Take the axially translating beam governed by Eq. (4.17) as an example. Use the
control strategy developed and apply the control input as shown in Eq. (4.27), the
governing equation with the control input for the cable can be given by the following
expression.
Z 1  2
@w0 @2 w0
wR 0 D  dx C U C F .w; w/
P ; (4.28)
0 @x @x 2

With the application of the third-order Galerkin discretization, Eq. (4.29) may
have the following form:
8
ˆ
ˆ wP 1;1 D w1;2
ˆ
ˆ
ˆ
ˆ wP D 2 . C 1 / C u1 C f1 .W; t /
ˆ
<
1;2 1
wP 2;1 D w2;2
; (4.29)
ˆ wP 2;2 D 2 .2 C 2 / C u2 C f2 .W; t /
ˆ
ˆ
ˆ
ˆ
ˆ wP 3;1 D w3;2
:̂
wP 3;2 D 2 .3 C 3 / C u3 C f3 .W; t /

where u1 , u2 , u3 , u4 , u5, and u6 are derived as follows through the third-order

Galerkin method,

2 2
u1 D U; u2 D 0  U; u3 D U:
 3
It should be noticed that due to the pinned–pinned boundaries of the axially
translating cable, the application of the Galerkin method based on Eq. (4.19) makes
the coefficients of u2 equal to zero. In the next section, it will be demonstrated in
the numerical simulation that the actual response of the axially translating cable at
a selected point can be well synchronized with a desired reference signal in the case
that the coefficients of u2 equals to zero.

4.6 Vibration Control

To demonstrate the applicability and effectiveness of the control strategy developed

in the previous section, numerical simulations are conducted for controlling an
axially translating cable governed with Eq. (4.17). The nonlinear responses of the
cable are emphasized in this research. With the numerical simulations performed,
a chaotic motion is found when the cable is translating at certain rates. The
proposed active control strategy is found not only effectively reduces the amplitude
of the chaotic motion, but also stabilizes the motion so that the response of
94 L. Dai and L. Sun

the translating cable is controlled to a desired periodic motion. To facilitate the

numerical simulation, the fourth-order P–T method by Dai (2008) is implemented.
The parameters used for the simulations for the responses of the axially
translating cable are given as follows,

Q11 D 2:9  104 N; l D 0:5m; D 1:00kg=m;

and the constant axially translating rate is given as below,

v0 D 2m=s:

The non-dimensionalized initial conditions, corresponding to the displacements

described by Eqs. (4.29) after the implementation of the third-order Galerkin
method, are taken as

w1;1 .0/ D 0:001; w1;2 .0/ D 0:005;

w2;1 .0/ D 0:0001; w2;2 .0/ D 0:0025;

w3;1 .0/ D 0:00005; w3;2 .0/ D 0:00125:

If the response of a point at 0.3125m along the x-axis of the cable is selected,
based on Eqs. (4.18) and (4.29) the non-dimensional response of the selected point
wp can be derived as,

X
6
wp D 'n wn;1 D 0:92388w1;1  0:70711w2;1 C 0:38268w3;1 : (4.30)
nD1

4.6.1 Chaotic Motion

The response of the cable translating at the speed v0 D 2m/s is shown in Fig. 4.2,
corresponding to the non-dimensional time from t D 0 to t D 15,000. During this
period of time, one may notice: in Fig. 4.2, it is a chaotic motion discovered; the
maximum amplitude of the vibration of the cable can exceed 0.15. In considering
that the displacement shown in the figure is non-dimensional, the amplitude is very
large. Thus, reduction and stabilization of the chaotic motion may improve the
operation of the beam.
Besides, from Fig. 4.3a–c, it can be learned that although the contribution of
the first vibration mode in Fig. 4.3a is larger than those of the other two vibration
modes as shown in Fig. 4.3b, c, the contributions of the other two vibration modes
are obviously not negligible. Actually it can be learned that the other two vibration
modes also significantly contributes to the actual response of the selected point.
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 95

Fig. 4.2 The wave diagram of wp without the application of the control strategy

Thus, the development of a multi-dimensional dynamic system is necessary for the

accurate prediction of the dynamics of the extending nonlinear elastic cable.

4.6.2 Amplitude Synchronization

The control strategy will be applied to show its effectiveness and efficiency in the
amplitude synchronization. Three different desired amplitudes will be specified as
the amplitude of the reference signal.

Ar D 0.045

In the application of the control strategy, the desired amplitude of the reference is
set as

wr D 0:045 sin.0:0369t /:

and other the control parameters and the unknown external disturbance take the
following values:
 
D 3:5; kf s D 0:01; F .w; w/
P D 0:001 sin wp :

As shown in Fig. 4.4, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.4 and 4.5.
In Fig. 4.4 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 15,000. It can be seen from Fig. 4.4 that a short period of
96 L. Dai and L. Sun

Fig. 4.3 (a) The contribution of the first vibration mode to wp without the application of the
control strategy. (b) The contribution of the second vibration mode to wp without the application
of the control strategy. (c) The wave diagram of w3 without the application of the control strategy
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 97

Fig. 4.4 The wave diagram of wp with the application of the control strategy

time is needed for stabilizing the beam after the application of the control strategy.
After the short period, the chaotic motion will then become a periodic one, of which
the amplitude is 0.045.
From Fig. 4.5a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
Figure 4.6 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
with the application of the control strategy. As shown in the figure, the maximum
amplitude of the cable is very close to that of the reference signal.
Figure 4.7 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.

Ar D 0.015

In the application of the control strategy, the desired amplitude of the reference is
set as

wr D 0:015 sin.0:0369t /:

and other the control parameters and the unknown external disturbance take the
following values:
 
D 2; kf s D 0:01; F .w; w/
P D 0:001 sin wp :
98 L. Dai and L. Sun

Fig. 4.5 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy

As shown in Fig. 4.8, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.8 and 4.9.
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 99

Fig. 4.6 The wave diagram of wp (the blue continuous line) and the reference signal wr (the green
dash line)

Fig. 4.7 The control input

Fig. 4.8 The wave diagram of wp with the application of the control strategy
100 L. Dai and L. Sun

Fig. 4.9 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy

In Fig. 4.8 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 15,000. It can be seen from Fig. 4.8, a short period of
time is needed for stabilizing the beam after the application of the control strategy.
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 101

Fig. 4.10 The wave diagram of wp (the blue continuous line) and the reference signal wr (the
green dash line)

After the short period, the chaotic motion will then become a periodic one, of which
the amplitude is 0.015.
From Fig. 4.9a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
Figure 4.10 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
with the application of the control strategy. As shown in the figure, the maximum
amplitude of the cable is very close to that of the reference signal.
Figure 4.11 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.

Ar D 0.005

In the application of the control strategy, the desired amplitude of the reference is
set as

wr D 0:005 sin.0:0369t /:

and other the control parameters and the unknown external disturbance take the
following values:
102 L. Dai and L. Sun

Fig. 4.11 The control input

Fig. 4.12 The wave diagram of wp with the application of the control strategy

 
D 2; kf s D 0:01; F .w; w/
P D 0:001 sin wp :

As shown in Fig. 4.12, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.12 and 4.13.
In Fig. 4.12 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 15,000. It can be seen from Fig. 4.12, a short period of
time is needed for stabilizing the beam after the application of the control strategy.
After the short period, the chaotic motion will then become a periodic one, of which
the amplitude is 0.005.
From Fig. 4.13a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
Figure 4.14 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 103

Fig. 4.13 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy
104 L. Dai and L. Sun

Fig. 4.14 The wave diagram of wp (the blue continuous line) and the reference signal wr (the
green dash line)

Fig. 4.15 The control input

well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
with the application of the control strategy. As shown in the figure, the maximum
amplitude of the cable is very close to that of the reference signal.
Figure 4.15 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 105

Fig. 4.16 The wave diagram of wp with the application of the control strategy

4.6.3 Frequency Synchronization

The control strategy will be applied to show its effectiveness and efficiency in the
frequency synchronization. Three different desired frequencies will be specified as
the amplitude of the reference signal.

!r D 0.0277

In the application of the control strategy, the desired frequency of the reference is
set as

wr D 0:03 sin.0:0277t /:

and other the control parameters and the unknown external disturbance take the
following values:
 
D 3; kf s D 0:01; F .w; w/
P D 0:001 sin wp :

As shown in Fig. 4.16, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.16 and 4.17.
In Fig. 4.16 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 15,000. It can be seen from Fig. 4.16, a short period of
time is needed for stabilizing the beam after the application of the control strategy.
After the short period, the chaotic motion will then become a periodic one, of which
the frequency is 0.0277.
From Fig. 4.17a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
106 L. Dai and L. Sun

Fig. 4.17 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 107

Fig. 4.18 The wave diagram of wp (the blue continuous line) and the reference signal wr (the
green dash line)

Fig. 4.19 The control input

Figure 4.18 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
with the application of the control strategy. As shown in the figure, the frequency of
the cable is very close to that of the reference signal.
Figure 4.19 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.
108 L. Dai and L. Sun

Fig. 4.20 The wave diagram of wp with the application of the control strategy

!r D 0.0092

In the application of the control strategy, the desired frequency of the reference is
set as

wr D 0:03 sin.0:0092t /:

and other the control parameters and the unknown external disturbance take the
following values:
 
D 3; kf s D 0:01; F .w; w/
P D 0:001 sin wp :

As shown in Fig. 4.20, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.20 and 4.21.
In Fig. 4.20 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 1,500. It can be seen from Fig. 4.20, a short period of
time is needed for stabilizing the beam after the application of the control strategy.
After the short period, the chaotic motion will then become a periodic one, of which
the frequency is 0.0092.
From Fig. 4.21a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
Figure 4.22 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 109

Fig. 4.21 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy
110 L. Dai and L. Sun

Fig. 4.22 The wave diagram of wp (the blue continuous line) and the reference signal wr (the
green dash line)

Fig. 4.23 The control input

with the application of the control strategy. As shown in the figure, the frequency of
the cable is very close to that of the reference signal.
Figure 4.23 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 111

Fig. 4.24 The wave diagram of wp with the application of the control strategy

!r D 0.0031

In the application of the control strategy, the desired frequency of the reference is
set as

wr D 0:03 sin.0:0031t /:

and other the control parameters and the unknown external disturbance take the
following values:
 
D 3; kf s D 0:01; F .w; w/
P D 0:001 sin wp :

As shown in Fig. 4.24, the proposed control strategy is applied at t D 7,200. After
the application of the control strategy, the responses of the cable at the translating
speed v0 D 2m/s are shown in Figs. 4.24 and 4.25.
In Fig. 4.24 the responses of the selected point, wp , is shown, for the period of
time from t D 7,200 to t D 15,000. It can be seen from Fig. 4.24, a short period of
time is needed for stabilizing the beam after the application of the control strategy.
After the short period, the chaotic motion will then become a periodic one, of which
the frequency is 0.0031.
From Fig. 4.25a–c, the displacements are shown in terms of w1 , w2 , and w3 .
Based on these figures, through the application of the control strategy, each of the
displacements of the axially translating cable is gradually stabilized from a chaotic
motion into a periodic one.
Figure 4.26 shows the comparison between the actual response of the beam wp
and the reference signal wr applied. One may notice that the reference signal wr is
well periodic with respect to time t. One may also see from the figure, the maximum
amplitude of the cable’s response slightly varies after the stabilization of the cable
112 L. Dai and L. Sun

Fig. 4.25 (a) The contribution of the first vibration mode to wp with the application of the
control strategy. (b) The contribution of the second vibration mode to wp with the application
of the control strategy. (c) The contribution of the third vibration mode to wp with the application
of the control strategy
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 113

Fig. 4.26 The wave diagram of wp (the blue continuous line) and the reference signal wr (the
green dash line)

Fig. 4.27 The control input

with the application of the control strategy. As shown in the figure, the frequency of
the cable is very close to that of the reference signal.
Figure 4.27 shows the control input U. Initially, the control input reaches a
peak very quickly after the application of the control strategy. Once the system
is stabilized, the control input displays a periodic wave diagram as shown in the
figure and the maximum value of the control input is significantly decreased to a
very small value.
114 L. Dai and L. Sun

Key Symbols

v0 The constant rate that the cable is axially moving at

l The length of the cable
u0 (x(t), t) The displacement along the horizontal axis
w0 (x(t), t) The displacement along the vertical axis
t The dimensional time
x(t) The position of any point of the translating cable
i The unit vector of the fixed Cartesian coordinate along the horizontal
axis
k The unit vector of the fixed Cartesian coordinate along the vertical
axis
r The position vector of any point of the translating cable without
deformation
R The displacement field of the axially translating cable
V The volume of the cable
T The kinetic energy of the cable
The mass of the cable per unit length
"11 The strain along the horizontal axis
U The total strain energy of the cable
Q11 The elastic coefficient along the direction of "11
W The virtual work done by force
L The Lagrangian function
t The non-dimensional time
 The non-dimensional coefficients corresponding to time
x The non-dimensional position of any point of the translating cable
v0 The non-dimensional rate the cable is moving at
w0 The non-dimensional displacement along the vertical axis
 The non-dimensional coefficients corresponding to the geometric
nonlinearity
'n .x/
 The comparison function
wn t The nth vibration mode
ˆ The general form of a nonlinear governing equation
U The total control input
P .w; w/ P The unknown external disturbance
 i (W, t) The corresponding expression of ˆ .w; w; P t / via the Galerkin method
ui The corresponding expression of U via the Galerkin method
fi (W, t) The corresponding expression of P .w; w/ P via the Galerkin method
wp The non-dimensional response of a selected point of the cable
xp The non-dimensional location of a selected point of the cable
wr The reference signal
wr The amplitude of the reference signal
!r The frequency of the reference signal
Ueq A part of the control input
4 Active Vibration Control for Nonlinear Axially Translating Cable Systems. . . 115

Ur A part of the control input

Ufs The control variable via fuzzy rule
The control parameters governing the sliding surface
kfs The control parameter corresponding to P .w; w/P
RC The positive real number

References

Abou-Rayan AM, Nayfeh AH, Mook DT, Nayfeh AM (1993) Nonlinear response of a parametri-
cally excited bucked beam. Nonliear Dyn 4:499–525
Carrera E, Giunta G, Petrolo M (2011) Beam structures: classical and advanced theories. Wiley,
New York
Dai L (2008) Nonlinear dynamics of piece constant systems and implementation of piecewise
constant arguments. World Scientific, New Jersey
Fung RF, Lu PY, Tseng CC (1998) Non-linearly dynamic modelling of an axially moving beam
with a tip mass. J Sound Vib 291:559–571
Ghayesh MH, Amabili M (2013) Nonlinear dynamics of an axially moving Timoshenko beam with
an internal resonance. Nonlinear Dyn 73:39–52
Haghighi HH, Markazi AHD (2009) Chaos prediction and control in MEMS resonators. Commun
Nonlinear Sci 15:3091–3099
Humer A (2013) Dynamic modeling of beams with non-material, deformation-dependent boundary
conditions. J Sound Vib 332:622–641
Kuo CL, Shieh CS, Lin CH, Shih SP (2007) Design of an adaptive fuzzy sliding-mode controller
for chaos synchronization. Int J Nonlinear Sci 8:631–636
Luo AC, Hamidzadeh HR (2004) Equilibrium and buckling stability for axially traveling plates.
Commun Nonlinear Sci Numer Simulat 9:343–3360
Nayfeh AH, Mook DT (1979) Nonlinear oscillation. Wiley-Interscience, New York
Tabarrok B, Leech CM, Kim YI (1974) On the dynamics of an axially moving beam. J Franklin
Inst 297:201–220
Tang JL, Ren GX, Zhu WD, Ren H (2011) Dynamics of variable-length tethers with application to
tethered satellite deployment. Commun Nonlinear Sci Numer Simulat 16:3411–3424
Utkin VI (1992) Sliding modes in control and optimization. Springer, Berlin
Wickert JA (1992) Non-linear vibration of a travelling tensioned beam. Int J Mech 27:503–517
Yang KJ, Hong KS, Matsuno F (2005) Energy-based control of axially translating beams: varying
tension, varying speed, and disturbance adaptation. IEEE Trans Control Syst Technol 13:1045–
1054
Yau HT, Kuo CL (2006) Fuzzy sliding mode control for a class of chaos synchronization with
uncertainties. Int J Nonlinear Sci Numer Simulat 7:333–338
Yau HT, Wang CC, Hsieh CT, Cho CC (2010) Nonlinear analysis and control of the uncertain
micro-electro-mechanical system by using a fuzzy sliding mode control design. Comput Math
Appl 61:1912–1916
Zhu WD (2002) Control volume and system formulations for translating media and stationary
media with moving boundaries. J Sound Vib 254:189–201
Zhu WD, Ni J (2000) Energetic and stability of translating media with an arbitrarily varying length.
J Vibrat Acoust 122:295–304
Chapter 5
Nonlinear Initial Value Ordinary Differential
Equations

Mohammad M. Aghdam, Ali Fallah, and Poorya Haghi

Keywords Multi-step method • Nonlinear differential equation • Initial value

problem • Error analysis

5.1 Introduction

Ordinary differential equations (ODEs) present the mathematical models of many

physical problems in different fields of science, engineering, and economy. ODEs
are categorized into initial value problems (IVP) and boundary value problem
(BVP). BVPs usually explain the equilibrium state of the systems such as steady
state heat or mass transfer while IVPs present the dynamic behavior of the system
such as motion of system particles subjected to external load or transient heat/mass
transfer. In this chapter, IVPs are considered. An initial value ODE in the general
case is in the form of:

dy
D f .x; y/ ; x Œa; b ; y.a/ D y0 (5.1)
dx
where f (x, y) can be linear or nonlinear function of x and y, and y0 is initial condition.
It is worth mentioning that, higher order ODEs and systems of ODEs can also be
shown in the form of Eq. (5.1). Thus, the presented method is applicable to systems
of ODEs of any order.

M.M. Aghdam () • P. Haghi

Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran
e-mail: [email protected]; [email protected]
A. Fallah
Department of Mechanical Engineering, Sharif University of Technology, Tehran, Iran
e-mail: [email protected]

© Springer International Publishing Switzerland 2015 117

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__5
118 M.M. Aghdam et al.

Except in some special cases such as constant coefficient linear ODEs, analytical
solutions for ODEs are generally not available. Therefore, various numerical
methods are proposed for the solution of initial value ODEs such as methods based
on Taylor series, one-step methods such as Euler method, multi-stage single-step
methods such as Runge–Kutta methods and linear multi-step methods. Single-step
methods only require information about the solution at one previous point such as
x D xn  1 to compute the solution at an advanced point such as x D xn . For instance,
Euler method is a single-step formula for the solution of IVPs that predicts the
solution of the problem at x D xn , only from slope of the y(x) at x D xn  1 (Süli and
Mayers 2003):
yn  yn1
yn0 D D f .xn1 ; yn1 / D fn1 ! yn D yn1 C hfn1 (5.2)
h
where h is the step size, i.e. h D xn  xn  1 and yn  1 D y(xn  1 ), fn  1 D f (xn  1 , yn  1 ).
Equation (5.2) is called explicit formula because the right hand side of (5.2) can be
explicitly computed from known values of yn  1 . In spite of the simplicity of the
Euler method, (5.2), it is not generally employed in applied cases. There are two
main reasons why Euler’s method is not generally used in scientific computing.
Firstly, the truncation error per step associated with this method is far larger
than those associated with other, more advanced methods for a given value of h.
Secondly, Euler’s method is too prone to numerical instabilities. The main reason
that Euler’s method has such large truncation errors per step is that the method only
evaluates derivatives at the beginning of the interval: i.e., at xn  1 to compute the
solution from xn  1 to xn . Therefore, the method is very asymmetric with respect to
the beginning and the end of the interval.
Multi-stage methods such as Runge–Kutta were developed to overcome this
shortcoming. In multi-stage methods, more symmetric integration is employed by
evaluating the derivatives in some intermediate points in the considered interval.
For instance, in the second-order Runge–Kutta method, derivatives are evaluated
by using an Euler-like trial stage to the midpoint of the interval and then using the
values of both x and y at the midpoint to make the actual step across the interval, see
Fig. 5.1. The second-order Runge–Kutta formula is (Süli and Mayers 2003):

k1 D hfn1
k2 D hf xn1 C h2 ; yn1 C k1
2
(5.3)
yn D yn1 C k2

Therefore, the Euler’s method is equivalent to a first-order Runge–Kutta method.

Obviously, there is no need to stop at a second-order method and it is possible
to achieve higher order of accuracy using more trial stages per interval. The most
common Runge–Kutta method is the fourth-order Runge–Kutta (RK4). It is worth
mentioning that multi-stage methods such as Runge–Kutta methods are still single-
step methods because only the information of the previous point is used to obtain the
current point value. In spite of higher order of accuracy of the multi-stage methods
with respect to simple single-step methods, they need many derivative calculations
in each step.
5 Nonlinear Initial Value Ordinary Differential Equations 119

Fig. 5.1 Comparison between Euler method and second-order Runge–Kutta method

Fig. 5.2 Domain of k-step method

Multi-step methods use previous point information to increase the accuracy of

the results instead of using trial stages in each interval, see Fig. 5.2.
Consider the general linear or nonlinear first-order ordinary differential equation
(5.1). The integral form of (5.1) can be expressed as:

Zx
y.x/ D y0 C f .x; y/ dx (5.4)
a

In multi-step techniques, numerical solution is obtained by using a discretized

version of (5.4) as:
xZ
nC1

ynC1 D yn C f .x; y/ dx (5.5)

xn
120 M.M. Aghdam et al.

Various multi-step methods use different ways to approximate the integral term
in (5.5). For instance, Adams–Moulton method, one of the most common multi-
step methods, employs an approximation of f (x, y) in (5.5) by using the Newton
backward difference method to obtain a multi-step formulation.
Although the existing multi-step methods such as Adams–Moulton method are
accurate and useful, they also have their own limitations in terms of accuracy
and stability at large step sizes and weak performance in the case of stiff ODEs.
Therefore, providing more accurate and/or stable predictions while increasing step
size to reduce computational cost would be interesting for relevant researchers.
In order to achieve this goal, in the present work, the well-known Bernstein
polynomials as the basis for Bézier curves are employed. Bernstein polynomials
(special case of B-spline) are piecewise polynomials that represent a great variety of
functions. They can be differentiated and integrated without difficulty and construct
spline curves to approximate functions with desired accuracy. There has been
widespread interest in the use of these polynomials to obtain solutions for various
types of differential equations (Caglar and Caglar 2006; Jator and Sinkala 2007;
Khalifa et al. 2008; Caglar et al. 2008; Zerarka and Nine 2008; Bhattia and Brackenb
2007). In these studies, the solution is expanded in terms of a set of continuous
Bernstein or B-spline polynomials over a closed interval and then by employing
the weighted residual methods such as Galerkin or collocation to determine the
expansion coefficients to construct a solution. However, in the current work, the
Bernstein polynomials and Bézier curve are used to evaluate the integral in (5.5) by
providing a general k-step formula.
Results of the presented formula show improved stability in comparison with
Adams–Moulton method for the same step size. This leads to significant reduction
of computational time. In addition, stability and convergence of the presented
formula are discussed through common stability and error analysis of multi-step
methods.

5.2 Bézier Curves

Before computer graphics ever existed, engineers designed aircraft wings and
automobile bodies using splines. A spline is a long flexible piece of wood or
plastic with a rectangular cross section held in place at various positions by heavy
lead weights with a protrusion called duck, where the duck holds the spline in a
fixed position against the drawing board (Beach 1991). The spline then conforms
to a natural shape between the ducks. By moving the ducks around, the designer
can change the shape of the spline. The drawbacks are obvious. Recording duck
positions and maintaining the drafting equipment necessary for many complex parts
will take up square footage in a storage facility, costs that would be absorbed by a
consumer. A not so obvious drawback is that when analyzed mathematically, there
is no closed form solution (Buss 2003).
5 Nonlinear Initial Value Ordinary Differential Equations 121

In order to overcome these shortages, in the 1960s a mathematician and engineer

named Pierre Bézier introduced a family of smooth curves based on the Bernstein
basis polynomials. Bézier used the curves mainly to design various parts of
automobile bodies such as exterior car panels for his work at Renault in the 1960s.
The advent of Bézier curves revolutionized the design process and helped the
designers draw smooth looking curves on a computer screen, and use less physical
storage space for design materials. Bézier’s contribution to computer graphics has
opened the road for computer-aided design (CAD) software like Maya, Blender, 3D
Max, and Corel (Gerald et al. 1999; Lorentz et al. 1986; Farin 2002).
The nth-degree Bézier curve for (n C 1) given points P0 , P1 , : : : , Pn can be
written as:

X
n
B.u/ D Bi;n .u/Pi ; u 2 Œ0; 1 (5.6)
iD0

where the points Pi are called control points for the Bézier curves and Bi,n are known
as Bernstein basis polynomials of degree n defined as:
 
n
Bi;n .u/ D .1  u/ni ui ; i D 0; 1; : : : ; n (5.7)
i

It is worth mentioning here that quadratic and cubic Bézier curves are the most com-
mon types of Bézier curves since higher degree curves are more computationally
expensive to evaluate. Figure 5.3 and Table 5.1 show quadratic and cubic Bernstein
basis polynomials.
The location of control points in the 2D Cartesian coordinate system is:

xi
Pi D (5.8)
yi

Using (5.8), one can rewrite (5.6) in the parametric form of:

X
n
x.u/ D Bi;n .u/xi (5.9.1)
iD0
Xn
y.u/ D Bi;n .u/yi (5.9.2)
iD0

Bézier curves are widely used in computer graphics and computer-aided design,
mainly to provide smooth curves. It should be noted that similar to the least-square
curves, Bézier curves are not really interpolating curves, as they do not normally
pass through all of the control points. However, they have the important property
of staying within the polygon determined by the given points as shown in Fig. 5.4.
122 M.M. Aghdam et al.

a
1
0.9 B0,2(u) B1,2(u) B2,2(u)
0.8
0.7
0.6
Bi,n(u)

0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
u
b
1
0.9 B0,3(u) B1,3(u) B2,3(u) B3,3(u)
0.8
0.7
0.6
Bi,n(u)

0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
u

Fig. 5.3 Bernstein polynomials (a) quadratic n D 2, (b) cubic n D 3

Table 5.1 Linear, quadratic, and cubic Bernstein polynomials

n Bi,n (u)
1 B0,1 (u) D 1  u, B1,1 (u) D u
2 B0,2 (u) D (1  u)2 , B1,2 (u) D 2u(1  u), B2,2 (u) D u2
3 B0,3 (u) D (1  u)3 , B1,3 (u) D 3u(1  u)2 , B2,3 (u) D 3u2 (1  u), B3,3 (u) D u3
5 Nonlinear Initial Value Ordinary Differential Equations 123

Fig. 5.4 Comparing Newton interpolation with Bézier curves and Convex hull property

More details of Bézier curves and their properties can be found elsewhere (Gerald
et al. 1999; Lorentz et al. 1986; Farin 2002).

5.3 Technique

In the present study, Bézier curves of any order are used to develop a general k-step
formula that becomes unstable at larger step sizes compared to the existing multi-
step methods. To this end, in the integral of (5.5) is approximated by mth-degree
Bézier curves using control points which are defined as:
 
Pnj D
xnj
D  xn  j h  ; j D n  m:::n (5.10)
fnj f xnj ; ynj

Using Eqs. (9) and (5.10), one can write the new parametric form of mth-degree
Bézier curves as:

X
m
x.u/ D Bi;m .u/xnmCi D ¿.u/ (5.11.1)
iD0

X
m
f .x; y/ D Bi;m .u/fnmCi D '.u/ (5.11.2)
iD0

Finally, after approximation of the integral term in (5.5), one may rewrite (5.5) as:
Z xnC1 Z ¿1 .xnC1 /
ynC1 D yn C f .x; y/ dx Š yn C '.u/¿0 .u/d u (5.12)
xn ¿1 .x n/

Using general formula (5.12) together with various orders of Bézier curves in (5.6)
leads to different multi-step formulas. For instance, assuming quadratic Bézier
124 M.M. Aghdam et al.

curve leads to three-step formula. Different degrees of Bézier curves correspond

to different step formulas that are:
Two-step formula:

h
ynC1 D yn C .3fn  fn1 / (5.13)
2
Three-step formula:

h
ynC1 D yn C .19fn  8fn1 C fn2 / (5.14)
12
Four-step formula:

h
ynC1 D yn C .175fn  81fn1 C 15fn2  fn3 / (5.15)
108
and general m C 1-step formula:
Z   !
Xm mC1
m m mi i
ynC1 D yn C h m .1  u/ u d u fnmCi (5.16)
1 i
iD0

5.4 Consistency, Stability, and Convergence Analyses

The main concepts in the analysis of multi-step methods, and indeed any numerical
method for differential equations, are consistency, stability, and convergence. The
basic theory for analysis of multi-step methods has been developed by Dahlquist
(1963) and is presented in the classical book of Henrici (1962).
It is useful to employ a general framework for the presentation of general m C 1-
step methods as:

amC1 ynC1 C am yn C    C a0 yn.mC1/ D h bmC1 fnC1 C    C b0 fn.mC1/

(5.17)

Associated with (5.17) are also two characteristics polynomials of the m C 1-step
formula (5.17) as:

p.z/ D amC1 zmC1 C am zm C    C a1 z C a0 (5.18)

q.z/ D bmC1 zmC1 C bm zm C    C b1 z C b0 (5.19)

5 Nonlinear Initial Value Ordinary Differential Equations 125

A multi-step formula is in the implicit form if bm C 1 ¤ 0 and is in explicit

configuration if bm C 1 D 0. Thus, all presented multi-step formulas in (5.13) through
(5.16) are in explicit form.
Definition 1 For a given m C 1-step formula Eq. (5.17) the polynomials

X
mC1
p.z/ D ai zi
iD0
(5.20)
X
mC1
q.z/ D bi z i
iD0

are called first and second characteristic polynomials, respectively.

A multi-step method is called consistent, if it reproduces the exact solution for
the differential equation y0 D 1, when exact starting approximations are used.
Definition 2 (Süli and Mayers 2003) The multi-step formula (5.17) is consistent
if p(1) D 0 and p0 (1) D q(1).
Theorem 1 General m C 1-step formula (5.16) is consistent.
Proof From (5.16) and (5.18), it is clear that for the m C 1-step formula

p.z/ D zmC1  zm (5.21)

Using (5.21) to determine p(1) and p0 (1), one can easily conclude for the m C 1-step
formula (5.16) that


p.1/ D 0
(5.22)
p 0 .1/ D 1

Therefore, in order for m C 1-step formula (5.16) to be consistent, the value of

q(1) must be equal to one. Using general m C 1-step formula (5.16) in conjunction
with (5.20), one can determine coefficients ofq(z) in (5.20) as:

mC1 0
Z 
m    Xmi    
m mi i m @ mi j m
bi D m .1  u/ u d u D .1/
i i
j D0
j i Cj C1
1
" iCj C1 #!
mC1
1 ; i D 0:::m (5.23)
m
126 M.M. Aghdam et al.

Substitution of (5.23) in (5.20), one can derive q(1) as:

0 0 #11
X X   mi     " i Cj C1
m @X m  i
m m
@ m mC1
q.1/ D bi D .1/ j
 1 AA
i D0 i D0
i
j D0
j i Cj C1 m
(5.24)
After some mathematical simplification, (5.24) can be rewritten as:

X
m X
m
bi D ci (5.25)
iD0 iD0

where
    " mC1i #
m mi m mC1
ci D .1/  1 .1 C .1//mi
mi mC1i m
(5.26)

It can easily be shown that the last term in (5.26) is zero for i D 0 : : : m  1.
However, for i D m the value of ci in (5.26) becomes one. Therefore, considering
(5.24) and (5.26) results in:

X
m X
m
q.1/ D bi D c i D cm D 1 (5.27)
iD0 iD0

Consequently, comparing (5.22) and (5.27) with Definition 2, it can be concluded

that the general m C 1-step formula (5.16) is consistent.
The numerical solution of a single-step method depends on the initial condition,
y0 while numerical solution of a k-step method depends on all the k starting values
y1 , y2 , : : : , yk . Thus, it is important to know whether the numerical solution is stable
with respect to perturbations in the starting values or not. A multi-step formula
is stable (zero-stable) for a certain differential equation on a given step size, h, if
a perturbation in the starting values of size " causes numerical solution over that
interval to change by no more than ˛" for some value of ˛ which does not depend
on the step size h. This is called “zero-stability” because it is enough to check the
condition for the differential equation y’ D 0 (Süli and Mayers 2003).
Definition 3 (Süli and Mayers 2003) The multi-step formula in general form of
(5.17) is stable (zero-stable) if all roots of p(z) lie in the disk jzj  1 and if each root
of modulus one is simple.
Theorem 2 General m C 1-step formula (5.16) is zero-stable.
Proof Considering (5.21), roots of the first characteristic polynomial of general
m C 1-step formula (5.16) are:
5 Nonlinear Initial Value Ordinary Differential Equations 127


0 i D 1:::m
zi D (5.28)
1 i DmC1

Using Definition 3, one may conclude that the m C 1-step formula (5.16) is zero-
stable.
Accuracy of results of the multi-step methods depends on the starting point
values and the value of step size. It is important to be assured that as the step size
value decreases, difference between the results of the multi-step formula and the
exact solution also decreases, and prediction tends to the exact solution. Consider
multi-step formula (5.17) is employed for the solution of IVP (5.1) in domain
a < x < b and the sequence of meshes as fa D x0 < x1 < x2 : : : < xn D bg. As the
number of intervals increased, i.e., n ! 1 the step size tends to zeros, h ! 0.
The multi-step method is said to be convergent if as h ! 0 the yn ! y(tn ). Swedish
mathematician Dahlquist (1963) proved the relation between the concepts of zero-
stability, consistency, and convergence of a multi-step method by the theorem named
after him as Dahlquist’s equivalence theorem.
Theorem 3 (Dahlquist’s Equivalence Theorem) (Dahlquist 1963) For the multi-
step formula (5.17) to be convergent, it is necessary and sufficient that the formula
(5.17) to be zero-stable and consistent.
Proof Proof of this theorem is beyond of scope of the presented chapter as details
can be found in Dahlquist (1963), Henrici (1962), and Gautschi (1997).
Theorem 4 General m C 1-step formula (5.16) is convergent.
Proof Since the general m C 1-step formula (5.16) is both zero-stable and con-
sistent, based on Dahlquist’s equivalence theorem the general m C 1-step formula
(5.16) is also convergent.
Definition 4 (Süli and Mayers 2003) A convergent multi-step formula (5.17)
is said to be strongly stable if z D 1 is the only root of modulus 1 of its first
characteristic polynomial. If there is more than one such root, it is said to be
relatively stable.
Theorem 5 General m C 1-step formula (5.16) is strongly stable.
Proof According to (5.28), p(z) have m zero roots while the only one nonzero root
is z D 1, thus the general m C 1-step formula (5.16) is strongly stable.

5.5 Numerical Results and Discussions

In this section, efficiency of the presented multi-step method in terms of accu-

racy, convergence, and computational cost is investigated using various numerical
examples. It should be noted that all considered examples are nonlinear ODEs
128 M.M. Aghdam et al.

and therefore, analytical solutions are not available. Predictions of the present
work with four-step formula (5.15) are compared with numerical results of ODE45
MATLAB. ODE45 is a single-step solver based on an explicit fourth- or fifth-
order Runge–Kutta formula for the solution of ordinary differential equations in
MATLAB software (Guide et al. 2004). Furthermore, results of the well-known
four-step Adams–Moulton method are also included mainly to compare stability
of the presented method.
In examples, three different levels of the step size h are chosen. These step sizes
are carefully selected to show three different stages. At the first stage, both presented
and Adams–Moulton numeric methods are stable and accurate while in the second
stage, instability initiates in the Adams–Moulton technique and finally at the third
stage unstable behavior can be observed in the presented method. Furthermore,
in order to examine accuracy of the method, error for both numeric techniques is
reported in some examples.
In both methods, four-step formula is employed which indeed information of the
first four points is needed to start solution procedure. Since y0 is the only known
initial condition, the ODE45 results are used to obtain the first three points, i.e.,
y1 , y2 , y3 for both presented and the Adams–Moulton method.

5.5.1 Example 1

As first example, a nonlinear mass–spring–damper system as shown in Fig. 5.5 is

considered. Governing equation of such system is the well-known Duffing equation
(Nayfeh et al. 2007):

d 2y dy
m Cc C KL y C KNL y 3 D 0 (5.29)
dt 2 dt
where m, c, KL , and KNL are mass, damping coefficient, linear, and nonlinear spring
stiffnesses, respectively.
Introducing new variables y1 D y and y2 D dy dt
, the second-order governing
nonlinear ODE (5.29) can be converted to a first order system of nonlinear ODE as:

Fig. 5.5 Nonlinear

mass–spring–damper system
5 Nonlinear Initial Value Ordinary Differential Equations 129

(
dy1
D y2
dy2  dt  (5.30)
dt
D  m1 cy2 C KL y1 C KNL y13

Figure 5.6 shows predictions of presented and Adams–Moulton methods together

with numerical solution obtained by ODE45 for different values of step sizes in
the case of nonlinear mass–spring system with damping. Numerical results are
obtained for m D 1, c D 0.5, KL D KNL D 1 and initial condition y(0) D 1, y0 (0) D 0.
In each case, errors of the Adams–Moulton (yAdams  yODE45 ) and presented method
(yBezier  yODE45 ) predictions are provided in the figures. Results show that for small
values of step size (h D 0.1), both methods are accurate and stable while for higher
step size (h D 0.3), error of the Adams–Moulton grows rapidly. For step sizes
greater than h D 0.3, the Adams–Moulton becomes unstable whereas the presented
method is still stable with reasonable level of accuracy for h D 0.5.

5.5.2 Example 2

In the second example, the Lotka–Volterra equation, which is, also known as the
predator–prey equation is considered. This equation describes the dynamics of
biological systems in which two species interact, one as a predator and the other
as prey. The populations change through time according to the pair of equations
(Berryman 1992):
8
< dx
dt
D ax  bxy
(5.31)
: dy D cy C dxy
dt

where x(t) and y(t) denote population of the prey and predator species, respectively
and a, b, c, and d are positive constants.
Solution of predator–prey equation with a D 1, b D 0.03, c D 0.4, d D 0.01 using
presented and Adams–Moulton methods is shown in Fig. 5.7. For h D 0.1, results for
both methods are accurate and stable. As the step size growth to h D 0.3, Adams–
Moulton shows instability while the presented method become unstable in larger
step size of h D 0.5.

5.5.3 Example 3

Consider a Van der Pol oscillator, which is a non-conservative oscillator with

nonlinear damping and governing equation of motion as (Nayfeh et al. 2007):

d 2y   dy
  1  y2 Cy D0 (5.32)
dt 2 dt
130 M.M. Aghdam et al.

a
1
ODE45
0.5 Adams-Moulton
Bézier
0
y

–0.5
–1
0 5 10 15
t
0.01
Adams-Moulton
0.005 Bézier
Error

0
–0.005
–0.01
0 5 10 15
t

b 1
ODE45
0.5 Adams-Moulton
Bézier
0
y

–0.5
–1
0 5 10 15
t
0.4
Adams-Moulton
0.2 Bézier
Error

0
–0.2
–0.4
0 5 10 15
t

c 1
0.5
0
y

ODE45
–0.5 Bézier
–1
0 5 10 15
t
0.2
0.1
Error

0
–0.1 Adams-Moulton
Bézier
–0.2
0 5 10 15
t

Fig. 5.6 Solution of nonlinear mass–spring system with damping for (a) h D 0.1, (b) h D 0.3,
(c) h D 0.5
5 Nonlinear Initial Value Ordinary Differential Equations 131

a 100

80 ODE45
Adams-Moulton
60 Bézier
x(t)

40

20

0
0 2 4 6 8 10 12 14 16 18 20
t(s)
60

50

40
y(t)

30 ODE45
20
Adams-Moulton
Bézier
10
0 2 4 6 8 10 12 14 16 18 20
t(s)

b 150
ODE45
Adams-Moulton
100
Bézier
x(t)

50

0
0 2 4 6 8 10 12 14 16 18 20
t(s)

80
ODE45
60 Adams-Moulton
Bézier
40
y(t)

20

0
0 2 4 6 8 10 12 14 16 18 20
t(s)

c 150
ODE45
100 Adams-Moulton
Bézier
x(t)

50

–50
0 2 4 6 8 10 12 14 16 18 20
t(s)

100

50
y(t)

0 ODE45
Adams-Moulton
Bézier
–50
0 2 4 6 8 10 12 14 16 18 20
t(s)

Fig. 5.7 Solution of Lotka–Volterra equation for (a) h D 0.1, (b) h D 0.3, (c) h D 0.5
132 M.M. Aghdam et al.

where  is a scalar parameter indicating the nonlinearity and the strength of the
2
damping. In the case of  D 0, Eq. (5.32) reduces to ddt y2 C y D 0 which has the
solution y D c1 sin(t) C c2 cos(t). For small values of , one obtain weekly nonlinear
oscillation, i.e., oscillation which slightly differ from simple harmonic motion. On
the other hand, however, for high values of  relaxation oscillation can be observed
which shows strongly nonlinear oscillations exhibiting sharp periodic jumps. In the
case of high damping, the Van der Pol equation becomes stiff. Most techniques show
weak performance for solution of stiff ODEs. Here the performance of the presented
and Adams–Moulton methods in two different conditions of low and high damping
is investigated.
Introducing new variables y1 D y and y2 D dy dt
, the second order governing
nonlinear ODE is converted to the following first order system of nonlinear ODE:
(
dy1
dt
D y2
dy2  (5.33)
dt
D  1  y12 y2 C y1

Solution of the Van der Pol equation in the case of  D 0.5 and  D 5 using
presented multi-step method and Adams–Moulton is presented in Figs. 5.8 and 5.9,
respectively. Included in the figures are also results of the ODE45 for comparison.
In the case of lower damping, i.e.,  D 0.5, predictions of both presented and
Adams–Moulton for h D 0.1 are stable and accurate. As the step size increases
to h D 0.2, Adams–Moulton results become unstable while the presented method
shows unstable behavior at h D 0.4.
Furthermore, for higher damping of  D 5, smaller step size is used. For
h D 0.001, both methods are accurate and stable. As the step size growth to
h D 0.005, instability initiates in the results of Adams–Moulton while the presented
method becomes unstable at step sizes ten times greater, i.e., h D 0.05. Results show
that even in the case of stiff ODEs, the presented multi-step method is more accurate
and stable in comparison with the well-known Adams–Moulton method.

5.6 Conclusion

Multi-step methods such as Adams–Moulton are common techniques for solution

of initial value ODEs. Besides their accuracy and stability, they also have their own
limitations such instability in large step sizes or in the case of stiff differential
equation. Thus, multi-step methods with better performance in terms of accuracy
and convergence particularly in the case of stiff equations are appealing.
In the present study, a new general multi-step formula developed based on the
Bézier curves. The consistency, stability, and convergence of the presented formula
are proved by using the common stability and error analysis theorems of multi-step
methods. The other advantage of the method is the presentation of a closed form
general m C 1-step formula.
5 Nonlinear Initial Value Ordinary Differential Equations 133

a 4
ODE45
2 Adams-Moulton
y(t)

0 Bézier

–2

–4
0 5 10 15
t(s)
0.2
0
Error

–0.2
Adams-Moulton
–0.4 Bézier
–0.6
0 5 10 15
t(s)

b 4
ODE45
2 Adams-Moulton
Bézier
y(t)

–2

–4
0 5 10 15
t(s)
1
0
–1
Error

–2 Adams-Moulton
Bézier
–3
–4
0 5 10 15
t(s)

c 4
ODE45
2 Bézier
y(t)

–2

–4
0 5 10 15
t(s)
0.1

0.05 Bézier
Error

–0.05

–0.1
0 5 10 15
t(s)

Fig. 5.8 Solution of Van der Pol equation in case of low damping  D 0.5 for (a) h D 0.1,
(b) h D 0.2, (c) h D 0.4
134 M.M. Aghdam et al.

a 4

2
y(t)

0 ODE45
–2 Adams-Moulton
Bézier
–4
0 5 10 15
t(s)
0.15
0.1 Adams-Moulton
0.05 Bézier
Error

0
–0.05
–0.1
0 5 10 15
t(s)

b
4

2
y(t)

0 ODE45
–2 Adams-Moulton
Bézier
–4
0 5 10 15
t(s)
2
Adams-Moulton
1 Bézier
Error

–1
0 5 10 15
t(s)

c
4
ODE45
2
Bézier
y(t)

–2

-4
0 5 10 15
t(s)
2

1 Bézier
Error

–1
0 5 10 15
t(s)

Fig. 5.9 Solution of Van der Pol equation in case of high damping  D 5 for (a) h D 0.001,
(b) h D 0.005, (c) h D 0.05
5 Nonlinear Initial Value Ordinary Differential Equations 135

Effects of the step size on the accuracy and stability of the presented method are
examined through numerical results of various nonlinear and stiff systems. Results
revealed that the presented method shows improved stability behavior in comparison
with the well-known Adams–Moulton method, which yields to significant reduction
of computational cost. Due to simplicity, minimum computational efforts, and
accuracy, the presented multi-step method is expected to be used in nonlinear
problems in various science and engineering applications.

Key Symbols

Bi,n (u) The ith Bernstein basis polynomials of degree n

Pi Bézier curve control point
p(z) First characteristic polynomials of multi-step formula
q(z) Second characteristic polynomials of multi-step formula
fn Value of f (x, y) at point (xn , yn )
h Step size
xn The nth discretized point x coordinate
yn The nth discretized point y coordinate and value of y (x) at point (xn )
x0 x coordinate of the first point
y0 Initial condition

References

Beach RC (1991) An introduction to curves and surfaces of computer-aided design. Van Nostrand
Reinhold, New York
Berryman AA (1992) The origins and evolution of predator–prey theory. Ecology 73(5):1530–1535
Bhattia MI, Brackenb P (2007) Solutions of differential equations in a Bernstein polynomial basis.
Comput Appl Math 205:272–280
Buss SR (2003) 3D computer graphics-a mathematical introduction with open GL. Cambridge
University Press, Cambridge, UK
Caglar N, Caglar H (2006) B-spline solution of singular boundary value problems. Appl Math
Comput 182:1509–1513
Caglar H, Ozer M, Caglar N (2008) The numerical solution of the one-dimensional heat equation
by using third degree B-spline functions. Chaos Solit Fract 38:1197–1201
Dahlquist GG (1963) A special stability problem for linear multi-step methods. BIT Numer Math
3(1):27–43
Farin G (2002) Curves and surfaces for CAGD. A practical guide, 5th edn. Elsevier Ltd., San
Francisco, CA, USA
Gautschi W (1997) Numerical analysis: an introduction. Birkhäuser, Boston
Gerald CF, Wheatley PO (1999) Applied numerical analysis, 6th edn. Addison-Wesley, Reading
Henrici P (1962) Discrete variable methods in ordinary differential equations. Wiley, New York
Jator S, Sinkala Z (2007) A high order B-spline collocation method for linear boundary value
problems. Appl Math Comput 191:100–116
136 M.M. Aghdam et al.

Khalifa AK, Raslan KR, Alzubaidi HM (2008) A collocation method with cubic B-splines for
solving the MRLW equation. Comput Appl Math 212:406–418
Lorentz GG (1986) Bernstein polynomials, 2nd edn. Chelsea Publishing Co., New York
MATLAB Reference Guide Version 7.0 (2004) The Math Works Inc
Nayfeh AH, Mook DT (2007) Nonlinear oscillations. Wiley, New York
Süli E, Mayers D (2003) An introduction to numerical analysis. Cambridge University Press,
Cambridge, UK
Zerarka A, Nine B (2008) Solutions of the Von Kàrmàn equations via the non-variational Galerkin-
B-spline approach. Commun Nonlinear Sci Numer Simulat 13:2320–2327
Chapter 6
The Loss Tangent of Visco-Elastic Models

Franz Konstantin Fuss

Keywords Visco-elasticity • Elasticity • Viscosity • Loss tangent • Standard

linear solid • Power model • Log model • Stress strain • Transient part • Steady
state • Laplace transform • Inverse Laplace transform • Convolution integral

6.1 Introduction

Visco-elastic models are characterised by phenomena such as

– Velocity dependency: increase of stiffness with strain rate,
– Stress relaxation: decay of stress with time at constant strain,
– Creep: increase of strain with time at constant stress, and
– Loss of stored (elastic) energy due to inner friction resulting in unequal loading
and unloading stress–strain curves, the area between the two curves correspond-
ing to the dissipated energy (hysteresis).
Visco-elastic models can be classified in various ways, e.g.
(1) Linear or non-linear models
– Linear models with at least one elastic and one viscous element in parallel
or in series;

F.K. Fuss ()

School of Aerospace, Mechanical, and Manufacturing Engineering, RMIT University,
Melbourne, VIC, Australia
e-mail: [email protected]

© Springer International Publishing Switzerland 2015 137

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__6
138 F.K. Fuss

– QLV/quasi-linear “viscous” models such as Prony series (Wiechert model;

Wiechert 1889, 1893);
– Non-linear models such as logarithmic and power law models.
or
(2) According to the decay function of stress relaxation, which can be
– Exponential: linear three-element model (standard linear solid),
– Power: non-linear power law model, or
– Logarithmic: non-linear logarithmic law model
Examples for the latter two non-linear models are:
– Power law model: biological tissues such as ligaments (Provenzano et al. 2001),
foams with non-negative stiffness (Fuss 2009), cork in cricket balls (Fuss 2008a,
2012);
– Logarithmic law model: solid polymers (Findley et al. 1989), polymer golf balls
(Fuss 2008b, 2012).
The loss tangent, tan ı, is defined as the tangent of the phase angle ı, which, in
turn, is the ratio of loss modulus E00 to storage modulus E0 .
E 00
tan ı D (6.1)
E0
where
0
E0 D cos ı (6.2)
"0
0
E 00 D sin ı (6.3)
"0

and  0 and "0 are the peak amplitudes of stress  and strain ", respectively.
The complex modulus E* is defined as
0 iı 0
E D e D .cos ı C i sin ı/ D E 0 C iE 00 D jE  j eiı (6.4)
"0 "0
p
where i D 1, and jE* j is the dynamic modulus, the magnitude of E* , i.e. the
resultant of loss modulus E00 and storage modulus E0
0
jE  j D (6.5)
"0
The loss tangent, tan ı, is usually determined by subjecting a material or structure
to sinusoidal strain "

" D "0 sin .2 f t / (6.6)

6 The Loss Tangent of Visco-Elastic Models 139

Fig. 6.1 Sinusoidal strain σ0

curve (blue) imposed on a
visco-elastic material and
ε0
resulting stress curve (pink);
¢ 0 : stress amplitude, maximal
stress; "0 : strain amplitude,

stress, strain
maximal strain; •: phase shift

time

where f is the cyclic frequency (angular frequency ! D 2 f ). The resulting reaction

stress  is equally sinusoidal, but out of phase with respect to the strain by the phase
angle ı

 D 0 sin .2 f t C ı/ (6.7)

A positive phase angle ı causes the stress peak to occur earlier than the strain peak
(Fig. 6.1), resulting in the typical hysteresis effect of visco-elastic materials when
plotting stress against strain (Fig. 6.2). The area of the hysteresis loop corresponds
to the energy dissipated into the material as thermal energy.
When subjecting a material to cyclic (sinusoidal) strain, the peak stress,  0 ,
increases during the first cycles (transient part) until it reaches an equilibrium or
steady state (Fig. 6.3). The phase angle ı is measured once the steady state has
set in.
It is evident, that the energy dissipated by inner friction depends on the viscosity
parameter . However, as the loss tangent is the ratio of loss to storage modulus,
the strain rate independent elasticity parameter E is expected to influence the loss
tangent too. Lastly, as the modulus (Young’s and tangent) increases with strain rate
and thus with frequency f, the latter could contribute to the loss tangent as well.
The objective of this Book Chapter is to explore in how far the viscosity
parameter , the strain rate independent elasticity parameter E and the strain
frequency f affect the loss tangent. The aim is to derive the loss tangent at steady
state of the three visco-elastic models mentioned above, in order to understand the
interaction between elasticity, viscosity and frequency and their effect on energy
loss. The function of stress relaxation with time is not the only difference between
the three models mentioned above. A further objective of this paper is to analyse
the basic differences of these models and to understand their applicability and
constraints.
140 F.K. Fuss

σ0
σ 0cosδ

Stress

ε 0cosδ
ε0
0
strain

Fig. 6.2 Hysteresis loop of stress–strain ellipse; ¢ 0 : maximal stress; "0 : maximal strain; •: phase
angle

transient part steady state

stress

time

Fig. 6.3 Stress–time curve (red) during the first load cycles; blue curve: transient component;
green curve: steadys state component; the stress–time curve (red) is the sum of transient and steady
state components
6 The Loss Tangent of Visco-Elastic Models 141

6.2 Analysis

6.2.1 The Standard Linear Solid (Zener Model)

The standard linear solid (SLS of Voight form) consists of two Hookean springs and
a viscous damper, where the spring with the modulus E1 is connected in series with
a Kelvin–Voight model, with spring of modulus E2 and damper of viscosity constant
 connected in parallel (Fig. 6.2).
From Fig. 6.4

" D "1 C "2 (6.8)

 D "1 E1 (6.9)

 D "2 E2 C "P2  (6.10)

Taking the Laplace transform of Eqs. (6.8)–(6.10), eliminating "1 and "2 by
substitution, and solving for b
 yields the constitutive equation of the standard linear
solid
E1 E2 C sE1
b
 Db
" (6.11)
E1 C E2 C s

where the caret (ˆ) denotes the transformed parameter.

The equation for stress relaxation results from applying a constant strain "c to the
model through a Heaviside function H(t)

" D "c H.t / (6.12)

Fig. 6.4 Standard linear

solid of Voight form; :
stress; ": strain; : viscosity
constant; E1 : modulus of
series spring; E2 : modulus of
parallel spring; "1 : strain of
series spring; "2 : strain of
Kelvin–Voight model
142 F.K. Fuss

the Laplace transform of which is

"c
b
"D (6.13)
s
By substituting Eq. (6.13) into Eq. (6.11), we obtain

E1 E2 C sE1
b
 D "c (6.14)
s.E1 C E2 C s/

the inverse Laplace transform of which yields the function of stress relaxation

.E1 CE2 /
 E2 C E1 et 
D E1 (6.15)
"c E1 C E2

where the stress  is normalised to the constant strain "c . Equation (6.15) proves the
exponential decay of stress with time, as mentioned above in the Introduction.
The loss tangent results from applying the sinusoidal strain of Eq. (6.6) to the
constitutive Eq. (6.11). The Laplace transform of Eq. (6.6) is

2 f
b
" D "0 (6.16)
s2 C .2 f /2

where "0 is the peak amplitude of the strain.

By substituting Eq. (6.16) into the constitutive Eq. (6.11), we obtain
  
2 f E1 E2 C sE1
b
 D "0 (6.17)
s 2 C .2 f /2 E1 C E2 C s

After rearranging

E2 C s
b
 D "0 2 f E1
s3 C s2E 1 C s2E 2 C s.2 f /2 C E1 .2 f /2 C E2 .2 f /2
(6.18)

the inverse Laplace transform of Eq. (6.18) yields

E1 CE2
E1 2 f et  C E1 2 f cos .2 f t/ C E1 E2 C E2 2 C 2 .2 f /2 sin .2 f t/
 D "0 E1
E1 2 C 2E1 E2 C E2 2 C 2 .2 f /2
(6.19)
The numerator of Eq. (6.19) comprises of a transient part (exponential function)
and the steady state part (sine and cosine functions). At large times, if t ! 1, the
transient part, i.e. the exponential term of the numerator, vanishes and the steady
state sets in:
6 The Loss Tangent of Visco-Elastic Models 143

"0 E1 h i
D E1 2 f cos .2 f t / C E1 E2 C E2 2 C 2 .2 f /2 sin .2 f t /
.E1 C E2 /2 C 2 .2 f /2
(6.20)

In order to obtain the loss tangent and the peak stress  0 , we apply the addition rules
to Eq. (6.7)

 D 0 sin .ı/ cos .2 f t / C 0 cos .ı/ sin .2 f t / (6.21)

where the unknowns are the two components of the peak stress  0 :  0 sin ı and  0
cos ı.
The loss tangent thus results from the ratio of the two components

0 sin ı
tan ı D (6.22)
0 cos ı

and the resultant of the peak stress  0 is obtained from

q
0 D .0 sin ı/2 C .0 cos ı/2 (6.23)

From Eq. (6.20) it follows that

"0 E1
0 sin ı D E1 2 f (6.24)
.E1 C E2 /2 C 2 .2 f /2

and
"0 E1
0 cos ı D E1 E2 C E2 2 C 2 .2 f /2 (6.25)
.E1 C E2 /2 C 2 .2 f /2

The loss tangent thus is

E1 2 f
tan ı D (6.26)
E1 E2 C E2 2 C 2 .2 f /2

and the peak stress  0 results from

r
"0 E1 2
0 D 2 2
E1 2 2 .2 f /2 C E1 E2 C E2 2 C 2 .2 f /2
.E1 C E2 / C 2 .2 f/
(6.27)

In both Eqs. (6.26) and (6.27) f and  are linked together and always occur as the
product f.
As the maximal strain rate "P0 of a sinusoidal strain function equals "0 2 f, the
strain rate dependency of  0 is given by
144 F.K. Fuss

v
u
u  2 "  2 #2
0 D
"0 E1 tE 2 2 "P0 C E E C E 2 C 2 "P0
2 1 1 2 2
"P0 "0 "0
.E1 C E2 /2 C 2 "0
(6.28)
The peak stress  0 increases with f and . If f or  ! 1,  0 asymptotes to "0 E1 . If f
or  ! 0,  0 ! "0 (E1 –1 C E2 –1 )–1 . The limits of  0 are evident when considering
that the peak strain rate changes with f. Thus, at zero strain rate or zero , the
standard linear solid reduces to two springs in series and the effective modulus of
the model becomes (E1 –1 C E2 –1 )–1 . At infinite strain rate or infinite , the damper
becomes rigid and the effective modulus of the model is just E1 .
The peak stress  0 increases with E1 . If f or  ! 1 and E1 ! 0 or 1, both the
effective modulus and  0 become equally 0 or 1, respectively. If f or  ! 0 and
E1 ! 0 or 1, the effective modulus and  0 become zero in the former case, and E2
and "0 E2 in the latter.
If E2 ! 0, the standard linear solid reduces to a Maxwell model (with a spring
and a damper in series), the modulus of which is (1) 1 or (2) E1 or (3) 0, if (1) E1 ,
and  or f, are 1 or (2)  or f are 1 or (3) E1 , and  or f are 0, respectively.
At  > 0 and f < 1, and E2 ! 0, the peak stress reaches

2 f "0 E1 
0 D q (6.29)
E1 2 C 2 .2 f /2

If E2 ! 1, the modulus of the standard linear solid reduces to E1 .  0 shows a local

minimum at a certain E2 (Fig. 6.5)
0

q
E1 2 C 42 .2 f /2  E1
E20 D (6.30)
2
resulting from equating the first E2 derivative of Eq. (6.27) with zero and solving for
E2 . Equation (6.30) is the only positive and real result of the fourth order polynomial
nature of the first E2 derivative of Eq. (6.27).
The loss tangent, Eq. (6.26), reveals that the phase angle ı depends on all four
parameters, E1 , E2 , , and f, where the latter two always occur as the product f.
If the frequency f ! 1 or 0, the phase angle ı ! 0 in both cases. Thus, we expect
a local maximum of ı at a certain frequency f0 (actually shown by Findley et al.
1989). Taking the first derivative of tan ı with respect to f in Eq. (6.26), equating it
with zero and solving for f provides
p
E1 E2 C E2 2
f0 D (6.31)
2 
6 The Loss Tangent of Visco-Elastic Models 145

fη = 100
fη = 10
fη = 20
fη = 0.5
0.8 fη = 0.2
fη = 2
fη = 1

0.6 E2
fη = 0.1 0
σ0

0.4
fη = 0.05

0.2
fη = 0.02

fη = 0.01
fη = 0
0
0.001 0.01 0.1 1 10 100 1000
E2

Fig. 6.5 Maximal stress ¢ 0 against modulus E2 of parallel spring; f : frequency; ˜: viscosity
constant; E2 : E2 at which a local minimum of  0 is found
0

Replacing f by f0 in Eq. (6.26),

E1
.tan ı/f0 D p (6.32)
2 E1 E2 C E2 2

and equating its denominator with zero yields the scenario at which ı D 0.5 , which
is E1 D –E2 and E2 D 0. In both cases, however, f0 D 0, which means that a standard
linear solid can never reach ı D 0.5 or an infinite loss tangent.
Rearranging Eqs. (6.31) and (6.32) shows that f0 depends on two ratios, R1 and
R2 , whereas tan ı at f0 (but also any other f at the same R1 ) depends only on R2
(Fig. 6.6)
R1
f0 D (6.33)
2

R2
.tan ı/f0 D p (6.34)
2 1 C R2
where
p
E 2 1 C R2
R1 D (6.35)

146 F.K. Fuss

E1E2 + E22
R1 = = 0.1 1 10
η f0
1
R2 = E1 / E2 = 10
0.9

0.8

0.7
phase angle δ (rad)

0.6

0.5

0.4 R2 = E1 / E2 = 1
f0
0.3

0.2

0.1 R2 = E1 / E2 = 0.1
f0
0
1E-005

0.0001

0.001

0.01

0.1

10

100

1000

10000

frequency f (Hz)

Fig. 6.6 Phase angle • against frequency f ; ˜: viscosity constant; E1 and E2 : moduli of series and
parallel spring; f0 : f at which a local maximum of ı is found

and
E1
R2 D (6.36)
E2

Equation (6.34) shows that tan ı at f0 is a constant and independent of f0 (Fig. 6.6).
Changing  has the same effect as changing f has. If  ! 1 or 0, the phase angle
ı ! 0 in both cases and we obtain a local maximum of ı at a certain 0
p
E1 E2 C E2 2
0 D (6.37)
2 f

Equation (6.37) is similar to Eq. (6.31) and the same principles apply.
The phase angle ı and tan ı increase with E1 . If E1 ! 0, tan ı ! 0. If E1 ! 1,
tan ı asymptotes to
6 The Loss Tangent of Visco-Elastic Models 147

2 f
lim .tan ı/ D (6.38)
E1 !1 E2

The phase angle ı and tan ı decrease with increasing E2 , asymptoting to 0 if

E2 ! 1. If E2 ! 0, tan ı reaches a limit of

E1
lim .tan ı/ D (6.39)
E2 !0 2 f

Lakes (2009) derived the loss tangent of the SLS of Maxwell form (with a spring
connected in parallel with a Maxell model).

6.2.2 The Power Law Model

The power law model is characterised by a power decay of stress  with time t:

D Et  (6.40)
"c

Stress  is normalised to the constant strain "c applied by the Heaviside function
H(t) of Eq. (6.12).
Taking Laplace transform of Eq. (6.40) yields

 . C 1/
b
 D "c E (6.41)
s C1
where  denotes the Gamma function (Fuss 2008a, 2012).
By substituting Eq. (6.13) into Eq. (6.41), we obtain the constitutive equation of
the power law of non-linear visco-elasticity (Fuss 2008a, 2012):

b
 D s b
"E . C 1/ (6.42)

By substituting the sinusoidal strain of Eq. (6.16) into the constitutive Eq. (6.42),
we obtain

2 f
b
 D "0 E . C 1/ s  (6.43)
s2 C .2 f /2

Taking the inverse Laplace transform of Eq. (6.43), we obtain a general fractional
derivative of the sine function
d
 D "0 E . C 1/ sin .2 f t / (6.44)
dt 
148 F.K. Fuss

where  is the th time derivative of the strain function of Eq. (6.6) times a constant.

The generalised solution of dtd  sin.t / results from applying the inverse operation
of the Riemann–Liouville fractional integration (with lower limit D 0),
 1 
d  t t 3
sin.t / D sin t C  C  C : : : ; (6.45)
dt  2  ./  .  2/

i.e. Eq. 6.10.3 of Oldham and Spanier (1974)

consisting of steady state (sine function) and transient parts (Maclaurin series in
square brackets). If  ! 0 or 1, the denominators of the transient part approach ˙ 1
(Gamma function of negative integers), the transient part reduces to 0, and a sine or
cosine function, respectively, remains.
By replacing the lower limit of the inverse operation of the Riemann–Liouville
fractional integration by –1, which equals the inverse operation of the Weyl integral
(Weyl 1917), we obtain the steady state part of the fractional derivative of Eq. (6.44)

 D "0 E . C 1/ .2 f / sin 2 f t C  (6.46)

2
Comparing Eq. (6.7) with the steady state Eq. (6.46), it becomes evident that the
phase angle ı is

ıD (6.47)
2
the loss tangent is

tan ı D tan  (6.48)

2
and the peak stress  0 is

0 D "0 E . C 1/ .2 f / (6.49)

The loss tangent depends solely on , independent of E and f. If  ! 1, both tan ı

and  0 ! 1, as both tan(0.5 ) and (0) are infinite. Thus, 0   < 1.
The peak stress  0 increases non-linearly with f and linearly with E and f  . The
peak stress  0 decreases non-linearly with , reaching a limit of "0 E, i.e. a Hookean
spring, if  ! 0. If  ! 1,  0 reaches infinity.
As the maximal strain rate "P0 of a sinusoidal strain function equals "0 2 f, the
strain rate dependency of  0 of Eq. (6.49) is given by

E . C 1/ E . C 1/ "0 1 

0 D "0 1
.2 f / .2 f /1 D "P0 1
D "P0 E . C 1/ "0 1 (6.50)
.2 f / "P0
6 The Loss Tangent of Visco-Elastic Models 149

In contrast to the fractional derivative approach shown above, Lakes (2009)

solved the loss tangent of the power model from the ratio of the constitutive
equations of loss to storage modulus.

6.2.3 The Logarithmic Law Model

The logarithmic law model is characterised by a logarithmic decay of stress  with

time t:

D E   log.t / (6.51)
"c

where “log” denotes the natural logarithm. Stress  is normalised to the constant
strain "c applied by the Heaviside function H(t) of Eq. (6.12).
Taking Laplace transform of Eq. (6.51) yields
 
E ” log s
b
 D "c  "c    (6.52)
s s s

where ” denotes the Euler–Mascheroni constant, i.e. 0.577215665 : : : (Fuss 2008b,

2012).
By substituting Eq. (6.13) into Eq. (6.52), we obtain the constitutive equation of
the logarithmic law of non-linear visco-elasticity (Fuss 2008b, 2012):

b
 Db
"E Cb
" .” C log s/ (6.53)

By substituting the sinusoidal strain of Eq. (6.16) into the constitutive Eq. (6.53),
we obtain

2"0 f
b
D .E C ” C  log s/ (6.54)
s 2 C .2f /2

after rearranging
  
2f ” log s s2f
b
 D "0 E  "0    (6.55)
s 2 C .2f /2 s s s 2 C .2f /2

and taking inverse Laplace transform, we obtain


 D "0 E sin .2f t /  "0 2f Œlog t Œcos .2f t / (6.56)

where * denotes a convolution.

150 F.K. Fuss

Applying the convolution integral, the convolution log (t) * cos (!t), where
! D 2 f, is solved accordingly:
Z t
Œlog t Œcos .!t / D log . / cos Œ! .t  / d (6.57)
0

where  is the dummy variable of the convolution integral.

Decomposition of cos(!t – ! ) according to the addition rules and subsequent
partial integration yields:

Œlog t Œcos .!t / D! 1 ŒCi .! / sin .!t / Si .! / cos .!t /  log ./ sin .!t !/ t0
(6.58)

where Ci and Si denote cosine and sine integrals respectively (definition of Ci and
Si according to Abramowitz and Stegun 1972).
Solving Eq. (6.58) from 0 to t:
h

Œlog t Œcos .!t / D ! 1 Ci .!t / sin .!t /  Si .!t / cos .!t /  log.t / sin.0/
i
Ci .!0/ sin .!t / C Si .!0/ cos .!t / C log.0/ sin .!t /
(6.59)

Apparently, Eq. (6.59) contains an indeterminate form, as both log(0) and Ci(0) are
1, and thus the term log(0) sin(!t) – Ci(!0) sin(!t), or sin(!t) [log(0) – Ci(0)],
delivers sin(!t) (1–1).
However,

Cin.t / D ” C log.t /  Ci.t / (6.60)

where Cin denotes an alternative cosine integral (definition of Cin according to

Schelkunoff 1944). As Cin(0) D 0, log(0) – Ci(0) D –”.
Yet, the argument of the cosine integral in Eq. (6.58) is ! , in contrast to the one
of the natural logarithm, which is just  . Thus we have to consider the multiplier !.
This multiplier leads to a convergence constant other than simply –”, if t ! 0.

Ci .!t / D ” C log .!t /  Cin .!t / D ” C log.t / C log .!/  Cin .!t / (6.61)

As Cin(0) D 0,

log.0/  Ci.0/ D ”  log .!/ (6.62)

Thus,

lim Œlog.t /  Ci .!t / D  log .!/  ” (6.63)

t!0
6 The Loss Tangent of Visco-Elastic Models 151

Equation (6.59) can now be solved, considering that Si(0) D 0

Œlog t  Œcos .!t / D! 1 fCi .!t / sin .!t / Si .!t / cos .!t /  sin .!t / Œlog .!/C” g
(6.64)
The solution of Eq. (6.56) after taking inverse Laplace follows

 D"0 E sin .!t / "0  fCi .!t / sin .!t / Si .!t / cos .!t /  sin .!t / Œ.log .!/ C”/ g
(6.65)

The steady state of Eq. (6.65) sets in at large times, or t ! 1. The transient part of
Eqs. (6.64) and (6.65) comprises of the cosine and sine integrals. When considering
the values at infinity of cosine and sine integrals, which are 0 and 0.5 respectively,
the convolution of Eqs. (6.64) and (6.65) yields at large times (steady state equation)
n  o
lim Œlog.t / cos .!t / D ! 1  cos .!t /  sin .!t / Œlog .!/ C ” (6.66)
t!1 2
and

 D "0 E sin .!t / C "0  cos .!t / C "0  sin .!t / Œlog .!/ C ” (6.67)
2
respectively.
After rearranging according to the procedure applied for determining the loss
tangent of the standard linear solid, we obtain
 
tan ı D (6.68)
2 E C  Œlog .2f / C ”

and
q
2
0 D "0 0:252  2 C ŒE C  log .2f / C ” (6.69)

As the maximal strain rate "P0 of a sinusoidal strain function equals "0 2 f, the strain
rate dependency of  0 after rearranging Eq. (6.69) is given by
s
   2
"P0
0 D "0 0:252  2 C E C  log C ” (6.70)
"0

The stress amplitude  0 increases with E, , and f. This fact is obvious, as E

is the strain rate independent elasticity parameter, i.e. the modulus or stiffness;
the viscosity parameter  is linked to the strain rate, i.e. at a given strain rate,
 increases with ; and the frequency f is linearly proportional to the strain rate
applied periodically to the model, i.e. at a given ,  increases with f.
If  ! 0,  0 ! "0 E, the stress equation of a Hookean spring. If  ! 1,  0 ! 1.
152 F.K. Fuss

η/E = 0.5 η/E = 1

100

η = 10
10
stress amplitude σ0

η=1
1

η = 0.1
0.1
1E-005 0.0001 0.001 0.01 0.1 1 10
frequency (Hz)

Fig. 6.7 Peak stress ¢ 0 against frequency f ; ˜: viscosity constant; E: velocity independent
elasticity parameter

If f ! 0 or 1,  0 ! 1. This means that  0 has a local minimum at a certain

frequency f0 (Fig. 6.7). As the first frequency derivative of Eq. (6.69) contains the
arguments f, log(2 f ), and log2 (2 f ), there is no closed-form analytical solution,
and the frequency, at which  0 is at a minimum, can only be obtained numerically.
Figure 6.7 shows that  0 increases with , and f0 increases with the ratio /E. The
ratio /E is identical to the viscosity of a log law model, not to be confused with the
viscosity constant .
The peak stress  0 increases almost linearly with E.
If E ! 1,  0 ! 1. If E ! 0 (purely viscous material)
q
2
lim 0 D "0  0:25 2 C Œlog .2f / C ” (6.71)
E!0

and  0 increases with  and f.

The loss tangent, Eq. (6.68), reveals that the phase angle ı depends on E, , and f.
Increasing the elasticity parameter E reduces the phase angle ı, thereby asymp-
toting to 0, when E approaches infinity. This result is evident, as a perfectly rigid
solid (E D 1) does not deform and thus there are no losses (ı D 0).
Reducing E increases tan ı. If E approaches zero (purely viscous material), tan ı
reaches a limit of
6 The Loss Tangent of Visco-Elastic Models 153

 1
lim .tan ı/ D (6.72)
E!0 2 log .2f / C ”

a constant, which is a function of f but independent of . The magnitude of the

viscosity  does not matter in this case, as the material is anyway purely viscous.
The phase angle ı becomes 0.5 if

1
f D (6.73)
2 e”

Thus, if E D 0 and the loss tangent is positive, f must be  0.08936 : : : Hz.

If f < 0.08936 : : : Hz, ı > 0.5 , and E0 (storage modulus) and tan ı are negative,
which is impossible, as the stored energy is zero (as E D 0), and the energy
dissipated cannot be negative.
If the frequency f ! 1 or 0, the loss tangent ! 0 in both cases, and the phase
angle ı ! 0 or , respectively. Thus, ı crosses /2 at a certain frequency f0 .
Equating the denominator of Eq. (3.20) with zero yields f0 at which ı D 0.5 :
E
e  ”
f0 D (6.74)
2
The ratio E/ in Eq. (6.68) equals the reciprocal value of the viscosity of a log law
model. The higher the viscosity, the smaller is the ratio E/. Figure 6.8 shows E/
against the frequency, the f0 curve at which ı D 0.5 , and the E/ and frequency
ranges at which ı is < or > 0.5 . If E/ ! 0, f0 approaches the value of Eq. (6.73),
which is f0 D 0.089359 Hz, i.e. a cycle period of 11.2 s. From Fig. 6.8, the loss
tangent, and this the storage modulus, is negative at small E/ (high viscosity) and
small frequencies (large cycle periods with small strain rates). This is in accordance
with a negative (elastic) modulus of the log law model at very small strain rates
(Eq. 5.73 of Fuss 2012).
The reciprocal of Eq. (6.68)
  
2 E
cot ı D C log .2f / C ” (6.75)
 

shows that the higher E/, the higher is cot ı. Thus, the higher /E, the higher is
tan ı. This explains that the viscosity of a log law model corresponds to the ratio
/E, and not to the viscosity constant . Figure 6.9 shows that the higher /E, the
higher is the phase angle ı. The viscosity constant  alone has no influence on ı.
154 F.K. Fuss

100

10
tan δ < π/2

f0 1
1 f0 =
2π eγ
E/η

0.1

0.01
tan δ > π/2

0.001

0.0001
0 0.02 0.04 0.06 0.08 0.1
frequency (Hz)

Fig. 6.8 Ratio of E/˜ against frequency f ; ˜: viscosity constant; E: velocity independent elasticity
parameter; tan •: loss tangent; f0 : frequency at which tan • D /2

2.5
E = 10
η=1
2
phase angle δ (rad)

η/E = 0.1

δ = π/2
1.5
E = 2000 E=1
η = 100 η = 10
η/E = 0.05 η/E = 10
1

0.5

–20 –10 0 10 20
log frequency

Fig. 6.9 Phase angle • against log frequency; ˜: viscosity constant; E: velocity independent
elasticity parameter
6 The Loss Tangent of Visco-Elastic Models 155

6.3 Summary

Material models are always simplified descriptions to assist calculations and assess-
ment of mechanical behaviour. This means that a material does not necessarily have
to follow the behaviour of its model. For example, the negative storage modulus of
a log law model at small E/ and small frequencies may well be a mathematical
phenomenon but does not necessarily exist in reality. The standard linear solid is
certainly oversimplified with three elements (2 springs, 1 damper) and is usually
expanded to more elements for better linear material characterisation (Prony series;
Wiechert model; Wiechert 1889, 1893; Fuss 2012).

6.3.1 Loss Tangent and Viscosity

SLS: tan ı and ı depend on E, , and f ; but at the same f, tan ı and ı at the same R1
depend only on R2 (Eqs. 6.35 and 6.36), i.e. the ratio of moduli of series to parallel
spring, but not on the viscosity constant .
Power law model: tan ı and ı depend on  only; 0   < 1.
Log law model: tan ı and ı depend on E, , and f ; but at the same f, larger E/
have larger tan ı and ı.

6.3.2 Relationship Between Frequency and Viscosity Constant

SLS: f and  are linked together and always appear as the product f. Equations
(6.26) and (6.27).
Power law model:  has no relationship with f in tan ı (as f does not influence
tan ı), whereas for  0 ,  appears in the gamma function, and is the exponent of f,
i.e. f .
Log law model: the viscosity constant appears as a stand alone , and as the
product of  and log 2 f.

6.3.3 Transient and Steady State Parts

SLS: transient part: exponential function; steady state part: sine and cosine
functions.
Power law model: transient part: Maclaurin series; steady state part: sine function
with  /2 phase shift (resulting in sine and cosine functions after applying addition
rules).
156 F.K. Fuss

Log law model: transient part: cosine and sine integrals; steady state part: sine
and cosine functions.

6.3.4 Negative Storage Modulus if tan ı > /2

SLS: tan ı < /2

Power law model: tan ı < /2
Log law model: tan ı can be > /2 at small E/ (high viscosity) and small
frequencies (large cycle periods with small strain rates).

Key Symbols

Ci Cosine integral function

Cin Alternative cosine integral function
cos Cosine function
cot Cotangent function
d Differential operator
E Modulus (velocity independent elasticity parameter)
E1 , E2 Moduli of springs of a standard linear solid
E0 Storage modulus
E00 Loss modulus
E* Complex modulus
jE*j Dynamic modulus
e Exponential function
f Frequency
H Heaviside (unit step) function
p
i 1
lim Limit
log Natural logarithm
R1 , R2 Parameter ratios of a standard linear solid
s Complex variable of transformed functions
Si Sine integral function
sin Sine function
t Time
tan Tangent function
tan ı Loss tangent
 Gamma function
Euler–Mascheroni constant (0.577215665)
ı Phase shift angle
" Strain
"c Constant strain
6 The Loss Tangent of Visco-Elastic Models 157

"0 Amplitude of strain, peak strain

b
" Transformed strain
"P0 Peak strain rate
 Viscosity constant
pi (3.14159 : : : )
 Stress
b
 Transformed stress
0 Amplitude of stress, peak stress
 Dummy variable of convolution integral
! Angular frequency ! D 2 f
* Convolution operator
1 Infinity

References

Abramowitz M, Stegun IA (1972) Handbook of mathematical functions, 9th edn. Dover

Publishing, Mineola
Findley WN, Lai JS, Onaran K (1989) Creep and relaxation of nonlinear viscoelastic materials.
Dover Publications, New York
Fuss FK (2008a) Cricket balls: construction, non-linear visco-elastic properties, quality control and
implications for the game. Sports Technol 1(1):41–55. doi:10.1002/jst.8
Fuss FK (2008b) Logarithmic visco-elastic impact modelling of golf balls. In: Estivalet M, Brisson
B (eds) The engineering of sport 7. Springer, Paris, pp 45–51
Fuss FK (2009) The collapse timing and rate of closed cell foams: revealing a conceptual
misunderstanding of foam mechanics. In: Alam F, Smith LV, Subic A, Fuss FK, Ujihashi S
(eds) The impact of technology on sport III. RMIT Press, Melbourne, pp 659–664
Fuss FK (2012) Nonlinear visco-elastic materials: stress relaxation and strain rate dependency. In:
Dai L, Jazar RN (eds) Nonlinear approaches in engineering applications. Springer, New York,
pp 135–170. doi:10.1007/978-1-4614-1469-8_5
Lakes R (2009) Viscoelastic materials. Cambridge University Press, Cambridge
Oldham KB, Spanier J (1974) The fractional calculus. Academic, New York
Provenzano P, Lakes R, Keenan T, Vanderby R (2001) Nonlinear ligament viscoelasticity. Ann
Biomed Eng 29(10):908–14
Schelkunoff SA (1944) Proposed symbols for the modified cosine and exponential integral. Q Appl
Math 2:90
Weyl H (1917) Bemerkungen zum Begriff des Differentialquotienten gebrochener Ordnung.
Vierteljahrsschrift der Naturforschenden Gesellschaft in Zürich 62:296–302
Wiechert E (1889) Ueber elastische Nachwirkung. Dissertation, Königsberg University,
Königsberg
Wiechert E (1893) Gesetze der elastischen Nachwirkung für constante Temperatur. Ann Phys
286:335–348, 546–570
 Part II
Practical Nonlinearity
Chapter 7
Optimization of Hood Design to Minimise
Pedestrian Head Injury in Impact

Revathi Krishnamoorthy, Monir Takla, Aleksandar Subic, and Derek Scott

Keywords Pedestrian protection • Head impact • Finite element analysis • Hood

panels • Crash safety

7.1 Introduction

The term pedestrian generally refers to a person walking, running, skating, skate-
boarding and commuting using similar devices.
There is a recent trend for urban design to provide for more walkable com-
munities. Present-day awareness of health, fitness and greenhouse gas emissions
has increased the popularity of cycling and walking in urban areas. Therefore, the
probability of pedestrian accidents has also increased.
Automobiles have evolved to be a successful medium for transportation due to
substantial inventions and developments in various countries. The concept of large-
scale production-line manufacturing enabled the affordable pricing of automobiles,
which has increased the demand for them globally.
In 1900, the United States of America was the only country manufacturing cars
and built only 4,192 passenger cars (Elert 2001). There were no buses or trucks at
that time. As of 2010, there are more than 600 million passenger cars worldwide
(Worldometers 2010). These numbers are increasing rapidly. It is estimated that, if
the present trends continue, the number of cars in the world will double by 2030.
As the demand for automobiles has increased, automotive design is improving
every day in many aspects of comfort, technology, efficiency, performance and

R. Krishnamoorthy () • M. Takla • A. Subic • D. Scott

School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University,
Bundoora, PO Box 71, Melbourne, VIC 3083, Australia
e-mail: [email protected]; [email protected]; [email protected];
[email protected]

© Springer International Publishing Switzerland 2015 161

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__7
162 R. Krishnamoorthy et al.

Fig. 7.1 Unsafe conditions for pedestrians (World Health Organization 2011)

safety. However, as many countries and manufacturers have been participating in

the manufacture of cars, there are significant differences in the quality and safety
performances offered by the vehicles.
The World Health Organization photograph (Fig. 7.1) portrays the unsafe
conditions for pedestrians, who risk their lives in order to fulfil their day-to-day
responsibilities.
In 1965, Ryan and McLean (1965) described the sequence of events when a
passenger car collides with an adult pedestrian standing erect.
Those sequences are as follows:
• Contact between the bumper and the lower legs of the pedestrian
• Contact between the leading edge of the hood and upper legs/pelvis of the
pedestrian
• Contact between the pedestrian’s head/upper torso and the top surface of the
hood/windscreen/windscreen frame (Fig. 7.2)
• Contact between the pedestrian and the ground.
They also speculated that it would be possible to minimise injuries in a
Pedestrian–Passenger Car Front–end Collision (PPCFC) by changing the frontal
shape of a car. Their study however did not suggest any design solutions, as they did
not have enough data.
In those days, the belief was that the only way to reduce pedestrian fatalities and
injuries was to prevent pedestrian–vehicle collisions. Consideration of modification
of vehicle design for pedestrian protection was not an option at that time. From this
sequence of events, it can be stated that typically the colliding vehicle runs under
the pedestrian and the severity of injuries vastly depend on the vehicle shape and
certain characteristics such as energy absorption.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 163

Fig. 7.2 Kinematics of a pedestrian in PPCFC (Synaptic analysis consulting group inc)

7.2 Background Research and Benchmarking

A collision between a pedestrian and a passenger car causes injuries to the whole
body. Recovery from head injuries however takes substantial time and these injuries
are more likely to be fatal. In 2003, Iskander Farooq and his team compiled the
data from the International Road Traffic and Accident Database (IRTAD) and the
German In-Depth Accident Study (GIDAS). They found that 62 % of pedestrian
fatalities occur due to head injuries.
The statistics show that about 92 % of pedestrian fatalities in the US are due to
the impact with the front-end of a car (National highway traffic safety administration
2009) as presented in Fig. 7.3. The pedestrian’s head contact with the vehicle front-
end or the ground could cause significant head injuries. However, studies show that
head impact with the vehicle front-end is likely to cause more damage than head
impact with the ground.
Other studies on injuries in children pedestrians show that the injuries to different
regions of the body vary based on a child’s age. Head and chest impacts are the
sources of fatal injuries for younger children. Contact between the pelvis and the
abdomen with the leading edge of the hood can cause fatal injuries for an older age
group.
164 R. Krishnamoorthy et al.

Front
Right side
Left side
Rear
Other/unknown

Fig. 7.3 Direction of impact for pedestrian fatalities due to collision with a passenger car in the
US (National highway traffic safety administration 2009)

Table 7.1 Distribution of pedestrian injury AIS 2–6 (Mizuno 2003)

Body region USA (%) Germany (%) Japan (%) Australia (%) Total (%)
Head 32:7 29:9 28:9 39:3 31:4
Face 3:7 5:2 2:2 3:7 4:2
Neck 0:0 1:7 4:7 3:1 1:4
Chest 9:4 11:7 8:6 10:4 10:3
Abdomen 7:7 3:4 4:7 4:9 5:4
Pelvis 5:3 7:9 4:4 4:9 6:3
Arms 7:9 8:2 9:2 8:0 8:2
Legs 33:3 31:6 37:2 25:8 32:6
Unknown 0:0 0:4 0:0 0:0 0:2
Total 100 100 100 100 100

According to the International Harm Reduction Association (IHRA) accident

data, head and leg injuries account for two thirds of pedestrian injuries. Approxi-
mately 84 % of AIS 2–6 injuries are caused by pedestrian contact with the vehicle
front-end. In Australia (Table 7.1), 39.3 % of AIS 2–6 injuries occur due to head
impact.
Head impact is the most prominent factor in pedestrian fatalities and injuries.
Thus, it represents a key hurdle in the pedestrian protection assessment criteria. Due
to these two critical features, it was decided to focus on improving vehicle design
to reduce head injuries and to develop a methodology for optimising hood panels
of passenger cars to ensure that the pedestrian Head Injury Criterion (HIC) falls
below the threshold values specified by both the GTR-9 and the consumer metric,
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 165

the Australasian New Car Assessment Program (ANCAP). This chapter presents the
development of a hood configuration that provides robust and homogeneous HIC for
different impact positions in the central area of the hood of a large sedan, taking into
consideration of the limited space available for deformation.

7.3 Vehicle Design Requirements

7.3.1 Regulatory Vehicle Design Requirement for Pedestrian

Protection

The GTR-9 has adopted the test procedures developed by the IHRA. Active and
passive components are included so as to ensure the same level of pedestrian
protection as 2003/102/EC-Phase II. The active component mandates fitting of
Brake Assist, a system designed to sense an emergency braking situation and
assist the driver in achieving the maximum possible deceleration in the prevailing
conditions, thus reducing the braking distance. The passive Component includes
a series of subsystem tests where the resulting injury levels must be below the
prescribed limits. This component of GTR-9 requires a careful vehicle design that
reduces the probability of severe head and lower leg injury to pedestrians in a
collision with the car.
Europe, Japan, China, United States of America, India, Korea and Australia are
the few among the many countries plan to adopt these regulations as vehicle design
regulatory requirements for pedestrian protection. Negotiations are in progress with
automakers and governments globally regarding the timing for enforcing GTR-9.
The GTR-9 test protocol required four subsystem tests to evaluate the vehicle
design for protection of pedestrians in a collision with a passenger car. They are:
• Lower leg form fired at 40 km h to bumper system
• Upper leg form fired at 20–40 km h (depending on vehicle shape) to the hood
leading edge
• Child head form fired at 35 km h to the hood top, at an angle of 50ı to the
horizontal plane
• Adult head form fired onto the hood top at an angle of 65ı to the horizontal plane,
at 35 km h.

Head Impact

The total head impact area is determined in a similar manner to the European
Commission Directive. Two thirds of the total impact area zoned as ‘HIC1000’ and
one-third zoned as ‘HIC1700’. The HIC1000 zone must contain at least half of the
child impact area. This regulation tests the protection offered to both the child and
adult head in this impact area. The impact area for the child head is between 1000
166 R. Krishnamoorthy et al.

WAD and 1700 WAD. The adult head impact area is between 1700 WAD and 2100
WAD. The impact area does not include the windscreen.

Lower Leg Impact

The bottom face of the lower leg impactor is positioned 25 mm above the ground
level. The area between the two bumper corners is the lower leg impact area. The
resulting injury values for the lower leg within the lower leg impact area must be
below a set of prescribed limits in the evaluation of vehicle design. The bounds for
the injury values are:
• Tibia acceleration is less than or equal to 170 g
• Knee-bending angle is less than or equal to 19ı
• Knee shear displacement is less than or equal to 6 mm.

Upper Leg Impact

The upper leg impact is a guided impact and the impact energy, impact angle and
impact velocity are calculated based on the Hood Leading Edge and Upper Bumper
Reference Line positions. Currently, there are no legal limits for this test and it is
conducted for monitoring purposes only.

7.3.2 Consumer Metric Vehicle Design Requirement

for Pedestrian Protection

The various new car assessment programs, usually made up of a consortium

of government and non-government safety organisations, have established testing
programs in many countries to evaluate new passenger vehicles for safety perfor-
mance, including pedestrian protection performance, and published the results for
consumers.
The pedestrian protection performance rating reported by Euro NCAP is one of
the most influential consumer metrics in the European market. The EEVC WG17
pedestrian protection test procedures serve as a base for these test and assessment
procedures. The Euro NCAP protocol version 5.3.1, adopted from February 2010,
has harmonised the test procedures with the GTR-9. Euro NCAP also proposes four
tests through vehicle subsystems (Fig. 7.4) similar to those of the GTR-9.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 167

Fig. 7.4 Pedestrian protection technical requirements (Euro NCAP 2010)

7.4 Evaluating Pedestrian Protection Offered by a Vehicle

In 1989, EEVC WG10 (Hardy et al. 2007) proposed the impactor test method for
vehicle pedestrian impacts. They developed four different types of subsystem test.
As stated earlier, so far all the regulatory requirements and consumer metrics
tests that evaluate pedestrian protection use this subsystem test only (Fig. 7.5).
The Institut National de Recherche sur les Transports et leur Sécurité (INRETS)
created the original leg form impactor (Hardy et al. 2007). However, EEVC WG
10 and EEVC WG 17 accepted the lower leg impactor developed by Transport
Research Laboratory (TRL) (Transport Research Laboratory 2012) to be used in
EEVC pedestrian protection testing, due to its high level of performance and
repeatability. These TRL-PLI impactors represent the 50th percentile male leg
designed for right side impact testing. Accelerometers and potentiometer in the leg
form impactor collect the acceleration and the relative rotation to assess the severity
of injuries. The TRL-PLI has a rigid femur and tibia covered with foam to represent
soft tissues and skin (Fig. 7.6).
In 2000, to improve the accuracy of the leg form, JAMA and the Japan
Automobile Research Institute (JARI) initiated the development of a Biofidelic
168 R. Krishnamoorthy et al.

Fig. 7.5 Subsystem test method (Hardy et al. 2007)

Fig. 7.6 Comparison between TRL-PLI and Flex-PLI (World Health Organization 2010)
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 169

Fig. 7.7 TRL-PLI lower leg impactor

Fig. 7.8 Leg impact and head impact test machines (Centre of automotive safety research 2007)

Flexible Pedestrian Leg form Impactor, Flex-PLI (Fig. 7.7). This impactor uses four
femur joints and five tibia joints to improve the dynamic response.
Flex-TEG, a technical evaluation group consisting of government and industry
members, is evaluating the possibility of using the Flex-PLI impactor for theGTR-9.
The linear guidance propulsion system launches the lower leg impactor in a free
flight mode when it has attained steady desirable velocity. This test simulates the
human lower leg (tibia, knee joint and femur) in a passenger car front-end impact
with a pedestrian.
In 1995, Germany’s federal highway research institute (BASt) conducted upper
leg form tests. Later TRL improved this upper leg form impactor. The EEVC
accepted this upper leg form impactor in its tests, even though JARI reported that
the measured forces are still lower than the inertia forces. During the test, a linear
guided launcher fires the upper leg impactor onto a static vehicle (Fig. 7.8). This
test simulates the upper leg covered with foam to replicate flesh and skin and rotates
around a friction-loaded pivot to mimic femur and pelvis fractures.
Originally, production of the head form impactors used a plastic sphere covered
by a rubber skin. EEVC WG17 updated the head form specifications to an
aluminium sphere covered with a 12 mm PVC skin without changing the outer
170 R. Krishnamoorthy et al.

diameter and mass for continuity of measurement. The head impactor is launched
in the same manner as the lower leg impactor. The head form contains a tri-axial
accelerometer that is used to measure the resultant acceleration throughout the
impact. The HIC value is calculated using the measured resultant acceleration.
In 2005, a team from the University of Virginia and General Motors Corporation
(Kam et al. 2005) conducted a study to develop full-scale pedestrian impact tests
using Post Mortem Human Surrogates (PMHS). This study was conducted utilising
existing test procedures, real-world data and MADYMO simulations to create an
experimental test system with full-scale pedestrians. In this study, the PMHS was
positioned laterally in mid-gait stance with the struck leg extended to the rear and
both hands extended to the front. The PMHS was attached to an electromagnetic
release mechanism, which was triggered to release the PMHS just before impact.
After the impact, a catcher mechanism would absorb the kinetic energy from the
PMHS to avoid any secondary injuries because of contact with the ground. This
study did not lead to the development of a full-scale pedestrian dummy.
In contrast, Honda developed a crash test dummy to understand the kinematics
of vehicle pedestrian impacts in 1998 with a second-generation released in 2000,
the POLAR II. This dummy gathers measurements from the head, cervical spine,
thorax, abdomen, pelvis, femur, knee and tibia. In 2008, Honda introduced the third-
generation pedestrian dummy, POLAR III (Fig. 7.9). This dummy has an improved
ability to evaluate bone fractures in the lower back and upper leg compared to the
POLAR II. Lower back and upper leg injuries are common in pedestrian impact
with higher vehicles like SUVs.
At present, both NCAP and regulatory tests use subsystem testing due to their
simplicity, repeatability and the ability to predict injury values accurate enough
for pedestrian protection. Since the number of tests required to assess pedestrian
protection is large, testing with full-scale pedestrian dummies would be expensive
and time consuming. The author is not aware of any current plans to examine
pedestrian kinematics using full-scale pedestrian impact tests.

7.5 Problem Identification

7.5.1 Peak Acceleration

Numerical tests were conducted in this work to obtain accelerations, time histories
and impactor displacements in x, y and z directions. The resultant accelerations
calculated from these numerical tests were used in calculating the HIC values. As
explained by Krishnamoorthy et al. (2013) in Advanced Materials Research, the
HIC calculation involves an exponent of 2.5 applied to the average head acceleration
measured over a window of 15 ms. The result of this formulation is that small
changes in the average acceleration can lead to large changes in the final HIC value.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 171

Fig. 7.9 POLAR III pedestrian dummy (Honda Worldwide 2008)

The average acceleration of any given head acceleration curve can be

increased by:
• Simply increasing the magnitude of a single peak acceleration
• Increasing the duration or width of a single peak
• Increasing the time difference between the first and an adjacent peak
• Increasing the magnitude of the acceleration of the first peak and the adjacent
peaks.
The accelerations calculated from the numerical tests conducted in this work
are shown in Figs. 7.10 and 7.11. It is evident from the curves that the HIC
value increases with the magnitude of resultant acceleration and increases with
the recording period for similar peak acceleration. Therefore, a low magnitude and
small duration of peak acceleration are required to keep the severity of head injuries
less than the limits specified by the GTR-9 and ANCAP.
Decreasing the structural stiffness of the hood panels decreases the magnitude
of acceleration and in turn increases the intrusion of the hood panels, otherwise
known as the deformation space. It is this interplay between the measured average
acceleration (and hence HIC) and the structural characteristics of the hood that
requires careful development if we are to minimise our requirement for the highly
valuable deformation space. Thus, optimisation of the structural stiffness of the hood
panels is essential to meet the requirements for head impact regulations.
172 R. Krishnamoorthy et al.

Fig. 7.10 Comparison of peak acceleration magnitude

Fig. 7.11 Comparison of peak acceleration duration

7.5.2 Deformation Space

Pedestrian protection requires a soft, energy-absorbing hood assembly, whereas

durability, noise and vibration require a hood assembly with higher structural
stiffness. The properties of the direct and indirect contact surfaces of the head in
a crash influence the severity of head injuries. The indirect contact surfaces are the
surfaces under the hood. The three important parameters that influence the severity
of pedestrian head impact injuries are the available deformation space, structural
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 173

stiffness and structural inertia. In this instance, the structural inertia is the resistance
given by the hood panel to changing its state to the changes in velocity and is
proportional to its mass.
Among these parameters, the available deformation space between the hood
assembly and the hard components within the engine bay is critical. The intensity
of the contact between the hood assembly and the under hood components with
high structural stiffness (secondary impact) increases the severity of pedestrian head
injuries. The deformation space required to absorb the energy varies with the head
impact mass, impact speed and impact angle. Intensity of the secondary impact can
be reduced either by increasing the structural stiffness and structural inertia of the
inner and outer hood panels or by increasing the deformation space.
Increasing the structural stiffness of the hood increases the magnitude of accel-
eration during impact due to reducing the hood deformation. Therefore, excessive
stiffness of the hood causes severe acceleration, which in turn increases the severity
of head injuries. Accordingly, increasing the structural stiffness to reduce the
intensity of the secondary impact is only possible up to a point, beyond which the
severity of injuries due to the contact between the hood assembly and the head
impactor (primary impact) becomes excessive. The simulations from the numerical
tests conducted in this work with increased structural stiffness for the inner and
outer hood panels are shown in Fig. 7.12. It shows that the deformation of the hood
panels is reduced whereas; the structural inertia of the panels has increased.
The resultant acceleration calculated for the simulation shown in Fig. 7.12 is
presented in Fig. 7.13, which exhibits a dominant initial peak resulting from high
inertial loading due to the structural inertia of the inner and outer hood panels.
Similarly, the simulation and resultant acceleration from the numerical tests
conducted in this work with the lack of deformation space are presented in Figs. 7.14
and 7.15. These figures show that the lack of deformation space leads to direct
contact of the head form with the hard components approximately 12 ms from the
start of the impact. This causes severe acceleration, hence increasing the risk of head
injury.

Fig. 7.12 Primary impact of head with hood panels with excessive stiffness
174 R. Krishnamoorthy et al.

Fig. 7.13 Effect of inertial loading of the head form on HIC value

Fig. 7.14 Secondary impact of head with hard components within engine bay

Fig. 7.15 Waveform with secondary impact

While increasing the deformation space improves pedestrian protection, it may

conflict with other vehicle design requirements. For example, the positioning of
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 175

the hard components, such as the engine and the suspension, should meet ground
clearance requirements. In addition, the relative height difference between the hood
and the driver has to meet visibility requirements. Therefore, when the hood is
elevated to meet the deformation space requirements for pedestrian protection,
the driver also needs to be elevated to meet the visibility requirements. Thus,
vehicle manufacturers need to increase the height of the car for occupant comfort,
ergonomics and safety, which will compromise aerodynamics, vehicle stability and
fuel economy. It might also have adverse effects on vehicle styling.
Therefore, in order to reduce the pedestrian head injury risk, it is important
to optimise the hood assembly by balancing its structural stiffness for minimal
deformation or risk a series of far reaching flow on effects into a whole range of
areas such as engine bay packaging, occupant comfort and visibility, styling and
even aerodynamics.

7.5.3 Optimal Waveform

A waveform in this study is the shape of a curve describing the relationship between
the resultant acceleration and time. Meeting the requirements of the HIC value with
the lowest deformation space requires balancing hood stiffness and deformation
space. The parameters that affect the HIC value are impact velocity, available
deformation space, magnitude of peak acceleration and pulse duration. The resultant
acceleration and pulse duration are the key factors that affect the HIC value. These
factors are influenced by the impact velocity and the available deformation space.
In regulatory and consumer metrics tests, the impact velocity is given as a constant.
Therefore, the main factors that affect the HIC value, which is calculated from the
wave form, are the structural stiffness and the deformation space.
In 1995, Lim et al. explored options to develop a methodology for estimating
the minimum deformation space required in an impact to a pedestrian head. They
used generic waveforms and idealised waveforms of square waves, sine waves and
half-sine waves in formulating equations for deformation space requirements.
They used the ‘area ratio’ method to determine the waveform efficiency. In
this method, the ratios of areas under the acceleration–intrusion curves, shown
in Fig. 7.16, are used to determine the efficiency of the waveform, as shown in
Eq. (7.1).

A1
D (7.1)
A2

In which ˜ D efficiency, A1 D area enclosed by the acceleration–displacement curve

of the given waveform and the horizontal plane, A2 D area of the shape that
encloses the geometrical shape of the acceleration–displacement curve of the given
waveform.
176 R. Krishnamoorthy et al.

Deceleration vs. Displacement

120

100
Deceleration [g]
80

60

40

20

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
Displacement [in]

Fig. 7.16 Waveform efficiency (˜) of a half-sine waveform (Lim et al. 1995)

With this method, the square waveform was determined as the most efficient
waveform, with 100 % efficiency (Fig. 7.16). Lim and his team calculated the
efficiency of the other waveforms by benchmarking against the square waveform.
Thus, the waveform efficiency of a sine waveform was determined as 63.66 % and
of a half-sine waveform as 50 %.
In 2007, Wu and Beaudet (2007) challenged the optimality of aiming at the
square waveform to achieve the minimal deformation space required to obtain a
given HIC value. They derived a realistic optimal waveform that is more practical
based on the theoretical optimal waveform, which is WSTC.
As shown in Eq. (7.2) below, HIC value calculation is the average of head
acceleration a(t) over an interval t1 to t2 , followed by a(t) is raised to the 2.5th power
and multiplied by the interval. The condition for selection of the interval is that it
should provide the maximum result within the 15 ms window:
82 32:5 9
ˆ
< Zt2 >
=
4 1 5
HIC D max a.t /dt .t2  t1 / (7.2)
:̂ t2  t1 >
;
t1

To achieve the maximum result for HIC value for a given magnitude of peak
acceleration, the acceleration between t1 and t2 should be equal, which produces
a square waveform. Therefore, assuming a(t1 ) D a(t2 ), Wu and Beaudet derived the
formula to calculate the maximum intrusion of the hood, which produces a square
waveform in a pedestrian head impact as shown in Eq. (7.3):
8
v0 3
d D 11:12  103 2
(7.3)
HIC 3
Where d D maximum intrusion in metres, v0 D impact velocity in m/s.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 177

Table 7.2 Comparison of deformation space requirements for various

waveforms
Waveform Deformation space in impact direction (mm)
Optimal 50:50
Square 67:32
Sine 79:49
Half-sine 96:38
Isosceles triangle 83:97
Ramp-up triangle 111:94
Ramp-down triangle 55:98

Wu and Beaudet stated that numerous other waveforms show same HIC value
with smaller deformation space requirements. They concluded that the square
waveform is ideal when the criterion is either maximum head acceleration or force.
Since HIC is the criterion for head impact, the function derived is directly related to
WSTC, which proposes that the head can tolerate large head acceleration for a very
short duration and smaller head acceleration for a longer duration. Therefore, they
considered WSTC as the optimal waveform and derived the equation for calculating
the head travel or deformation space as stated in Eq. (7.4):
Z 8
v0 3
dD v  dt D 8:342  103  2
(7.4)
HIC 3
Comparison of the deformation space requirement for a HIC value of 1000 at
40 km/h impact velocity of the optimal waveform and other waveforms reveals
that the optimal waveform requires the least amount of deformation space and the
second best is the ramp-down triangle waveform. Ramp-up triangle waveform in
Table 7.2 refers to a triangular waveform with late peak in which the resultant
acceleration gradually increases over time and drops down over a short period.
Ramp-down triangle waveform is the triangular waveform with early peak in which
the resultant acceleration increases from zero to peak within a short period and
decreases gradually. Wu and Beaudet did not consider rebound in their calculations.
The optimal waveform (WSTC) suggests that head acceleration could be infinite
at time zero, but in reality, this is impossible. Therefore, Wu and Beaudet considered
the optimal waveform as the theoretically optimal waveform. When tp equal
a 2 ms, delay is applied to obtain peak acceleration, the theoretically optimal
waveform becomes realistic and efficient, thus realistic optimal waveform as shown
in Fig. 7.17.
With the realistic optimal waveform, they stated that as design guidance, the
deformation space required to obtain a HIC value less than or equal to 1000 is
60 mm for a child head at 40 km/h impact velocity. For adult head impact, they
suggested that the required deformation space is 12 % more than that required for
child head impact, due to the steeper impact angle.
178 R. Krishnamoorthy et al.

Fig. 7.17 Realistic optimal

waveform for pedestrian head
impact (Wu and Beaudet
2007)

In the research presented in this chapter, the impact performance offered by a

vehicle to the child head has been tested with 60 mm of deformation space as a
starting point. Wu and Beaudet derived the value of the required deformation space
using theoretical calculations ignoring rebound and other parameters that could
influence the head impact performance. In practice, the value of the deformation
space required to meet the compliance threshold homogeneously will be higher than
60 mm.

7.5.4 Influence of Hood Design

The structure and stiffness of the components in a hood assembly (Fig. 7.18) of
a car generally influence the severity of head injury in a pedestrian hood impact.
To reduce the severity of head injuries closer to the edges of the hood, the hood
assembly has to be weaker in this area. This will increase the deflection of the hood
assembly in the middle, thus increasing the severity of head injury closer to the
edges of the hood. The size of the hood also has an effect on the probability of
severe head injury because, as the hood size is increased, the deflection in the middle
of the hood also increases. Thus, more deformation space is required to reduce the
intensity of the secondary impact.
The challenge of reducing the severity of head impact near the fender can
be addressed by changing the hood design from an inlaid hood (Fig. 7.19) to a
wraparound hood (Fig. 7.20). However, the wraparound hood design raises issues
related to repair costs in low-speed impacts because it is more exposed to damage.
The length of the hood also plays a vital role in pedestrian protection in a
passenger car front-end collision with a pedestrian. In the case of a shorter hood
length, the pedestrian is likely to come into contact with the A-pillar, generally
an area of very high stiffness, resulting in increased risk of head injury. As the
length of the hood increases, the challenge to meet the durability, noise and vibration
guidelines become more difficult.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 179

Fig. 7.18 Underside view of a hood assembly

Fig. 7.19 Inlaid hood (A2Mac1 2010)

The choice of material and thickness of the outer hood panel is important in
reducing the probability of severe head injuries. A combination of material and
thickness with better energy absorption is favourable for pedestrian protection.
180 R. Krishnamoorthy et al.

Fig. 7.20 Wraparound hood (A2Mac1 2010)

The preferred material for pedestrian protection should also satisfy quality, durabil-
ity, noise and vibration requirements. At present, the materials generally considered
for the outer hood panel include steel, aluminium, plastic, carbon fibre and
fibreglass.
The inner hood panel acts as an energy absorber and provides torsional and
bending stiffness to the hood assembly. Therefore, the choice of structure, material
and thickness for this component also plays a vital role in reducing the severity of
head impact.
Many concepts for the structure of the inner hood panel have been considered
by automakers around the world such as multi-cone, multi X pattern, frame, single
skin, double skin etc.
The choice of glue that is used between the inner and outer hood panels also has
an important role in reducing the severity of head injuries. The stiffness of the hood
assembly also increases with the number of gluing points and glue length between
the inner and outer hood panels.
Therefore, optimising the central area of a large hood poses several challenges.
This work includes developing a hood configuration that provides a robust and
homogeneous head impact performance for different impact positions in the central
area of the hood of a large passenger car while minimising the deformation space
required.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 181

7.6 Methodology

7.6.1 Parametric Geometric Model Creation

In this research, the designs of the hood assembly were created as three-dimensional
parametric geometric models using the chunky solid method in Unigraphics NX5
CAD software. The design boundaries were set up using datum planes at set
coordinates. This methodology enables faster design modifications in the hood
assembly.
Once a design alternative was developed, checking for fitment and clearance was
carried out by virtually assembling the hood to the original donor vehicle. The
mid-surface was extracted from these geometric models to enable the FE model
development to assess the head impact performance.

7.6.2 Finite Element Model Creation

The FE models were developed using ANSA V13 with the mid-surfaces extracted
from the geometric models. PRIMER was utilised to replicate the hood interface
conditions by assembling these FE models to the donor vehicle. The section,
material properties, joints and constraints to the components of the hood assembly
to the replicated design solution intensions were assigned using PRIMER as shown
in Fig. 7.21. The FE model of the head impactor was positioned at the intended
impact position and angle as shown in Fig. 7.22. The head impact condition of
the pedestrian was simulated using LS-DYNA-3D Computer Aided Engineering
(CAE) software as exhibited in Fig. 7.23. It shows different orientations of the
head impactor at 5 ms intervals for one data point. The HYPERGRAPH and
HYPERVIEW programs were used to post-process the results which are shown in
Figs. 7.24, 7.25, and 7.26.
The energy data from the finite element analysis is presented in Fig. 7.27. In
general, the energy absorption data is used as a quality check for the finite element
model. In the finite element model constructed and utilised in this work, the energy
input is from the mass and velocity of the impactor. There are no other components
such as airbag is included in the model that can influence change in total energy
within the duration of the analysis. Therefore, the total energy remains constant for
the duration of the head impact analysis as shown in Fig. 7.27. The kinetic and
internal energy plots demonstrate that the major portion of the kinetic energy is
converted into internal energy. Thus, it can be stated that the finite element model
does not contain errors either in contact definition or in constraints, which can
influence energy fluctuations.
182 R. Krishnamoorthy et al.

Fig. 7.21 FEA model for optimisation

Fig. 7.22 FEA model with head impactor for optimisation

Fig. 7.23 FEA model simulation using LS-DYNA for optimisation

7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 183

Fig. 7.24 Contour plot of hood displacement using HYPERVIEW

Fig. 7.25 Plot of hood displacement vs. time using HYPERGRAPH

7.6.3 Head Impact Model

Currently, finite element models of head impactors, complete anthropomorphic test

dummies and complete human body models are being used for head impact analysis
in the safety community. In this research, only the child head impactor considered
as this is the impactor used in the GTR-9 and NCAP tests to assess the protection
offered by a vehicle in a child head impact.
The FE model for simulating the head impact contains the hood assembly and the
interfacing components to constrain the hood assembly. It contains a validated head
impactor model and the prescribed requirements for the impactor such as mass,
184 R. Krishnamoorthy et al.

Fig. 7.26 Plot of resultant acceleration vs. time using HYPERGRAPH

Fig. 7.27 Energy plot from the finite element model

diameter, angle, position and initial velocity of the impact to match the ANCAP
pedestrian protection testing protocol.
To minimise the CPU time for solving the FE model, the FE model was
developed for only the hood assembly and the front tie bar assembly. The front
tie bar assembly has been clipped, as shown in Fig. 7.28. The interface between the
modelled parts and the rest of the car body has been modelled by attaching it to the
ground through applying appropriate boundary conditions.
The deforming components have been modelled accurately with the necessary
mesh quality. This requires checking of the element length, aspect ratio, warpage
and angles of the mesh in the FE model. The position and properties of the
connections between the components such as the welds, joints, clinches and
adhesives have been matched to the physical vehicle build. Validated material
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 185

Fig. 7.28 Pedestrian head impact finite element model

models have been used to assign the material properties of various components used
in this research. Stress–strain curves at specific strain rates have been included for
the deforming components to improve the predictive accuracy of the model.
The interaction between the vehicle and impactor has been defined using the
contact definitions in LS-DYNA. The coefficient of friction has been assigned to
define the friction between the vehicle components and the impactor components as
well as the friction between the vehicle and the impactor.

7.6.4 Correlating Finite Element Simulations

with Experimental Results

The experimental results of a large sedan already in the market and tested by
ANCAP for pedestrian protection have been utilised for correlation. ANCAP tested
this vehicle to assess the level of pedestrian protection offered using the Euro
NCAP pedestrian subsystem testing protocol version 5.3. The pedestrian protection
experimental tests were conducted at the Centre of Automotive Safety Research,
Adelaide.
Figures 7.29 and 7.30 show good correlation between experimental test results
and CAE simulations in both measured head acceleration and the deformation
pattern of the hood outer surface.
The correlation was used to calibrate the FE models to obtain good predictive
accuracy in the numerical tests for the optimal hood design solution.
186 R. Krishnamoorthy et al.

Fig. 7.29 Correlation between ANCAP test and FE simulation

Fig. 7.30 Correlation of hood deformation between ANCAP test and FE simulation

7.6.5 Computation of Hood Intrusion

The accelerometer, which measures the acceleration in three directions, is located

at the mass centre of the impactor. In the FE model, a node has been placed at
this location to represent the accelerometer and enable computation of the impactor
displacement in both local and global coordinates, as shown in Fig. 7.31.
When the displacement is determined in the impact axis using the node on the
accelerometer, the displacement (Intrusion A) is more than the calculated intrusion
in impact direction (Intrusion T), as shown Fig. 7.32.
This is due to the rotation of the accelerometer axis along with the impactor
rotation in rebound as shown in Fig. 7.33. Therefore, the impactor intrusion in the
vertical axis, which is the Z direction, was computed in global coordinates.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 187

Global coordinate
Accelerometer coordinate

Accelerometer

Impact direction

Fig. 7.31 Coordinates for intrusion computation

The intrusion computed in the Z direction was multiplied by 1/sin(™), which

was calculated using trigonometry, to derive the intrusion in the impact direction
(Intrusion T) as exhibited in Fig. 7.34.
Although ‘Intrusion T’ was calculated from the displacement of the centre of
the head impactor in the global Z direction, there could be a variation between the
actual intrusion and ‘Intrusion T’ if the first point of contact or impact point is
different to the measured point. In addition, an assumption has been made that there
is no compression of the head form in the impact. In this research, the ‘Intrusion T’
measurement has been utilised as the actual deformation space required in all the
calculations and comparisons.
188 R. Krishnamoorthy et al.

Fig. 7.32 Comparison of impactor intrusion

Fig. 7.33 Accelerometer rotation due to impactor rebound

7.7 Vehicle Design for Pedestrian Head Impact Protection

7.7.1 Design Space Definition

The design space is the available space into which the hood assembly should fit
without affecting the down vision plane, according to the regulatory requirement
ECE-R125 and positioning of the components within the engine bay.
Figure 7.35 illustrates the available packaging space for inner and outer hood
panels. The inner and outer hood panels are the two main components in the hood
assembly that affect pedestrian protection performance.
Therefore, the scope of this research is to optimise the inner and outer hood
panels to meet the HIC value requirements with minimal deformation, as shown in
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 189

Z
A - Intrusion measured
Impact point T - Intrusion T
Θ - Impact angle

Impact direction

Measured point Projected point

A T X

Fig. 7.34 ‘Intrusion T’ calculation method

Fig. 7.35 Design space for Down vision plane

hood assembly
ECE-R1
5

Plane to define hard

components with
n engine bay

Fig. 7.36. However, modifications that may be required in interfacing components

to improve the pedestrian safety performance offered by the vehicle are not included
in the scope of this research.

7.7.2 Impact Area Definition

The head impact area on the hood was defined as per the GTR-9 test area definition.
The Hood Leading Edge (HLE—Fig. 7.37) or 1000 Wrap Around Distance (WAD),
Hood Rear Reference Line (HRRL—Fig. 7.38) and Side Reference Lines (Fig. 7.39)
determine the boundaries for head impact area. WAD divides the head impact area
into adult and child head impact zones. WAD is the distance measured from the
ground using a flexible tape held taut on the outer surface of the vehicle front
structure in a vertical longitudinal plane.
190 R. Krishnamoorthy et al.

Fig. 7.36 Scope of the project

Hood Leading
Edge

Straight edge
1,000 mm long

50°
600 mm

Fig. 7.37 Hood leading edge (United Nations Economic Commission for Europe 1998)

In the GTR-9 protocol, the accident data and technical feasibility of regulating
within the test area are the basis for WAD boundaries. The child impact area is
defined from 1000 WAD or HLE to 1700 WAD and the adult head impact area are
defined from 1700 WAD to HRRL or 2100 WAD. The resulting test area is as shown
in Fig. 7.40.
A range of impact points has also been selected for the numerical simulations,
varying in fore–aft as well as in cross-car directions of the vehicle at regular intervals
in order to analyse the central area of the hood, as shown in Fig. 7.41. These impact
positions are within the child head impact zone. This area has been selected because
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 191

Hood Rear Reference Line

Sphere

Fig. 7.38 Hood rear reference line (United Nations Economic Commission for Europe 1998)

Hood Side
Reference Line

Straight edge
700 mm long

45°

Fig. 7.39 Hood side reference line (United Nations Economic Commission for Europe 1998)

high hood intrusions occur in the central area of the hood and the interfaces around
the edges prevent deformation.
The impactor speed has been set to the ANCAP requirement, 40 km/h that is
higher than the GTR-9 test speed, which is 35 km/h, to analyse the most severe load
case.
192 R. Krishnamoorthy et al.

Fig. 7.40 Head impact test

area on hood

Fig. 7.41 Impact positions

considered for optimisation
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 193

Beam design Cut-out design Multi-one design Skeleton design Grid design

Multi cone hood inner design

Beam hood inner design
Skeleton structure hood inner design
Cut–out hood inner design Aluminium hood inner outer panels
Grid hood inner design
Steel hood inner and outer panels
Plastic hood inner outer
Composite hood inner and outer

0.4 mm to 1.5 mm gauge variation for hood

inner and outer

Fig. 7.42 Concepts for improving pedestrian protection performance

7.8 Parameters for Improving Pedestrian Protection

7.8.1 Concepts Generation

Various concepts for the inner hood structure were considered. The selection of
concepts was done referring to literature and the hoods available in the market that
could be considered in the hood design for a large sedan as shown in Fig. 7.42. The
inner hood structures considered for improving pedestrian protection performance
are the beam, cut-out, skeleton, grid and multi-cone patterns. The beam pattern has
beams positioned in the fore–aft direction of the vehicle, which provides structural
stiffness to the hood assembly. The cut-out pattern is a mass-saving and cost-
saving option in which the centre portion of the inner hood has a large opening.
The skeleton pattern is beams nested in an A-shape. This pattern allows improving
the structural stiffness of the hood in the centre, but outboard openings reduce the
mass and structural stiffness. The multi-cone pattern is cone shaped forms nested at
constant distances. This design provides homogeneous structural stiffness. The grid
pattern consists of longitudinal and lateral beams arranged at regular intervals. Thus,
this pattern also should provide homogeneous structural stiffness and the cut-outs
enable mass reduction.
Aluminium and steel were considered in the analysis for the inner and outer
hood material because they generally meet the requirements for noise and vibration,
194 R. Krishnamoorthy et al.

Concept
Concept

Steel(current)

Multi-cone design
(concepts3)
Aluminium
(concept1)

(concept2)

Conposite

Skeleton design
Cut-out design
Plastic

Beamdesign

Grid design
(concept1)

(concept2)

(concept3)

(concept4)
Criteria
Criteria
Impact on cost – + –
Impact on mass + + + Impact on cost + – – –
Ease of assembly installation S + S Impact on mass + – – –
Durability D S – – Ease of assembly installation S S S S
Ability ot with stand engine Complexity of design
D
A
compartment heat S – – (manufacturability) A S S S S
T
T
Hood stiffness consistency U S – – Durability S + + +
U
Compatibility with high-speed M Hood stiffness consistency – + + +
M
crash conditions S – – Compatibility with real-world impacts – + + +
Noise and vibration S – – Noise and vibration – + + +
Long-term reliability S – –
Long-term reliability – + + +
Appearence S – –
Appearence – + + +
∑+ (Sum of positives) 1 3 1
∑+ (Sum of positives) 2 6 6 6
∑– (Sum of negatives) 1 7 8 ∑– (Sum of negatives) 5 2 2 2
∑S (Sum of “same”) 8 0 1
∑S (Sum of “sames”) 3 2 2 2

+ = Better than datum

– = Worse than datum
S = Same as datum

Fig. 7.43 Pugh matrix

Fig. 7.44 Inner hood geometries chosen for developing an optimal solution

durability and high-speed crash performances. A thickness range of 0.4–1.5 mm

was considered for the inner and outer hood thickness, values less than 0.4 mm will
not meet the durability and forming requirements while those above 1.5 mm will
result in high structural stiffness and structural inertia as well as increased mass.

7.8.2 Concepts Evaluation

The concepts mentioned in the previous section were analysed through the Pugh
matrix as shown in the previous page in Fig. 7.43. The comparison of the concepts
to various performance criteria, multi-cone, skeleton and grid patterns for inner hood
structure (Fig. 7.44) ended up with the same merits, as well as aluminium and steel
for the material. Therefore, these configurations were chosen for further evaluation.
The hood attachment and boundary strategies, though may improve pedestrian
protection performance, were not been taken into consideration throughout the
analysis.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 195

Fig. 7.45 Control factors in scope for refinement

Table 7.3 Variables considered for refinement

Variable Values of variable Type of variable
Inner hood geometry Skeleton pattern, multi-cone pattern grid pattern Discrete
Inner/outer hood material Aluminium steel Discrete
Inner hood gauge 0.4–1.5 mm Continuous
Outer hood gauge 0.4–1.5 mm Continuous
Impact position 1–8 Discrete

7.8.3 Control Factors in Scope

Design parameters such as thickness and material for the inner and outer hood
panels as well as inner hood structure were considered for improving the pedestrian
protection performance. Figure 7.45 shows the list of control factors that are in scope
to modify the concepts selected through the Pugh matrix.
The variables considered in this study have been summarised in Table 7.3;
tolerances otherwise known as noise factors that might have some influence on the
resulting HIC value and the deformation of hood panels have not been considered
in this research work for simplicity purposes.
196 R. Krishnamoorthy et al.

Fig. 7.46 Parameter diagram

7.8.4 Concepts Refinement

The parameter diagram as shown below in Fig. 7.46 clearly outlines the relationship
between the input and output parameters.

7.9 Optimisation of Hood Panels

7.9.1 Methodology

Strength-two orthogonal array design of experiments was used in this work to create
a matrix with the variables mentioned in Table 7.3. This matrix was utilised to create
a multitude of hood designs by varying the values of these design parameters. Non-
linear LS DYNA models were created to replicate pedestrian protection head impact
physical testing for all the configurations as per the Design of Experiments (DOE)
matrix. The accelerometer time histories were gathered as output for all numerical
tests, and from that, the HIC values were calculated using Eq. (7.2). The vertical
intrusion of the head impactor from the node displacement was also collected to
calculate the intrusion in impact direction. Accordingly, the efficiency was derived
for all hood configurations. Finally, the mean value of efficiency (˜), the mean
value of HIC (HIC) and the standard deviation of HIC (¢HIC) were derived for all
the hood configurations while considering all impact positions.
The ˜ provides average efficiency between the impact positions for a given
configuration of the hood assembly. Similarly, ¢HIC from the calculated mean
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 197

Fig. 7.47 Predictability

of response surface 2,050

Predicted HIC value

1,550

1,050

550

50
50 550 1,050 1,550 2,050
Actual HIC value

value for a given configuration of the hood assembly between the impact positions
was utilised to quantify the homogeneous behaviour of the hood assembly. The
HIC provides the average HIC value between all the impact positions for a given
configuration of the hood assembly. The aim was to obtain the lowest value of ¢HIC
thus achieving better homogeneous head impact performance with the highest value
of ˜ for HIC values less than 1000.
The Kriging response surface based approach was used as an interpolation
method to predict the values for the variables considered in the improvement of
pedestrian protection. The Kriging interpolation method was named after a South
African mining engineer DG Krige and it has been used as a fundamental tool in the
field of mining (Bohling 2005). In recent times, this method has been successfully
implemented in a variety of applications. In the Kriging approach, the response
surface is mapped with limited sample data in the design space followed by utilising
this surface to estimate the values of variables at locations where the sample data
is unavailable. In this research, response surfaces were mapped with the numerical
results for ˜, HIC and ¢HIC.
Figure 7.47 shows the predictive accuracy of the HIC response surface. The
analysis of variance (ANOVA) technique was used to assess the relationship
between the inputs and outputs as well as to identify the important variables in this
study (Gelman 2004). The ANOVA charts showed that the material, thickness of
the inner hood panel, thickness of the outer hood panel and the hood inner structure
contribute a substantial percentage in determining the ˜, HIC and ¢HIC values.
The ANOVA graph for the HIC was presented in Fig. 7.48.
In Fig. 7.48, HOODMATL is the hood panel material, OTRGAUGE is the
thickness assigned for the outer hood panel, INRGAUGE is the thickness assigned
for the inner hood panel and INRGEOM is the type of structure of the inner hood.
The Monte Carlo method, which is a mathematical technique for statistical
sampling was utilised to check whether any other configuration might provide a
better output than the ones considered in the sample. This method generates a cloud
of input variables that obey the defined properties for which the output was predicted
from the existing response surfaces.
198 R. Krishnamoorthy et al.

Fig. 7.48 ANOVA chart for Percent Contribution to HIC value

mean value of HIC 0% 20% 40% 60% 80% 100%
HOODMATL
INRGAUGE
OTRGAUGE
INRGEOM
HOODMATL *…
INRGEOM *…
HOODMATL *…
OTRGAUGE *…
HOODMATL *…
INRGEOM *…

The sum of the mean value of HIC and its standard deviation (HIC C ¢HIC)
provides 67 % confidence to this research, assuming normal distribution
(Narasimhan 1996). The aim of this study is to obtain 95 % confidence by setting the
HIC threshold condition to the sum of the mean value of HIC and twice its standard
deviation (HIC C 2¢HIC) to be less than or equal to 1000. Aiming for 99 %,
robustness requires the HIC threshold condition to be the sum of the mean value
of HIC and three times its standard deviation (HIC C 3¢HIC). For a production
representative hood, it would be necessary to achieve all the impact positions less
than the HIC value of 1000. This would be achieved through the detailed design
work on a specific hood design and its interfacing conditions. Since this research
focuses on a more general study looking at material, thickness and construction of
hood panels, the 95 % confidence limit is used.
The severity of pedestrian head injuries could be influenced either by the primary
impact or by the secondary impact. These concepts were refined in two stages. The
first step was the selection of the type of the inner hood structure and inner and outer
hood material first by considering only primary impact followed by considering both
the primary and secondary impacts. The second step was the selection of the outer
hood panel thickness and the inner hood panel thickness and structure. The two-
stage optimisation approach was utilised in this research to reduce the number of
simulations and modelling complexity.

7.9.2 Selection of Hood Structure and Material

The components within the engine bay were not considered in this analysis model;
thus, only primary impact has been taken into consideration.
In this research, the efficiency of a hood design was calculated as the ratio of the
actual deformation of hood assembly (‘Intrusion T’) and the optimal deformation
calculated from the HIC value of the corresponding analytical test. The optimal
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 199

190

170

150
σHIC

130

110

90

70
0.76 0.78 0.80 0.82 0.84 0.86
μη
Steel – Multi – cone Steel – Skeleton Steel – Grid
Aluminium – Multi – cone Optimal – confirmation Optimal – predicted

Fig. 7.49 Evaluation of efficiency

deformation is the theoretical minimum deformation needed for a given HIC value
and it was calculated using Eq. (7.5) (Wu and Beaudet 2007) as shown below:
8
v0 3
d D 8:342  103 2
(7.5)
HIC 3
Where, d is the deformation and v0 is the impact velocity.
In numerical testing, the actual deformation was computed in the Z direction and
used to derive the ‘Intrusion T’ as discussed earlier. The equation to calculate the
efficiency (˜) of the hood assembly is:

Optimal deformation
˜D (7.6)
Intrusion T
Thus, a hood configuration with maximum efficiency for an impact had minimum
‘Intrusion T’ for a given HIC value.
Figure 7.49 shows the ˜ against respective ¢HIC. Among these configurations,
the optimal hood configuration that provides the highest ˜ was identified. The
numerical tests conducted to check the predictive accuracy of this optimal solution
show that the results are comparable. This showed that the deviation between the
predicted and numerical values is within 0.01 and therefore it could be stated that
accurate predictions can be made from the response surfaces.
200 R. Krishnamoorthy et al.

Fig. 7.50 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point A

Fig. 7.51 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point B

Numerical tests were conducted for eight impact locations (A–H) as shown in
Fig. 7.41 to derive the values for this data point and the details of the results from
the numerical tests are shown in figures ranging from Figs. 7.50, 7.51, 7.52, 7.53,
7.54, 7.55, 7.56, and 7.57. The details included in these figures are,
• Plot of resultant acceleration against time
• Plot of the displacement of the impactor in global Z direction against time
• Contour plot of the effective plastic strain
• Contour plot of the magnitude of the deformation of the hood panels.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 201

Fig. 7.52 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point C

Fig. 7.53 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point D

The results exhibited in figures ranging from Figs. 7.50, 7.51, 7.52, 7.53, 7.54,
7.55, 7.56, and 7.57 show that the deformation of the hood panels is not excessive,
even while secondary impact is not considered, due to the optimisation of the
structural stiffness of the hood panels.
HIC values less than the threshold value (1000) are feasible with various
combinations of inner and outer hood thicknesses. The efficiency variation is only
3 % for various combinations of the inner and outer hood thickness between 0.5 and
1.4 mm.
202 R. Krishnamoorthy et al.

Fig. 7.54 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point E

Fig. 7.55 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point F

Thus, thinner inner and thicker outer hood thicknesses could be chosen within
this range taking into consideration the torsional and bending stiffness, durability,
noise and vibration performances.
Figure 7.58 shows the mean value of efficiency plotted against the combined
inner and outer hood panel thickness. Although it could be concluded that similar
efficiency is possible with various combinations of inner and outer hood thicknesses,
efficiency varies with the combined thickness of inner and outer hood, as shown in
Fig. 7.58. The optimal efficiency is achievable with a combined thickness of about
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 203

Fig. 7.56 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point G

Fig. 7.57 HIC value, effective plastic strain, displacement of the impactor and hood with
consideration to primary impact only for point H

1.76 mm. A trend line in this plot will show that the efficiency continues to increase
with higher gauges but the HIC values will exceed the required limit.
The Eq. (7.5) that was utilised to derive the optimal deformation was derived
from the theoretical optimal waveform, which is the WSTC where the resultant
acceleration is infinite at zero millisecond and is impractical to obtain. In practice,
it takes a minimum of 1–2 ms to reach the peak acceleration in a pedestrian head
impact. Therefore, it could be concluded that in practice it is impossible to achieve
100 % efficiency with any hood configuration.
204 R. Krishnamoorthy et al.

0.84

0.83

0.82
μη

0.81

0.80

0.79
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
Sum of inner and outer hood thickness

Fig. 7.58 Efficiency and combined hood panel thickness relationship

From the optimal acceleration curve, the calculated optimal deformation space
required for HIC value that is less than 1000 is 50.5 mm. The results presented
in Fig. 7.58 show that 84 % efficiency is possible with a steel multi-cone hood
structure with a combined thickness of 1.76 mm for inner and outer hood panels.
The deformation space required for a hood assembly with 84 % efficiency has been
found to be approximately 60 mm. This is in agreement with suggestions given by
Wu and Beaudet, who also derived equations to calculate the optimal deformation
space as shown Eq. (7.5). Thus, a hood configuration with 84 % efficiency could be
considered as the practical limit of efficiency as well as optimal hood configuration.
To check the effect of allowing 70 mm deformation space, impact has been
simulated with different values of deformation space within the range from 60
to 74 mm. Same variables and qualifiers such as the mean value of HIC and its
standard deviation have been used in the analysis. Comparison of HIC C 2¢HIC
for various deformation space as shown in Fig. 7.59 indicates that a minimum of
about 69 mm deformation space is required to obtain robust and homogeneous head
impact performance when considering both primary and secondary impacts.
Therefore, it can be stated that a multi-cone hood inner structure made of steel
with approximately 1.6 mm of combined thickness of thinner inner and thicker outer
hood panels requires a deformation space of about 69 mm to obtain robust and
homogeneous head impact performance.

7.9.3 Optimisation of Panel Thickness for Aluminium Material

From the previous optimisation steps, it was identified that multi-cone structure
for inner hood gave optimal head impact performance for steel and aluminium
hood panels. However, it was also noticed that a 70 mm deformation space was
not enough to obtain the HIC value less than 1000 when hood panels are made of
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 205

1200

1150

1100
μHIC+2σHIC

1050

1000

950

900
60mm 62mm 64mm 66mm 68mm 70mm 72mm 74mm

Fig. 7.59 Influence of deformation space on HIC value (steel)

aluminium. Thus, another study was conducted with deformation space of 80 mm

for aluminium hood panels with multi-cone inner hood structure.
Figure 7.60 shows the mean value of HIC plotted against its standard deviation.
The head impact performance results from this study shows that 80 mm is enough
to obtain the mean value of HIC less than 1000, although it is not enough to provide
robust head impact performance.
Therefore, analysis was carried out by increasing the deformation space from
80 to 85 mm. The plot of inner and outer hood panel thickness against the
HIC C 2¢HIC in Fig. 7.61 shows that it is possible to obtain robust head impact
performance with HIC value less than the compliance threshold with a 85 mm
deformation space. Inner hood thickness of 0.58 mm and outer hood thickness of
1.5 mm show optimal head impact performance with aluminium hood panels.
Figure 7.62 shows that the HIC value increases rapidly when the combined
thickness of hood panels is below the optimal panel thickness, due to the intensity of
the secondary impact. The HIC value slowly ramps up when the combined thickness
is higher than the optimal thickness.
Figure 7.63 presents a close-up view of the optimal performance from Fig. 7.62.
From this graph, it could be stated that combined hood panel thickness of 2.1 mm
with aluminium material shows optimal head impact performance. Figure 7.64
presents the combined inner and outer hood panel thickness used in various vehicle
models. As mentioned earlier, vehicle manufacturers balance the deformation space,
mass, noise, vibration, durability and pedestrian protection requirements when
selecting the inner and outer hood panel thicknesses.
206 R. Krishnamoorthy et al.

500

450

400

350

300
σHIC

250

200

150

100

50
800 900 1000 1100 1200 1300 1400 1500
μHIC

Head impact performance

Sum of the mean value and twice the standard deviation of HIC

Fig. 7.60 Head impact performance with 80 mm deformation space

2000

1800

1600
μHIC+2σHIC

1400

1200

1000

800
0.4 0.6 0.8 1.0 1.2 1.4 1.6
Hood panel thickness (mm)

Inner hood panel thickness Outer hood panel thickness

Fig. 7.61 Influence of hood panel thickness on HIC value (aluminium)

To check the effect of allowing 85 mm deformation space, impact has been

simulated with different values of deformation space within the range from 80
to 85 mm. Same variables and qualifiers such as the mean value of HIC and its
standard deviation have been used in the analysis. Comparison of HIC C 2¢HIC
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 207

30000

25000

20000
μHIC+2σHIC

15000

10000

5000

0
0.5 1.0 1.5 2.0 2.5 3.0
Combined hood panel thickness (mm)

Fig. 7.62 Influence of combined hood panel thickness on HIC value (aluminium)

2200

2000

1800
μHIC+2σHIC

1600

1400

1200

1000

800
1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9
Combined hood panel thickness (mm)

Fig. 7.63 Combined hood panel thickness vs. the sum of the mean value of HIC and twice its
standard deviation (aluminium)

for various deformation space as shown in Fig. 7.65 indicates that a minimum of
about 84 mm deformation space is required to obtain robust and homogeneous head
impact performance when considering both primary and secondary impacts.
Therefore, it can be stated that multi-cone inner hood structure made of
aluminium with approximately 2.1 mm of combined thickness of thinner inner and
thicker outer hood panels require a deformation space of about 84 mm to obtain
robust and homogeneous head impact performance.
208 R. Krishnamoorthy et al.

Volvo S60 ( 2010 )

RENAULT Laguna ( 2008 )

Vehicle model

OPEL Insignia ( 2009 )

MERCEDES E-Class ( 2009 )

BMW 7 Series ( 2008 )

AUDI A8 ( 2006 )

1.80 1.90 2.00 2.10 2.20 2.30 2.40

Combined thickness (mm)

Fig. 7.64 Combined thickness of aluminium outer and inner hood panels in various vehicles

1,100.0

1,050.0
μHIC+2σHIC

1,000.0

950.0

900.0
80mm 81mm 82mm 83mm 84mm 85mm

Fig. 7.65 Influence of deformation space on HIC value (aluminium)

7.9.4 Final Design Concept

The plots of the resultant acceleration against time curve, known as waveform, for
steel and aluminium hood assemblies were compared to the theoretical and realistic
optimal waveforms.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 209

300
Theoretical optimal waveform
Realistic optimal waveform
250 Resultant accelaration - optimal (steel)
Resultant accelaration - optimal (aluminium)
200 HIC: 687.118
Acceleration (g)

150 HIC: 869.277

100

50

0
0 5 10 15 20 25 30
Time (ms)

Fig. 7.66 Waveform comparison-70 mm deformation space

The realistic optimal waveform is the combination of peak acceleration ramping

up within 1 or 2 ms from the start of head impact followed by acceleration ramping
down to a lower value that minimises the head injury. This waveform requires
minimum deformation space for a given HIC value, in this instance 1000. The basis
of the realistic optimal waveform is the theoretical optimal waveform in which
the resultant acceleration reaches infinity at the beginning of the impact, but is
impractical to obtain.
Figure 7.66 presents the head impact performance of a steel hood configuration
with multi-cone inner hood structure with 0.5 mm for inner hood panel thickness
and 1.1 mm for outer hood panel thickness. The aluminium hood assembly was
analysed with a combined thickness of inner and outer panels of 2.7 mm. It shows
that similar magnitudes of peak acceleration are possible with steel and aluminium
by varying the panel thickness. Aluminium material however provides a higher HIC
value compared to steel due to an increase in acceleration in the later part of the
impact duration.
The peak acceleration of approximately 175 g within 2 ms is feasible due to the
structural stiffness of the thicker outer hood panel, and the acceleration ramps down
as the head contacts the thinner inner hood panel in the optimal hood configuration.
The plots in Fig. 7.67 show that the optimal hood configuration with aluminium
material does not match the realistic optimal waveform, because it is not optimal for
deformation space.
210 R. Krishnamoorthy et al.

300
Theoretical optimal waveform
Realistic optimal waveform
250 Resultant accelaration - optimal (aluminium)

200
Acceleration (g)

150

100

50
HIC: 750.333

0
0 5 10 15 20 25 30
Time (ms)

Fig. 7.67 Waveform comparison aluminium with 85 mm deformation space

7.10 Discussion

Since the magnitude of peak acceleration and rebound velocity have substantial
influence on the HIC value of pedestrian when colliding with the front of a passenger
car, the factors affecting the magnitude of peak acceleration were investigated by
creating an ANOVA chart for peak acceleration.
In Fig. 7.68, HOODMATL is the hood panel material, OTRGAUGE is the
thickness assigned for the outer hood panel, INRGAUGE is the thickness assigned
for the inner hood panel and INRGEOM is the type of structure of the inner
hood. The ANOVA chart showed that hood material had the highest effect on
the magnitude of peak acceleration. Outer hood panel thickness was the next
contributing factor, followed by inner hood panel thickness and inner hood structure
respectively, as shown in Fig. 7.68. These findings support the above mentioned
results, where increase in panel thickness was required for aluminium hood panels
to match the peak acceleration of steel panels.
To understand the reason for the increase in HIC value of pedestrian when
colliding with hood panels made of aluminium, the resultant acceleration was
plotted against vertical intrusion, as shown in Fig. 7.69. Although both steel and
aluminium result in similar values for peak acceleration and total vertical intrusion,
the value of HIC varied between them.
This prompted a comparison between the average values of the rebound velocity
of the hood for various impact positions for both materials as presented in Fig. 7.70.
This shows that in general, the rebound velocity for aluminium is approximately 2.2
times higher than that for steel.
The integration of the resultant acceleration, which is velocity, is used in the
calculation of HIC value. Thus, an increase in the rebound velocity increases the
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 211

Percent contribution to peak acceleration

0% 20% 40% 60% 80% 100%

HOODMATL

OTRGAUGE

INRGAUGE

INRGEOM

HOODMATL * OTRGAUGE

HOODMATL * INRGAUGE

INRGEOM * INRGAUGE

HOODMATL * INRGEOM

INRGEOM * OTRGAUGE

OTRGAUGE * INRGAUGE

Fig. 7.68 ANOVA chart for peak acceleration

160
Steel
140 Aluminium

120
Resultant accelation (g)

100

80

60

40

20

–20
0 10 20 30 40 50 60 70 80 90
Vertical intrusion (mm)

Fig. 7.69 Resultant acceleration against vertical intrusion

HIC value. Therefore, further exploration has been carried out to understand the
exact reason for this increase in rebound velocity in aluminium hood panels.
To understand the basic concepts behind the difference in rebound velocities,
a simple example has been studied theoretically. The strain energy for steel and
212 R. Krishnamoorthy et al.

6.0

5.0
Rebound velocity (km/h)

4.0

3.0

2.0

1.0

0.0
Steel Aluminium

Fig. 7.70 Comparison of rebound velocities

aluminium was calculated for a simplified beam with a rectangular cross section to
demonstrate the difference in stored elastic energy. To simplify the calculation, the
beam is assumed to be loaded to its elastic bending moment.

ZL
M2
Ue D dx (7.7)
2EI
O

Where, Ue is the strain energy in the beam due to the load, M is the bending moment,
E is the Young’s modulus and I is the sectional modulus.
Equation (7.8) below shows the distribution of stress in a beam:

My
y D (7.8)
I
Where, y is equal to half of the thickness of the beam, that is, t/2, assuming t is the
thickness of the beam. Therefore, it can be written as follows:

y bt 2
M D (7.9)
6
Where, ¢y is the stress at yield point and b is the width of the beam.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 213

Substitution of Eq. (7.10) in Eq. (7.8) provides the following:

y 2 bt l
Ue D (7.10)
6E
In which, l is the length of the beam.
If we assume same length and width of the beam for aluminium and steel, the
calculation of the ratio of elastic strain energy between aluminium and steel is as
follows:

Ue of alumi ni um y.al/ 2 tal Est

D  (7.11)
Ue of st eel Eal y.st/ 2 tst

Where, tal is the thickness value assigned for aluminium, in this case, 2.7 mm; tst
is the thickness value assigned for steel, in this case 1.6 mm, Eal is the Young’s
modulus value for aluminium that is 70,000 MPa and Est is the young’s modulus
value for steel that is 200,000 MPa.
The yield strength of the aluminium material used in this example is assumed to
have ¢y(al) while the yield strength of the steel material is assumed to be ¢y(st). It
was assumed in this work that ¢y(al) D ¢y(st) as the aluminium material selected
for hood panels has yield strength close to that of steel grade usually used to
manufacture hood panels.
When these values were substituted in Eq. (7.11), the ratio of the elastic strain
energy of steel to aluminium was 1:4.8.
Therefore, it could be stated that when loaded to same level of stress, hood panels
made with aluminium store 4.8 more times the elastic strain energy stored by steel.
This causes an increase in rebound velocity by using the equation shown below:

mv2
Ue D (7.12)
2
Where, m is the mass and v is the velocity of the head.
So, 4.8 times increase in elastic strain energy leads to 2.2 times increase in
rebound velocity. This coincides with the average value of rebound velocities
comparison, shown in Fig. 7.70.
In Eq. (7.10), the stored elastic strain energy is proportional to the thickness.
Thus, the value of rebound velocity could be reduced by reducing the thickness.
However, this change will increase the deformation space due to the reduction in
structural stiffness. Since the structural stiffness of aluminium is lower than steel,
reduction in panel thickness to reduce the rebound velocity will increase the defor-
mation of hood made of aluminium. Therefore, it could be stated that aluminium
requires more deformation space in comparison to steel to meet pedestrian head
injury requirements.
Steel hood panels exhibit lower rebound velocity, which matches the realistic
optimal waveform. Therefore, the preferred choice of material for hood panels is
214 R. Krishnamoorthy et al.

steel if the main objective is to reduce the deformation space. The benefits of
minimising the deformation space are significant. They include the freedom of
styling, improved aerodynamics, and hence improvements in vehicle stability and
fuel economy. If the main objective is to reduce mass, then the preferred choice for
hood panels is aluminium; however, this will require more deformation space.
The area for aluminium and steel hood panels were the same in this study.
Thus, when 2.7 mm for aluminium and 1.6 mm for steel was substituted for panel
thickness value along with their respective density values, the result showed 42 %
less mass for aluminium hood assembly compared to steel hood assembly.

7.11 Conclusion

The main objective of this work was to develop a methodology for optimising
the hood assembly for minimal deformation with robust and homogeneous impact
behaviour while meeting pedestrian head injury requirements. In practice, the design
of the outer and inner hood panels must meet all general design requirements in
addition to pedestrian head impact performance. Achievement of design criteria for
homogeneous impact performance for the pedestrian head could aid in meeting the
general design requirements.
In this research, the important design parameters were identified and a design
of experiments matrix was created. Then, three-dimensional geometric models
and FE models were created to the design parameters defined in the design of
experiments matrix. The results from the numerical experiments were utilised to
create the response surfaces for the mean value of efficiency (˜), the mean value
of HIC (HIC) and the standard deviation of HIC (¢HIC). The Analysis of Variance
(ANOVA) technique was used to quantify the contribution of design parameters
in determining HIC value and efficiency. Then these data were filtered with the
condition that the sum of the mean value of HIC and twice its standard deviation
(HIC C 2¢HIC) should be less than or equal to 1000. This was done to match
the head injury value specified by the GTR-9 and ANCAP and to produce 95 %
confidence in the results of this optimisation, assuming normal distribution.
The filtered data was compared using ¢HIC as a qualifier for homogeneous
performance and ˜ as a qualifier for minimal deformation of outer and inner
hood panels. This resulted in the identification of the multi-cone pattern for hood
inner structure as the best structure for both steel and aluminium compared to
other structures. Further refinement indicated that steel was the preferred choice
of material for the hood assembly to obtain robust and homogeneous head impact
performance with minimal deformation compared to other materials.
Findings from this study showed that it is possible to obtain HIC values less than
the threshold value, while achieving similar efficiencies for various combinations
of thinner inner and thicker outer hood panels. However, efficiency variation was
noticed when changing the value of the combined thickness of inner and outer hood
panels.
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 215

The results from a previous study showed that it is possible to obtain HIC less
than 1000 with 60 mm deformation space with a multi-cone structure made of steel
of about 1.8 mm of combined inner and outer hood thickness, but this is not enough
to obtain robust and homogeneous behaviour less than the HIC threshold value when
secondary impact is considered.
Therefore, impact has been simulated with different values of deformation space
within the range from 60 to 74 mm. Same variables and qualifiers such as the mean
value of HIC and its standard deviation have been utilised in the analysis. The results
indicated that a minimum of about 69 mm deformation space was required to obtain
robust and homogeneous head impact performance when considering both primary
and secondary impact.
In conclusion, multi-cone inner hood structure made of steel for inner and outer
hood panels is the preferred choice for hood assembly when minimal deformation
is the main requirement for the vehicle design team.
Another study was conducted to quantify the minimum deformation space
required for the aluminium hood system. The multi-cone structure for the inner
hood required the least amount of deformation space in comparison to other
geometries. Impact simulation with 80 mm of deformation space showed that it
is possible to obtain HIC less than 1000, but this is not enough to obtain robust
and homogeneous behaviour below the HIC threshold value when secondary impact
is considered. Thus, impact has been simulated with different values of deformation
space within the range from 80 to 85 mm. The same variables and qualifiers such
as the mean value of HIC and its standard deviation have been utilised in this
analysis. The results indicated that a minimum of about 84 mm deformation space
was required to obtain robust and homogeneous head impact performance when
considering both primary and secondary impact.
In conclusion, a robust and homogeneous pedestrian head impact performance is
possible from the following design parameters for aluminium hood panels: multi-
cone structure for inner hood, and approximately 2.1 mm combined thickness with
thinner inner and thicker outer hood combination.

7.12 Recommendations for Further Work

Vehicle design for pedestrian protection requires improvements in many compo-

nents in the vehicle front-end such as the lower bumper system, upper bumper
system, interfaces of hood assembly to the vehicle and hood assembly. To optimise
the vehicle front-ends to improve pedestrian protection performance while meeting
all the other requirements of vehicle design is a very large scope of work. In this
research, the development of optimisation methodology of large-size hoods for
robust and homogeneous performance with minimal deformation was completed
while meeting pedestrian performance requirements for head impact.
The analysis in this research was carried out with a 3.5 kg child head impactor
fired at 50ı impact angle from the horizontal plane. In summary, the hood
216 R. Krishnamoorthy et al.

configuration was optimised for that given mass and angle of impact, and if an area
required both a child head impactor (3.5 kg) and an adult head impactor (4.5 kg),
compromises would need to be made to accommodate various masses and angles of
impact. Further studies, however may be required to optimise medium and smaller
size hoods. Similarly, work may be required to optimise components such as hood
hinges, A-pillar, base of windscreen, wiper assembly, fender brackets, under hood
components, glue between inner and outer hood panels and hood latch.
Although many key parameters were identified and optimised, further work
would be required to maximise the head impact performance in pedestrian pro-
tection offered by vehicles without compromising all other requirements such
as occupant safety, quality, fuel economy, durability and low repair costs. How-
ever, a foundation had been laid for continued research in improving the pedes-
trian protection offered by vehicle front-ends in a pedestrian collision with a
passenger car.

Key Symbols

A Acceleration (g)
T Duration of acceleration (s)
a(t) Resultant acceleration (g)
t1 Time at the beginning of the impact (s)
t2 Time at the end of the impact (s)
˜ Efficiency
A1 Area enclosed by acceleration–displacement curve
A2 Area of the shape that encloses acceleration–displacement curve
d Maximum intrusion (m)
v0 Initial impact velocity (m/s)
˜ Mean value of efficiency
HIC Mean value of HIC
¢ HIC Standard deviation of HIC
Ue Strain energy (J)
M Bending moment (Nm)
E Young’s modulus (MPa)
I Sectional modulus (mm3 )
t Thickness of the beam (mm)
b Width of the beam (mm)
l Length of the beam (mm)
¢y Stress at yield point (MPa)
tal Thickness value assigned for aluminium (mm)
tst Thickness value assigned for steel (mm)
Eal Young’s modulus value for aluminium (MPa)
Est Young’s modulus value for steel (MPa)
¢ y(al) Stress at yield point for aluminium (MPa)
7 Optimization of Hood Design to Minimise Pedestrian Head Injury in Impact 217

¢ y(st) Stress at yield point for steel (MPa)

m Mass of the head impactor (kg)
v Velocity of the head impactor (m/s)

References

A2Mac1 (2010) Automotive benchmarking, viewed 22 March 2010. http://a2mac1.com/home

Bohling G (2005) Kriging, University of Kansas, viewed 28 July 2011. http://people.ku.edu/~
gbohling/cpe940/Kriging.pdf
Centre of automotive safety research (2007) Head impact test machine, The University of Adelaide,
viewed 17 December 2009. http://casr.adelaide.edu.au/impact.html
Elert G (2001) Number of cars, Glenn Elert, viewed 16 June 2010. http://hypertextbook.com/facts/
2001/MarinaStasenko.shtml
European Commission (2010) Road safety, viewed 13 February 2010. http://ec.europa.eu/
transport/road_safety/specialist/statistics/trends
Gelman A (2004) Analysis of variance-why it is more important than ever, Columbia University,
viewed 28 July 2011. http://www.stat.columbia.edu/~gelman/research/published/banova7.pdf
Hardy BJ, Lawrence GJL, Knight IM, Simmons ICP, Carroll JA, Coley G, Bartlett RS (2007) A
study of possible future developments of methods to protect pedestrians and other vulnerable
road users, UPR/VE/061/07, TRL Limited
Honda Worldwide (2008) Honda to begin using third generation pedestrian dummy in crash tests
to enable evaluation of injuries to the lower back and upper leg, viewed 17 December 2009.
http://world.honda.com/news/2008/4080918POLAR-III/
Kam CY, Kerrigan J, Meissner M, Drinkwater C, Murphy D, Bolton J, Arregui C, Kendall R,
Ivarsson J, Crandall J, Deng B, Wang JT, Kerkeling C, Hahn W (2005) Design of a full-scale
impact system for analysis of vehicle pedestrian collisions, SAE Technical Paper 2005-01-
1875, SAE World Congress, Detroit, Michigan, SAE International, Warrendale, Pennsylvania,
USA
Krishnamoorthy R, Takla M, Subic A, Scott D (2013) Design optimisation of passenger car hood
panels for improved pedestrian protection. Adv Mater Res 633:62–76
Lim GG, Chou CC, Patel RN, Shahab SA, Patel PJ (1995) Estimating the minimum space to
meet federal interior head impact requirement, SAE Technical Paper 950333, SAE International
Congress and Exposition, Detroit, Michigan, SAE International, Warrendale, Pennsylvania,
USA
Mizuno Y (2003) Summary of IHRA pedestrian safety WG activities (2003) – proposed test
methods to evaluate pedestrian protection offered by passenger cars
Narasimhan B (1996) The normal distribution, Stanford University, viewed 28 July 2011. http://
www-stat.stanford.edu/~naras/jsm/NormalDensity/NormalDensity.html
National highway traffic safety administration (2009) viewed 10 October 2011. http://www-fars.
nhtsa.dot.gov/People/PeoplePedestrians.aspx
Ryan GA, McLean AJ (1965) Pedestrian survival. Safe car guide 2009, Australian New Car
Assessment Program, viewed 25 November 2009, http://www.safecarguide.com/exp/ancap/
ancap.htm
Synaptic analysis consulting group inc viewed 23 February (2011). http://www.synapacg.com/
casestudies_pedestrian.html
Transport Research Laboratory (2012) Pedestrian Safety, Transport Research Laboratory, viewed
15 October 2012. http://www.trl.co.uk/research_development/improving_safety/user_safety/
pedestrian_safety.htm
218 R. Krishnamoorthy et al.

United Nations Economic Commission for Europe (1998) Global technical regulation No. 9
(Pedestrian safety), United Nations viewed 12 July 2011. http://live.unece.org/trans/main/
wp29/wp29wgs/wp29gen/wp29registry/gtr9.html
World Health Organization (2010) Global status report on road safety: time for action, World
Health Organization, viewed 12 August 2010. http://www.who.int/violence_injury_prevention/
road_safety_status/2009
World Health Organization (2011) Violence and injury prevention and disability, World Health
Organization, viewed 10 October 2011. http://www.who.int/violence_injury_prevention/road_
safety_status/multimedia/pictures/en/index.html
Worldometers (2010) World statistics updated in real time, Worldometers, viewed 15 July 2011.
http://www.worldometers.info/cars/
Wu J, Beaudet B (2007) Optimization of head impact waveform to minimize HIC. World Congress
Chapter 8
Nonlinear Finite Element and Post-Buckling
of Large Diameter Thin Walled Tubes

Toh Yen Pang, Dale Waterson, Rees Veltjens, and Tristan Garcia

Keywords Finite element analysis • Thin wall structure • Axial crush testing
• Post-buckling • Non-linear

8.1 Introduction

The post-buckling and crushing behaviour of structures is of great importance in

many engineering disciplines. In particular, post-bucking behaviour is of interest
when considering highly dynamic collision events; where both the integrity of
a structure may be compromised and when predicting response to assess safety
aspects (such as a motor vehicle collision). In these cases, the way in which a
structural member fails can strongly influence the manner in which forces are
distributed throughout the remainder of the structure and transferred to passengers
involved in an impact (Alghamdi 2001). Virtually all modes of modern day trans-
portation (including sea, land and air) consist of thin walled structural components.
The benefits of these structures are clear; with the reduction in weight there is the
potential to dramatically improve the performance and efficiency of such vehicles.
However, with the increase in performance of each of these vehicles, as well as the
increase in societal awareness with regards to safety and design standards, there is an
increasingly large requirement for these structures to have the capacity to dissipate
a large proportion of impact energy to reduce the trauma imposed on occupants
during a collision.

T.Y. Pang () • D. Waterson • R. Veltjens • T. Garcia

School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University,
Bundoora East Campus, Bundoora, Melbourne, VIC, Australia
e-mail: [email protected]; [email protected]; [email protected];
[email protected]

© Springer International Publishing Switzerland 2015 219

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__8
220 T.Y. Pang et al.

In recognition of the role that thin walled structures play in collision safety,
there has been a plethora of research conducted with regard to this particular
field. This includes (but is certainly not limited to) the manner in which thin wall
structures may collapse based on their geometry (Song et al. 2012) to how thin
walled structures may be utilised as impact energy absorbers (Abramowicz 2003).
Ultimately, the survivability of motor vehicle collisions is not only a function of
the amount of impact energy present, but also the manner in which that energy is
dissipated. This impact energy needs to be distributed in a controlled manner such
that the deceleration of the vehicle improves the chances of occupant survivability
(Abramowicz 2003). In addition to this, there is the need to verify the functionality
of designs in collision incidents, including both experimental and computational
analysis of structures. In order to enhance the collision behaviour of thin walled
structures, analysis into the collapse behaviour of individual thin walled structural
components needs to be undertaken.
This study aims to investigate the crush behaviour of a large diameter, thin
walled cylindrical tube under a compressive axial load. Experimental analysis will
be conducted to determine the cylinder’s modes of failure under a constant rate of
compression and gain data relating to the forces required to crush the specimen. This
will then be followed by a numerical finite element analysis, which will be used to
compare and verify the ability to correctly and adequately predict the results of such
an experiment.

8.2 Experimental Test

8.2.1 Specimens and Test Machines

The specimens used during the experiment consisted of two thin walled aluminium
cylindrical tubes (shown in Fig. 8.1), each with the length of 300 mm, outside
diameter of 100 mm and wall thickness of 3.0 mm.
The experiment involved crushing each specimen in a vertical press which was
capable of providing a constant rate of compression. Each thin walled cylindrical
tube was placed between the top and base plates of the press. The test specimen was
then axially compressed by 150 mm. The force required to maintain the constant
rate of compression was measured using load cells and recorded by a computer.
The setup of the load press is shown in Fig. 8.2. The tests were conducted at a
compression rate of 20 mm/min.
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 221

Fig. 8.1 Thin walled

cylindrical tube used for the
experiment

Fig. 8.2 Experimental setup

of test specimen in the load
press

8.2.2 Results of Experimental Test

During the compression both specimens exhibited similar modes of axisymmetric

local buckling failure that involved the tube folding outwards and then back inward
as shown in Fig. 8.3. The tube rippled and folded neatly across the centreline of the
wall section.
Two out of the three tests displayed near identical buckling modes throughout
the full range of compression. One other specimen exhibited a similar failure mode
initially; however a non-symmetric failure mode became dominant for the remainder
of the range.
222 T.Y. Pang et al.

Fig. 8.3 Images of experiment demonstrating mode of failure

Fig. 8.4 Force vs. displacement graph of experiment

The force–displacement plots for the tests with axisymmetric failure modes are
shown together for comparison in Fig. 8.4. There is an extremely high level of
consistency in the experimental results. Both tests demonstrate a nearly identical
pattern in terms of the force required to compress the tube throughout the range of
the experiment. Each of the peaks corresponds neatly with the point at which each
tube begins to buckle, with the peak stress of approximately 190 kN.
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 223

8.3 Numerical Calibration

8.3.1 Finite Element (FE) Models

The FE code Abaqus/Explicit v6.11 was adopted to simulate the crushing process
of the cylindrical tubes. The cylindrical tube was meshed using four-node doubly
curved thin or thick shell S4R elements with a shell thickness of 3 mm.
The boundary conditions of the simulation were setup according to the experi-
mental buckling tests. The base plate was initially contacting the bottom edge of the
tube, and then moved axially upward to allow the top edge of the tube to contact
the top plate and then crush the tube. The upper and lower plates were modelled
as analytical rigid bodies. Explicit procedure was used with crushing velocity of
3.1 m/s.

8.3.2 Material Model

The shell material was Aluminium alloy 6060 T5, which is a ductile homogenous
metal and exhibits a very linear stress–strain relationship, i.e. elastic region, up to a
yield point. The mechanical properties for the elastic region are as follows: density
¡ D 270  103 kg/m3 , Young’s modulus E D 70 GPa, Poisson’s ratio  D 0.3, yield
stress ¢ o D 130 MPa.
After the yield point is reached, the metal starts to deform plastically, i.e.
large deformations that permanently deform the structure. The plastic stress–strain
relationship was derived using the Ramberg–Osgood equation given by (Wei and
Elgindi 2013):
 H
 0 
"plastic D C 1 (8.1)
E B 0

where B and H are the parameters given as B D 7 GPa, H D 3.0.

The elastic and plastic data was tabulated and input into the ABAQUS material
model. A maximum plastic strain of 20 % was used to capture the material failure
before the maximum strains could be reached. In order to realistically represent the
experimental conditions, damage criteria was also included in the material model.

8.3.3 Model for Ductile and Shear Failure Initiation

Fracture damage can occur via two main mechanisms; due to the nucleation, growth
and coalescence of voids, and shear fracture due to shear band localization. For
the analysis models using damage, there were three criteria of relevance: ductility,
224 T.Y. Pang et al.

Fig. 8.5 Stress–strain

behaviour of material damage
where softening occurs, as
well as degradation of
material

shear and Müschenborn–Sonne forming limit diagram (MSFLD) criteria (Dassault

Systèmes 2011a). Inputs for these criteria determine the point at which softening
occurs, as well as the rate of softening as the material is strained. This is illustrated
in Fig. 8.5, where beyond point c the capability of the material to carry load is
reduced and follows damage mechanism c–d until rupture occurs.

8.3.4 Ductile Damage and Failure Criterion

Abaqus offers a variety of choices of damage initiation criteria for ductile metals.
pl
The model assumes that the equivalent plastic strain at the onset of damage, "D , is
a function of stress triaxiality and strain rate:

"D ; "P
pl pl
(8.2)

where  D  p/q is the stress triaxiality, p is the pressure stress, q is the Mises
equivalent stress, and "P is the equivalent plastic strain rate (Dassault Systèmes
pl

2011b).
The ductility failure initiation criterion that is represented by a correlation to the
equivalent plastic strain rate (Dassault Systèmes 2011a; Hooputra et al. 2004) is
given by:
 
"C  
T sinh Œk0 .  / C "T sinh k0   
C
; "P
pl pl
"D D (8.3)
sinh Œk0 .  C /

where "C 
T and "T correspond to the equivalent plastic strain at ductile damage initi-
ation for equibiaxial tensile and equibiaxial compressive deformation, respectively.
C and  correspond to the stress triaxiality for isotropic materials in equibiaxial
tensile deformation state and compressive deformation state, respectively.  is the
ratio of the equivalent mean stress to the Mises equivalent stress. k0 is the coefficient.
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 225

8.3.5 Shear Failure Criterion

pl
For shear failure, it is assumed that the equivalent strain at facture "S is a function
of the shear stress ratio,  S , given by

1  ks 
S D (8.4)


where ks is a material parameter and  D  max / eq . The shear failure initiation

criterion that is represented by a correlation to the equivalent plastic strain is given
by:
   C 
"C
S sinh f S  S

C "
S sinh f S  S
S ; "P
pl pl
"S D   (8.5)
sinh f SC  S

where  C  C
S and  S correspond to the values  s , f D constant, and "S and "S


correspond to the equivalent plastic strain at shear damage initiation for equibiaxial
tensile and equibiaxial compressive deformation, respectively.
For this study, all the ductile and shear failure parameter data, in Eqs. (8.3)
and (8.5), for aluminium was taken from the ABAQUS example manual (Dassault
Systèmes 2011a). These parameters were manipulated to suit aluminium 6060 T5
and are covered further in the discussion section.

8.3.6 Damage Evolution and Element Removal

In Abaqus, damage evolution occurs once the damage initiation criteria are satisfied.
Plastic displacement-based linear damage evolution law is used for all damage
initiation criteria. The value of the plastic displacement at which the damage
variable reaches 1 is taken as 0.1. The default maximum degradation rule is used,
and the elements are removed from the mesh once the maximum degradation has
occurred (Dassault Systèmes 2011a).

8.3.7 Solution Procedure Using ABAQUS/Explicit

The ABAQUS/Explicit performs dynamic analysis, using a Lagrangian formulation

and integrating the equations of motion in time explicitly by means of central
differences. In the dynamic analysis, the basic equations of motion of a cylindrical
shell are expressed as:

MRu.t / C cPu.t / C Ku.t / D Fext .u.t // (8.6)

226 T.Y. Pang et al.

where [M] is the mass matrix, [c] the damping matrix, [K] the stiffness matrix
including non-linearity, u the nodal displacement vector, uP the nodal velocity vector,
uR the nodal acceleration vector and [Fext ] the external load vector (Bisagni 2005).
Following the standard equation of motion, we can write down the spatially dis-
cretised finite element equations of motion for the elastodynamic contact problem,
such as buckling behaviour of the cylindrical shells under axial compression loads,
as follows (Burlayenko and Sadowski 2014):

MRu.t / C cPu.t / C Ku.t / D Fext .u.t //  Fcont .u.t // (8.7)

Where: Fcont is the contact force vector.

When we define a contact/impact problem, which subjects to given initial
conditions u0 D u0 and uP 0 D v0 and a given load history [Fext (u(t))], the solution of
Eq. (8.7) is to hold for all time interval t 2 [0, T] (Burlayenko and Sadowski 2014;
Heinstein et al. 2000; Strömberg 2005). The time interval is first partitioned into i
subintervals [ti , tiC1 ] such that ti < tiC1 , and t1 D 0 and tiC1 D T.
Let the time increment be t D tiC1 – ti , and accelerations, velocities and dis-
placements referring to this time increment are denoted by uR iC1 , uP iC1 and ui C 1 ,
respectively. Hence, the discretized system of Eq. (8.7) at a certain time tiC1 can be
written as:

MRuiC1 C cPuiC1 C KuiC1 D Fext

i C1  FiC1
cont
(8.8)

8.3.8 Central Finite Difference

In FE analysis of time-dependent problems, the equation of motion is mostly solved

by using an explicit central difference integration rule (Heinstein et al. 2000; Sun
et al. 2000; Dassault Systèmes 2011c):

uiC1 D ui C tiC1 uP iC1=2 (8.9)

1
uP 1. D uP 1. C .tiC1 C ti / uR t (8.10)
iC
2 i
2 2

The acceleration at t1 D 0 from Eq. (8.7):

 
uR i D M1 Fext
i  cPui  Kui  Fcont
i (8.11)
. .
Where: i, i  1 2 and i C 1 2 refer to the increment number and mid-increment
numbers.
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 227

The explicit procedure is based on the implementation of an explicit integration

rule along with the use of diagonal element mass matrices, hence no couple
equations need be solved as the kinematic state is advanced using known values
of uP 1. and ui from the previous increment.
i
2

8.3.9 Contact Constraint

The tube was compressed freely and axially and there were a number of surfaces
which were identified as likely to contact during the simulation;
• The upper and lower edges of the tube, which would contact with the load press
plates.
• The side walls of the tube and the load plate, which would occur as the thin
walled tube buckled and folded over during the first failure.
• The self-contact of the tube, which would occur as each ripple of the wall folded
back onto itself.
Penalty friction formulation, with the Coulomb friction coefficient, was adopted
in the simulation to enforce the contact constraint between the master surface and
the slave surface.
Whenever contact takes place, the kinematic variables at the contact interface
between the two bodies can be split into (1) the normal, FcN , and (2) the tangential,
FcT , directions for a single constraint to the global solution as (Sheng et al. 2006):
   
Fcont
i D FcN i C FcT i (8.12)

For the penalty method, the matrix form of the normal contact is expressed by
(Heinstein et al. 2000; Wriggers 2006):

FcN D "N GN GNT (8.13)

Where: "N is the penalty parameter, GN is the gap function assembled into a
matrix, which depends on the deformation.
Discretizing Eq. (8.13) and its associated tangent matrix will produce:
 c  
FN i D "N Ai GN GNT i (8.14)

In the case of the kinematical relations for the tangential motion, FcT , in the contact
area, it is necessary to distinguish between stick and slip motion, where only
the classical Coulomb law is applied. Over a time increment t D tiC1 – ti , the
incremental tangential displacement is given by .GT˛ /i D T˛ i ui , where T˛
is the expression for the tangent matrix (Sheng et al. 2006; Wriggers 2006). First
consider the stick case, where
228 T.Y. Pang et al.

FcT stick D "T Ai T˛ i TT˛ i (8.15)

i

and in the case of slip:

FcT slip D "N Ai T˛ i GNT i (8.16)

i

Where: Ai is the contact area,  is the friction coefficient, and "T is the
regularisation parameter which could be viewed as an elastic or spring constant
applied in the tangential direction.
Refer to Wriggers (2006) for more discussion on the derivation and the explicit
form of tangential stick and slip.

8.3.10 Results of FE Analysis

When comparing the numerical results to the experimental results, reasonable

correlation between the results was obtained when the ductile and MSFLD damage
criterion were used together. The ductility criteria, which were adopted from the
ABAQUS example ‘Failure of a thin walled tubes’ (Dassault Systèmes 2011a), were
scaled to suit the material properties. A scaling factor of 2/3 was found to yield
results which closely matched the experimental response throughout the range of
compression.
The force magnitude response of the post-buckling regime exhibits very similar
behaviour to the experimental results for each peak value during folding (Fig. 8.6).
The initial peak stress of 190 MPa (when buckling begins to occur) is almost
identical to that which is exhibited by the experimental test specimens, in both
magnitude and the displacement at which initial buckling of the structure occurs.
The mode of failure and layered folding pattern exhibited by the FE model (using
shell elements) demonstrated reasonably good agreement with the experimental
results (Fig. 8.7). The failure mode started as local denting followed by global
folding of the thin wall, which is associated with large strains and extensive plastic
deformation. The close-up deformation is shown in Fig. 8.8.

8.4 Discussion

The equations of equilibrium governing the dynamic response of the cylindrical

shell were solved by an explicit integration operator, using the explicit central
difference formula implemented in Abaqus/Explicit (Dassault Systèmes 2011c).
Several researchers (Alghamdi 2001; Song et al. 2012; Heinstein et al. 2000; Bisagni
2000) have showed that the dynamic analysis using an explicit integration operator
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 229

Fig. 8.6 Comparison of force–displacement plots for numerical and experimental analyses

Fig. 8.7 Deformation of aluminium tubing using ductility and MSFLD damage criteria for
Aluminium 6060 T5
230 T.Y. Pang et al.

Fig. 8.8 Close-up deformation of aluminium tubing using ductility and MSFLD damage criteria
for Aluminium 6060 T5

is efficient to study the buckling behaviour of cylindrical shells. In the study,

only linear reduced integration elements were used, due to the inherent simplifi-
cation of the mass matrix. Linear reduced elements have the mass lumped to the
nodes (lumped mass matrix), which results in a simplified method of determining
acceleration from the inverse diagonal mass matrix. Hence, the computational cost
requirement is reduced.
Sensitivity checks were carried out for the cylindrical tube with global seed sizes
of 10, 7.5, 5 and 2.5 mm around the collapsing end. The finest mesh size of 2.5 mm
was found to yield satisfactory results; a difference of only 2 % in the plastic strain at
the onset of the failure was achieved when compared with the coarser size of 5 mm.
According to Giglio et al. (2012) a reasonable compromise between accuracy of
results and duration of the analysis for sheet metals convergence is achieved when
additional element division does not change the final plastic strain by more than 2 %.
The final model of the cylindrical tube in this study was built with 9,394 elements.
The mesh refinement showed the best correlation to the experimental results
using a biased mesh, especially at the end of the tube where the buckling initiated.
The biased meshing method was found to be important in modelling the failure
mode of the structure. Using biased mesh reduced the computational resources
required for analysis, as the total number of elements (and consequently the number
of equations) was reduced. Although using a biased mesh reduced the total number
of elements in the analysis, there was still sufficiently density of mesh at the
buckling end to produce accurate results.
The inclusion of damage criteria plays an important role in the collapse of
structures in ABAQUS. Without the inclusion of such failure modes, significant
issues arise with non-convergence and analysis error unless initial imperfections are
included in the simulation. However as the experimental tubing was undamaged
before the crushing occurred, it was considered that the model should be developed
without initial imperfections. Hence the use of ductility damage criteria was
included to initiate buckling failure. The ductility criteria were taken from the
literature (Dassault Systèmes 2011a; Hooputra et al. 2004) and entered into the
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 231

material model for the aluminium. Analysis without ductility showed poor results
where collapse was not present in the model. The best results for explicit modelling
with damage criteria were produced when a 2/3 scaling factor was used on ductility
criteria and used in conjunction with MSFLD criteria.
The inclusion of MSFLD criteria (Dassault Systèmes 2011a) had a significant
effect on the failure mode of the tube. The force–displacement response for the
model without MSFLD does not correlate well with the experiment for collapse
forces or the buckling pattern. Figure 8.7 displays the collapse mode for models
where the MSFLD criteria were included. The numerically simulated collapse
occurs with the same failure mode as the experiments, where five circular folds
occurred at one end of the tubing, hence the use of MSFLD criteria was considered
necessary.
The general contact definition with hard contact relationship and penalty enforce-
ment within the ABAQUS code (Dassault Systèmes 2011d) was used to model the
contact between the press plates and the tube. It was also found that the FE model
was sensitive to the coefficient of friction () used for the contact surfaces. The
crushing tube was studied for  D 0, 0.1, 0.2, 0.5 and 0.7. The final model required
a lower coefficient of friction,  D 0.2 to gain accurate results.
The dynamic buckling of cylindrical tube was challenging to simulate accurately
due to the strong nonlinearities caused by frictional contact, large deformation
(geometrical) and elasto-plasticity (materials) (Sheng et al. 2006). Extra efforts were
needed to select appropriate material properties and contact modelling methods in
order for the simulation to accurately represent the experimental results.
Overall, the finite element model was capable of accurately simulating the
crushing behaviour of the thin walled tube. It was able to predict, with high accuracy,
the force and displacement of the load press at which critical buckling would occur,
as well as the relative magnitude of post-buckling loads required at each subsequent
failure of the tube. It demonstrated peak forces corresponding to the preliminary and
secondary folds, which occurred during each subsequent buckle. The model was
able to adequately capture the behaviour of the collapsing structure for engineering
purposes.

8.5 Conclusion

ABAQUS explicit modelling is suitable for post-buckling analysis with large

deformations. Mesh refinement in the area of failure is required to allow the
analysis to converge on a solution. Additionally, biased meshing is useful to control
where the buckling collapse occurs in structures with identical ends, such as the
specimens analysed. Biased meshing also reduces the number of elements and
computations required to complete analysis. The inclusion of ductility damage
criteria and MSFLD criteria had a significant effect on the collapse of the cylin-
drical tube without initial imperfections, where experimental and numerical results
demonstrated good correlation.
232 T.Y. Pang et al.

Key Symbols

Ai Contact area
c Damping matrix
E Young’s modulus
t Time increment
F Force vector
FcN Matrix form of the normal contact
FcT Matrix form of the tangential contact
GN Gap function
K Stiffness matrix
M Mass matrix
T˛ Tangent matrix
u, uP , uR Nodal displacement, velocity and acceleration vector
"N Penalty parameter
"CT Equibiaxial tensile
"T Equibiaxial compressive deformation
pl
"D Equivalent plastic strain at the onset of damage
 Stress triaxiality
Poisson’s ratio
 Friction coefficient
S Shear stress ratio
Density
o . . Yield stress
i, i  1 2 , i C 1 2 Increment number and mid-increment numbers

References

Abramowicz W (2003) Thin-walled structures as impact energy absorbers. Thin Walled Struct
41:91–107
Alghamdi AAA (2001) Collapsible impact energy absorbers: an overview. Thin Walled Struct
39:189–213
Bisagni C (2000) Numerical analysis and experimental correlation of composite shell buckling and
post-buckling. Compos Part B Eng 31:655–67
Bisagni C (2005) Dynamic buckling of fiber composite shells under impulsive axial compression.
Thin Walled Struct 43:499–514
Burlayenko VN, Sadowski T (2014) Nonlinear dynamic analysis of harmonically excited debonded
sandwich plates using finite element modelling. Compos Struct 108:354–66
Dassault Systèmes (2011a) Example problems manual, volume I: static and dynamic analyses.
Dassault Systèmes Simulia Corp., Providence
Dassault Systèmes (2011b) Analysis user’s manual, volume III: materials. Dassault Systèmes
Simulia Corp., Providence
Dassault Systèmes (2011c) Analysis user’s manual, volume II: analysis. Dassault Systèmes
Simulia Corp., Providence
8 Nonlinear Finite Element and Post-Buckling of Large Diameter Thin Walled Tubes 233

Dassault Systèmes (2011d) Analysis user’s manual, volume V: prescribed conditions, constraints
& interaction. Dassault Systèmes Simulia Corp., Providence
Giglio M, Manes A, Viganò F (2012) Ductile fracture locus of Ti–6Al–4V titanium alloy. Int J
Mech Sci 54:121–35
Heinstein MW, Mello FJ, Attaway SW, Laursen TA (2000) Contact—impact modeling in explicit
transient dynamics. Comput Methods Appl Mech Eng 187:621–40
Hooputra H, Gese H, Dell H, Werner H (2004) A comprehensive failure model for crashworthiness
simulation of aluminium extrusions. Int J Crashworthiness 9:449–64
Sheng D, Wriggers P, Sloan SW (2006) Improved numerical algorithms for frictional contact in
pile penetration analysis. Comput Geotech 33:341–54
Song J, Chen Y, Lu G (2012) Axial crushing of thin-walled structures with origami patterns. Thin
Walled Struct 54:65–71
Strömberg N (2005) An implicit method for frictional contact, impact and rolling. Eur J Mech – A
Solids 24:1016–29
Sun JS, Lee KH, Lee HP (2000) Comparison of implicit and explicit finite element methods for
dynamic problems. J Mater Process Technol 105:110–8
Wei D, Elgindi MBM (2013) Finite element analysis of the Ramberg–Osgood bar. Am J Comput
Math 3:211–2116
Wriggers P (2006) Computational contact mechanics. Springer, Berlin
Chapter 9
Exhaust System Acoustic Modeling

Milan Simic

Keywords Exhaust system • Sound • Acoustic filter • Physical networks

• Transfer function • Power spectrum • Bode plot

9.1 Introduction

An exhaust system removes chemical and acoustic pollutions generated by a

combustion engine. Like all other subsystems in automotive engineering, exhaust
systems have evolved for all this years of automotive history, from simple cast
iron manifold, with pipes and a silencer to complex systems that include catalytic
converters, supertone silencer, muffler, and tailpipe tips. One of the key system
components is a catalytic converter which converts poison gasses pollution by
transforming it into carbon dioxide and water. System often has few exhaust pipes
and silencers.
All components of the system, regardless of their primary role of chemical
or noise pollution control, can be seen as a kind of acoustic filters. Mufflers are
particularly designed in such a way that acoustic waves, travelling through, are
maximally attenuated, obstructed, and subjected to self-interference. A study on
acoustic modeling and testing of exhaust and intake system components is presented
in Elnemr (2011). In addition to that, research of automotive exhaust system
hanger location, for the best prevention of vibration transfer to the vehicle body,
is conducted as presented in Jianwang et al. (2008). In our research, at this stage,
we have concentrated on the acoustic side of the system.

M. Simic ()
School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University,
Bundoora East Campus, Plenty Road, Bundoora, Melbourne, VIC 3083, Australia
e-mail: [email protected]

© Springer International Publishing Switzerland 2015 235

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__9
236 M. Simic

Fig. 9.1 An exhaust system structure

There is a large variety of system constructions for various vehicles. A generic

exhausted system is shown in Fig. 9.1. Dimensions are given as they are used in
our modeling. Some systems may have everything doubled, or only mufflers, or just
tailpipe. This is used to increase power and torque, i.e., to introduce less attenuation
of those key performance engine characteristics. For example, an investigation on
exhaust system parameters for fuel economy improvement of small gasoline engine
is presented in Peng et al. (2009). In addition to that there is still energy that can
be extracted from the exhaust system as shown in Capel et al. (2013). We have
measured the real signal, i.e., sound pressure, from the vehicle and designed exhaust
system model starting from that.
Our investigation of the system acoustics is based on the Physical Networks’
approach. Any engineering system can be expressed as a physical network (Sanford
1965). It can incorporate physical quantities of the same type, like mechanical sys-
tem with translator motion, or just electrical system. Modern engineering systems
are mainly comprehensive, including physical subsystems with different types of
quantities, like electromechanical, or more than that, mechatronics (Silva 2005).
Physical networks approach, modeling and simulations of acoustic, hydraulic, and
mechanical systems, using Electronics Circuits Analysis Program (ECAP) are
presented in Simic et al. (1978).
In this paper, the basic physical network principles are explained in Sect. 9.1.
Components of the exhaust physical network are introduced in Sect. 9.2. The whole
system model and simulation by a large ladder filter topology are presented in
Sect. 9.3. Transfer function, i.e., bode diagram of the model is given in the next
section. Finally frequency characteristic of the measured sound, i.e., power spectrum
of the sound from the real vehicle is compared to the model performances and
presented in the final section of this chapter.

9.2 Physical Networks

Long time ago, in the nineteenth century, scientists, physicists, and philosophers
have stated that the unity of the nature is expressed in the amazing analogies of
the differential equations used to represent various physical phenomena. The same
9 Exhaust System Acoustic Modeling 237

equations can be used for the study of hydrodynamics and for the electrical potential
theory. The theory of turbulence in liquids and the theory of friction in gases show
extreme analogies with the electromagnetic theory.
Our, nonlinear, exhaust system, as shown in Fig. 9.1, can be presented as a
cascaded structure of linear, one-dimensional physical subsystems with constant
parameters. Those building components, called physical networks, consist of ideal
linear elements with two terminals. Network elements represent mathematical
relationships between two dependent system variables in a physical system. Often,
real systems are extremely complex, so that they cannot be easily simplified, but
many systems are in that category, or have linear subsystems as their integral parts.
An equivalent physical network represents a sort of linear graph associated to
the system equations. The basic definitions, principles, and rules for solving system
equations represented by the network are independent of physical system that the
particular network is representing.

9.2.1 Network Variables

Physical networks have two basic time dependent variables: flow, f, and potential, p.
• Flow is a variable that flows through network elements and connection lines
• Potential is a variable measured across a network element, or between two
network points
• Potential of a point in the network depends on the chosen referent point.
Examples of physical variables are electrical current and potential. Current is
directly related to the flow of electrons as i D dq
dt
. Potential difference is called
voltage. Then we have force and velocity in a mechanical system with translation,
or torque and angular speed in a mechanical system with rotation. In a hydraulic
system we have flow and pressure. All those systems can be modeled using the
same type of ordinary differential equations (ODE).

9.2.2 Ordinary Differential Equations

We can use ODEs with constant coefficients for modeling various one-dimensional
physical systems. An ODE is a relation between two variables, t and y D y(t), and
2 n
the derivatives of y, dy
dt
; ddt y2 ; : : : ; ddt ny .

d ny d n1 y dy
An n
C A n1 n1 C    C A 1 C A 0 y D f.t/ (9.1)
dt dt dt
238 M. Simic

Equation (9.1) is called ordinary because only one independent variable exists,
which is usually time, and it is linear because only the first exponent of dependent
variable, or its derivatives is present. General solution, yG (t), of an ODE is given as

yG .t / D yH .t / C yP .t /;

where yH (t) is general solution of equation when f (t) D 0, and yP (t) is any solution
of the Eq. (9.1).

9.2.3 Network Elements

Physical network is a linear graph that includes network elements and connecting
lines between them. There are two types of elements: active and passive. Active
elements are flow, or potential sources, whose operations are expressed as func-
tions of time: F D F(t) and P D P(t). They supply energy to the system. Passive
elements are two connection points elements with defined relationship between
flow through and potential across elements’ terminals. They cannot supply more
energy to the system than what was already accumulated. That is expressed through
initial conditions. There are three types of passive elements used to express three
different relationships between network variables, flow and potential. Those basic
relationships are proportion, integration, and differentiation.
Examples of all three types of passive elements in generic physical network,
as well as in electrical and mechanical system with translation are given in the
Table 9.1. Expressions for the accumulated energy and power dissipation in each of
those systems are also given. In a generic physical system, two network quantities
relations as integration, proportion, and differentiation are associated to elements
labeled as A, B, and C, as shown in Table 9.1. The general name for all elements, A,
B, and C is impedance, Z, or admittance Y. Similarly, analog to those quantities, in
an electrical system we have inductivity, L, conductivity G, i.e., resistivity, R D 1/G,
and capacity C. Network variables are current, i, and voltage, u, which is electrical
potential difference. Finally for the translation we have k for stiffness, B for friction,
m for mass of the object. Network variables here are force, F, and velocity v.

Table 9.1 Analogies relationships

Relationship Prototype Electrical Translation
R R R
Integration f D A pdt i D L1 udt F D k vdt
f2 Li 2 F2
Accumulated energy 2A 2 2k
Proportion f D Bp i D Gu D R1 u f D Bv
f2 2 2
Power dissipation fp D B D p 2 B i u D i 2 R D uR F v D FB D v2 B
dp
Differentiation f D C dt i D C dtdu
F D m dt
dv
Cp 2 C u2 mv2
Accumulated energy 2 2 2
9 Exhaust System Acoustic Modeling 239

There are also other systems like thermal, fluids, or rotation where analogies
could be easily established as given in Sanford (1965) and Silva (2005).

9.2.4 Mono and Multi-dimensional Systems

Apart from already presented mono-dimensional systems, real physical systems

are often multi-dimensional. In that case, they have to be represented by partial
differential equations including more than one independent variable. Systems that
deal with the physical fields’ effects in 3D space, like gravity, or electromagnetic
field are multi-dimensional. Even when energy is distributed through one dimension,
but that transfer is the function of time, we have a multi-dimensional system.
In mono-dimensional dynamic systems, where independent variable is time, we
can assume that, the energy associated to the dependent variables is transferred
instantly through the system, to all remote parts. Accordingly, dependent variables
are just functions of time and not the distance from the energy source. We
should mention here that, there are also static mono-dimensional systems, where
independent variable is not time, but could be distance x. Examples of such systems
could be found in Statics, a branch of Mechanics.
The issue of system multi-dimensionality, caused by the energy transition time,
could appear in electrical systems of the small dimensions, when the emitting
frequency of the energy source is high, or when we have long distance energy
transmissions. While electrical energy travels relatively quickly many other forms
of energy are for the order of magnitude slower than that. Examples are acoustic
waves, forces, and waves in liquids, and especially heat energy transfer in solids.
Those systems have to be treated as multi-dimensional even when low frequencies
are involved and physical dimensions of the systems are small.
In order to get objective criteria for the best estimation of the system dimen-
sionality, when harmonic energy sources are applied, we should look at the source
signal, i.e., energy wavelength, . It is given as
c
D (9.2)
f
where c is the transfer speed and f is the frequency defined by the energy source.
Wavelength must be larger than the longest dimension in the physical system so
that the system can be treated as mono-dimensional. When that is not the case, then
system has to be treated as multi-dimensional since impedance is not concentrated
in one point only. It is distributed along the length of the system and following that,
partial differential equations have to be used, with two independent variables, time,
and the distance.
240 M. Simic

9.3 Acoustic Elements

Exhaust system is a nonlinear, multi-dimensional physical network. Since we are

now concentrated on the acoustics of the system, only, it is for us an acoustic
physical network. Nonlinearities come from the changes of the temperature and
the gas pressure along the system, as well as the physical dimensions and the shape
of the system structure.

9.3.1 From Multi to Mono-Dimensional System Model

Let us start with the dimensionality problem first. The longest dimension in the
systems is the length, which is 3,800 mm. The speed of sound, in the air, at 20 ı C is
equal to c20 D 343 m/s. Temperature of the gasses is higher than the temperature of
the environment, which influence the speed, but for the initial estimations we will
take the speed as c D c20 D 343 m/s. Now we need the frequencies of the emitted
sound. In order to get that, sound is recorded and power spectrum is extracted using
the program shown in Fig. 9.2.

Fig. 9.2 Power spectrum measurement program

9 Exhaust System Acoustic Modeling 241

Fig. 9.3 Exhaust system sound power spectrum

Powers spectrum of the sound is shown in Fig. 9.3. We can see that the majority
of energy is transferred by the frequency components in the rage of 10–1,000 Hz.
The maximum energy is carried by the components around 500 Hz.
Following that we can calculate wavelengths associated to those frequencies:
c 343 c 343
1 D D D 34:3m 2 D D D 0:343m (9.3)
f1 10 f2 1; 000
We will now split the system into the subsystems whose maximum linear dimen-
sions are small compared to the minimum wavelength in the range of interest, which
is 2 . In the first simulation we will adapt the subsystems’ lengths of 80, 160, and
200 mm as given by the vector V. Vector components, associated to the exhaust
system from the Fig. 9.1, are all given in mm. After comparing simulation results
by the real measurements, given in the Fig. 9.3, we will see if further subdivisions
are needed.
V D .200; 200; 200; 200; 200; 200; 80; 80; 80; 80; 80; 80; 200; 200; 200; 200; 200; 200; 200; 200/
(9.4)
242 M. Simic

Fig. 9.4 Basic acoustic subsystem

9.3.2 Acoustic Network Elements Linearization

Exhaust system, shown in the Fig. 9.1, can be represented as a cascaded network
of Ra La Ca subsystems, shown in Fig. 9.4, where Ra is a symbol for the acoustic
resistance, La is a symbol for the acoustic inductance, and Ca represents acoustic
capacity.
Acoustic resistivity is an element that dissipates acoustic energy in the system.
It incorporates all sound attenuation components, as found along the whole exhaust,
especially in catalytic converters and mufflers. For the simplicity of the initial
modeling we will neglect effects of the resistivity as a separate component and
represent basic subsystem without proportional element, i.e., resistivity, as shown
in the Fig. 9.4. At the same time we are going to include attenuation through the
effects of more rapid temperature decrease in the converter and mufflers chambers.
The fast temperature change is caused by the gas molecules kinetic energy drop,
when colliding with obstacles on the path through the various system components.
Acoustic capacity and inductivity are calculated using the following expressions:

Sl l
Ca D La D % (9.5)
%c 2 S

where c is the transfer speed, as before, S is the cross section surface area, l is the
length of the subsystem, and is the gas density (Sanford 1965; Simic et al. 1978).
Parameter l (length) takes the values from the components of the vector V, given by
Eq. (9.4).
The next nonlinearity issue to challenge is the temperature dependency of the
whole system that can be represented as changes in network elements La and Ca , as
the function of the temperature. Let us start with the ideal gas low equation
pV
T
D C onst
(9.6)
p1 V1 p0 V0
T1
D T0
9 Exhaust System Acoustic Modeling 243

where

p0 D 760mmHg D 0:1 106 N=m2 D 106 bar

is normal atmospheric pressure, while V0 is gas volume at the temperature of T0 ,

T0 D 273:15K D 0ı C:

Let us assume that the pressure difference at the borders of the subsystems is
4p D 200 bar. Then we have the following:

p1 D p0 C p
p1 106 C 200
D D 1:0002  1
p0 106

Following that we have simplified Eq. (9.6) as given below,

V1 V0 V0 T0
D or D (9.7)
T1 T0 V1 T1

Gas density is expressed as

m
%0 D f or T D T0
V0
m
%1 D f or T D T1
V1

Dividing last two equations we have

%1 V0
D (9.8)
%0 V1

From the Eqs. (9.7) and (9.8) we derive

V0 T0
%1 D % 0 D %0 (9.9)
V1 T1

Finally expression for the acoustic inductivity becomes

l l T0
La D % D %0 (9.10)
S S T1

Specific gas density is labeled as 0 at the temperature of T0 D 0 ı C D 273.15 ı K.

Working temperature is labeled as T1 . Gas density is another nonlinear quantity,
which depends on the engine cycle, but we will assume that it is constant and equal
to the air density.
244 M. Simic

Fig. 9.5 Gas temperature distribution along the system

Finally expression for the acoustic capacity would be:

Sl Sl Sl
Ca D D r 2
D (9.11)
%c 2 % 0 c0 2
%0 TT01 c0 TT10

We can see that the acoustic capacity is not the function of the temperature.
Gas burning temperature depends on the engine type but it is around 650 ı C
generally. At the exit, out of the tailpipe, temperature is still above ambient
temperature and we will adapt the value of 35 ı C. Temperature distribution is given
in the Fig. 9.5.
We can see that there are 20 subsystems, with the length distribution as given by
the vector V, Eq. (9.4), and with the temperature distribution as per Fig. 9.5, from
600 ı C at the start of the cascade down to 35 ı C at the end. In addition to that, each
subsystem has its cross section surface area as specified by Fig. 9.1. We have circles
with given diameters as ˚1 D 40 and ˚2 D 120 and a rectangle with 80  100, all
in mm.
In the subsystems which have higher temperatures, sound speed is higher too, and
accordingly the wavelength as per Eq. (9.2). As a consequence, the approximation
of the physical system is more accurate because of the better relationship between
wavelength and the maximum liner dimensions of the system.

9.4 Ladder Topology

We can now design the whole system. Figure 9.6 shows a segment of the ladder
network that consists of the subsystems as given in the Fig. 9.4.
In this physical network we have flow and pressure as network variables.
An Excel program is used to calculate distributed parameters. A fragment from
the calculations is demonstrated in Fig. 9.7.
9 Exhaust System Acoustic Modeling 245

Fig. 9.6 A segment from the acoustic ladder network

Fig. 9.7 Distributed parameters calculations

We have to clarify just one more design solution. Header, as shown in Figs. 9.1
and 9.8, consists of four extractors with the diameter of 40 mm and the length of
200 mm. Those four pipe segments can be represented in the physical network
as four parallel connected acoustic inductors and four parallel connected acoustic
capacitors, as shown in Fig. 9.8. Energy sources are presented as series connection
with acoustic inductivities. Equivalent circuit or network structure is given in
Fig. 9.8. This subsystem is presented as parallel connection of elements because we
have about the same pressure difference on each of them and the total flow down the
pipe, i.e., the whole exhaust system is equal to the sum of four flows. This qualifies
such a system as a parallel structure in the physical networks. According to that,
total equivalent inductivity Lae and capacity Cae of the first subsystem are given as

La
Lae D and Cae D 4Ca (9.12)
4
To finalize this discussion we have to mention that the exhaust system behaves as a
sound source at the end of tailpipe. We need to know impedance of the open pipe.
It can be show that it is predominantly inductive. We also have to know that the gas
has a one-directional motion component, apart from oscillations, and that it moves
outside of the exhaust forming a hot gas cylinder. By observation we concluded that
this cylinder has the length of around 200 mm as the most of our subsystems. We
put that in the calculations together with the simulation of the open space by the
larger values for the last subsystem dimensions.
246 M. Simic

Fig. 9.8 Header and its equivalent physical network: Flow sources F(t) are represented by arrows,
acoustic inductors as La and capacitors as Ca. Equivalent circuit is given on the right where
equivalent source flow is 4F(t) and equivalent components are as given in the text and on the
figure

Fig. 9.9 Simulink model of the cascaded acoustic filter

We have used various modeling environments in the different stages of the

design. Modern technology enables easy integration and combination of results
from the different programming platforms. Our model in Simulink is shown in
Fig. 9.9. Ladder network shown in Fig. 9.6 is built on L half sections, while the
other one from the Fig. 9.9 is based on T section filter subsystems integration.

9.5 Transfer Function

Let us look at the basic L Half section of our ladder network as shown in Fig. 9.4.
Transfer function in Laplace domain of this La Ca filter is given as
1
uout 1
T D D Cs
D (9.13)
ui n 1
Cs
C Ls 1 C LC s 2

Applying this to the whole ladder network we come up with the acoustic transfer
function of the exhaust system.
9 Exhaust System Acoustic Modeling 247

Fig. 9.10 Bode plots of the exhaust system

1
Tal l D YiD20 (9.14)
.1 C Lai Cai s 2 /
iD1

Bode plot of the transfer function is calculated using MATLAB and the magni-
tude and phase shift graphs are shown in the Fig. 9.10.

9.6 Conclusion

Comparing bode plot graph from the Fig. 9.9 and the power spectrum diagram
from the real exhaust system, shown on the Fig. 9.3, we can see that our model
simulate exhaust system with extremely high precision. Figure 9.9 also presents
phase response of our acoustic ladder filter. Both transfer function diagrams have
frequency expressed in rad/s shown on x axis in logarithmic scale. Phase shift plot
is shown in degrees, while magnitude is presented in decibels of power. The range
presented is from 102 to 105 rad/s. Since we know that w[rad/s] D 2f [Hz], the
frequency range covered by the diagrams in Fig. 9.9, in Hz is from 16 to 1,600 Hz.
This was our range of interest according to the real, measured data from the vehicle,
Fig. 9.3.
248 M. Simic

This model is now a tool for the further investigation in acoustic properties and
design of the new exhaust systems. We can easily increase number of subsystems
and so achieve better relationship between wavelength and the maximum dimension
in the subsystem. In addition to that we could place temperature sensors along
the system to get more accurate temperature distribution. Microphones, pressure
sensors, infrared, and other could be also used to collect more real data (Ba et al.
2010).

Key Symbols

A, B, C Network elements
Ai Ordinary Differential Equation (ODE) Coefficient i, i D 0, 1, : : : , n
c Wave transfer speed
C Capacity
Ca Acoustic capacity
Cae Equivalent acoustic capacity
diy
dt
ith derivative of y, i D 1, 2, : : : , n
dp An infinitesimal change in potential
dq An infinitesimal change in electric charge
dq
dt
First derivative of charge with respect to time
dt An infinitesimal change in time
du An infinitesimal change in voltage
dv An infinitesimal change in velocity
E Equivalent sound source
f Flow
f Frequency
f (t) Function of time
F, F(t) Flow source
F Force
G Conductivity
i Electric current
k Spring stiffness
l Subsystem length
L Inductivity
La Acoustic inductance
Lae Equivalent acoustic inductance
m Mass
p Potential
P, P(t) Potential source
q Electric charge
R Resistivity
Ra Acoustic resistivity
s Complex argument
9 Exhaust System Acoustic Modeling 249

S Cross section surface area

t Time
T, T0 Temperature
T Transfer function
u Voltage
v Velocity
w Angular frequency
V, V0 Gas volume
V Lengths vector
y Variable
yG (t) ODE general solution
yH (t) Homogenous ODE solution
yP (t) A particular solution of an ODE
Y Admittance
Z Impedance
4p Pressure difference
 Wavelength
 pi constant
Q
iD20
Product from i D 1 to i D 20
iD1
Gas density
˚ Diameter

References

Ba F, Wan L (2010) Five components automobile exhaust measurement system. In: International
conference on educational and network technology (ICENT), pp 248–250
Capel EM, Ibrahim T, Nor NM (2013) Hybrid energy from exhaust system. In: IEEE seventh
international power engineering and optimization conference (PEOCO), pp 134–138
Elnemr Y (2011) Acoustic modeling and testing of exhaust and intake system components.
Licentiate Thesis, Department of Aeronautical and Vehicle Engineering, Royal Institute of
Technology, Stockholm
Jianwang S, Aimin D, Xiaoxiong J (2008) Research of automotive exhaust system hanger location.
In: IEEE Vehicle power and propulsion conference (VPPC’08), pp 1–3
Peng H, Yunqing L, Lifeng Z (2009) Study on exhaust system parameters for fuel economy
improvement of small gasoline engine. In: International technology and innovation conference
(ITIC 2009), pp 1–5
Sanford RS (1965) Physical networks. Prentice-Hall, Englewood Cliffs
Silva CW (2005) Mechatronics an integrated approach. CRC Press LLC, New York
Simic M (1978) Physical systems analysis using ECAP. Masters Research, Electronics Engineer-
ing, University of Nis, Nis, Serbia
Chapter 10
Nonlinear Approaches in Three Dimensional
Medical Image Registration

Ruwan Tennakoon, Alireza Bab-Hadiashar, and Zhenwei Cao

Keywords Non-rigid registration • Medical image analysis • Deformable

registration

10.1 Introduction

Image registration is the task of finding a spatial transformation T that aligns the
objects in two or more images capturing the same or related scene. It is one of
the most crucial problems of computer vision and has been studied for over three
decades. The underlying task is very general and has a wide range of applications in
the fields of medical imaging, machine vision, remote sensing, cartography, etc. In
this chapter we focus on the application of image registration in the fields of medical
imaging.
Although registration can involve two or more images, typical medical image
registration problem comprises of two images. The two images to be registered are
named fixed or target (IF (x)) and moving or source (IM (x)) images. The two images
can be from different times, captured using different sensors or from different
viewpoints (Zitova and Flusser 2003).

R. Tennakoon () • Z. Cao

Faculty of Science, Engineering and Technology, Swinburne University of Technology,
Hawthorn, VIC 3122, Australia
e-mail: [email protected]
A. Bab-Hadiashar
School of Aerospace, Mechanical and Manufacturing, RMIT University,
Melbourne, VIC 3001, Australia

© Springer International Publishing Switzerland 2015 251

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__10
252 R. Tennakoon et al.

Fig. 10.1 In this illustration the moving image needs to be both translated and rotated to align
with the fixed image

Mathematically the problem of image registration can be expressed as finding

the spatial mapping

T W F ! M

such that the following correspondence holds:

IF .x/ $ IM .T .x//

where F is the fixed image domain with dimensionality d and M the moving
image domain. The image dimensionality d can be either two (2D) or three (3D).
Hence, the transformation function can be a mapping from 2D ! 2D, 3D ! 2D or
3D ! 3D. Figure 10.1 shows an example of two images with overlap that can be
registered in one coordinate system by translating and rotating one image (moving)
to be aligned with the other (fixed) image.
Medical imaging consists of many modalities depending on the technique used
to capture the images. The common modalities are X-ray, computed tomography
(CT), magnetic resonant imaging (MRI), positron emission tomography (PET),
single proton emission tomography (SPECT) and ultrasound (US) imaging. These
modalities can be categorised into two main groups according to the purpose of the
images (Oliveira and Tavares 2014): Anatomical images and functional images. CT,
MRI, US and X-ray fall into the anatomical category where the purpose of the image
is to study the anatomy, while functional MRI (fMRI), PET and SPECT fall into the
functional category where the purpose is to study the functions performed by the
imaged organ. Figure 10.2 shows brain images from the same subject captured using
three different techniques (CT, MIT and PET). We can observe that each modality
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 253

Fig. 10.2 (a) CT, (b), MRI, (c) PET images of the same subject. Images are from the RIRE dataset
(http://www.insight-journal.org/rire)

provides a specific type of information. For example, MRI captures the soft tissue
in detail whereas the bone structures are clearer in the CT image. A registration task
may involve images from the same modality or different modalities. Depending on
the number of modalities involved, the registration task can be classified as either
mono-modal or multi-modal registration. In mono-modal registration both images
are from the same modality (e.g. time series CT image registration for change
detection), whereas in multi-modal registration the images are from two different
modalities (e.g. fusion of CT with MRI images for image guided radiotherapy).
The solution to the registration problem generally consists of three main parts:
transformation model, objective function and the optimisation procedure (Sotiras
et al. 2013; Oliveira and Tavares 2014). Other than the key parts there are several
other support modules such as the image interpolation, multi-resolution pyramids,
pre-processing and feature extraction. The techniques used in each module of
a given algorithm depend on the specific requirements of the given registration
problem.
Transformation Model. Transformation model defines the nature of the geometric
transformation between the two images. The selection of the model depends on the
expected deformations in the imaged object and the selection critically affects the
accuracy of the final registration outcome.
Objective Function. The objective function characterises the desired output of
the registration process. Typically an objective function consists of two parts: the
(dis-)similarity measure M, which quantifies the alignment between images and
the regularisation term R, which constrains the solution domain. A typical objective
function is as follows:

M . IF .x/; IM .T .x/// C  R.T / (10.1)

254 R. Tennakoon et al.

where  represents the trade-off between alignment and complying with the
constraints. The choice of appropriate measure in objective function depends on the
modalities involved in the given registration problem as well as the image quality,
texture and deformation involved.
Optimisation Algorithm. The optimisation method tries to find the parameters
of the transformation model that maximises (or minimises) the similarity (or dis-
similarity) measure. In a very few cases using landmark based objective function
the optimal parameters can be directly calculated but in general more sophisticated
optimisation algorithms are needed to find the optimal transformation model.
In addition to the three key parts described above, there are several support
modules involve in registration, such as image interpolation, hierarchical strategies
and pre-processing.
Interpolator. The mapped point from one image domain to the other may not
exactly coincide with grid position. Hence, an interpolator is used to calculate
the intensity of the non-grid positions using grid values. The most common
interpolation functions used in medical image registration are nearest neighbour,
linear and B-Spline interpolators.
Hierarchical Strategies. The complex and nonlinear objective functions in image
registration lead to many local minima. In order to escape from being trapped by
one of these local minima, most registration algorithms use hierarchical approaches.
These techniques can be classified into three main categories (Lester and Arridge
1999): increasing data complexity, increasing warp complexity and increasing
model complexity.
Image Pre-processing. In order to get the best performance, the images for
registration may need to be pre-processed before feeding them to the registration
algorithm. Smoothing, motion correction, segmentation of relevant organ and nor-
malisation are some of the pre-processing steps that can be performed. Although the
registration algorithms vary widely, in general the flow of a registration algorithm
can be summarised by flowcharts shown in Fig. 10.3.

10.1.1 Application of Image Registration in Medical Imaging

There are many applications of image registration in medical imaging field and some
of those are listed below. A detailed description of those applications is outside the
scope of this exposition and can be found in Holden (2008), Oliveira and Tavares
(2014) and Rueckert and Aljabar (2010).
• Disease Diagnosis: The early detection is the most effective way to reduce
mortality in many diseases. Computer Aided Diagnosis (CAD) systems are
developed to enhance the capabilities of medical practitioners by attracting
attention to certain areas in images, providing better comparisons and enabling
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 255

b
Fixed
Fixed image
ima
mage Moving Image

a
Fixed image Moving Image Feature
Feat
ature Feature
eat
ature
Extraction
Extraction
Ex Extraction
Extraction
Ex

Multi-resolution Multi-resolution Determining Feature

Pyramid Pyramid set correspondance

Model parameter
Objective Function Interpolation Search

Transformation Final model

Optimisation parameters
model

Fig. 10.3 Registration flow of a typical intensity information based registration algorithm

medical practitioners to measure tumour sizes effectively. These systems can

speed up the diagnostic process, reduce errors or misses and improve quantitative
evaluation. Image registration plays a major role in CAD by enabling to combine
information from multiple views, previous times, and multiple modalities (Giger
et al. 2008). CAD is used to diagnose breast cancer (Wei et al. 2009), lung cancer
(Guo et al. 2012; Shiraishi et al. 2006; Van Ginneken et al. 2001), Parkinson’s
disease (Illán et al. 2012), emphysema (Sluimer et al. 2006), brain tumour (Jaya
and Thanushkodi 2011) and many other common diseases.
• Progress monitoring: Detecting and measuring changes of an anomalous region
over time is vital to evaluate the success and to plan the future course of med-
ication. But due to orientation changes, organ motions and anatomical changes
during successive image acquisitions, direct comparisons are not possible. Image
registration helps to reduce the effects of these deformations in order to make
a realistic comparison (Ko and Betke 2001; Liang et al. 2013; Arzhaeva et al.
2010; Timp et al. 2007; Yuanjie et al. 2009).
• Image guided surgery: Image guidance provides valuable information about the
target position and optimal probe manipulations in minimal invasive treatment
procedures such as neurosurgery (Huang et al. 2012; Mostayed et al. 2013),
endoscopy (Burschka et al. 2004; Mirota et al. 2012), interventional cardiology
(Linte et al. 2010; Xishi et al. 2009), minimal invasive lung surgery (Sadeghi
Naini et al. 2010; Uneri et al. 2013) and robotic microsurgery (Gerber et al.
2013). Here registration plays an essential role in combining and augmenting
the high quality pre-operative images (CT, MRI, PET) with the intra-operative
images from patient on the operating table (X-ray, US, interventional corn
beam CT). A comprehensive review of registration techniques that are applied
specifically for minimal invasive therapy can be found in Rui et al. (2013).
256 R. Tennakoon et al.

• Image guided radiotherapy: In radiotherapy generous safety margins have to

be set around the target due to organ movements and target localisation errors.
These margins lead to a compromise in the dose delivered to the tumour hence,
accurate target localisation (defining the tumour volume) and alignment is very
important in order to maximise the radiation dose to the target while minimising
the effects on the surrounding organs. Image guided radiotherapy techniques
use on-line or off-line image registration. Off-line registration of images from
different modalities (CT/MRI, CT/PET) can improve target localisation whereas
organ motion can be determined by on-line registration of repeatedly acquired
images. This information is then used to adapt the treatment (patient positioning,
directing radiation beam) (van Herk 2007; Robertson et al. 2013; Worm et al.
2010).
In medical imaging applications the deformations involved in the images are
often highly nonlinear. Also the relationships between the voxel intensities of the
same structure in two different images can be nonlinear, especially for multi-modal
registration. Due to these reasons the resulting optimisation is highly nonlinear and
the main objective of this chapter is to explore the use of nonlinear methods in
medical image registration applications.
Section 10.2 explores the nonlinear similarity measures used in medical image
registration while Sect. 10.3 investigates the transformation models and Sect. 10.4
the solution to nonlinear registration problem. Finally Sect. 10.5 concludes the
chapter.

10.2 Similarity Measures

A similarity measure is part of the objective function which quantifies the alignment
between the images being registered. Registration algorithms can be broadly
classified into two main groups, depending on the nature of the information used
to calculate the similarity measure: extrinsic or intrinsic. Extrinsic methods use
artificial markers attached to the patient, prior to being imaged. These markers
are visible in all imaged modalities and will lead to efficient registration methods
where parameters can be calculated explicitly. However, due to the invasiveness
of attaching markers, restrictions on the number of markers that can be attached
and the difficulty in capturing non-rigid motions, these methods are limited to rigid
registration and pose estimation (Maintz and Viergever 1998). In contrast, intrinsic
methods rely only on information readily available in the image. Depending on the
feature space used in the similarity measure, registration algorithms are broadly
classified into three main categories (Sotiras et al. 2013): image feature based
methods, voxel information based methods and hybrid methods. Feature based
methods use landmarks or features extracted from the images in the registration
process while the voxel information based methods use the voxel information
directly. Hybrid methods are a combination of the two techniques.
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 257

10.2.1 Voxel Information Based Similarity Measures

In voxel information based registration, the similarity measure takes the intensity
information of the two images as input and outputs a value that relates to how well
the two images are aligned. It is important to note that in medical image registration
the images can be from the same modality (mono-modal registration) or they can be
from different modalities (multi-modal registration).
Firstly, we consider the mono-modal registration case where the images are
acquired using the same imaging technique. Examples of this type of registration
would be MRI–MRI registration or CT–CT registration. The simplest similarity
measure is based on the assumption that corresponding anatomical structures in
both images have similar voxel values except for Gaussian noise and this measure
is named sum of square difference (SSD) similarity measure. The mathematical
representation of this measure is as follows:

1 X
MSSD ./ D . IF .x/  IM .T .x///2 (10.2)
N
x 2 F \ M

where N is the number of voxels in the overlapping region of these images. In cases
when the images contain some voxels with large errors (outliers) SSD would fail, as
these high error values skew the objective function. This can be common in medical
imaging due to the injection of contrasting agents or position changes of the patient
in two time-lapse images (Hill et al. 2001). This problem can be mitigated by using
the sum of absolute difference (SAD) similarity measure instead, which is more
robust to outliers.
1 X
MSAD ./ D j IF .x/  IM .T .x//j (10.3)
N
x 2 F \ M

The assumption that the voxel values are identical except for Gaussian noise is
very restrictive and a more realistic assumption would be to hypothesise a linear
relationship between intensities. For this assumption the optimal similarity measure
is cross correlation or normalised cross correlation of image intensities which can
be represented as:
X
f. IF .x/  F / . IM .T .x//  M /g
MC C D qX x X I x 2 .F \ M /
. IF .x/  F /2 . IM .T .x//  M /2
x x
(10.4)

In the above measure F and M are the mean intensities of the fixed and the moving
images, respectively. It is important to note here that even though the relationship
between the intensity values of the two images is assumed to be linear, the resulting
cost functions will be nonlinear.
258 R. Tennakoon et al.

In some medical image registration applications, where the organ of interest

is elastic, the assumption of linear relationship between corresponding voxels is
unrealistic. Examples of this scenario would be the lung CT images. The density of
tissue, which is measured by CT, may vary depending on the volume of the air inside
the lung. Hence the intensity of the same anatomical structure would change during
the breathing cycle. In order to overcome this problem specifically in lung image
registration there have been methods developed that assume a nonlinear relationship
between voxel values, using the conservation of mass principle. Gorbunova et al.
(2012) assumed that the mass of parenchyma remains constant and the density of
the lung tissue is inversely proportional to the amount of air. The change in volume
was calculated using the determinant of the Jacobian (Yin et al. 2009) and this was
integrated to the model to obtain the following cost function.

1 X
MMP D . IF .x/  jJT .x/j : IM .T .x///2 (10.5)
N
x 2 . F \ M /

where jJT (x)j is the determinant of the Jacobian of the transform model.
Castillo et al. (2009, 2012) used the principle of mass preservation in an
optical
R flow setting. R They incorporated the mass preservation assumption, which
is,  IF (x) dx D  IM (x) dx, in an optical flow equation:

IF .x/ C r IM .x/  v C IM .x/div.v/ D 0 (10.6)

where v is the velocity field in the image. The integrated form of this equation
(Corpetti et al. 2002) leads to the following compressible flow similarity measure:

1 X
MCF D .ln .IF .x//  ln .IM .T .x/// C div .T .x/  x//2 : (10.7)
N
x 2N.x/

It is important to note that most similarity measures such as SSD and SAD
assume independence between intensities from voxel to voxel. This means that
if the same permutation of voxel indices is applied to both images the resulting
similarity measure will be unchanged. To account for the spatially varying intensity
distortions, Myronenko and Xubo (2010) developed a new similarity measure
named residual complexity. To arrive at the correct form of this measure, they first
introduced an intensity correction field that represents the differences of intensities
of the registered images. By analytically solving for this correction field and using
adaptive regularisation, they derived the following similarity measure:
 T 2 !
X
N
qn r
MRC D log C1 (10.8)
nD1
/
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 259

where r D (IF (x)  IM (T(x))) and qi is an eigenvector of Q where QƒQT is the

spectral decomposition of PT P where P is the regularisation operator of the
correlation field.
Secondly, we will consider the multi-modal image registration. Here, the voxel
information of the two images (to be registered) is generated through different
physical processes and as such the relationship between the voxel intensities of the
same structure can be complex. For this kind of applications the use of information
theory measures such as Mutual Information (MI) proposed by Viola and Wells III
(1997) and Collignon et al. (1995) has gained a lot of attention. A detailed review
of these techniques was provided by Pluim et al. (2003). In these methods only
a probabilistic relationship between the voxel values is assumed and the mutual
information is designed to measure how well one image explains the other image.
The similarity measure based on mutual information can be expressed as:
   
MMI D H . IF / C H IM T  H IF ; IM T (10.9)

where H(I) is the Shannon entropy of the image and H( IF , IM T ) is the joint entropy
of two images. Figure 10.4 shows the joint histograms between a brain MRI image
and a rotated version of itself. These plots show that when the images are perfectly
registered the joint histogram is concentrated whereas when the mis-registered

Fig. 10.4 Joint histograms between the original image (top-left) and rotated versions of it. (Top-
right) No rotation, (bottom-left) rotated by 2ı , (bottom-right) rotated by 15ı
260 R. Tennakoon et al.

increases they become dispersed. The mutual information value for each case is
also shown in this figure and we can see that the more the images aligned, the larger
the mutual information values.
The above measure does not work well when the overlap of two images is low.
To solve this problem, the normalised mutual information (10.10) was proposed
(Studholme et al. 1999).
 
H . IF / C H IM T
MNMI D   (10.10)
H IF ; IM T

Cahill et al. (2008) further analysed the overlap invariance of several information
theory based similarity measures including NMI and found that in some cases
those methods would still fail. To make the NMI measure overlap invariant, they
introduced a modified form of NMI.
Another drawback of mutual information based methods is that it does not
consider the relevant spatial information. As a result, MI based similarity measures
may face problems in presence of local structures or spatially varying intensity
distortions. Various techniques have been proposed that integrate spatial information
in the MI cost function. For instance, Pluim et al. (2000) proposed a new measure
that combined mutual information with spatial information given by gradient
magnitude and orientation. The new measure was constructed by simply multiplying
the NMI cost function with proposed gradient measure. In contrast, Rueckert et al.
(2000) incorporate the spatial information using the second order entropy and joint
entropy in the calculation of NMI.
Another approach to include spatial information in the MI cost function is to
use local estimation of the mutual information by progressively subdividing the
images (Likar and Pernuš 2001; Andronache et al. 2006). Studholme et al. (2006)
combined these regional terms into one global cost function called Regional Mutual
Information (RMI). RMI incorporates the region information describing the global
relationship between intensities, as a third channel of information (in addition to
the two intensity channels) in MI calculation. Using additional channels in MI
calculation is computationally expensive and Klein et al. (2008) proposed to reduce
the computational complexity by using random sampling. Loeckx et al. (2010) also
incorporated a third spatial dimension to create the conditional mutual information
(cMI) similarity measure.
Local structural information can also be added to MI computation using the
graph based ’-MI techniques (Hero et al. 2002; Staring et al. 2009). Recently,
Rivaz and Collins (2012) used self-similarity, which measures the similarity of one
image patch to another patch in the same image in a weighted graph based ’-MI
implementation for non-rigid image registration.
Calculation of mutual information involves some nontrivial and sensitive
smoothing parameters for density approximations and the resulting cost function
is highly non-convex and contains many local minima. In order to overcome these
limitations, (Haber and Modersitzki 2006; Modersitzki 2008) proposed the use of
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 261

Normalised Gradient Fields (NGFs). The rationale behind this is that if the images
are mis-registered the angle between the two gradient fields at a given point will be
non-zero. The resulting similarity measure can be expressed as:

1X
MNF G D kn© .IF ; x/  n© .IM ; T .x//k (10.11)
2 x

where

rI.x/
n© .I; x/ D p (10.12)
rI .x/0 rI.x/ C ©2

Ruhaak et al. (2013) used this similarity measure in registration of full-body CT and
PET images.

10.2.2 Image Feature Based Similarity Measures

The next class of image registration algorithms is based on matching image

features. These methods first identify anatomically important point, curve or surface
landmarks of the images. Once the correspondences are established between the
landmarks, a similarity measure (SSD) that minimises the Euclidian distance
between points can be used to calculate the model parameters. A detailed review
of feature based registration methods was provided by Audette et al. (2000). The
interest in the use of pure feature based methods has gradually diminished since the
landmark or feature extraction processes introduce some errors that would propagate
through the algorithm. Moreover, the need for extrapolation/interpolation from a
sparse set of landmarks to recover the dense deformation field means that the
estimation of those fields away from the landmarks will be unreliable (Sotiras et
al. 2013).
Intensity based methods only account for intensity patterns and do not care about
the structure while feature based methods only use sparse landmarks. The hybrid
algorithms combine the two in order to achieve more robust registration results
(Rueckert and Aljabar 2010).

10.3 Transformation Model

The transformation model in image registration context is a spatial mapping from

one image domain to another. The transformation model that best matches a given
registration application depends on the underlying deformations that are to be
recovered. There are many transformation models that are used in medical image
262 R. Tennakoon et al.

processing. A transformation model or function can be either linear or nonlinear.

The definition of a linear transformation function is as follows:
Definition Let V and W be two vector spaces over the same scalar field F. A
function

T WV ! W

is said to be a linear transformation if the following two properties hold for any two
points x, y 2 V and scalar  2 F (Rynne and Youngson 2008):

T .x C y/ D T .x/ C T .y/

T .x/ D  T .x/:

In general linear transformations map a line to a line or a point. In contrast

nonlinear transformations map lines to curves. In most cases of medical image
registration the nature of the deformations is nonlinear due to the non-rigid nature
of organs and muscles. Hence, transformation functions that are used are mostly
nonlinear.
In image registration, the objective is to map the moving image to the coordinate
system of the fixed image which is called the forward transformation. However,
most algorithms due to practical advantages calculate the inverse transformation
(transformation from fixed image to moving image). The rationale behind this
choice (Sotiras et al. 2013) is that if a forward transformation is calculated, in
order to generate the transformed image each grid point in M needs to be
transformed using the calculated transformation function followed by a scattered
data interpolation and due to the large number of points the computational cost of
the interpolation would be huge. On the other hand, if the inverse transformation
is calculated, the values of transformed image can be calculated by transforming
each point in the new grid to M and picking the relevant value from the moving
image (when the transformation leads to a non-grid position in the moving image,
intensity interpolation can be done using neighbouring voxels), which has far less
computational workload. Figure 10.5 further elaborates this point.
Transformation models can be first grouped into two main categories: rigid and
non-rigid. The simplest type of transformation function used in medical imaging is
the rigid transformation. Rigid Transformation is a combination of a rotation and a
translation. In three-dimensions, rigid motion can be represented by three rotation
and three translation parameters. Rigid registration can be used to match images that
are of bone structures or enclosed by bony structures.
The rigid transformation falls within the broader category of affine transforma-
tions. This can be expressed using the following equation:
0
xŒd 1 D AŒd d  xŒd 1 C bŒd 1 (10.13)
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 263

a
Transformed Moving Image
Image

Scatter data
Moving Image
Interpolation

Fixed image Intensity

grid Interpolation

Transformed
Image

Fig. 10.5 Inverse transformation vs. forward transformation. The figure shows that using inverse
transform is computationally advantageous. (a) Inverse transform. (b) Forward transform

Affine transformation consists of a linear transformation (A  x) representing rota-

tion, scaling and shear and a translation vector (b). Except for the special case
where the translation part is zero the affine transformation in general is nonlinear,
contrary to its appearance that has caused some authors to classify it as linear
transform (Glocker et al. 2011; Lester and Arridge 1999). Due to the scaling and
shear components the general affine transformation is also categorised as non-rigid.
Affine registration is sometimes used in cases where images are distorted due to
physical characteristics of the acquisition system. It is also extensively used as a
pre-registration technique in complex registration algorithms with large degrees of
freedom (Fig. 10.6).
The limited degrees of freedom allowed by affine or the rigid transformation
is not adequate to represent the complex deformations involved in medical image
registration. Hence, non-rigid registration models with much higher degrees of
freedom are introduced. Holden (2008) and Sotiras et al. (2013) categorised
those non-rigid registration algorithms into three main groups: transformations
derived from physical models, transformations derived from interpolation theory
and constraints on transformation.

10.3.1 Transformations Derived from Physical Models

These models use the characteristics of physical phenomena such as elastic material
and viscous fluid flows to model the deformations in medical images.
264 R. Tennakoon et al.

Fig. 10.6 Original image (c) and the image after an affine transform (d). The transformation of
both the grid (a, b) and the intensities are shown

Elastic Model

These models assume that the images behave similar to elastic bodies under
deformation. Here, the external forces due to the mismatch in images (driving forces
in registration are calculated using the similarity measure) are balanced with the
internal forces that impose smoothness in the deformation field.
The elastic deformations of an object can be represented using Navier–Cauchy
equation:
 
r 2 u.x/ C . C / r r T u.x/ C f .x/ D 0 (10.14)

where f (x) is the external force that drives the registration, which is the similarity
measure, u(x) is the deformation at point x,  quantifies the stiffness of the material
and  is the Lamés first coefficient.
This Navier–Cauchy partial differential equation leads to an optimisation prob-
lem which can be solved in many ways such as: variational, finite difference, finite
element, basis function expansions and Fourier transform methods (Modersitzki
2004; Broit 1981; Gee et al. 1994; Christensen et al. 1994).
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 265

One of the drawbacks in using this model is that the resulting transformation
is not inversely consistent. This can be corrected either by adding a constraint
that penalises the inverse inconsistency (after calculating forward and backward
transformations) or by coupling the forward and backward transformations in
deriving the final transformation. Detailed description of this is provided later when
constraint on transformation is explained.
In the linear elastic model explained above, the second order terms of the
displacement field gradients are ignored. This means that the relationship between
stress and strain is assumed to be linear. This makes this model incapable of
handling large deformations. To overcome these issues nonlinear elastic models
were developed. These models were based on hyperelastic material properties
(Rabbitt et al. 1995) and St Venant–Kirchoff elasticity energy (Pennec et al. 2005).

Fluid Flow

In fluid flow models the images are assumed to behave as viscous fluids under
deformation. The advantage of these models is that they do not assume small
deformations.
The fluid flow transformation model can be represented by the following Navier–
Stokes partial deferential equation.
 
r 2 v.x/ C . C / r r T v.x/ C f .x; u/ D 0: (10.15)

This equation is very similar to the elastic model equation in the previous section.
The only difference is that in fluid flow model a velocity field is used instead of the
deformation field. In the above equation the first term imposes spatial smoothness of
the velocity field whereas the second term allows for expansions and contractions.
 and  are the viscosity coefficients.
Finding the solution to fluid flow models involves solving a large set of partial
differential equations. The earliest implementations used successive over-relaxation
which is computationally expensive (Christensen et al. 1994). Christensen et al.
(1996) used parallel computations to reduce the computational time. Some faster
ways of solving differential equations such as scale space convolution with a filter
(impulse response of the regularisation operator) (Bro-Nielsen and Gramkow 1996)
have also been proposed.

10.3.2 Transformations Derived from Interpolation Theory

In the transformation models that belong to this category the displacement field is
modelled using functions derived from interpolation and approximation theories.
The requirement to intercept sample values is relaxed in approximation assuming
errors in sample values (Holden 2008). In these methods firstly the basis function
266 R. Tennakoon et al.

centre points (or control points) are selected and then the parameters of each basis
function are adjusted (or calculated) to fit the deformation field.

Radial Basis Functions

Radial basis functions (RBFs) are the most commonly used interpolation theory
based transformation model in landmark based image registration. The value of
an RBF only depends on the distance from its control point. Several such basis
functions placed on the spatial domain of the image can be used to model the
deformation field of the image effectively. The transformation function based on
RBF can be defined as follows:

X
N
T .x/ D x C k R .kx  xk k/ (10.16)
kD1

where  k are the parameters that control the shape of the deformation field,
xk , k 2 1 : : : N are the RBF control points and R() is the RBF kernel. One of
the advantages of RBFs is the ability to place the control points in an irregular
grid which enables modelling non-uniform deformations with less parameters than
a grid-based control point placement scheme.
In landmark based registration, the RBF is combined with a polynomial basis.
The polynomial basis (x) is added to take care of the more global deformations
while the RBF part highlights the local changes. This transformation function can
be as follows:

X
N X
m
T .x/ D x C k R .kx  xk k/ C N Cj j .x/ (10.17)
kD1 j D1

Once the problem is in this setting, if the landmarks and their correspondences are
available a closed form solution can be found as the RBFs are positive definite.
Some of the RBFs commonly used in literature are given in Table 10.1. Here we
note that these RBFs are nonlinear.
The RBF described above has a global support region or in other words a given
control point effects the entire deformation field. This property has adverse effects
when there are outliers or the desired deformation field is local.

Table 10.1 Commonly used Thin plate spline ri 2 log ri

radial basis function kernels p
Multi-quadratic ri 2 C d 2
where r D kx  xi k
Inverse multi-quadratic (ri C d2 ) 1/2
2
 c2
Generalised multi-quadratic ri C c12
2
.ri =c1 /
Gaussian e
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 267

In medical image registration a given image may contain several local motions
due to the presence of different anatomical structures. In order to recover these types
of local motions RBFs that have compact spatial support are proposed. Examples
of these functions are Wendland Function and Wu’s compact support RBF (Rohde
et al. 2003).

Free Form Deformation

The main idea behind free form deformation (FFD) based image registration is to
match two images by deforming one image via manipulating a grid of control points
superimposed over that image. FFD is the most widely used transformation model in
medical image registration. Typically the control point grid is a rectangular grid with
arbitrary resolution which is coarser than that of the image. The control points are
moved in such a way that a similarity measure is maximised and the deformations
at each voxel are then calculated by interpolation using local supported kernels such
as B-Splines.
A B-Spline function is a compact support kernel which makes it suitable for
modelling deformation fields with local motions. The B-Spline kernel of order zero
is a rectangle function and the higher order kernels are calculated by convolution of
the zero order kernel with itself N times (N is the degree of the kernel). The first
four B-Spline kernels are shown in Fig. 10.7.
These spline kernels are stitched together over the image domain in a rectangular
grid in order to model the deformations of the image. Given the compact support of
the B-Spline kernel, the deformation of a particular point represented by a B-Spline
would only depend on several control points in the immediate neighbourhood of
that point. Using this property, for 3D control point grid with spacing [ı x , ı y , ı z ] the

β0
1 β1 = β0 ∗ β0
β =β ∗β
2 1 0
0.8
β =β ∗β
3 2 0

0.6

0.4

0.2

0
−1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4

Fig. 10.7 B-Spline kernels

268 R. Tennakoon et al.

multivariate extension of cubic B-Splines based FFD transformation model can be

written as follows (Rueckert et al. 1999):

X
3 X
3 X
3
T .x/ D x C Bl;3 .u/Bm;3 .v/Bn;3 .w/ diCl; j Cm; kCn (10.18)
lD0 mD0nD0

where i D bx/ı x c  1, j D by/ı y c  1, k D bz/ı z c  1, u D x/ı x  bx/ı x c, v D y/ı y 

by/ı y c and w D z/ı z  bz/ı z c.
The cubic B-Spline functions in the above equation are:

.1u/3
B0;3 .u/ D 6
.3u3 C6u2 C4/
B1;3 .u/ D 6
(10.19)
. 3u3 C3u2 C3uC1 /
B2;3 .u/ D 6
.u3 /
B3;3 .u/ D 6

The B-Spline kernels are also {2 smooth at the control points which provide conti-
nuity when the control points are moved. This enables the B-Spline transformation
model to generate smooth deformations. An example of a B-Spline transformation
is shown in Fig. 10.8 where the control points are placed in a regular grid.
Rueckert et al. (1999) used the B-Spline motion model after an affine pre-
registration step for non-rigid registration of contrast enhanced breast MRI images.
In their work NMI was used as the similarity measure. Kybic and Unser (2003)
used the B-Spline deformation model coupled with SSD similarity measure to
register images from MRI, SPECT, CT and ultrasound modalities. Their method
also included a soft landmark based constraint to guide registration in difficult
scenarios.
The control node displacements act as parameters in an FFD. Hence, the details
captured by the deformation field depend on the resolution of the control point
grid. Large spacing enables the modelling of global deformations whereas small
spacing represents highly local deformations. The number of control points also
directly affects the computational complexity of the resulting optimisation problem.
The optimal control point grid is decided based on the model flexibility and
computational complexity.
The B-Spline based FFD framework does not guarantee the preservation of
topology and there are several methods that can be employed to rectify this issue.
Noblet et al. (2005) enforced topology preservation by controlling the Jacobean
of the transformation and Kim (2004) suggested a sufficient condition for local
invariability. Chun and Fessler (2008) extended this and proposed a new penalty
based approach for topology preservation B-Spline transformation.
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 269

Fig. 10.8 Original image and the image after a B-Spline transform. The transformation of both
the grid and the intensities are shown together with the control point locations

Although in the basic implementation of B-Spline based FFD the control point
grid is uniform there has been some work that employs a non-uniform grid (Wang
and Jiang 2007; Jacobson and Murphy 2011).

Wavelet and Fourier

Fourier series and wavelet functions are extensively used to represent signals. In
Fourier representation the signal is characterised as a combination of sinusoidal
signals which localise the signal in frequency domain. On the other hand, the
wavelet representation localises the signal in both frequency and spatial domain
making it ideal to represent localised deformations (Wu et al. 2000).
One of the major attractions of these techniques in modelling deformation fields
is their ability to decompose in to multi-resolution.
270 R. Tennakoon et al.

Local Affine

In locally affine transformation model, the image domain is partitioned into

independent blocks and for each block a rigid or affine transformation is calculated
by maximising a voxel based similarity measure (Pitiot et al. 2003). An issue with
this local transformation is that it may not be continuous at the block edges. In
order to overcome this issue regularisation can be performed using either nonlinear
interpolation or filtering. Another possible solution is to use fuzzy regions instead
of the definitive independent blocks (Arsigny et al. 2005).
One of the parameters needed to be selected is the size of the block. This
parameter controls the degrees of freedom in the registration and if a small block
size is selected there will be inadequate information inside the block to find the
registration while a large block size may not capture local deformations. There
are several algorithms present in literature that can automatically calculate the
appropriate block size using image information such as structural information,
residual error and motion.

10.3.3 Constraints on Transformation

Constraints on transformation model limit the available solution space by imposing

constraints based on prior knowledge of the physical properties of the objects in the
registration problem.

Inverse Consistency

Given two images IF and IM to be registered, a basic assumption would be that

the transformations that map moving to fixed images TMF : (IM ! IF ) and fixed to
moving images TFM : (IF ! IM ) are directly related:

TMF D TFM 1 (10.20)

However, most of the existing algorithms are asymmetric, which violate the above
condition. Hence, the accuracy of these algorithms may vary depending on which
image is selected as the fixed image.
One option in overcoming this issue is to add an additional term to the
objective function that penalises the difference between the forward and reverse
transformation (Christensen and He 2001).
X  
RIC D TMF .x/  TFM 1 .x/ (10.21)
x 2
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 271

Another solution is to transform both images to the same domain and construct the
final transform by taking the inverse of one part and adding it to the other.

Topology Preservation

The aim of topology preservation is to keep the connected structures together and
maintain the neighbourhood relationship between structures. This will prevent the
disappearance of existing structures and the appearance of new structures in the
image (Musse et al. 2001). In order to preserve topology the transformation function
must be continuous, bijective and the determinant of the Jacobean must be positive.

Diffeomorphic Transformation

A diffeomorphic transformation maps a differentiable manifold to another differen-

tiable manifold such that both the function and its inverse are differentiable. This
means that a diffeomorphic transformation has a unique inverse which maps that
fixed image back to the moving image. The diffeomorphic assumption is violated
in some cases such as images before a surgery and after where some anatomical
structures are changed (Christensen 1999). Diffeomorphic mapping also guarantees
topology preservation.
Continuum mechanical methods such as linear elasticity and viscous fluid
were used to enforce diffeomorphism by Christensen et al. (Christensen 1999;
Christensen et al. 1996).
Vercauteren et al. (2007) used the demons algorithm combined with Lie group
framework on diffeomorphism and an optimisation procedure on Lie groups to
provide non-parametric free-form diffeomorphic transformation.

10.4 Optimisation

The optimisation procedure tries to find the transformation model that best aligns
the images according to an objective function. The objective function ‰ employed
by most image registration scenarios is nonlinear as well as non-convex.

‰ ./ D M . IF .x/; IM .T .x/// C R .T / (10.22)

Here M is the similarity measure which quantifies the alignment between the two
images and R is a regularisation term that incorporates a prior known property of
the deformation. The similarity form for most problems is organised such that when
the alignment is at the highest the value of the similarity measure is the minimum
(used like a dissimilarity measure).
272 R. Tennakoon et al.

This leads to a nonlinear optimisation problem of the form:

b
 D arg min Œ‰ ./ (10.23)


where  are the parameters of the transformation and b  is the optimal parameters
that best align the images.
In some cases the parameters of the registration can be calculated directly by
solving a set of linear equations. An example of such a scenario is landmark (with
correspondences) based registration using RBF.
Other than these specific cases most registration problems need an iterative
approach to find the optimum transformation model. These optimisation algorithms
can be categorised into two main parts: continuous optimisation and discrete
optimisation.

10.4.1 Continuous Optimisation

The continuous optimisation methods can be used in problems where the parameters
are real valued and the objective function is differentiable. These methods update the
parameters using an iterative method that follows the equation:

nC1 D n C ˛n gn .n / (10.24)

where  are the parameters, n is the index of the current iteration, ˛ n is the step
length and gn ( n ) is the search direction. There are several methods that belong to
this category and they differ based on the method used to calculate step length and
search direction.
In gradient descent (or steepest descent) the cost function gradients at the current
position is selected as search direction. Once the direction is set, a line search
method or a gradually decreasing function can be used as the step length (Klein et
al. 2007). The gradient decent descent has a low convergence rate. The nonlinear
conjugate gradient method provides a better convergence where the search for
direction is along the gradient component from which previous gradient direction is
excluded. In both of the above algorithms the knowledge of a mathematical model is
needed in order to calculate the gradients. However, in Powell’s conjugate gradient
method no such information is needed as it does not calculate gradients.
In Quasi-Newton methods Hessian of the objective function is set as the step
length. Because the calculation and inversion of the Hessian is computationally
expensive an approximation of the Hessian is used commonly. Some well-known
methods to approximate the Hessian are Symmetric-Rank-1, Davidon–Fletcher–
Powell and Broyden–Fletcher–Goldfarb–Shanno.
Gauss–Newton method is a gradient based optimisation method that can be
employed when the objective function is in the form of sum of squares. Calculation
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 273

of the Hessian, which is computationally expensive, is not needed for this method.
Here Hessian is approximated using the Jacobean and the search direction is given
by (JT ()J()) 1 r  . Levenberg–Marquardt method is also used to optimise a
function defines as sum of squares but here the parameter updates are adaptively
changed between gradient descent update and Gauss–Newton update. The search
direction is given by (JT ()J() C  diag(JT ()J())) 1 r  . The advantage of this
method is that it can converge to the solution starting form a location far away from
the final solution.
Another very popular method in medical image is the stochastic optimisation
where an approximation of the gradient is used as the search direction together with
a diminishing step size. The gradient approximation can be calculated either using
analytical gradients using only a subsample of points, Kiefer–Wolfowitz method
or Simultaneous Perturbation. Klein et al. (2007) showed that out of the above
methods analytical gradients calculated using only a subset of points selected by
uniform sampling is the most efficient. Uniform sampling gives equal significance
to all the voxels in an image and since the true function in this case is not distributed
uniformly, using only a subset of voxels produces a biased estimate of the gradients.
To reduce the sampling bias they proposed to renew the set of chosen samples
at every iteration of the optimisation routine giving equal significance to all the
voxels. To improve the registration accuracy Bhagalia et al. (2009) introduced the
idea of using the importance sampling (IS) technique to select a subset of voxels.
Tennakoon et al. (2014) proposed the use of rank ordered statistics based robust
segmentation technique on the intensity difference image to identify the relevant
voxels for registration and used these in calculating the gradients.

10.4.2 Discrete Optimisation

Discrete optimisation schemes are used when the parameters of the optimisation
problem are in discrete forms. For instance, the image registration problem can be
seen as a discrete optimisation problem when posed as a metric labelling problem.
However generally this will result in a non-convex energy with many local minima
and the optimisation would involve a large set of parameters in which finding a
solution is computationally expensive. Kleinberg and Tardos (2002) have shown that
original registration problem when posed as metric labelling problem is equivalent
to a discrete Markov Random Fields (MRFs) minimisation problem, where the aim
is to assign each node (voxel or parameter) a label l 2 L (where L is a discrete
label set) which minimises an energy function E(l) of the following form.
X   X  
E.l/ D Dp lp C  Vp;q lp ; lq (10.25)
pP fp;qgN
„ ƒ‚ … „ ƒ‚ …
Data T erm Smoothness T erm
274 R. Tennakoon et al.

In this equation Dp (lp ) is the cost of assigning label lp to node p and Vp,q (lp , lq ) is the
cost of assigning labels lp , lq to neighbouring pixels p, q. Recently many techniques
emerged that can solve this problem efficiently and these methods can be segmented
into three main groups: Graph-cut based methods, message passing methods and
linear programming methods.
Graph-cut based methods use the max-flow min-cut principle (Ford and Fulk-
erson 2010). Here a graph is constructed where each node is connected to its
neighbours and two special nodes called the source and the sink. Once the graph is
constructed the minimum cut that segments this graph into two partitions is given by
the saturated edges at the maximum flow from source to sink. Boykov et al. (2001)
extended this idea to multiple labels. They introduced two algorithms, the ’-“-swap
and ’-expansion for energies where Vp,q () is semi metric and metric, respectively.
These methods use large moves compared to standard algorithms making them
computationally efficient and also there is a theoretical guarantee of convergence
within a known factor (vicinity) of the global minimum. So and Chung (2009) and
So et al. (2011) formulated the image registration problem as a discrete labelling
problem and assigned each voxel a discrete displacement label. They used an energy
function which comprised of SAD as the data term and a smoothness term that
penalises the difference in displacement in adjacent voxels. They used ’-expansion
algorithm to derive the optimal labelling.
Graph-cut methods can only be applied to a limited set of energy functions
(Kolmogorov and Zabin 2004). However, the MRFs problem discussed above
can be at least approximately solved using the max-product Belief Propagation
(BP) technique introduced by Pearl (1988). BP uses local messages passed along
the edges of a graphical model to derive the solution. The solution will only
be exact in case the graph is a tree, but if the graph contains loops only an
approximate solution can also be found. Generalised BP algorithm proposed by
Yedidia et al. (2000) achieves better convergence than the ordinary BP. Max-product
tree-reweighted (TRW) message passing introduced by Wainwright et al. (2005) is a
more generalised version that efficiently finds the solution to any type of graph and
this approach was further modified by Kolmogorov (2006). A detailed description of
the BP algorithm is provided in Szeliski et al. (2008) and Kschischang et al. (2001).
Shekhovtsov et al. (2008) and Lee et al. (2008) used TRW algorithm in 2D and
3D intensity based image registrations, respectively. They decomposed the graph
into single dimensional graphs in order to reduce the computations. Kwon et al.
(2011) used TRW to optimise an energy that contains a dense local descriptor as
data term and higher order smoothness prior.
Linear programming based methods try to improve the efficiency by solving a
linear programming relaxation of the original labelling problem. These methods
are generally faster than graph-cut based methods and the regularisation term does
not have to be metric as in ’-expansion. Glocker et al. (2008) used the FFD
model with B-Splines to reduce the dimensionality of the problem and used an
energy function that is projected onto the control points (which comprised of an
arbitrary similarity measure and a smoothness measure that penalises the difference
in adjacent parameter labels). Then the problem is modelled as an equivalent integer
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 275

program which was subjected to linear program relaxation. The optimal labelling
is then found by using the fast primal-dual algorithm (Komodakis et al. 2007).
Additionally they used a hierarchical warping scheme to account for the large
deformation and imposed a hard constraint to preserve the diffeomorphism. In
Glocker et al. (2011) the authors extended this work to integrate both iconic and
geometric registration into a unified formulation and proposed two algorithms. The
first one is a computationally efficient implementation while the second uses a tight
relaxation.

10.5 Summary

Image registration of medical images is a fundamental task in image processing and

it has been studied for over three decades. Due to the non-rigid motions associated
with soft tissue, different types of modalities involved in capturing those images
and large number of parameters needed to model the associated deformation image
registration problem, the associated objective function, transformation models and
optimisation procedures are highly nonlinear. This chapter outlines the nonlinear
methods used in the key segments of solving the medical image registration
problem.

Key Symbols

T Spatial transformation
x Voxel index
IF (x) Fixed image
IM (x) Moving image
F Fixed image domain
M Moving image domain
 Transformation parameter vector
u(x) Deformation at voxel x
v(x) Velocity at voxel x
‰() Objective function
M() Similarity measure
R() Regularisation term
J() Jacobean of the transformation
276 R. Tennakoon et al.

References

Andronache A, Cattin P, Székely G (2006) Local intensity mapping for hierarchical non-rigid
registration of multi-modal images using the cross-correlation coefficient. Biomed Image
Regist 4057:26–33 [Springer]
Arsigny V, Pennec X, Ayache N (2005) Polyrigid and polyaffine transformations: a novel geomet-
rical tool to deal with non-rigid deformations—application to the registration of histological
slices. Med Image Anal 9:507–523
Arzhaeva Y, Prokop M, Murphy K, Van Rikxoort EM, de Jong PA, Gietema HA, Viergever MA,
Van Ginneken B (2010) Automated estimation of progression of interstitial lung disease in CT
images. Med Phys 37:63–73
Audette MA, Ferrie FP, Peters TM (2000) An algorithmic overview of surface registration
techniques for medical imaging. Med Image Anal 4:201–217
Bhagalia R, Fessler JA, Boklye K (2009) Accelerated nonrigid intensity-based image registration
using importance sampling. IEEE Trans Med Imaging 28:1208–1216
Boykov Y, Veksler O, Zabih R (2001) Fast approximate energy minimization via graph cuts. IEEE
Trans Pattern Anal Mach Intell 23:1222–1239
Broit C (1981) Optimal registration of deformed images. University of Pennsylvania
Bro-Nielsen M, Gramkow C (1996) Fast fluid registration of medical images. In: Höhne K, Kikinis
R (eds) Visualization in biomedical computing. Springer, Berlin
Burschka D, Li M, Taylor R, Hager GD (2004) Scale-invariant registration of monocular
endoscopic images to CT-scans for sinus surgery. In: Medical image computing and computer-
assisted intervention—MICCAI 2004. Springer
Cahill ND, Schnabel JA, Noble JA, Hawkes DJ (2008) Revisiting overlap invariance in medical
image alignment. In: IEEE computer society conference on computer vision and pattern
recognition workshops, 2008 (CVPRW’08). IEEE, pp 1–8
Castillo E, Castillo R, Zhang Y, Guerrero T (2009) Compressible image registration for thoracic
computed tomography images. J Med Biol Eng 29:222–233
Castillo E, Castillo R, White B, Rojo J, Guerrero T (2012) Least median of squares filtering
of locally optimal point matches for compressible flow image registration. Phys Med Biol
57:4827–4833
Christensen GE (1999) Consistent linear-elastic transformations for image matching. In: Kuba
A, Šáamal M, Todd-Pokropek A (eds) Information processing in medical imaging. Springer,
Berlin
Christensen G, He J (2001) Consistent nonlinear elastic image registration. In: IEEE workshop on
mathematical methods in biomedical image analysis, 2001 (MMBIA 2001). IEEE, pp 37–43
Christensen GE, Rabbitt RD, Miller MI (1994) 3D brain mapping using a deformable neu-
roanatomy. Phys Med Biol 39:609
Christensen GE, Rabbitt RD, Miller MI (1996) Deformable templates using large deformation
kinematics. IEEE Trans Image Process 5:1435–1447
Chun SY, Fessler JA (2008) Regularized methods for topology-preserving smooth nonrigid image
registration using b-spline basis. In: 5th IEEE international symposium on biomedical imaging:
from nano to macro, 2008 (ISBI 2008). IEEE, pp 1099–1102
Collignon A, Maes F, Delaere D, Vandermeulen D, Suetens P, Marchal G (1995) Automated multi-
modality image registration based on information theory. Inf Process Med Imaging 3:263–274
Corpetti T, Mémin É, Pérez P (2002) Dense estimation of fluid flows. IEEE Trans Pattern Anal
Mach Intell 24:365–380
Ford D, Fulkerson DR (2010) Flows in networks. Princeton University Press, Princeton
Gee JC, Haynor DR, Reivich M, Bajcsy RK (1994) Finite element approach to warping of brain
images. In: Medical imaging 1994. International Society for Optics and Photonics, pp 327–337
Gerber N, Gavaghan KA, Bell BJ, Williamson TM, Weisstanner C, Caversaccio MD, Weber S
(2013) High-accuracy patient-to-image registration for the facilitation of image-guided robotic
microsurgery on the head. IEEE Trans Biomed Eng 60:960–968
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 277

Giger ML, Chan H-P, Boone J (2008) Anniversary paper: history and status of CAD and
quantitative image analysis: the role of medical physics and AAPM. Med Phys 35:5799–5820
Glocker B, Komodakis N, Tziritas G, NAVAB N, Paragios N (2008) Dense image registration
through MRFs and efficient linear programming. Med Image Anal 12:731–741
Glocker B, Sotiras A, Komodakis N, Paragios N (2011) Deformable medical image registration:
setting the state of the art with discrete methods. Annu Rev Biomed Eng 13:219–244
Gorbunova V, Sporring J, Lo P, Loeve M, Tiddens HA, Nielsen M, Dirksen A, De Bruijne M (2012)
Mass preserving image registration for lung CT. Med Image Anal 16:786–795
Guo W, Li Q, Boyce SJ, Mcadams HP, Shiraishi J, Doi K, Samei E (2012) A computerized scheme
for lung nodule detection in multiprojection chest radiography. Med Phys 39:2001–2012
Haber E, Modersitzki J (2006) Intensity gradient based registration and fusion of multi-modal
images. In: Medical image computing and computer-assisted intervention—MICCAI 2006.
Springer
Hero A, Ma B, Michel OJ, Gorman J (2002) Applications of entropic spanning graphs. IEEE Signal
Process Mag 19:85–95
Hill DL, Batchelor PG, Holden M, Hawkes DJ (2001) Medical image registration. Phys Med Biol
46:R1
Holden M (2008) A review of geometric transformations for nonrigid body registration. IEEE
Trans Med Imaging 27:111–128
Huang CH, Hsieh CH, Lee JD, Huang WC, Lee ST, Wu CT, Sun YN, Wu YT (2012) A CT-
ultrasound-coregistered augmented reality enhanced image-guided surgery system and its
preliminary study on brain-shift estimation. J Instrum 7, P08016
Illán IA, Górriz JM, Ramírez J, Segovia F, Jiménez-Hoyuela JM, Ortega Lozano SJ (2012)
Automatic assistance to Parkinsons disease diagnosis in DaTSCAN SPECT imaging. Med Phys
39:5971–5980
Jacobson TJ, Murphy MJ (2011) Optimized knot placement for B-splines in deformable image
registration. Med Phys 38:4579–4582
Jaya J, Thanushkodi K (2011) Implementation of computer aided diagnosis system based on
parallel approach of ant based medical image segmentation. J Comput Sci 7:291–297
Kim J (2004) Intensity based image registration using robust similarity measure and constrained
optimization: applications for radiation therapy. Citeseer
Klein S, Staring M, Pluim JPW (2007) Evaluation of optimization methods for nonrigid medical
image registration using mutual information and B-splines. IEEE Trans Image Process
16:2879–2890
Klein S, Van der Heide UA, Lips IM, Van Vulpen M, Staring M, Pluim JP (2008) Automatic
segmentation of the prostate in 3D MR images by atlas matching using localized mutual
information. Med Phys 35:1407–1417
Kleinberg J, Tardos E (2002) Approximation algorithms for classification problems with pairwise
relationships: metric labeling and Markov random fields. J ACM 49:616–639
Ko JP, Betke M (2001) Chest CT: automated nodule detection and assessment of change over
time-preliminary experience. Radiology 218(1):267–273
Kolmogorov V (2006) Convergent tree-reweighted message passing for energy minimization.
IEEE Trans Pattern Anal Mach Intell 28:1568–1583
Kolmogorov V, Zabin R (2004) What energy functions can be minimized via graph cuts? IEEE
Trans Pattern Anal Mach Intell 26:147–159
Komodakis N, Tziritas G, Paragios N (2007) Fast, approximately optimal solutions for single
and dynamic MRFs. In: IEEE conference on computer vision and pattern recognition, 2007
(CVPR’07), 17–22 June 2007, pp 1–8
Kschischang FR, Frey BJ, Loeliger H-A (2001) Factor graphs and the sum-product algorithm.
IEEE Trans Inf Theory 47:498–519
Kwon D, Lee KJ, Yun ID, Lee SU (2011) Nonrigid image registration using higher-order MRF
model with dense local descriptor. In: Computer vision and pattern recognition workshop, 2011
Kybic J, Unser M (2003) Fast parametric elastic image registration. IEEE Trans Image Process
12:1427–1442
278 R. Tennakoon et al.

Lee KJ, Kwon D, Yun ID, Lee SU (2008) Deformable 3D volume registration using efficient MRFs
model with decomposed nodes. In: BMVC 2008, pp 1–10
Lester H, Arridge SR (1999) A survey of hierarchical non-linear medical image registration.
Pattern Recogn 32:129–149
Liang Y, Ryan NS, Schott JM, Fox NC (2013) Imaging the onset and progression of Alzheimer’s
disease: implications for prevention trials. J Alzheimers Dis 33:S305–S312
Likar B, Pernuš F (2001) A hierarchical approach to elastic registration based on mutual
information. Image Vis Comput 19:33–44
Linte CA, White J, Eagleson R, Guiraudon GM, Peters TM (2010) Virtual and augmented medical
imaging environments: enabling technology for minimally invasive cardiac interventional
guidance. IEEE Rev Biomed Eng 3:25–47
Loeckx D, Slagmolen P, Maes F, Vandermeulen D, Suetens P (2010) Nonrigid image registration
using conditional mutual information. IEEE Trans Med Imaging 29:19–29
Maintz JBA, Viergever MA (1998) A survey of medical image registration. Med Image Anal 2:
1–36
Mirota DJ, Hanzi W, Taylor RH, Ishii M, Gallia GL, Hager GD (2012) A system for video-based
navigation for endoscopic endonasal skull base surgery. IEEE Trans Med Imaging 31:963–976
Modersitzki J (2004) Numerical methods for image registration (numerical mathematics and
scientific computation). Oxford University Press, Oxford
Modersitzki J (2008) FLIRT with rigidity—image registration with a local non-rigidity penalty. Int
J Comput Vis 76:153–163
Mostayed A, Garlapati RR, Joldes GR, Wittek A, Roy A, Kikinis R, Warfield SK, Miller K (2013)
Biomechanical model as a registration tool for image-guided neurosurgery: evaluation against
bspline registration. Ann Biomed Eng 41:2409–2425
Musse O, Heitz F, Armspach J-P (2001) Topology preserving deformable image matching using
constrained hierarchical parametric models. IEEE Trans Image Process 10:1081–1093
Myronenko A, Xubo S (2010) Intensity-based image registration by minimizing residual complex-
ity. IEEE Trans Med Imaging 29:1882–1891
Noblet V, Heinrich C, Heitz F, Armspach J-P (2005) 3-D deformable image registration: a
topology preservation scheme based on hierarchical deformation models and interval analysis
optimization. IEEE Trans Image Process 14:553–566
Oliveira FP, Tavares JM (2014) Medical image registration: a review. Comput Methods Biomech
Biomed Eng 17:73–93
Pearl J (1988) Probabilistic reasoning in intelligent systems: networks of plausible inference.
Morgan Kaufmann, San Francisco
Pennec X, Stefanescu R, Arsigny V, Fillard P, Ayache N (2005) Riemannian elasticity: a
statistical regularization framework for non-linear registration. In: Medical image computing
and computer-assisted intervention—MICCAI 2005. Springer
Pitiot A, Malandain G, Bardinet E, Thompson PM (2003) Piecewise affine registration of biological
images. In: Biomedical image registration. Springer
Pluim JP, Maintz JA, Viergever MA (2000) Image registration by maximization of combined
mutual information and gradient information. In: Medical image computing and computer-
assisted intervention—MICCAI 2000. Springer, pp 452–461
Pluim JP, Maintz JA, Viergever MA (2003) Mutual-information-based registration of medical
images: a survey. IEEE Trans Med Imaging 22:986–1004
Rabbitt RD, Weiss JA, Christensen GE, Miller MI (1995) Mapping of hyperelastic deformable
templates using the finite element method. In: SPIE’s 1995 international symposium on optical
science, engineering, and instrumentation. International Society for Optics and Photonics, pp
252–265
Rivaz H, Collins DL (2012) Self-similarity weighted mutual information: a new nonrigid image
registration metric. Med Image Comput Comput Assist Interv 15:91–98
Robertson S, Weiss E, Hugo GD (2013) Deformable mesh registration for the validation of
automatic target localization algorithms. Med Phys 40(7):071721
10 Nonlinear Approaches in Three Dimensional Medical Image Registration 279

Rohde GK, Aldroubi A, Dawant BM (2003) The adaptive bases algorithm for intensity-based
nonrigid image registration. IEEE Trans Med Imaging 22:1470–1479
Rueckert D, Aljabar P (2010) Nonrigid registration of medical images: theory, methods, and
applications [applications corner]. IEEE Signal Process Mag 27:113–119
Rueckert D, Sonoda LI, Hayes C, Hill DLG, Leach MO, Hawkes DJ (1999) Nonrigid registration
using free-form deformations: application to breast MR images. IEEE Trans Med Imaging
18:712–721
Rueckert D, Clarkson M, Hill D, Hawkes DJ (2000) Non-rigid registration using higher-order
mutual information. In: Medical imaging 2000. International Society for Optics and Photonics,
pp 438–447
Ruhaak J, Konig L, Hallmann M, Papenberg N, Heldmann S, Schumacher H, Fischer B (2013)
A fully parallel algorithm for multimodal image registration using normalized gradient fields.
In: 2013 IEEE 10th international symposium on biomedical imaging (ISBI), 7–11 April 2013,
pp 572–575
Rui L, Li Z, Ying S, Shun M, Chefd’hotel C (2013) A review of recent advances in registration
techniques applied to minimally invasive therapy. IEEE Trans Multimed 15:983–1000
Rynne BP, Youngson MA (2008) Linear functional analysis. Springer, London
Sadeghi Naini A, Patel RV, Samani A (2010) CT-enhanced ultrasound image of a totally deflated
lung for image-guided minimally invasive tumor ablative procedures. IEEE Trans Biomed Eng
57:2627–2630
Shekhovtsov A, Kovtun I, Hlaváč V (2008) Efficient MRF deformation model for non-rigid image
matching. Comput Vis Image Underst 112:91–99
Shiraishi J, Abe H, Li F, Engelmann R, Macmahon H, Doi K (2006) Computer-aided diagnosis
for the detection and classification of lung cancers on chest radiographs. ROC analysis of
radiologists’ performance. Acad Radiol 13:995–1003
Sluimer I, Schilham A, Prokop M, Van Ginneken B (2006) Computer analysis of computed
tomography scans of the lung: a survey. IEEE Trans Med Imaging 25:385–405
So R, Chung A (2009) Multi-level non-rigid image registration using graph-cuts. In: IEEE
international conference on acoustics, speech and signal processing, 2009 (ICASSP 2009),
19–24 April 2009, pp 397–400
So RWK, Tang TWH, Chung ACS (2011) Non-rigid image registration of brain magnetic
resonance images using graph-cuts. Pattern Recogn 44:2450–2467
Sotiras A, Davatzikos C, Paragios N (2013) Deformable medical image registration: a survey. IEEE
Trans Med Imaging 32:1153–1190
Staring M, van der Heide UA, Klein S, Viergever MA, Pluim JP (2009) Registration of cervical
MRI using multifeature mutual information. IEEE Trans Med Imaging 28:1412–1421
Studholme C, Hill DL, Hawkes DJ (1999) An overlap invariant entropy measure of 3D medical
image alignment. Pattern Recogn 32:71–86
Studholme C, Drapaca C, Iordanova B, Cardenas V (2006) Deformation-based mapping of volume
change from serial brain MRI in the presence of local tissue contrast change. IEEE Trans Med
Imaging 25:626–639
Szeliski R, Zabih R, Scharstein D, Veksler O, Kolmogorov V, Agarwala A, Tappen M, Rother C
(2008) A comparative study of energy minimization methods for Markov random fields with
smoothness-based priors. IEEE Trans Pattern Anal Mach Intell 30:1068–1080
Tennakoon RB, Bab-Hadiashar A, Cao Z, De Bruijne M (2014) Nonrigid registration of volumetric
images using ranked order statistics. IEEE Trans Med Imaging 33:422–432
Timp S, Varela C, Karssemeijer N (2007) Temporal change analysis for characterization of mass
lesions in mammography. IEEE Trans Med Imaging 26:945–953
Uneri A, Nithiananthan S, Schafer S, Otake Y, Stayman JW, Kleinszig G, Sussman MS, Prince
JL, Siewerdsen JH (2013) Deformable registration of the inflated and deflated lung in cone-
beam CT-guided thoracic surgery: initial investigation of a combined model- and image-driven
approach. Med Phys 40
Van Ginneken B, Ter Haar Romeny BM, Viergever MA (2001) Computer-aided diagnosis in chest
radiography: a survey. IEEE Trans Med Imaging 20:1228–1241
280 R. Tennakoon et al.

van Herk M (2007) Different styles of image-guided radiotherapy. Semin Radiat Oncol 17:
258–267
Vercauteren T, Pennec X, Perchant A, Ayache N (2007) Non-parametric diffeomorphic image
registration with the demons algorithm. In: Medical image computing and computer-assisted
intervention—MICCAI 2007. Springer
Viola P, Wells WM III (1997) Alignment by maximization of mutual information. Int J Comput
Vis 24:137–154
Wainwright MJ, Jaakkola TS, Willsky AS (2005) MAP estimation via agreement on trees:
message-passing and linear programming. IEEE Trans Inf Theory 51:3697–3717
Wang J, Jiang T (2007) Nonrigid registration of brain MRI using NURBS. Pattern Recogn Lett
28:214–223
Wei J, Chan HP, Sahiner B, Zhou C, Hadjiiski LM, Roubidoux MA, Helvie MA (2009) Computer-
aided detection of breast masses on mammograms: dual system approach with two-view
analysis. Med Phys 36:4451–4460
Worm ES, Hansen AT, Petersen JB, Muren LP, Præstegaard LH, Høyer M (2010) Inter- and
intrafractional localisation errors in cone-beam CT guided stereotactic radiation therapy of
tumours in the liver and lung. Acta Oncol 49:1177–1183
Wu Y-T, Kanade T, Li C-C, Cohn J (2000) Image registration using wavelet-based motion model.
Int J Comput Vis 38:129–152
Xishi H, Jing R, Guiraudon G, Boughner D, Peters TM (2009) Rapid dynamic image registration
of the beating heart for diagnosis and surgical navigation. IEEE Trans Med Imaging 28:
1802–1814
Yedidia JS, Freeman WT, Weiss Y (2000) Generalized belief propagation. In: NIPS 2000,
pp 689–695
Yin Y, Hoffman EA, Lin C-L (2009) Mass preserving nonrigid registration of CT lung images
using cubic B-spline. Med Phys 36:4213
Yuanjie Z, Kambhamettu C, Bauer T, Steiner K (2009) Accurate estimation of pulmonary
nodule’s growth rate in CT images with nonrigid registration and precise nodule detection
and segmentation. In: IEEE computer society conference on computer vision and pattern
recognition workshops 2009 (CVPR workshops 2009), 20–25 June 2009, pp 101–108
Zitova B, Flusser J (2003) Image registration methods: a survey. Image Vis Comput 21:977–1000
Chapter 11
Pose Estimation of a Dioptric Imaging Sensor
with a Circle-Projecting Collimated Laser
Moving Inside a Pipeline

Abbasali Dehghan Tezerjani, Mehran Mehrandezh, and Raman Paranjape

11.1 Introduction

Pipelines are the primary source for the transport of material such as water, oil,
and gas in the world. Therefore, quality control at the manufacturing stage and/or
pro-active timely inspection of them, when in service, are critical. Failure in pipes
can have severe adversarial environmental and economic shocks. It is estimated that
over 50 % of the pipeline infrastructure in North America are older than 60 years,
Cohen (2012). Development of non-destructive testing methods for quality control
of pipes, at the manufacturing stage, and inspection, when pipes in use, is paramount
to the industry. This research is a timely response to a worldwide need for tools that
can do consistent reporting on pipe’s condition with minimum man-hour required.
Robotic inspection systems would be a perfect fit to this task.
An image-based optics laser sensor, targeted towards the inspection of pipes, was
developed in our group, Tezerjani et al. (2013). The information obtained via images
can be used to (1) detect defects, (2) classify them, (3) size them (i.e., making metric
measurements on defects via images), and (4) map them (i.e., localize them inside
the pipe with respect to a reference point). The focus of this research paper is to
provide a guideline for precise pose estimation of an optics laser imaging sensor
required for sizing the defects and providing metric measurements from images.
In this context, precise image-based metric measurements would require: (1) an
accurate estimation of the pose of the imaging sensor with respect to the detected
defects, and (2) an accurate projection model to convert highly distorted 2D images
to a scaled Axonometric projection space.

A.D. Tezerjani () • M. Mehrandezh • R. Paranjape

Faculty of Engineering and Applied Science, University of Regina,
3737 Wascana Pky, Regina, SK, Canada S4S 0A2
e-mail: [email protected]; [email protected]; [email protected]

© Springer International Publishing Switzerland 2015 281

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__11
282 A.D. Tezerjani et al.

Vision-based pose estimation of imaging sensors has been studied for the past
three decades. The seminal work of Dementhon and Davis (1995) on real-time
pose estimation of a perspective camera pioneers the research. However, there is
a huge body of literature on improving the image-based pose-estimation algorithms
with the objective to make them less sensitive to the image noise, and lens
imperfection/distortion, e.g. Simo-Serra et al. (2012) and Josephson and Byrod
(2009).
In general, the pose estimation of imaging sensors was applied primarily in
three areas: (1) visual servoing of unmanned systems, e.g. Tahri et al. (2010),
(2) 3D digital reconstruction of objects from sequential images and/or video
streams, e.g. Spica et al. (2014) and Chang (2002) and (3) Augmented reality, e.g.
Hagbi et al. (2011). In this research, the pose estimation is targeted towards the 3D
reconstruction of an object (i.e., pipe’s surface) from sequential images obtained
via an optics laser (i.e., a laser in conjunction with a camera). The technical term
widely used in the literature is: structure from motion (SFM), Spica et al. (2014) and
Chang (2002). A super-resolution 3D reconstruction of an object can be achieved
using the optics laser. The accuracy can be comparable to that in military-grade
inertial measurement units (IMUs).
Image-based pose estimation in unstructured environments through an SFM
paradigm remains a challenge Daniilidis and Spetsakis (1997). Feature detection,
and feature matching/adjustment are paramount in the accuracy of the pose esti-
mations. Usually, one would require significant image overlaps between sequential
images to identify false detection or outliers Iketani et al. (2007). Dynamic features
can also affect the results, therefore, intensive statistical analysis would be needed
to identify and bundle static features and to separate them from dynamic features,
Alvarez et al. (2007). An alternative to the SFM in unstructured environment is to
use structured light patterns, such as a grid-dot or circular laser patterns projected on
objects Andrews et al. (2013). M. Vincez et al. use a laser projector with a grid-dot
pattern to estimate the relative pose between a spray-painting nozzle attached to a
robot’s end-effector and the surface to be painted with applications in automotive
industry Vincze et al. (2002). Similar work has been done for robotic welding
Pipeline and Gas Journal (1999).
S. Winkelbach et al. use optics laser for 3D reconstruction of objects Winkelbach
et al. (2006). However, they assume that the imaging tool is calibrated offline
and that it remains stationary. Also, these imaging sensors can be used to scan
small objects only. For scanning large objects, such as that in a pipeline, the
imaging tool needs to move long distances. Any disturbance in the imaging sensor’s
motion would affect the quality of the digital reconstruction of the objects. Further
continuous pose estimation techniques would be needed for better accuracy.
More specifically, image-based inspection of pipelines is gaining significant
attention by the research institutes and industry, e.g. Duran et al. (2007, 2003) and
Matsui et al. (2010). Duran et al. used a laser projecting device with a circular
pattern and a perspective camera to detect defects Duran et al. (2007). Their problem
11 4-DOF Pose Estimation of a Pipe Crawling Robot 283

formulation is based on the assumption that the laser/camera would remain at the
center point of the pipe all the time. Besides, their work revolved around defect
detection and classification only.
A large number of companies making pipe inspection tools use optics laser to
detect and classify defects in pipes as well, e.g. Maverick Inspection Ltd. (2009) and
Raush Electronics USA (2014). However, the precise sizing of the detected defects
would require additional time-consuming a priori system setup, such as alignment of
the scanning device in the pipe. A similar concept was used by Unnikrishnan et al.
(2009) to estimate the pose of an optics laser scanning device for the inspection
of underwater infrastructure. The pose estimations were used only to navigate their
underwater vehicle system. Therefore, a low-resolution image would suffice. In their
work, a look-up table, containing the mapping between the laser ring image and the
pose of the robot, was generated through a training step. Usually, the training is
carried out on a set of representative tests and then the results are extrapolated to
un-rehearsed scenarios. This can be highly susceptible to the image noise, and the
quality and diversity of the date used for training. A much higher accuracy would
be needed in estimating the pose of the imaging sensor for 3D reconstruction of a
surface in high resolution.
To the best of our knowledge, this is the first time a systematic algorithm is
developed for pose estimation of a panoramic optics laser imaging sensor inside
a pipe. This would minimize the man-hour needed to set up the device, simply
because no alignment would be required. Also, the algorithm accounts for unwanted
disturbances in the sensor’s motion. Therefore, the pose estimation can be done
more accurately and, consequently, this will improve the accuracy of the image-
based metric measurements of the defects.
The layout of the paper is as follows: Sect. 11.2 formulates the projection geom-
etry of the proposed imaging platform, and introduces the required components for
the pose estimation. Section 11.3 presents a novel approach for mapping the pose
estimation problem into the numerical optimization problem. Performance of the
proposed pose estimation technique is tested using the synthesized images and the
experimental results are presented in Sect. 11.4. In this section, the performance
of the proposed algorithm is discussed, and effects of different parameters on the
performance of the pose estimation algorithm are studied.

11.2 Problem Formulation

Figure 11.1a–c show the robotic inspection system, the imaging sensor mounted on
the robot, and a snapshot of the robot inside a PVC pipe, respectively. The main
components of the imaging unit, namely (1) the camera with a fish-eye lens, (2)
a laser projector, (3) a 45ı conic mirror, and (4) connectors and holding brackets
are shown in Fig. 11.2. The laser dot reflected on the mirror will be projected as
a perfect circle on the pipe’s surface as long as the laser/camera is centered inside
the pipe. The image of the circle projected on the pipe’s wall not be necessarily
284 A.D. Tezerjani et al.

Fig. 11.1 Robotic pipe inspection system. (a) Robotic inspection system. (b) Imaging sensor
mounted on the robot. (c) Snapshot of the robot inside a PVC pipe

Fig. 11.2 Main components of the imaging unit

a conic because of the lens distortion. Its overall shape would depend on the
radial/tangential distortions in the fish-eye lens and also the aspect ratio of the
camera’s CCD array and/or the aspect ratio of the display unit (i.e., the computer
screen). The image of the projected laser would be also skewed depending on the
lateral and vertical sway motions in the laser/camera and also its pan and tilt angles.
However, it is noteworthy that the axial movement of the imaging sensor inside
the pipe and its rolling motion will not affect the image. The objective is to estimate
four degrees of motion, namely two translational sway motions along horizontal and
vertical axes and two pan and tilt angles from a set of image moments of the laser
image seen by the camera. The selection of the image moments will be discussed
below.
First, let us define the notations used for representing the pose of the sensor
and also the image moments. For this purpose, two rectangular coordinate systems
are introduced. The “0” frame denotes a fixed frame whose z-axis is along the
longitudinal axis of the pipe (see Fig. 11.3). The “C” frame, also seen in Fig. 11.3,
denotes a frame whose origin is attached to the center of projection of the camera.
The pose of the imaging sensor is represented by a quadruplet; < xc ; yc ; p ; t >,
where xc and yc denote the lateral and vertical sway in camera’s motion with respect
11 4-DOF Pose Estimation of a Pipe Crawling Robot 285

Fig. 11.3 Proposed imaging platform and its configuration inside the pipe

to the fixed frame, respectively. Also, p and t denote the pan and tilt rotational
angles around the x and y axes of the fixed frame, respectively (see Fig. 11.3).
The image moments are defined as the coefficients of a high-order polynomial
optimally fit to the laser image. It is noteworthy that the laser image seen by the
camera would not be a conic in general, therefore, it cannot be represented by a
quadratic polynomial. Extra care must be taken to make sure that the functional
relationship between the pose quadruplet and the image moments is of full rank.
Furthermore, due to the radially symmetric shape of the pipe, finite number of
disperse solutions to the pose estimation problem may exist. A binary filter would
be needed to select the right solution.
Model reduction by the separation of variables: The relationship between the
pose quadruplet and the image moments will be highly nonlinear and coupled,
therefore, nonlinear optimization techniques must be adopted. However, some
variables can be decoupled through a separation of variable technique. This model
reduction can speed up the convergence of the optimization process while assuring
that a local optimum solution will be reached within the initial simplex defined by
the user. This is in particular important to avoid jumping between finite possible
solutions that may exist.
In the proposed model reduction process, first the relationship between the
parametric equation of the 3D curve, which is the result of the intersection of the
laser cone with the pipe’s wall and the pose quadruplet, is derived. In the second
step, the relationship between the parametric equations of the same 3D curve with
the image moments is derived. Solutions obtained via optimization techniques from
286 A.D. Tezerjani et al.

Fig. 11.4 Projection

geometry of the imaging
platform

one stage are traversed to the second and a back-propagation correcting mechanism
is used, based on a least square error of a fitting function, iteratively till a solution is
converged.
Stage I—Calculating the intersection curve between the laser cone and the pipe’s
wall using the pose quadruplet:
As seen in Fig. 11.4, the omnidirectional laser projects a ring perpendicular to
its central axis on the pipe’s wall. The image of the ring collinear with the laser
projector at the distance d is captured by the fish-eye camera. This image is used
to estimate the relative pose of the camera, < xc ; yc ; p ; t > inside the pipe.
Figure 11.4 demonstrates the projection geometry of the imaging platform. Here,
xc , and xl represent the coordinates of the camera pinhole and the laser projector
head, respectively. Given the pose quadruplet < xc ; yc ; p ; t >, the coordinate of
the laser projector head is calculated as follows:
0 1 0 1
0 xc  d sin t
xl D xc C Rx .p / Ry .t / @ 0 A D @yc C d cos t sin p A (11.1)
d d cos p cos t

where Rx and Ry denote the rotation matrices about the X and Y axes. The equation
of the ring plane is calculated as following:

x !
l  xc  Œx; y; z  xl D 0 )
T

x sin t  y cos t sin p  z cos p cos t C  D 0; (11.2)

11 4-DOF Pose Estimation of a Pipe Crawling Robot 287

where  D d  xc sin t C yc cos t sin p . We define the 3D curve p. /, as the

intersection of the ring plane and the pipe, where  2 Œ0; 2 refers to the point
angle with respect to the X axis in the X Y plane (see Fig. 11.4). Given the equation
of the ring plane, parametric 3D coordinates of the intersection curve p. / are
calculated as:
0 1
rp cos 
B rp sin  C
p. / D B C
@  C rp cos  sin t  rp cos t sin p sin  A ;  2 Œ0 2 (11.3)
cos p cos t

Given the 3D coordinates of the intersection curve, p. / , the incident ray angle
of the fish-eye lens corresponding to each point on this curve, is calculated as
follows:
! !
./ D†.p./  xc ; 
xl  xc / (11.4)

It should be added that the angle  on the fixed frame maps to the angle on the
camera’s frame by an offset angle ı. Equation (11.5) shows the formulation of the
offset angle based on the pose quadruplet. Further details about the offset angle and
the way it is calculated can be found in Appendix 1.

D  C ı;
!
d sin p C yc C cos t C rp sin p sin t  xc sin p sin t
ı D tan1    (11.5)
cos p rp  xc C d sin t

Stage II: Calculating the incident ray angle of the intersection points with respect
to the fish-eye lens, using the image moments:

Employing the widely used equi-distant projection model for the fish-eye lens, one
can relate the pixel coordinates of the image of the intersection curve, p. / to the
lens focal length, f , and angles  and as in Eq. (11.6). Further details on this can
be found in Fleck (1995).
   
u. / cos
D f  ./  (11.6)
v. / sin

This projection model was numerically simulated for a particular case, using a
real camera with the equidistant fish-eye lens with a focal length of f D 0:14 cm,
and the following configurations: xc D 6, yc D 4, p D =6, t D =8,
d D 5 cm, and rp D 20 cm. Figure 11.5 shows the z-coordinates of p. /, incident
ray angle, ./, the distance from p. / to the fish-eye lens, and the image of
the laser ring in the image plane. The negative values of the z-coordinates of the
288 A.D. Tezerjani et al.

Z coordinate of the ring point p(θ) w.r.t. the fish-eye lens incident ray φ (θ) w.r.t. the fish-eye lens
25 85

20
80
15
zp(θ) in cm

10

¢(θ)
75
5
0
70
–5

–10 65
0 100 200 300 0 60 120 180 240 300 360
θ θ

distance from the ring point p(θ) w.r.t. the fish-eye lens image of the laser ring inside the image plan of an equidistant lens
35 0.2

30
0.1

25
Ip(θ)-ccI

0
v

20

–0.1
15

10 –0.2
0 60 120 180 240 300 360 –0.2 –0.1 0 0.1 0.2
θ u

Fig. 11.5 Projection of the laser ring on the image plane of an equidistant fish-eye lens for a
sample camera pose and configuration f D 0:14 cm, tx D 6, ty D 4, ˛x D =6, ˛y D pi=8,
d D 5 cm, and rp D 20 cm

intersection curve, p. /, correspond to the points behind the camera pinhole. As this
figure shows, the laser’s image is not always conic or even congruent.
In the next section, we formulate a nonlinear optimization algorithm for estimat-
ing the pose quadruplet using the ring’s image.

11.3 Solution Methodology

To estimate the relative pose of the camera inside the pipe, we employ a curve fitting
technique, and convert the pose estimation problem to an optimization problem
using the following steps.
Step 1—Defining the reference signal:
Based on the formulated projection geometry of the proposed imaging platform, we
define the reference signal ref ./ as follows:

ref ./ Dk p./  xc k2 D d 2 = cos2 ./;  2 Œ0 2 (11.7)

11 4-DOF Pose Estimation of a Pipe Crawling Robot 289

Step 2—Deriving the analytic form of the reference signal

The expanded analytical form of the reference signal, refa . /, using Eq. (11.3) is:

refa . / D .rp cos   xc /2 C .rp sin   yc /2

 
 C rp cos  sin t  rp cos t sin p sin  2
C (11.8)
cos p cos t

With some mathematical manipulation it can be shown that the analytical

reference signal can be also written as:

refa . / Da1 C a2 sin  C a3 cos  C a4 sin  cos  C a5 sin2  (11.9)

One may refer to Appendix 2 for the details of the derivation of the coefficients
a1 ; : : : ; a5 .
Step 3—Calculating the reference observation signal from the image moments
Having the calibrated fish-eye camera, the direction of incident ray ./ cor-
responding to the image coordinates .u; v/T is calculated using the intrinsic
parameters of the camera, regardless of the calibration technique used (Gennery
2006; Kannala and Brandt 2004; Scaramuzza et al. 2006). Here,  denotes the
orientation of the incident ray in the image plane. Kannala and Brandt (2004)
formulated the mapping from the image coordinate to the direction of incident ray
angle for the calibrated fish-eye lenses. Assuming the fish-eye camera calibrated,
this mapping will be as follows:
 1  
cos 1 u. /
./ D f (11.10)
sin v. /

From Eq. (11.10), another reference signal, this time obtained from the image,
denoted by refo . / is calculated as follows:

refo . / D d 2 = cos2 .  ı/ (11.11)

where the offset angle ı is calculated using Eq. (11.5). The signal refo is called the
reference observation signal.
Step 4—Fitting a trigonometric function to the observation reference signal
As Eq. (11.9) shows, the reference signal is a trigonometric function of . It means
that the reference observation signal can also be represented in this form. In order
to compare the analytical and observation forms of the reference signal, we fit the
290 A.D. Tezerjani et al.

reference observation signal to the function given in Eq. (11.9) using the least square
fitting algorithm. The fitted signal can be then represented as:

refo . / Do1 C o2 sin  C o3 cos  C o4 sin  cos  C o5 sin2  (11.12)

Step 5—Calculating the objective function

Given Eqs. (11.9) and (11.12), an objective function is defined as follows:

Z 2
J.xc ; yc ; p ; t ; / D Œrefo . /  refa . / 2 d ; (11.13)
0

where  represents all constant parameters in the equations (i.e., d and rp ).

By substituting the analytical and observation reference signals from Eqs. (11.9)
and (11.12) into Eq. (11.13) and doing some mathematical manipulation, the
objective function can be simplified to:
 
 8 e1 2 C 8 e1 e5 C 4 e2 2 C 4 e3 2 C e4 2 C 3 e5 2
J.xc ; yc ; p ; t ; / D ;
4
(11.14)

where ei D ai  oi . This objective function is used in the optimization process to

estimate the optimal value of the pose quadruplet. In a short
˚ form, the optimization 
problem can be mathematically written as: min J.xc ; yc ; p ; t ; / .
QD<xc ;yc ;p ;t ;>

Step 6—Minimizing the performance index, J using Nelder–Mead method

The optimization problem is defined as finding the optimal values of the pose
quadruplet, Q D< xc ; yc ; p ; t >, which minimizes the value of the objective
function J.xc ; yc ; p ; t ; /. The optimization algorithm starts with an initial seed
for the pose quadruplet. The optimal value of the pose is calculated by minimizing
the value of the objective function through iterations using the Nelder–Mead
simplex method (Lagarias et al. 1998).
Algorithm 1 represents the optimization process, in a nutshell:

Algorithm 1 Estimating the camera pose quadruplet

1: Select an initial seed for the current camera pose Q D< xc ; yc ; p ; t >.
2: Calculate the offset angle ı using equation (11.5).
3: Calculate the reference observation signal refo . / using the equations (11.10) and (11.12) and
the offset angle ı calculated in step 2.
4: Fit the trigonometric function to the reference observation signal refo . /.
5: Calculate the value of the objective function J using
˚ the equation (11.14). 
6: Find the optimal pose quadruplet min J.xc ; yc ; p ; t ; / using the NM’s simplex
QD<xc ;yc ;p ;t >
method.
11 4-DOF Pose Estimation of a Pipe Crawling Robot 291

Analysis of the proposed method shows that the backward mapping from the
laser ring image to the camera pose is not a one-to-one mapping. In fact, each
specific image of the laser ring is obtainable through two different camera poses.
For the numerical optimization technique used in this work, the initial seed value
plays a major role in convergence to the right solution. Selecting an appropriate
value for the initial seed not only improves the success rate of the optimization
algorithm but also affects the computation time of the algorithm. In the next section,
we study the effect of the initial seed through some experiments and present an
initial seed selection method to improve the success rate of the algorithm.

11.4 Experiment Results and Discussion

In order to evaluate the performance of the proposed algorithm for estimating the
camera pose inside the pipe, a simple GUI which generates the synthetic images
based on the camera pose, the pipe radius, the camera–laser distance, and the
projection model of the fish-eye lens was developed. Figure 11.6 shows a snapshot
of the developed GUI. Each image generated by this tool is passed to the pose
estimation algorithm, and the relative pose of the camera inside the pipe is estimated.
The numerically estimated camera pose is compared to its real value and the
difference is reported.

Fig. 11.6 Snapshot of developed GUI for generating sample data

292 A.D. Tezerjani et al.

a Reference signals after 15 iteration, J=1.22e+5

b Reference signals after 50 iteration, J=3.009e+4
1500 1500
ref0 ref0
Q= < –1.994 2.9160 0.278 0.0477 > Q= < –0.7621 2.9676 0.3989 –0.1042 >
1400 ref
a
1400 ref
a

1300 1300
1200 1200
1100 1100
ref(θ)

ref(θ)
1000 1000
900 900
800 800
700 700
600 600
500 500
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
θ θ
c Reference signals after 100 iteration, J=8.4e+3 d Reference signals after 140 iteration, J=0.1934
1500 1500
Q= < –2.0188 4.6497 0.3040 –0.2798 > ref0 ref0
Q= < –1.2995 5.6057 0.250 –0.2158 >
1400 ref
a
1400 ref
a

1300 1300
1200 1200
1100 1100
ref(θ)

ref(θ)

1000 1000
900 900
800 800
700 700
600 600
500 500
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
θ θ

Fig. 11.7 Observation and analytical reference signals during the optimization process, where the
actual pose is Qact ual D< xc D 1:2988 yc D 5:6033 p D 0:2496 t D 0:2153 >: (a) after
15 iterations, (b) after 50 iterations, (c) after 100 iterations, and (d) after 140 iterations

The nonlinear functions used in the proposed pose estimation technique were
cyclic with respect to the pan/tilt angles with a period of 2 radians. However,
they are not cyclic in terms of the lateral/vertical sway motion in the camera. In
fact, there are only two possible identical solutions to the sway motions that would
yield exactly the same image. In other words, the solutions to the sway motion
estimation can have dual jumps. These jumps, and also the cyclic solutions of the
pan/tilt estimates, were detected numerically using a binary filter. No false return
was ever perceived in the large number of numerical simulations carried out in this
research.
Figure 11.7 shows the time-elapsed results of the optimization process. Insets
show the variation of the reference signal with  at different iterations. After about
140 iterations the optimization process was terminated meeting the stop criterion.
Figure 11.8a shows the error of each component of the estimated camera pose
algorithm with respect to the number of iterations. As this figure shows, the rate of
change of error in all components of the estimated camera pose declines below 1 %
after 500 iterations.
11 4-DOF Pose Estimation of a Pipe Crawling Robot 293

a 40
b 100
ex
c
35
ey 90
c
30
eα
80

% success rate
25 eβ
Error %

20 70

15 60
10
50
5
40
0 0 20 40 60 80 100
0 100 200 300 400 500 600
offset value: (x0=xa ± offset* R/100)
Number of iterations

Fig. 11.8 Performance of the pose estimation algorithm vs. number of iterations and initial seed
value. (a) Pose estimation error vs. the number of iterations. (b) Success rate of the pose estimation
algorithm vs. the offset value of the initial seed for the camera pose

Figure 11.8b shows the success rate of the pose estimation algorithm with respect
to the initial seed value. Here, R denotes the range of the corresponding camera pose
component. As this figure shows, choosing the initial seed with less than 57 % offset
still gives more than 90 % success rate.
One immediate solution to the initial seed selection is to use the last-known
camera pose as the initial seed. Assuming the slow movement of the robot between
two successive frames captured by the camera, the last-known configuration of the
robot would remain reasonably close to the current actual camera pose. In this
experiment, all the components of the camera pose are changing simultaneously in
a spiral model. To examine this, the actual camera pose in 60 consecutive frames is
synthetically generated based on a spiral model. Then, the actual value of the camera
pose in each frame is used for generating the image moments. These image moments
are used as the input of the pose estimation algorithm. Figure 11.9 compares the
actual and estimated trajectories of the camera pose using the proposed algorithm,
where the last estimated pose is used as the initial seed for pose estimation. This
figure shows the actual trajectory that camera has traveled and its corresponding
estimated trajectory. Here, the estimated pose does not converge to the actual pose
in a few cases. Further elaboration of this result reveals that one of the camera pose
components on these spots is close to zero, and consequently, its rate of change is too
high. It means that in these spots, the last-known camera pose is not a good initial
seed for estimating the camera pose in the current frame. Except these spots, the
pose estimation error is negligible and the trajectory of the estimated camera pose
is the same as the trajectory of the actual camera pose. This experiment indicates
that selecting the last-known camera pose yields a high success rate of the pose
estimation algorithm.
294 A.D. Tezerjani et al.

a Comparison of the actual and estimated camera pose trajectories in xc - yc plane b Comparison of the actual and estimated camera pose trajectories in θp - θt plane
0.3
actual trajectroy actual trajectroy
estimated trajectroy estimated trajectroy
0.25
0.5
0.2
0 frame no =60 frame no =1 0.15

0.1
–0.5
yc

θt
0.05
–1 frame no =1
0 frame no =60

–1.5 –0.05

–0.1
–2
–0.15

–2.5 –0.2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4
xc θp

Fig. 11.9 Comparison of the actual and estimated camera pose for 60 consecutive frames of
the camera moving in a spiral path. In this experiment the last estimated pose (last-known
configuration) is used as the initial seed for the pose estimation on the next frame. (a) Trajectories
in X-Y plane. (b) Trajectories in pan-tilt plane

a b × 10
4

0.35 14
initial seed offset: 40% initial seed offset: 40%
initial seed offset: 60% initial seed offset: 60%
Final value of objective function (J)

0.3 12
Computation time (sec)

0.25 10

0.2 8

0.15 6

0.1 4

0.05 2

0 0
10 60 110 160 210 260 310 10 60 110 160 210 260 310
Number of iterations Number of iterations

Fig. 11.10 Computation time and optimization error vs. number of iterations for two initial seeds
with 40 and 60 % offset. (a) Computation time vs. number of iterations. (b) Value of the objective
function J vs. number of iterations

The computation time is one of the major challenges for the iteration-based
optimization algorithms. In the proposed algorithm, the computation time depends
on the number of iteration, and the initial seed value. Figure 11.10 shows the
computation time and the optimization error vs. number of iterations for two
different initial seeds with different offsets. Here, the computation time increases
linearly by the number of iterations. However, when the algorithm converges to the
actual pose, the rising trend of the computation time stops. It means that increasing
the number of iterations has no significant effect on the computation time, after that
the optimization algorithm converges to the actual pose. Also, as seen in this figure,
11 4-DOF Pose Estimation of a Pipe Crawling Robot 295

the lower offset of the initial seed with respect to the actual pose yields in lower
computation time of the algorithm in converging to the actual pose.
For real-time pose estimation, the analytical calculation of the derivatives would
speed up the Nelder–Mead’s convergence. However, in real-world applications,
there will also be a trade-off between the speed of convergence and the accuracy
of results when having noise in the image. This can be verified only experimentally.
However, a numerical calculation of derivative was adopted to have a more realistic
optimization approach encountered in real scenarios.

11.5 Conclusions and Future Work

In this work, a new optimization method for estimating the 4-DOF pose of a sensor
module comprised of a conical laser projector and fish-eye camera was proposed.
We formulated the projection geometry of our proposed imaging platform, and
developed a GUI for generating the synthesized image set, to evaluate our method.
We did different experiments for finding the correlation between the initial seed
value and the success rate of the algorithms. Our experiments reveal high success
rates of the proposed pose estimation approach, when the initial seed is not too
far from the actual pose. Our experiments show that the initial seed value and the
success rates of the algorithms are highly correlated. Regardless of this correlation,
our experiments imply that the proposed approach has high performance when the
initial seed is selected with low offset from the actual pose. Based on this result, we
proposed an efficient way of choosing the initial seed, using the last known pose of
the robot, in order to minimize the effect of the initial seed value on the performance
of the algorithms.
A learning-based method was also adopted to solve the backward mapping
for the pose estimation problem, however the results were not satisfactory. The
culprit in ANN-based approach to solve an optimization problem with multiple
possible solutions of this caliber is that the “random-walk" nature of the ANN can
intermittently converge to multiple non-local optimal solutions. Even after setting a
small simplex, the random walks associated with the ANN may go beyond a local
search. Further rudimentary attractors/penalty functions would be required to force
the ANN-based search take place within the initial simplex.
The main contribution of this work comparing with similar techniques is that
no prerequisite calibration or setup is required. As the future step, we will do a
sensitivity analysis on un-modeled errors in the system, in a simulation environment
first and then verify the results via experimentation.

Acknowledgement The partial funding provided by the Natural Sciences and Engineering
Research Council of Canada (NSERC) is greatly acknowledged.
296 A.D. Tezerjani et al.

Appendix 1: Calculating the Offset Angle

In order to calculate the offset angle between the fixed frame and the camera frame,
one needs to rotate a sample intersection point p.0 / on the pipe’s inner wall, with
respect to the camera orientation, < p d >, in opposite direction. To ease the
calculation, we chose the point p. D 0/ as the sample point. The coordinates of
the point are calculated using Eq. (11.3) as following:
 !T
d C rp sin.t /  xc sin.t / C yc cos.t / sin p
p. D 0/ D rp ; 0 ;  
cos p cos.t /
(11.15)
Now, by rotating this point around the pipe fixed frame in opposite direction, the
coordinate of the corresponding point on the pipe wall, cp. D 0/ is calculated as
following.

cp. D 0/ D RX .p / RY .t / p.0/; (11.16)

where RX and RY are the rotation matrices along the X and Y axes. The offset angle
is calculated as projected angle of the point cp0 in the X  Y plane as following:
 
ycp.D0/
ı D atan
xcp.D0/
!
d sin p C yc C cos t C rp sin p sin t  xc sin p sin t
D atan    (11.17)
cos p rp  xc C d sin t

Having the offset angle ı, the intersection point p./ on the pipe’s inner wall is
mapped to the corresponding point on the image plain with the relative orientation
along the x axis equal to  C ı.

Appendix 2: Calculating the Analytic Form

of the Reference Signal

In order to represent the analytic form of the reference signal, we expand Eq. (11.8),
and separate the terms as the trigonometric functions of the variable . By doing
this, the reference function is represented by the following form:

ref a . / Df1 C f2 sin  C f3 cos  C f4 sin  cos  C f5 sin2   2 Œ0 2

(11.18)
11 4-DOF Pose Estimation of a Pipe Crawling Robot 297

where f1 , f2 ,: : :, and f5 are the parametric functions of the camera pose <
xc ; yc ; p ; t >. These coefficients are calculated as following:

2m21 C m22 C m23

f1 D xc2 C yc2 C rp2 C
2m4
2m1 m3 2m1 m2
f2 D 2yc rp C f3 D 2xc rp C
m4 m4
m2 m3 m22  m23
f4 D f5 D
m4 2m4

where

m1 D d  xc sin t C yc cos t sin p m2 D rp sin t

m3 D rp cos t sin p m4 D cos2 p cos2 t (11.19)

It shows that the refa is the parametric trigonometric function of the variable .

References

Alvarez L, Deriche R, Papadopoulo T, Sánchez J (2007) Symmetrical dense optical flow estimation
with occlusions detection. Int J Comput Vis 75(3):371–385
Andrews DP, Bedford J, Bryan PG (2013) A comparison of laser scanning and structure
from motion as applied to the great barn at Harmondsworth, UK. ISPRS Int Arch
Photogramm Remote Sens Spat Inf Sci XL-5/W2:31–36 [Online]. Available: http://www.int-
arch-photogramm-remote-sens-spatial-inf-sci.net/XL-5-W2/31/2013/
Chang P (2002) Robust tracking and structure from motion with sampling method. Ph.D.
dissertation, Robotic Institute, Carnegie Mellon University
Cohen BR (2012) Fixing America’s crumbling underground water infrastructure. Last viewed:
March-2014 [Online]. Available: http://cei.org/issue-analysis/fixing-americas-crumbling-
underground-water-infrastructure
Daniilidis K, Spetsakis ME (1997) Understanding noise sensitivity in structure from motion. In:
Visual navigation: from biological systems to unmanned ground vehicles. Citeseer [Online].
Available: citeseer.ist.psu.edu/daniilidis97understanding.html
Dementhon DF, Davis LS (1995) Model-based object pose in 25 lines of code. Int J Comput Vis
15(1–2):123–141
Duran O, Althoefer K, Seneviratne LD (2003) Pipe inspection using a laser-based transducer and
automated analysis techniques. IEEE/ASME Trans Mechatron 8(3):401–409
Duran O, Althoefer K, Seneviratne LD (2007) Automated pipe defect detection and categorization
using camera/laser-based profiler and artificial neural network. IEEE Trans Autom Sci Eng
4(1):118–126
Fleck MM (1994) Perspective projection, the wrong imaging model. Technical report 95-01.
Computer Science, University of Iowa
Gennery D (2006) Generalized camera calibration including fish-eye lenses. Int J Comput Vis
68(3), 239–266 [Online]. Available: http://dx.doi.org/10.1007/s11263-006-5168-1
Hagbi N, Bergig O, El-Sana J, Billinghurst M (2011) Shape recognition and pose estimation for
mobile augmented reality. IEEE Trans Vis Comput Grap 17(10):1369–1379
298 A.D. Tezerjani et al.

Iketani A, Sato T, Ikeda S, Kanbara M, Nakajima N, Yokoya N (2007) Video mosaicing based on
structure from motion for distortion-free document digitization. In: Computer Vision–ACCV
2007. Springer, Berlin, pp 73–84
Josephson K, Byrod M (2009) Pose estimation with radial distortion and unknown focal length.
In: IEEE Conference on Computer Vision and Pattern Recognition, 2009 (CVPR 2009). IEEE,
New York, pp 2419–2426
Kannala J, Brandt S (2004) A generic camera calibration method for fish-eye lenses. In:
Proceedings of the 17th International Conference on Pattern Recognition, 2004 (ICPR 2004)
1:10–13
Lagarias JC, Reeds JA, Wright MH, Wright PE (1998) Convergence properties of the Nelder-Mead
simplex method in low dimensions. SIAM J Optim 9:112–147
Matsui K, Yamashita A, Kaneko T (2010) 3-d shape measurement of pipe by range finder con-
structed with omni-directional laser and omni-directional camera. In: 2010 IEEE International
Conference on Robotics and Automation (ICRA). IEEE, New York, pp 2537–2542
Maverick Inspection Ltd. (2009) Laser pipe profiling. http://www.maverickinspection.com/
Services/LaserPipeProfiling/tabid/135/Default.aspx. Accessed Feb 2014
Pipeline and Gas Journal (1999) Crc-evans automatic welding introduces laser vision. http://www.
pipelineandgasjournal.com/crc- evans- automatic-welding-introduces-laser-vision-welding-sy
stem. Accessed 1 March 2014
Raush Electronics USA (2014) Raush laser pipe profiling system. http://rauschusa.com/products/
laserprofile. Accessed Feb 2014
Scaramuzza D, Martinelli A, Siegwart R (2006) A flexible technique for accurate omnidirectional
camera calibration and structure from motion. In: IEEE International Conference on Computer
Vision Systems, 2006 (ICVS ’06), pp 45–45
Simo-Serra E, Ramisa A, Alenya G, Torras C, Moreno-Noguer F (2012) Single image 3d human
pose estimation from noisy observations. In: 2012 IEEE Conference on Computer Vision and
Pattern Recognition (CVPR). IEEE, New York, pp 2673–2680
Spica R, Giordano PR, Chaumette F (2014) Active structure from motion for spherical and
cylindrical targets. In: Proceeding of the IEEE International Conference on Robotics and
Automation (ICRA). IEEE, Hong Kong
Tahri O, Mezouar Y, Chaumette F, Araujo H (2010) Visual servoing and pose estimation with
cameras obeying the unified model. In: Visual Servoing via Advanced Numerical Methods.
Springer, Berlin, pp 231–250
Tezerjani AD, Mehrandezh M, Paranjape R (2013) Design and development of an optics laser
sensor for quality control and inspection of pipes. Faculty of Engineering and Applied Science,
University of Regina, internal document, September 2013
Unnikrishnan P, Thornton B, Ura T, Nose Y, A conical laser light-sectioning method for navigation
of autonomous underwater vehicles for internal inspection of pipelines. In: OCEANS 2009-
EUROPE. IEEE, New York, pp 1–9
Vincze M, Pichler A, Biegelbauer G, Hausler K, Andersen H, Madsen O, Kristiansen M (2002)
Automatic robotic spray painting of low volume high variant parts. In: Proceedings of the 33rd
ISR (International Symposium on Robotics), 2002
Winkelbach S, Molkenstruck S, Wahl FM (2006) Low-cost laser range scanner and fast surface
registration approach. In: Pattern Recognition. Springer, New York, pp 718–728
Chapter 12
Analysis of Tide Gauge Sea Level Time Series

Albert Parker

Keywords Data mining • Statistical analysis • Curve fitting • Forecast

12.1 Introduction

Global warming refers to a continuing rise in the average temperature of Earth’s

oceans and air and sea at Earth’s surface as a result of the increasing anthropogenic
carbon dioxide emission. According to the IPCC (http://www.climatechange2013.
org/report/), ‘Human influence has been detected in warming of the atmosphere and
the ocean, in changes in the global water cycle, in reductions in snow and ice, in
global mean sea level rise, and in changes in some climate extremes. This evidence
for human influence has grown since AR4. It is extremely likely (95–100%) that
human influence has been the dominant cause of the observed warming since the
mid-20th century.’ Because of this strong political focus on the climate, the analysis
of time series of measured parameters linked to the climate has become a central
point in the new science of climate change mostly made of modelling results.
Linear and nonlinear regressions and curve fittings have been used to analyse the
time series of climate parameters. Classic statistical tools may determine the values
of parameters for an equation whose form may be specified to better represent the
experimental evidence with a best fit of a set of data values. Linear analyses are
widespread, but polynomial, exponential, periodic, and general multiple nonlinear
functions are similarly or even more helpful. In addition to computing the optimal
values of the parameters to best fit the function to the data, it is also important to

A. Parker ()
School of Aerospace, Mechanical and Manufacturing Engineering,
RMIT University, Melbourne, VIC, Australia
e-mail: [email protected]

© Springer International Publishing Switzerland 2015 299

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5__12
300 A. Parker

compare the data points with the fitted equation. The distribution of residual values
and the computation of the auxiliary statistic help in understanding the quality of
the fitting.
Worth of mention, statistic is applied to a set of data generally assumed to
be a truly accurate representation of the tracked phenomena. However, every
measurement has his error, but random and systematic, and there may be significant
biases due to many factors not under control, for example the modified heat release
and heat storage close to a thermometer location or the land motion where a tide
gauge is located.
Unfortunately, the time series of climate parameters are often characterised by
scattered measurements of poor quality. Statistic is not helpful to address these
issues, and the reliability of the results from the analysis drastically reduces. Many
‘observational’ parameters such as the global air and sea surface temperature or
the global sea level are not truly measured products, but the result of a complex
computational procedure from supporting experimental information is particularly
poor especially for the past. What has not been measured cannot be guessed with
the accuracy needed in the climate analyses.
This chapter provides some simple tools common to the most part of the
statistical packages and their sample use to time series of sea levels. According
to the global warming theory, the increased carbon dioxide emission produces an
increased heat uptake, and the most part of this heat uptake goes to warm the oceans
that consequently increase in volume mostly by thermal dilatation and in less extent
by the addition of fresh water from the melting of ices. By applying the proposed
technique, the reader may understand if there is any effect of global warming on the
recorded time series.

12.2 A Technique for Analysing Sea Levels

Sea levels are very well known to oscillate in time following many periodicities,
with the detectable longest periodicity obviously linked to the length of the
measured record. The sea level measured by tide gauges is not the absolute sea
level, as the land locally may move upwards or downwards because of isostasy or
subsidy. Other perturbing factors like the change of the instrument or the relocation
of the instrument or works about the tide gauge location may all reduce the quality
of the record. At the best, tide gauge records are available from the mid-late 1800s
and in very few selected locations worldwide. Usually, monthly average mean sea
levels are considered for each individual location.
Generally speaking (http://www.fig.net/pub/monthly_articles/july_2010/july_
2010_hannah.pdf), the issues with the analysis of tide gauges include tide gauge
errors, the differences between the reading of the tide gauge and the actual sea level;
error in the datum of the tide gauge; analysis errors, as the consideration of records
that are short or incomplete; geophysical effects, including land motion and harbour
construction. This chapter is only dealing with the analysis the data recorded by the
tide gauge as a true representation of the monthly mean sea level.
12 Analysis of Tide Gauge Sea Level Time Series 301

12.3 Tide Gauge Measurements

Tide gauges as the one at Fort Denison in Sydney, NSW, Australia or the one in
Freemantle, Western Australia, Australia are measuring tides since the 1800s. Tides
have been measured at Fort Denison since 1857 on completion of the fort, initially
using a bench mark cut into the stonework. From 1867 consecutive measurements
with different instruments were used as tide measuring technology developed.
Measurements in Freemantle are more recent.
Sydney has two tide gauge records in Fort Denison, SYDNEY, FORT DENISON
[PSMSL; http://www.psmsl.org/data/obtaining/, ID 65, latitude 33.850, longitude
151.233, Time span of data 1886–1993, Completeness (%) 100] and SYDNEY,
FORT DENISON 2 [PSMSL; http://www.psmsl.org/data/obtaining/, ID 196, lati-
tude 33.855, longitude 151.226, Time span of data 1914–2012, Completeness (%)
98]. The two tide gauges have almost 80 years of successful overlapping and the
integration of the two records in a composite record spanning 1886–2012 without
any gaps does not pose any quality issue.
Fremantle has only one tide gauge, FREMANTLE [PSMSL; http://www.psmsl.
org/data/obtaining/, ID 111, latitude 32.065556, longitude 115.748139, Time span
of data 1897–2012, Completeness (%) 92]. This tide gauge has significant gaps, and
the way the gaps are filled interpolating neighbouring years may change the result
of a statistical analysis. In a first instance the missed data are simply neglected.
Tide gauges do not measure the absolute sea level but the sea level relative to
the land. Being the land subject to isostasy or subsidy, sea levels measured by tide
gauges may rise or fall simply because the land is moving downwards or upwards.
Figure 12.1 (from http://en.wikipedia.org/wiki/Glacial_isostatic_adjustment)
presents computations for the glacial isostatic adjustment. The land may clearly
be affected by velocities comparable with the velocity of rise or fall of the oceans.
The figure presents a modelling result of present mass change due to post-glacial
rebound and the reloading of the ocean basins with seawater. Blue and purple areas
indicate rising due to the removal of the ice sheets. Yellow and red areas indicate
falling as mantle material moved away from these areas in order to supply the rising
areas, and because of the collapse of the fore bulges around the ice sheets. Models
are absolutely not accurate but they give an idea of a general trend.
Additional land motions exist because of more local phenomena, both natural
and man-made, as it is the case of the large subsidence along the coast of the US
states of Texas or Louisiana or in the Pacific atolls.
GPS monitoring (http://www.sonel.org/) and local surveying may help under-
standing the land motion contribution to the tide gauge signal, but in the limit of the
accuracy of the measurement of the vertical position, the stability of the positioning
of the tide gauge vs. the GPS datum, the recalibration issues and the reduced length
of the record.
The present distribution of the PSMSL stations (http://www.psmsl.org/data/
obtaining/) is apparently sufficient with about 2,000 stations in the PSMSL database
for which monthly and annual mean sea level information is available. However,
302 A. Parker

18.0

16.5

vertical crustal motions in mm per year via GIA theory 15.0

13.5

12.0

10.5

9.0

7.5

6.0

4.5

3.0

1.5

0.0

–1.5

–3.0

–4.5

–6.0

GMT 2010 Oct 26 18:37:27 deg ord 70 paulson GIA appx. uplift rates in mm per yr

Fig. 12.1 Glacial isostatic adjustment (fromhttp://en.wikipedia.org/wiki/

Glacial_isostatic_adjustment)

many stations have historically only been measured for some months or years and
the longer records are very scattered and located only in few selected areas. The
geographical distribution of the longer records contains significant geographical
bias towards the Northern Hemisphere (Europe, Japan and USA). The tide gauge
records do not cover the same time period of time, and many of them have quality
issues. Only the long-term time series may be used to infer sea level trends if there
are no quality issues.
Figure 12.2 (from http://www.psmsl.org/data/obtaining/) shows the locations of
sites with not less than 40 years of data. The coverage of the map is everything but
good. The coverage drops dramatically requiring at least 60 years of data that is
the minimum necessary to understand a trend (Parker 2013a, b, c, d, e, f, g, 2014;
Parker et al. 2013). The figure also presents the number of PSMSL stations with
data available from Northern and Southern Hemisphere (NH and SH respectively).
Brown colour corresponds to the mean sea level data available from NH and blue
colour corresponds to the mean sea level data available from SH. The most part of
the world oceans is definitively not covered enough.
Sydney is the only tide gauge covering the South Pacific since more than
100 years. Sydney and Fremantle are the two tide gauges covering the southern
hemisphere extending more than 100 years.
12 Analysis of Tide Gauge Sea Level Time Series 303

700

600

500
Number of locations

400

300

200

100

0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000

Fig. 12.2 Demography of PSMSL stations (from http://www.psmsl.org/data/obtaining/, updated:

22 August 2012): stations with at least 40 years of data (top) and stations from Northern and
Southern Hemisphere (bottom)

12.4 Linear Model

Traditionally, tide gauge data have been analysed by using a linear model (Parker
2013a, b, c, d, e, f, g, 2014; Parker et al. 2013). This basic linear model is the simple
linear regression which has only one independent variable:

y D ˇ0 C ˇ1  x C " (12.1)

where “0 , “1 are coefficients and " is the random error. Linear regression fits a
straight line to a given data set. The fitting model is written as:
304 A. Parker

y D ˇ0 C ˇ1  x C "
"  N 0;  2

where ˇ 0 is the y intercept, ˇ 1 is the slope, and " is the error term. The error
term represents the unexplained variation in the dependent variable and is usually
assumed to have a mean value of zero.
To estimate the parameters, the chi-square minimization or “weighted least-
square” method is used. The sum of the squares of the deviations between the
theoretical curve and the experimental points for a range of independent variables
is minimised. After fitting, the model may then be evaluated using hypothesis tests
and plotting of residuals.
Figure 12.3 presents the sea levels results for the Sydney and Fremantle tide
gauges from the basic linear analysis. The figure presents the sea levels in Sydney
on the left and Fremantle on the right. Figure 12.3a shows the monthly sea levels
(from http://www.psmsl.org/data/obtaining/), Fig. 12.3b the linear trend, Fig. 12.3c
the oscillations of the monthly sea levels about the linear trend and Fig. 12.3d the
land motion in nearby GPS stations (from http://www.sonel.org/).
For what concerns Fremantle, the nearby GNSS Stations from SONEL are
located at a distance from the tide gauge of 32,074 m. For what concerns Sydney,
the nearby GNSS Stations from SONEL are located at a distance from the tide
gauge of 10,775 m. In addition to the vertical motion of the tide gauge, other factors
as harbour constructions or the use of successive instruments as tide measuring
technology developed may have an influence on the recorded signal.
The classic approach to analyse sea level data is to use a linear fit. In Eq. (12.1),
x is the time, y is the sea level, and “0 and “1 are constants to compute the average
sea level rise over the period of observation.
The R-squared value of the fitting is not a measure of the accuracy of the trend
line but only a parameter related to the legitimate oscillations about the linear trend.
The coefficient of determination compares the estimated and actual y-values, and
ranges in value from 0 to 1. If it is 1, there is no difference between the estimated
y-value and the actual y-value. If it is 0, the regression equation is not helpful in
predicting a y-value. Being very well known that the sea levels oscillate, it should
not be a surprise that R2 is not unity.
The present SLR is 0.652 mm/year for Sydney and 1.658 mm/year for Fremantle.
The vertical motion of the Sydney and Fremantle tide gauges is unknown, but the
nearby GPS stations show significant subsidy over the last decade with rates of
0.89 ˙ 0.65 mm/year and 2.99 ˙ 0.34 mm/year respectively.
The presence or absence of acceleration in the measured signal is detectable by
observing the graph of the monthly deviations vs. this linear trend. The monthly
sea levels oscillates about the linear trend, and these oscillations are periodical.
Therefore, there is no visual evidence of acceleration in the recorded signal.
The accuracy and reliability of each measurement is not an issue that may be
addressed with a statistical analysis. The only accuracy parameter that may be
considered in the statistical analysis is the percentage of gaps (missed data) in
12 Analysis of Tide Gauge Sea Level Time Series 305

Fig. 12.3 Sea levels in Sydney (left) and Fremantle (right): (a) monthly sea levels; (b) linear
trend; (c) oscillations of monthly sea levels about the linear trend; (d) land motion in nearby GPS
stations
306 A. Parker

the time series and the length of the record. Biasing to the instrument reading for
multiple reasons is difficult to be detected by simply considering the recorded data.
Because temperatures and sea levels are very well known to oscillate with up to
quasi-60 years detected in the available instrumental data, the length of the record
and the time when the measurement starts play a significant role in determining
the longer term sea level rate of rise. For example, in a location with a perfect
60 years periodical oscillation, while a record of 60 years length may include a
complete oscillation, with less than 60 years the rate of rise may be overestimated
or underestimated depending on the start of the measurements about a peak or a
valley of the oscillation. Worth of mention, also records longer than 60 years have a
result that depends on a lesser extent on the starting point.
This classic approach requires long term records to avoid extrapolation of the sea
level rate. The linear approximation of the variation of sea level from only a multi-
decadal oscillation may lead to wrong results. classic approach requires records of
enough length to avoid computing as a longer term sea level rate of rise the variation
of sea level from a valley to a peak of a multi-decadal oscillation, therefore using
as the longer term linear trend a wrong curve (Parker 2013a, b, c, d, e, f, g, 2014;
Parker et al. 2013). The time histories of linear fitting parameters are an additional
tool that may help clarifying the influence of the natural oscillations on the result of
the basic linear model.

12.5 Time Histories of Linear Fitting Parameters

A fully satisfying monitoring technique may be developed by using linear fittings to

compute reference velocity and acceleration of sea levels and detect the periodicities
of the natural oscillations. The dependent y-values are the monthly average sea
levels and the independent x-values are the time in years. According to Eq. (12.1),
the rate of rise of sea level SLR is equal to “1 . From a recorded time series xi , yi ,
i D 1, : : : , N with i D 1 the oldest data point and N the latest data point, it is possible
to compute by linear fitting many rates of rise SLRj,k depending on the initial and
final points j, k selected in the distribution.
The calculations for the sea level rate of rise SLRj,k is based on the formula:

X
k
.xi  x/  .yi  y/
iDj
SLRj;k D (12.2)
X
k
2
.xi  x/
iDj

In this equation x and y are the sample means and j and k are the indices of the
first and last record of the measured distribution considered for the particular SLR
estimation.
12 Analysis of Tide Gauge Sea Level Time Series 307

The LINEST function in Microsoft Excel gives the desired statistics for a line
by using the “least squares” method to calculate a straight line that best fits the data
and returns an array that describes the line.
 
SLRj;k D LINEST yj W yk ; xj W xk

If at a certain time xk , xj is taken as (xk  20) to compute the SLR20 with time
windows of 20 years, it is taken as (xk  30) to compute the SLR30 with time
windows of 30 years, and it is taken as (xk  60) when computing the SLR60 with a
time window of 60 years. xj is then simply taken as x1 when computing the SLRA
by using all the data available at any time. This way, from a measured distribution
xi , yi for i D 1, : : : , N it is possible to estimate the time histories of SLR20 , SLR30 ,
SLR60 , and SLRA (Parker 2013a, b, c, d, e, f, g, 2014; Parker et al. 2013).
Providing that more than 60–70 years of continuously recorded data, without
any quality issues, are available in a given location, the SLRA,m returns a reasonable
estimation of the velocity of sea level change at the present time xm . The acceleration
may then be computed as

SLRA;m  SLRA;m1
SLAm D (12.3)
xm  xm1

This conventional velocity and acceleration will clearly oscillate, and their time
history, rather than a single value, will be a parameter of interest (Parker 2013a, b,
c, d, e, f, g, 2014; Parker et al. 2013).
Alternative acceleration parameters may be obtained by using a polynomial
fitting of the data in the recorded time series xi , yi , i D 1, : : : , N. A second order
polynomial fitting (parabolic fitting) returns the average acceleration over the length
of the considered time window as twice the second order coefficient. The LINEST
function may be combined with other functions to calculate the statistics for other
types of models that are linear in the unknown parameters, including polynomial,
logarithmic, exponential, and power series. In the specific of the parabolic fitting, it
is possible to define an acceleration parameter.
 
SLAj;k D 2  LINEST yj W yk ; xj W xk O f1; 2g

More than one value of SLA is indeed the time histories of SLA20 , SLA30 , SLA60 ,
and SLAA that may be of interest.
The availability of a long record permits to understand which conclusions could
have been inferred from measurements in the same location if the measurements
could have started much later. At a certain time xN , the graph of SLRk,N vs. the
record length xN  xk gives an idea of the errors in evaluating the rate of rise in
shorter records.
In a case with non-accelerating tide gauge records as the norm, SLR1,N returns the
present sea level rate of rise, and the graphs of SLRj,k and SLAk help to confirm the
lack of any acceleration and understand the longer periodicities in the oscillations. In
308 A. Parker

a case of accelerating tide gauge records, the approach would confirm the presence
of acceleration in the form of a constantly increasing SLRj,k and a constantly
positive SLAk rather than oscillating values about the longer term trend and the
zero.
Figure 12.4 presents the time histories for Sydney and Fremantle of the above
velocity and acceleration parameters based on the linear model. The figure presents
the sea levels in Sydney (left) and Fremantle (right). Figure 12.4a is the time
history of SLR computed with 20, 30 and 60 years windows. Figure 12.4b is the
time history of SLR computed by using all the data available at any time—this is the
conventional sea level velocity. Figure 12.4c is the time history of SLA computed
as the time rate of change of the above SLR. Figure 12.4d is the apparent present
velocity that could have been computed with the linear model in case the measuring
record could have started in different times.
Figure 12.4a evidences the presence of multi-decadal oscillations of different
periodicities in Sydney and Fremantle. Figure 12.4b shows the oscillating con-
ventional sea level velocity, and Fig. 12.4c the conventional sea level acceleration
oscillating between positive and negative values. Figure 12.4d shows the huge errors
that could have been made using short records to assess the rate of rise of sea levels.
Figure 12.5 presents for Sydney (left) and Fremantle (right) in Fig. 12.5a the
time history of SLA computed with 20, 30 and 60 years windows, in Fig. 12.5b the
time history of SLA computed by using all the data available at any time—this is
the average sea level acceleration since the start of the record and in Fig. 12.5c the
apparent present acceleration that could have been computed as twice the second
order coefficient of a polynomial fitting in case the measuring record could have
started in different times. The accelerations clearly evidence the many oscillations
similarly to the velocities and the positions.

12.6 Nonlinear Model

The linear model may be further refined by introducing a nonlinear model based on
multiple fittings. This model may be used to forecast future sea levels including the
effect of the periodic oscillations, obviously on the basis of what has been measured
up to a certain time.The nonlinear regression model is
!

y D f x;  C "

!

Where x is the independent variable and  D .™1; ™2; : : : ; ™j / are j parameters. In
this equation, fit parameters are related to the independent variable in a nonlinear
fashion.
When the relationship between dependent and independent variable is nonlinear
there are no explicit solutions for the equation parameters and parameter values are
estimated by using an iterative process.
12 Analysis of Tide Gauge Sea Level Time Series 309

Fig. 12.4 Sea levels in Sydney (left) and Fremantle (right): (a) time history of SLR computed with
20, 30 and 60 years windows; (b) time history of SLR computed by using all the data available at
any time—this is the conventional sea level velocity; (c) time history of SLA computed as the time
rate of change of the above SLR; (d) apparent present velocity that could have been computed with
the linear model in case the measuring record could have started in different times
310 A. Parker

Fig. 12.5 Sea levels in Sydney (left) and Fremantle (right): (a) time history of SLA computed with
20, 30 and 60 years windows; (b) time history of SLA computed by using all the data available at
any time—this is the average sea level acceleration since the start of the record; (c) apparent present
acceleration that could have been computed as twice the second order coefficient of a polynomial
fitting in case the measuring record could have started in different times

A common measure of the quality of nonlinear fitting parameters is the chi-

square value:
X
n
_ 2
 D
2
wi yi  y i
iD1

_
Here wi is the weighting coefficient, yi are the experimental data points and y i are
the theoretical points. The “best-fit” parameter values are obtained by minimising
chi-square.
We propose here a method to analyse tide gauge records by fitting with the
available data with a line and multiple sinusoidal functions and use these data to
fill the gaps eventually available in the record. We apply these methods to the tide
gauge records of Sydney and Fremantle.
12 Analysis of Tide Gauge Sea Level Time Series 311

We do use a line and multiple sinusoidal functions fitting to better describe the
tide gauge behaviour. The fitting curve is:

Xn   
x  xc;i
y D .y0 C a  x/ C y0;i C Ai  sin   (12.4)
iD1
wi

In this equation, n is the number of sinus functions and y0 , a, y0,i , Ai , xc,i , wi are the
fitting coefficients.
The composite tide gauge record for Sydney does not pose any quality issue
having no gaps. Fremantle has significant gaps, and the way the gaps are filled
interpolating neighbouring years may change the result of a spectral analysis of
the time series and affect the periodical oscillations detected.
In the present approach, the parameters of Eq. (12.4) are computed first from the
measured data, then the gaps in the measured data are filled with the estimation from
Eq. (12.4) and new parameters of Eq. (12.4) are computed. This iterative procedure
is the best approach to avoid introducing unrealistic periodicities of oscillations.
The Solver suite of commands in Microsoft Excel may be used to compute
the fitting coefficients y0,i , Ai , xc,i , wi by minimising the chi-square value. Solver
is a process of changing the values in cells to see how those changes affect the
outcome of formulas on the worksheet. Solver works with a group of variable cells
that participate in computing the formulas in the objective and constraint cells.
However, adds-in as XLSTAT (Parker 2013e) may certainly make easier to perform
the nonlinear fitting or run a spectral analysis to compute the spectral density or the
periodogram of the time series.
Figure 12.6 presents the results for Sydney and Freemantle. The figure presents
the periodogram from the spectral analysis of the monthly departures vs. the linear
trend of Fig. 12.3 then the measured monthly sea levels and the values computed
with the fitting with a line and multiple sinusoidal functions of Eq. (12.4).
The spectral analyses of the Sydney and Fremantle time series are significantly
different. The spectral analysis allows to transform a time series into its coordinates
in the space of frequencies, and then to analyse its characteristics in this space.
From these coordinates it is possible to extract the magnitude and the phase and
build representations such as the periodogram or the spectral density.
The spectral representation of a time series fXt g, (t D 1, : : : , n), decomposes
fXt g into a sum of sinusoidal components with uncorrelated random coefficients.
From there we can obtain decomposition the auto-covariance and autocorrelation
functions into sinusoids. The spectral density corresponds to the transform of a
continuous time series. However, we usually have only access to a limited number
of equally spaced data, and therefore, we need to obtain first the discrete Fourier
coordinates (cosine and sine transforms), and then the periodogram. From the
periodogram, using a smoothing function, we can obtain a spectral density estimate
which is a better estimator of the spectrum.
The iterative process to compute the parameters of the sinus fitting in Eq. (12.4)
may return periodicities different from the periodogram especially in the longer
periodicities because of lack of data and noise and the different algorithm. Noise
312 A. Parker

Fig. 12.6 Sea levels in Sydney (left) and Fremantle (right): (a) periodogram of the monthly
oscillations about the linear trend; (b), (c) and (d) measured and computed monthly sea levels
by the proposed multiple nonlinear fitting over different time windows—better accuracy may be
obtained by using other periodic functions

in the measurements and length of the record limit the accuracy of the detection
especially of the longer periodicities.
Table 12.1 presents the coefficients of Eq. (12.4) for Sydney and Fremantle.
The periodicities ofthe sinusoidal oscillations of (12.4) considered in Sydney
are 1.00, 0.50, 58.07, 20.04, 12.93, 4.19, 30.85 and 18.39 years, with decreas-
ing amplitudes 37.54, 21.56, 13.23, 12.68, 10.58, 6.76, 6.56 and 3.96 mm.
Table 12.1 Coefficients of Eq. (12.4) for Sydney and Fremantle
Sydney
y0 5703.6 y0i 0.0392 0.0241 0.1997 0.0267 0.0001 0.0544 0.003 0.0092
a 0.652 xci 1.2521 0.102 1509.7 120.9778 756.3 79.2851 31.4733 19.3755
Wi 0.5003 0.25 15.4263 29.0358 10.0214 9.1953 6.4644 2.0947
Ai 37.5435 21.5583 6.5628 13.2312 12.6756 3.9577 10.5844 6.7593
12 Analysis of Tide Gauge Sea Level Time Series

Fremantle
y0 3457.3 y0i 0.0081 0.2473 0.0081 0.008 0.0084 2.7812 0.0081 0.3093
a 1.6565 xci 0.6338 4.7161 17.0552 12.3597 2.3347 312.3075 70.4469 13.3771
Wi 0.4999 0.2301 2.793 1.8029 2.0694 34.0483 6.2119 1.8847
Ai 101.4107 3.5149 25.5496 21.2954 21.8308 23.2554 19.25 13.468
313
314 A. Parker

Especially the longer periodicities are sensitive to the parameters of the iterative
nonlinear fitting procedure. The periodicities of the sinusoidal oscillations of (12.4)
considered in Fremantle are 1.00, 5.59, 68.10, 4.14, 3.61, 12.42, 3.77 and 0.46 years,
with decreasing amplitudes 101.41, 25.55, 23.26, 21.83, 21.30, 19.25, 13.47 and
3.51 mm.
The oscillations are mostly different in amplitude and period (but also in phase)
in Fremantle (Indian Ocean) and Sydney (Pacific Ocean). Apart from the common
1 and 0.5 years periodicities and the close quasi-60 years oscillation, then Sydney
has also strong oscillations of periodicity about 20 years that are not detectable in
Fremantle. The longer periodicity of Sydney is very close to the record length and
clearly indicates the need of more data to detect the periodicities above the quasi-
60 years that is already difficult to clarify when less than two full cycles available.
The shape of the periodic function may be optimised for a better accuracy
(reduction of the residual error).
The reconstruction of the tide gauge record for small gaps as in the case of
Fremantle does not pose particular issues. However, when the missed information
exceeds the recorded information, the reliability of the infilling techniques drasti-
cally reduces.
Breakpoint alignment may be necessary in cases of abrupt changes in a record
resulting from perturbing events as relocation of the tide gauge, destruction of the
tide gauge, construction works close to the tide gauge and any other biasing issue
that may be identified in the record.
The extension of the tide gauge record to cover more than the length of the
measurements is not recommended.
Testing of the local statistic vs. a regional expectation is similarly not rec-
ommended, being the sea level behaviour still very far from being completely
understood locally and globally.

12.7 Conclusions

A simple but reliable methodology has been shown to compute the rate of rise of
sea levels and assess the presence or absence of accelerations or decelerations.
The methodology builds on the linear fittings of the available data over different
time windows and the analysis of the time histories of velocity and acceleration
parameters.
Multiple nonlinear fittings may help to infill gaps in incomplete records, detect
the periodicities of the natural oscillations about a longer term trend and eventually
predict the future trends on the basis of what has been recorded as to date.
This forecasting procedure is legitimate if a specific tide gauge of enough quality
and length and all the neighbouring tide gauge of about same quality and length are
all acceleration free.
12 Analysis of Tide Gauge Sea Level Time Series 315

Key Symbols

y Sea level
x Time
“ Fitting coefficient
" Error
¢ Standard deviation
SLR Sea level velocity
SLA Sea level acceleration
w Weight or fitting coefficient
a Fitting coefficient
A Fitting coefficient
xc Fitting coefficient

References

Parker A (2013a) Natural oscillations and trends in long-term tide gauge records from the pacific.
Pattern Recognit Phys 1:1–13. doi:10.5194/prp-1-1-2013
Parker A (2013b) Oscillations of sea level rise along the Atlantic coast of North America north of
Cape Hatteras. Nat Hazards 65(1):991–997. doi:10.1007/s11069-012-0354-7
Parker A (2013c) Sea level trends at locations of the united states with more than 100 years of
recording. Nat Hazards 65(1):1011–1021. doi:10.1007/s11069-012-0400-5
Parker A (2013d) The non-linear, naturally oscillating pattern of sea-levels in the Chesapeake Bay,
East Coast, USA. Nonlinear Eng 2(1–2):1–10. doi:10.1515/nleng-2013-0008
Parker A (2013e) Apparent hot and cold spots of acceleration along the Atlantic and Pacific coasts
of the United States. Nonlinear Eng. doi:10.1515/nleng-2013-0012
Parker A (2013f) Why global warming went missing since the year 2000. Nonlinear Eng.
doi:10.1515/nleng-2013-0017
Parker A (2013g) Minimum 60 years of recording are needed to compute the sea level rate of rise
in the western south pacific. Nonlinear Eng. doi:10.1515/nleng-2013-0011
Parker A (2014) Confirming the lack of any sea level acceleration around the Australian coastline.
Nonlinear Engineering 3(2):99–105, doi: 10.1515/nleng-2013-0025
Parker A, Saad Saleem M, Lawson M (2013) Sea-level trend analysis for coastal management.
Ocean Coastal Manage 73:63–81. doi:10.1016/j.ocecoaman.2012.12.005
Index

A deployment process, 82
Acoustic filter, 235, 246 description, 81
Active vibration control. See Axially dynamic behavior and stabilities, 82
translating cable systems equations of motion
Amplitude synchronization constant velocity, 87–88
application, control strategy, 98–100 displacement field, 85
chaotic motion, 100–103 Hamilton’s principle, 86
control strategy, 95, 97, 98, 102 kinetic energy, 85
input control, 97, 99, 101, 102, 104 Lagrangian function, 86
reference signal, 97, 99, 101, 102, 104 model, 84, 85
Autodriver algorithm, 10, 29 nonlinear differential governing
Autonomous vehicle steering dynamics equation, 89
changes, transient and steady-state strain energy, 86
responses von Karman-type equations, 85
differences, instantaneous and FSMC, 83, 84
steady-state values, 11–13, 15, 16 Galerkin method, discretization, 90
equations of motion, 10–11 non-dimensionalization, 89
global frame, angular orientation, 13–14 pinned–pinned boundary, 90–91
instantaneous and steady-state paths, SMC, 83
16, 17 spacecraft antenna, 82
integration, 15 vibration control (see Vibration control)
magnification, stage depicts, 16–18 volume control, 83
proximity, 29
realistic step change, 10
rotation center, 14–15 B
steady-state time history, 11, 12 Bézier curves
time history, 11 Bernstein polynomials, 121, 122
understeer passenger car, 9 CAD software, 121
velocity components, 15 2D Cartesian coordinate system, 121
SSDS (see Steady-state dynamic steering linear, quadratic and cubic, 121, 122
(SSDS)) Newton interpolation, 121, 123
Axial crush testing, 220–221 parametric form, 123
Axially translating cable systems step formulas, 124
aerospace and aeronautical structures, 81 wood/plastic, protrusion, 120
control strategy development, 91–93 Bode plot, 247

© Springer International Publishing Switzerland 2015 317

L. Dai, R.N. Jazar (eds.), Nonlinear Approaches in Engineering Applications,
DOI 10.1007/978-3-319-09462-5
318 Index

C linearization, network elements

Chaos capacity, 242
motion, 94–96 gas density, 243, 244
piecewise linear systems, 61 pressure difference, 243
Timoshenko beams, 82 relationship, 244
Circle-projecting collimated laser resistivity, 242
analytic form, reference signal, 292–293 subsystem, 242
offset angle calculation, 292 temperature dependency, 242–243
pipelines (see Pipelines) multi to mono-dimensional system,
Control strategy development 240–241
desired/reference vibration, 92 physical networks (see Physical networks)
fuzzy rule, 92 structure, 236
Galerkin method, 91–92 transfer function (see Transfer function,
governing equation, 91, 93 exhaust system)
third-order Galerkin discretization, 93
Convolution integral, 150
Coordinate transformation, 36 F
Crash safety Feature based methods, 260
high-speed, 194 Finite element analysis
severity, head injuries, 172 Abaqus/Explicit v6.11, 223, 225–226
testing, 170 ANCAP test and FE simulation, 185, 186
Curve fittings central finite difference, 226–227
linear and nonlinear regressions, 295 close-up deformation, 228, 230
pipeline, 284 contact surfaces, 231
correlation, hood deformation, 185, 186
ductility and MSFLD damage criteria,
D Aluminium 6060 T5, 228, 229
Data mining, 197 energy plot, 181, 184
Deformable registration force–displacement plots, 228, 229
HIC value requirements, 188–189 head impactor, 181, 182
hood assembly, 198 HYPERVIEW programs
optimal waveform, 177 acceleration vs. time, 181, 184
Dioptric imaging sensor. See Pipelines hood displacement vs. time, 181, 183
Dynamic systems of multi-dimensions, 83, implementation, 228
84, 95 LS-DYNA-3D, 181, 182
MSFLD damage criterion, 228
NCAP pedestrian subsystem testing
E protocol, 185
Elasticity, loss tangent. See Visco-elasticity optimisation, 181, 182
Error analysis PRIMER, 181
multi-step methods, 120 Forecast
and stability, 132 sea levels, 304
Exact methods tide gauge, 310
continuity and compatibility, 63–64 Fourier series and wavelet functions, 268
linear equation, 67–68 Free form deformation (FFD)
one-to-one relationship, 64 B-Spline transformation, 266–268
secondary suspension, 63 control node displacements, 268
steady state periodic response, 63 medical image registration, 265
transcendental equations, 65–68 Frequency response
Exhaust system acoustic modeling bilinear mass-spring vibrator, 60
control, chemical pollutions, 235 nonlinear vibrating systems, 71
ECAP, 236 primary suspension, 68
intake system components, 235 stiffness values, 59
ladder topology, 244–246 transcendental equations, 59
Index 319

Frequency synchronization HOODMATL, 197

application, control strategy, 105, 106, 108, and INRGEOM, 197
109, 111, 112 Kriging approach, 197
control input, 107, 110, 113 Monte Carlo method, 197
reference signal, 105, 107, 108, 110 nonlinear LS DYNA models, 196
FSMC. See Fuzzy sliding mode control predictability, response surface, 197
(FSMC) primary impact, head injuries, 198
Fuzzy sliding mode control (FSMC), 83, 84 selection, structure and material
actual and optimal deformation,
198–199
G evaluation, efficiency, 199
Galerkin method, 90–92 HIC values, 201, 204
hood panel thickness relationship,
202–204
H multi-cone hood inner structure, 204
Head impact model numerical tests, 200–203
ANCAP pedestrian protection testing optimal deformation space, 204
protocol, 184 primary impact, 198, 200
child head impactor, 183 structural stiffness, 201
finite element model, 184–185 theoretical optimal waveform, 203
GTR-9 and NCAP tests, 183 WSTC, 203
hood assembly, 183 standard deviation, 198
requirements, 183–184 strength-two orthogonal array design, 196
Hood design thickness, aluminium material
assembly, 178, 179 deformation space, 207, 208
central area, 180 HIC value, 205–207
inlaid, 178, 179 80 mm deformation space, 205, 206
inner panel, 180 multi-cone structure, 204–205
length, 178 outer and inner, 204, 208
material and thickness, outer panel, Hood rear reference line (HRRL), 189, 191
179–180
passenger car front-end collision, 178
structure, 180 I
wraparound, 178, 180 Initial value problem (IVP), 117, 118, 127
Hood intrusion computation Intensity based methods, 260
accelerometer rotation, 186, 188 Interpolation theory
coordination, 186, 187 FFD, 265–268
direction, 186 Fourier series and wavelet functions, 268
displacement, impactor, 186, 188 locally affine, 268
‘Intrusion T’ calculation method, 187, 189 RBFs, 264–265
measurements, acceleration, 186 Inverse Laplace transform
Hood leading edge (HEL), 189, 190 logarithmic law model, 149, 151
Hood panels SLS, 142
aluminium and steel, 213–214 transfer function, 246
ANOVA chart, HIC mean value, 197, 198
final design
aluminium with 85 mm deformation L
space, 209–210 Ladder topology, acoustic modeling
head impact performance, steel hood energy sources, 245, 246
configuration, 209 header and equivalent physical network,
optimal hood configuration, 209 245, 246
peak acceleration, 209 parameters calculations, 244, 245
waveforms, 208–209 pipe segments, 245
HIC values, 196–197 Simulink model, 246
320 Index

Laplace transform Multi-step method

power law model, 147–149 Adams–Moulton method, 120, 132
SLS, 141 Runge–Kutta, 118
stress relaxation, 142 stability and error analysis, 120
Logarithmic law model
addition rules, 150
Hookean spring, 151 N
of non-linear visco-elasticity, 149 Nonlinear behaviors, 81
storage modulus, 153 Nonlinear differential equation
viscosity constant, 153, 154 Adams–Moulton method, 120, 128
Loss tangent Bézier curves (see Bézier curves)
visco-elastic models (see Visco-elasticity) B-spline polynomials, 120
and viscosity, 155 BVPs and IVP, 117
consistency, stability and convergence
characteristics polynomials, 124–125
M
Dahlquist’s equivalence theorem,
Medical image registration
127
applications, 254–256
k-step method, 126
computer vision, 251
mC1-step formula, 125–126
CT, MRI and PET, 252, 253
multi-step method, 125, 127
image interpolation, 253
Euler vs. Runge–Kutta method, 118, 119
intensity information based registration
harmonic motion, 132
algorithm, 254, 255
k-step method, 119
interpolator, 254
Lotka–Volterra equation, 129, 131
mono-modal registration, 253
mass–spring–damper system, 128–130
multi resolution pyramids, 253
multi-stage methods, 118
objective function, 253–254
ODE45, 128
optimization method, 254, 255
ODEs (see Ordinary differential equations
pre-processing and feature extraction, 253
(ODEs))
similarity measures
Runge–Kutta methods, 118
feature based methods, 260
Van der Pol oscillator, 129, 133–134
voxel information based registration,
Nonlinear finite element and post-buckling
256–259
ABAQUS code, 230, 231
spatial mapping, 251
axisymmetric local buckling failure, 221,
transformation model (see Transformation
222
model)
collision safety, 220
translation and rotation, image, 252
and crushing behaviour, 219
Multiple coordinate system kinematics
cylindrical tube, 229–231
Euler derivative transformation formula
damage criteria, 230–231
classical mechanics, 36
dynamic response, cylindrical shell, 228,
global-fixed and rotating body-fixed
230
frame, 34–35
failure mode, 221, 222
kinematics, 36
force vs. displacement graph, 222
local/simple, 34
identical buckling modes, 221
time-derivative vector, 34
MSFLD, 231
kinematic chain rule
numerical calibration
angular acceleration, 40, 41
central finite difference, 226–227
magnitudinal and convective rate of
contact constraint, 227–228
change, 41
damage evolution and element removal,
multiple angular velocity vectors, 38–40
225
three angular velocities, 41
ductile damage and failure, 224
Razi acceleration, 32
FE models (see Finite element analysis)
rigid body motion, 33
initiation, ductile and shear failure,
second derivative, 36–37
223–224
three frames, 37–38
Index 321

material, 223 differences, rebound velocities, 211–212

shear failure criterion, 225 EEVC WG10, 167
solution procedure, ABAQUS/explicit, EEVC WG17, 169–170
225–226 elastic strain energy, aluminium material,
sheet metals convergence, 230 213
specimens and test machines fatalities and injuries, 162–164
aluminium cylindrical tubes, 220, 221 filtered data, 214
load press, 220–221 finite element model, 181–183
structure, thin wall (see Thin wall structure) Flex-PLI and TRL-PLI, 167, 168
transportation, 219 GIDAS, 163
Nonlinear vibration isolator HIC values, 164, 214–215
blades and magnetics, 82 hood intrusion, 186–189
cable system, 84 HOODMATL, 210
frequency response, 71 inner hood geometries, 194
FSMC strategy, 84 integration, 210–211
multi-dimensional systems, 84 IRTAD, 163
Non-rigid registration leg and head impact test machines, 169
B-Spline motion model, 267 lower leg impactor, TRL-PLI, 169
interpolation and approximation theories, minimum deformation space, 215
264–268 multi-cone inner hood structure, 215
physical phenomena, 263–264 NCAP and regulatory tests, 170
scaling and shear components, 262 optimal waveform, 175–178
transformation constrains, 269–270 optimisation, hood panels (see Hood
Numerical methods panels)
multi-step methods, 124 panel thickness, 210
Runge–Kutta, 118 parametric geometric model, 181
passenger car collides, 162
peak acceleration
O duration, 171, 172
Ordinary differential equations (ODEs) HIC calculation, 170–171
first order system, 128 magnitude, 171, 172
ODE45, 128 and rebound velocity, 210
physical networks, 237–238 performance, 193–194
plastic sphere coverage, 169
PMHS, 170
P POLAR II and III, 170
Pedestrian protection, head injury PPCFC, 162, 163
AIS 2–6 injuries, 164 Pugh matrix, 194
ANCAP, 164–165 recommendations, 215–216
ANOVA chart, peak acceleration, 210, 211 refinement, 196
automobiles, 161–162 requirements, vehicle design (see Vehicle
car manufacturing, 161 design, pedestrian protection)
and chest impacts, 163 research, design parameters, 214
control factors, refinement scope, 195 resultant acceleration vs. vertical intrusion,
cycling and walking, urban areas, 161 210, 211
deformation space standard deviation, HIC, 214
direct and indirect contact surfaces, 172 stress distribution, beam, 212–213
hood panels, excessive stiffness, 173 subsystem test method, 167, 168
inertial loading, 173, 174 technical, 169
secondary impact, 173, 174 tri-axial accelerometer, 170
structural stiffness, 172–173 unsafe conditions, 162
vehicle design requirements, 174–175 variables, 195
design, hood, 178–180 vehicle front-end, 163
322 Index

Physical networks trigonometric function, 285–286

approach, engineering system, 236 underwater infrastructure inspection, 279
ECAP, 236 variables separation, model reduction, 281
elements, 238–239 vision-based pose estimation, 278
mono and multi-dimensional systems, water, oil and gas transport, 277
239 Post-buckling. See Nonlinear finite element
ODE, 237–238 and post-buckling
structure, 236, 237 Power law model
variables, 237 fractional derivative approach, 149
Piecewise linear systems Riemann–Liouville fractional integration,
asymmetric systems, 75–77 148
DOF vibrating system, 60 Power spectrum
double/multiple hitting, 72 measurement program, 240
dual rate bilinear vibrating system, 59, 60 system sound exhaustion, 241
exact solution (see Exact methods)
gap-damper engagement, 59, 60
R
implicit plot, 73
Radial basis functions (RBFs), 264–265,
mechanical model, 61
270
motion, system, 62
Razi acceleration
nondimensional form, 62
classical expression, 33
nonlinearity range, 73–75
coordinate frames, 41–42
secondary suspension, 61
inertial, 41, 43–45
stability analysis, 61
mechanical interpretation
steady response, 73, 75
angular velocity vectors, 50–52
time response
compound rotation motion, 49, 50
excitation frequency, 69
derivative transformation method, 48
frequency, 68, 69
inertial and non-inertial frame, 48, 49
implicit plot, 69
Newton’s second law of motion, 48
primary suspension, 68
three coordinate frames system, 48–49
steady response, 70
total acceleration vector, 49–51
transcendental equations, 59
mixed derivative transformation formula,
vibration isolator, 70
31–33
Pipelines
transformation, multiple coordinate frames,
computation time and optimization error
31, 45–47
vs. number of iterations, 290
Rigid body kinematics
imaging platform and configuration, 280,
angular velocity, 48
281
application, 52
imaging unit components, 279, 280
compound rotation motion, 33, 44, 45
inspection tools, 279
coordinate transformation, 36
laser ring projection, 283, 284
inertial acceleration, 46, 47
learning-based method, 291
linear acceleration vector, 46
metric measurements, image-based, 277
mechanics, Razi acceleration, 33
model reduction process, 281
non-inertial rotating coordinate frames, 33
Nelder–Mead method, 286, 291
non-destructive testing methods, 277
numerical optimization technique, 287 S
optimization techniques, nonlinear, 281 Sliding mode control (SMC), 83
pose estimation algorithm vs. initial seed SLS. See Standard linear solid (SLS)
value, 289 Sound
projection geometry, imaging, 282 attenuation components, 242
reference observation signal, 284, 285 measures, 236
robotic inspection systems, 277, 279, 280 and power spectrum, 240, 241
time-elapsed results, optimization process, pressure, 236
288 real vehicle, 236
Index 323

source, 245 T
speed of, 240, 244 Thin wall structure
SSDS. See Steady-state dynamic steering aluminium cylindrical tubes, 220, 221
(SSDS) collision safety, 220
Standard linear solid (SLS) crushing behaviour, 220, 231
Kelvin–Voight model, 141 failure, 228
loss tangent, 142, 143 global folding, 228
Maxell model, 147 impact, energy absorbers, 220
maximal stress vs. modulus, parallel spring, Tide gauge sea level time series
144, 145 classic statistical tools, 295
phase angle vs. frequency, 145, 146 global warming, 295
Statistical analysis linear model
accuracy and reliability, 300 multi-decadal oscillation, 302
dynamic features, 278 in Sydney, 300, 301
tide gauge, 297 measurement technology
Steady state glacial isostatic adjustment, 297, 298
fractional derivative, 148 PSMSL station demography, 298,
SLS, 142 299
stress–time curve, 139, 140 nonlinear model
visco-elasticity, 155–156 breakpoint alignment, 310
Steady-state dynamic steering (SSDS) chisquare value, 304, 306
angles, front-wheel-steering vehicle, 20–21 coefficients, Sydney and Fremantle,
conditions, 24 307, 309
coordination, road curvature center, 21–22 fitting curve, 306–307
cycloidal motion lane change in Sydney and Freemantle, 307, 308
maneuver-moving, 19 statistical packages, 296
radius of curvature, 19, 20 technique, 296
description, 16 thermometer location/land motion, 296
final position, vehicles, 28–29 time histories, linear fitting parameters,
forward velocity, 19 302–304
kinematics angles Time histories, linear fitting parameters
actual road vs. steady-state road, 27, 28 acceleration parameters, 303
average, 23, 24 LINEST function, Microsoft Excel, 302
body frame, 23, 25 monitoring technique, 302
differences, 27, 28 non-accelerating tide gauge records,
global frame, 27 303
motion expression, body frame, 23, sea levels in Sydney, 304–306
25–26 Transfer function, exhaust system
outer and inner wheel, 22–23 acoustic filter, 246
rate of change, vehicle heading angle, Bode plots, 247
23, 25 ladder network, 246
side slip velocity, 23, 24 Transformation model
vs. steady-state, 27 continuous optimization methods, 270–271
steering road, 24, 26 diffeomorphic, 269–270
velocity, global frame, 23, 26 discrete optimization, 272
measurements, 29 forward vs. inverse, 261
output–input relationships, 17–18 hierarchical methods, 272
road curvature center, 19–20 interpolation theory
Steering dynamics. See Steady-state dynamic FFD, 265–268
steering (SSDS) Fourier series and wavelet functions,
Stress strain 268
loss tangent, 139 locally affine, 268
Riemann–Liouville fractional integration, RBFs, 264–265
148 inverse consistency, 269
324 Index

Transformation model (cont.) steady-state conditions, 4

physical characteristics steer angle, 4–5
elastic model, 263 rotation center, 8–9
fluid flow, 264 steady-state responses, 6–7
pre-registration technique, 262 steering angle
topology preservation, 269 changes, 9–16
Transient part steady-state, 16–29
power law model, 148 Vibration control
SLS, 142 amplitude synchronization (see Amplitude
visco-elasticity, 155–156 synchronization)
axially translating cable, 94
chaotic motion, 94–96
V frequency synchronization (see Frequency
Vehicle design, pedestrian protection synchronization)
components, GTR-9, 165 Galerkin method, 94
consumer metric, 166–167 numerical simulations, 93
design space Visco-elasticity
definition, 188 characteristics, 137
hood assembly, 188–189 decay function, stress relaxation, 138
modifications, interfacing components, hysteresis loop, stress–strain ellipse, 139,
189 140
project scope, 189, 190 linear/non-linear models, 137–138
regulatory requirement, 188 negative storage modulus, 156
head impact area sinusoidal strain, 139
adult and child, 165–166 stress–time curve, 139, 140
child, 190–191 transient and steady state parts, 155–156
GTR-9 protocol, 189, 190 Viscosity, loss tangent. see Visco-elasticity
HEL, 189, 190 Voxel information based registration
hood side reference line, 189, 191 Gaussian noise, 257
HRRL, 189, 191 image intensities, 257
impactor speed, 191 joint histograms, brain MRI image and
optimisation, 190, 192 rotated version, 259
WAD divides, 189 mono-model registration, 256
lower leg, 166 optical flow setting, 258
passenger car, 165 residual complexity, 258
upper leg, 166 sum of absolute difference (SAD), 257
Vehicle dynamics sum of square difference (SSD), 256
equations of motion
coefficients, 4, 5
forward velocity, 5 Z
planar bicycle car model, 3–4 Zener model. See Standard linear solid (SLS)