INTERNATIONAl.

CENTRE FOR MECHANICAL SCIENCES

COURSES AND LECTURES - No. 297

ROTORDYNA MICS 2
PROBLEMS IN
TURBO MACHINERY

EDITED BY
N. F. RIEGER
STRESS TECHNOLOGY INCORPORATED

SPRINGER-VERLAG WIEN GMBH

Le spesc di stampa di questo volume sono in parte coperte cia contributi
del Consiglio Nazionale delle Ricen:he.

This volume contains 3S4 illustrations.

This work is subject to copyright.

All rights are reserved,
whether the whole or part of the material is concerned
specif1Cally those of translation, reprintins, re-use of illustrations,
broadcastin& reproduction by photocopyinJ machine
or similar means, and storase in data banks.
© 1988 by Springer-VerlagWien
Originally published by Springer-Verlag Wien New York in 1988

ISBN 978-3-211-82091-9 ISBN 978-3-7091-2846-6 (eBook)

DOI 10.1007/978-3-7091-2846-6
 PREFACE

Current attention in turbomachinery design and operation is sharply focused on the

achievement of higher levels ofavailability and reliability. The usual trends towards minimum
weight and low vibration have become standard well-recognized criteria for turbomachines of
all types. The thrust of new work is towards increasing the number of operating hours per year
and to minimize outage and maintenance periods.
It is hoped that the technology presented in this volume, "Rotordynamics II- Problems of
Turbo machinery" will contribute towards the above objectives by its consolidated presentation
of existing material, and with the inclusion of newly prepared material on a range of new and
important topics. This latter is not elsewhere available in a single volume. The subject of
rotordynamics now covers a large number of contributing disciplines, and it is now soundly
based on long established principles. Nonetheless, rotordynamics continues to evolve new and
fast-developing aspects, such as the recent surge in instrumentation development, analytical
equipment, and in monitoring devices, all of which are part of the current revolution in micro-
electronics. Certain new devices such as magnetic bearings also offer great promise for future
application in space, computer, and medical applications. These devices have themselves given
rise to sub-sets of rotordynamics technology. In more conventional areas, new problems such
as torsional transient effects on machine survivability, and the recent emphasis on blading life
improvement have contributed to increased demands for structural component reliability, and
have required additional technology involving life prediction and component life extension.
Rotordynamics research is therefore much different, more detailed, and perhaps more
exacting than the situation which existed ten orFifteen years ago. It is therefore timely to focus
these newer developments into a single volume, written by many experts involved in developing
this new technology. Such a volume should provide designers and operators with a
concentration of new material which will, in turn, further contribute to the development of
more reliable and better functioning turbomachinery.
The initial chapters ofthis volume discuss the operating properties offluid-film bearings, and
the analysis of rotor behavior in such bearings. This fundamental material is utilized in Chapter
Four, where a discussion of problems of rotor balancing is presented. This section covers
balancing machines, rigid rotor balancing, flexible rotor balancing, and several balancing case
histories from practice. Chapter Five deals with techniques for identification of stiffness,
damping, and inertia coefficients for seals,for instances where the dynamic behavior offluid-
moving turbomachines requires the contributions of the interstage and gland seals to be
included in the analysis. Chapter Six is a discussion of the stability of rotors in two oil-film
bearings. The prediction of instability threshold speeds is demonstrated, and verified by
experimental studies.
The technology presented in the preceding six chapters is coordinated in Chapter Seven,
which deals with the computer analysis of rotor-bearing systems. A specific large computer
code called PALLA for the dynamic analysis of rotor structural systems is described. The
functioning of this code is demonstrated using several examples from practice. This code
analyzes several rotating shafts in fluid-film bearings and seals, mounted on a flexible
foundation. Although codes of this type have existed since Prahl's critical speed analysis in
/945. to this day it is still difficult to decide on the degree of complexity which the ideal or
optimum rotordynamic computer code should contain. PALLA provides the analyst with a
comprehensive tool for complex rotor-structure systems, to obtain response and stability
information in using well defined support properties.
Machine-soil interactions are discussed in Chapter Nine. The theory of such interactions is
established through the use of viscoelastic field equations, and applied, with experimental
verification, to the case of a turbomachine frame on a soil foundation.
Experimental methods for the study of rotor behavior in bearings are discussed in Chapter
Eight. Sensor technology is presented first, and when.the modern principles of this science have
been described, several experimental techniques for analysis of rotors in bearings and seals are
discussed with examples. Chapter Eight describes certain aspects ofpossible interactions which
may occur beteen a machine and its foundation. The practical aspects of this presentation
throw valuable light upon a complex subject which faces both analysts and experimentalists,
and complements the discussion of the PALLA code in Chapter Seven.
 The remaining three chapters deal with special topics which have now developed into
sophisticated sub-technologies. Chapter Ten discusses problems oftorsional shaft systems. The
analysis of turbine-generator torsional transient vibrations has recently undergone
considerable development due to major torsional transient problems which have occurred in
units around the world. Analy~ica/techniques for torsional systems using modal analysis are
discussed in the second section of Chapter Ten, and the complications introduced by the
presence of gears in a drive system are discussed in section three of this chapter. Problems of
turbine blades are discussed in Chapter Eleven. The first presentation deals with free vibrations
and forced vibrations ofblades, and the second section examines the current state-of-the-art for
blade excitation and damping. Case studies involving problems of turbine blades are discussed
in section three ofthis Chapter, and a consolidation ofexisting approaches for life evaluation of
blades is discussed in section four. Experimental and analytical studies on the damping
properties of steam turbine blades are described in part five of Chapter Eleven.
Chapter Twelve describes several special topics. The first of these is magnetic bearings for
which a comprehensive introduction to this subject with applications is presented in the first
section. The technology of magnetic bearings is currently being advanced by needs in space
technology, computers, and elsewhere. This section deals with control systems of magnetic
bearings, and the details presented for application ofsuch devices should provide designers and
users with much valuable guidance. The final section of Chapter Twelve deals with vibrations in
variable speed machines. This topic is of interest wit~ all machines which much traverse one or
more critical speeds during runup and rundown, and machines which operate over a broad
range of operating speed such as gas turbines, utility steam turbines, and pumps. The
importance of the rate at which a critical speed is traversed has been recognized since Lewis s
work in 1932. The technology of this chapter should find further application in life evaluation
techniques discussed above for such machines.
As Editor it is my pleasure to thank all authors who have contributed their labors, creativity,
and valuable time to prepare their sections of this volume. Warm thanks will doubtless also be
expressed by the many readers who scan these pages. Thanks are also due to the diligent staffat
CIS M. in particular Professor Giovanni Bianchi of the Politecnico di Milano whose idea it was
to publish these proceedings; to Professor Carlo Tasso who supervised the preparation of the
\'olume itself. and Signora Bertozzi who had the delicate task of guiding these many authors
towards a common goal. Our publisher Springer- Verlag also deserves our grateful thanks for
producing such a fine manuscript, and we also record our indebtedness to alithe secretaries
who worked long and hard to ensure that the quality of the manuscript was achieved to the
satisfaction of the authors. My own thanks go to Ms. Candace Rogers of Stress Technology
who did an outstanding job of personally checking all the manuscripts that went into this
volume. I extend my grateful thanks to all these people for their contributions.
It is the hope ofall those who have contributed to this volume that their labors have produced
a new state-of-the-art document which will serve as a guide for the creators of new
turbomachines and for those who must maintain the present machines.

Neville F. Rieger
Rochester, New York
 CONTENTS

Page

Preface
Chapter I
Introduction
by N.F. Rieger ................................................................ . I
Chapter 2
Bearing Properties
by E. Kramer ................................................................. 17
Chapter 3
Analysis of Rotors in Bearings
by E. Kramer ................................................................. 41
Chapter 4.1
Principles of Balancing and of Balancing Machines
by N.F. Rieger ................................................................ 67
Chapter 4.2
Flexible Rotor Balancing
by N.F. Rieger ................................................................ 95
Chapter 4.3
Case Histories in Balancing of High Speed Rotors
by N.F. Rieger ............................................................... 129
Chapter 5
Seal Properties
by R. Nordmann ............................................................. 153
Chapter 6
Stability of Rotors
by R. Nordmann ............................................................. 175
Chapter 7
Computer Analysis of Rotor Bearings- P.A.L.L.A.: A Package to Analyze the
Dynamic Behavior of a Rotor-Supporting Structure System
by G. Diana, A. Curani, B. Pizzigoni .......................................... .. 191
Chapter 8.1
Sensor Technology
by J. Tonnesen ............................................................•. . 261
Chapter 8.2
Experimental Techniques for Rotordynamics Analysis
by J. Tonnesen ............................................................... 269
Chapter 9
Interaction between a Rotor System and its Foundation
by L. Gaul . .........................................................•..•..... 283
Chapter 10.1
Problems of Turbine Generator Shaft Dynamics
by D. W. King, N.F. Rieger ...••...•..•......•.••.......•...................... 307
Chapter 10.2
Torsional Systems: Vibration Response by Means of Modal Analysis
by P. Schwibinger, R. Nordmann ............................. .................. 331
Chapter 10.3
Torsional Dynamics of Power Transmission Systems
by N.F. Rieger ............................. ............................. ..... 359
Chapter 11.1
Free and Forced Vibrations of Turbine Blades
by H. lrretier . ............................. ............................. ...... 397
Chapter 11.2
Flow Path Excitation Mechanisms for Turbomachine Blades
by N.F. Rieger ............................. ............................. ..... 423
Chapter 11.3
The Diagnosis and Correction of Steam Turbine Blade Problems
by N.F. Rieger ............................. ............................. ..... 453
Chapter 11.4
An Improved Procedure for Component Life Estimation with Applications
by N.F. Rieger ............................. ............................. ..... 485
Chapter 11.5
Damping Properties of Steam Turbine Blades
by N.F. Rieger ............................. ............................. .... . 515
Chapter 12.1
Magnetic Bearings
by G. Schweitzer ............................. ............................. ... 543
Chapter 12.2
Vibrations in Variable Speed Machines
by H. lrretier . ............................. ............................. ...... 571
 CHAPTER I

INTRODUCTION

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

The major problem areas of rotordynamics are

identified. The manner in which these problems are
addressed as Chapters of this book is described. A
number of commonly-used rotordynamic terms are
defined for reference purposes. The orbits of
several frequently observed whirl motions are
discussed and correlated with their causes, for
convenience in problem diagnosis. Literature
sources for further information on specific problems
including books, papers, and conference proceedings,
with a reference listing are presented. Prominent
references in the historical development of rotor-
dynamics are cited.

1.1 Problems and Scope of Rotordynamics

Rotor vibration problems may arise from a number of sources,

of which the most important are residual rotor unbalance and
rotor instability, Rotor unbalance causes a rotating force
in synchronism with running speed (synchronous unbalance),
whereas rotor instability is a self-excited vibration which
may arise from bearing fluid-film effects, electromagnetic
effects, flow effects, or from some combination of these
factors. Several other mechanisms have also led to rotor
vibration problems in the past. For example, hysteresis (dry
friction) effects have lead to whirl in lightly damped
machinery; different shaft lateral stiffnesses are a known
cause of unstable whirl in electrical machinery; and
non-linear foundation effects have led to vibrations in
certain speed ranges. In most instances, practical 'fixes'
can be applied to reduce or eliminate these undesirable
2 N.F. Rieger

vibrations once the problem is correctly diagnosed. The

design of smooth running rotating machinery requires that
each of these problem areas should be carefully eliminated
from the machine system specifications. In most instances
this can be done quite readily if adequate allowance is made
for the machine dynamic characteristics as part of the design.

The dynamic characteristics of greatest interest in rotating

machinery are:

a) Rotor critical speeds in the operating range.

b) Unbalance response amplitudes at critical speeds.

c) Threshold of resonant whip instability.

d) Bearing transmitted force.

e) System torsional critical •peeds.

f) Gear dynamic loads.

g) Disk natural frequency (compressor. turbine. gear).

h) Bucket. blade. and impeller natural frequencies and

modes.

i) Blade flutter fr~quencies.

j) Rotating stall and surge thresholds.

Many other important vibration topics such as noise and

structural vibration of rotating machinery could be added to
this 1 ist. Each of the above subjects has an extensive
published literature. In this book. the fundamentals of each
topic are presented with comments and discussions on important
recent contributions to the literature of each. No single
book presently covers the entire subject of rotor bearing
dynamics. but a thorough appreciation may be gathered from
those publications listed in the references to each chapter.

The properties of hydrodynamic bearings are discussed in

Chapter 2. The plain cylindrical bearing is considered
first. and expressions for its static performance are derived.
using the short bearing form of Ockvirk for convenience.
This theory is then extended to the dynamic case. and expres-
sions are developed for the linearized th~Jory of short
bearings. Properties of special bearing types. such as
partial. tilting pad. and multi-arc bearings are discussed.
This chapter provides a basis upon which the theory and
experience of rotors in flexible supports can be developed in
the chapters which follow.
Introduction 3

Chapter 3 contains an introduction to the analysis of rotors

in bearings. The basic theory of the Jeffcott-Foppl rotor is
developed in detail with emphasis on the principles involved.
This leads to the well-known expressions for critical speed
and unbalance response.

The principles of rotor balancing are discussed in Chapter

4. The fundamentals of rigid rotor balancing using two
correction planes are first presented and the development of
modern balancing machines of the Lawaczek-Heyman type is
discussed. Case histories of certain balancing problems
which have been encountered in practice are described.

Chapter S discusses the principles of hydrodynamic seals and

their influence on rotor performance. The representation of
the performance of certain types of seals in terms of stiff-
ness, damping, and inertia coefficients is discussed, with
examples which compare seal performance with test and
experimental data.

Chapter 6 deals with stability problems of rigid rotors in

bearings. The relationship between the eight linearized
coefficients and the instability threshold speed is
developed. Experience with in stability in practice is
presented, and experimental comparisons with predicted
threshold data based on matrix eigenvalue extraction
procedures is presented.

In Chapter 7 a general-purpose computer program for the

behavior of a rotor on a flexible foundation is discussed.
The structure of the code is described, with an example of
its use on a practical problem. The properties of several
computer programs which are a~·ailable from various sources
for rotordynamic analysis are discussed.

Practical procedures for obtaining test data on rotordynamic

performance of rotors in supports are discussed in Chapter
8. The technology and application of several types of
measuring sensors (capacitance, inductance, light sensors,
etc.) and their readout equipment are presented. The use of
such equipment for rotor performance measurements is discussed
with several practical test examples is also described.

The problem of interaction between a rotor and its supporting

foundation is discussed in Chapter 9. Many types of such
foundations exist, and their dynamic influence on rotor
performance is discussed, based on generalized findings from
a program of experiments, and correlation of results with
supporting theory.
4 N.F. Rieger

Torsional dynamics of drive trains is discussed in Chapter

10. The systems discussed include turbine-generator sets for
which an appropriate modal theory is derived and presented
with experimental correlation, and a program of tests on
practical turbine-generator sets with an analysis of test
results compared with transient response theory. Several
practical examples of geared industrial drive problems which
have involved torsional dynamic analysis for their solution
are discussed.

Chapter 11 provides a comprehensive discussion of turbo-

machine blade problems. Principles of vibration of such
blades are first discussed, followed by a discussion of the
types of excitation and damping forces which may be encoun-
tered in practice, together with their most probable sources.
Case studies of certain classical problems are given, together
with a discussion of current procedures for blade life
evaluation.

Certain special topics of current interest in the vibration

of rotating machinery are presented in Chapter 12. The
principles and application of magnetic bearings to rotating
machinery is an important topic which holds considerable
promise for future developments in high speed equipment.
Selected theoretical developments and supporting test
developments are described for both passive and active
magnetic rotor supports. A second topic is the problem of
the transient passage of a mechanical component through
resonance during runup and rundown operation. The principles
of this problem for a rotor passing through a critical speed,
and for a blade passing through an excitation harmonic are
described, with sample applications.

The extensive literature which deals with vibration problems

of rotors in bearings indicates the scope and sophistication
of the modern technology of this subject. This book provides
both a broad and up-to-date review of today's technology of
rotordynamics. The specialists who have written the various
chapters of this book are experts in each area. It is hoped
that this focus of timely expertise will be of value and
guidance to users of this book.

1.2 Definitions

Disk - A wheel, usually solid and axially slim, on which

mechanical work is performed or from which work is extracted.
Examples: turbine disk, compressor wheel.

~ - A beam, usually axisymmetric (commonly circular) on

which disks may be carried, either integrally or by shrink
fitting.
Introduction 5

~ - An assembly of disks on a shaft or simply a massive

shaft, mounted in supporting bearings. Two special classes
of rotors are: rigid rotor, flexible rotor.

P-igid rotor - A rotor which operates substantially below

its first bending critical speed. A rigid rotor can be
br0ugLt into, and will remain in, a state of satisfactory
balance at all operating speeds when balanced on any two
arbitrarily selected correction planes.

Flexible rotor - A rotor which operates close enough to,

or beyond its first bending critical speed for dynamic
effects to influence rotor deformations. Rotors which
cannot be classified as rigid rotors are considered to
be flexible rotors.

Dearing - Any low friction support which carries the rotor

and provides dynamic constraint in the transverse and/or
axial directions. The two main categories are fluid-film
bearings and rolling-element bearings. Types: journal
bearing, thrust bearing.

Fluid-film bearing - A bearing whose low friction

property derives from the thin fluid layer between the
rotor surface and the bearing metal. The fluid-film
layer may be generated by journal rotation (hydrodynamic
bearing) or by externally pressurized pumped fluid under
pressure {hydrostatic bearing).

Rolli ng-e 1 ement bearing - The low friction property

derives from mechanical rolling with marginal lubri-
cation, using ball or roller elements.

Journal - Specific portions of shaft surface from which rotor

applied loads are transmitted to bearing supports.

Pedestal - nearing support possessing mass elastic proper-

ties, mounted on the machine foundation.

Foundation - Machine support. May be rigid or possess mass

elastic properties.

System - The interacting combination of rotor, bearings,

pedestals and foundation which responds as a complex to
dynamic excitation.

E!U.L!. - Rotor transverse orbital motion about the static

equilibrium position, at any axial location.

Natural frequency - Any frequency of free vibration at which

a natural mode of the system assumes its maximum amplitude.
6 N.F. Rieger

Critical speed - Rotor speed at which local maximum amplitude

whirling occurs. Where the rotor has no significant gyro-
scopic effects, a critical speed occurs whenever the rotor
speed coincides with a system natural frequency.

Unbalance- Eccentricity of local e.g. of rotor from

undisturbed axis of rotation. Product of rotor local mass
times eccentricity of e.g. from shaft elastic axis. Also
expressed in terms of eccentricity along, e.g., in micro-
inches of eccentricity.

Unbalance response - Whirl maximum amplitude at a given

speed, caused by dynamic forcing action of rotating unbalance
on the system.

Sub-critical speed System vibration occurring at integer

sub-multiple of main system critical speed (1/2 w , 1/3 w ,
1/4 w ), • • • • • • 1/10 w ) arising from stiffness nog-
linearlty, such as 'flat' shaft.

Whirl instabilitY - Condition in which whirl radius increases

with time. May arise from variety of causes, e.g., bearings,
shaft hysteresis or geometry, foundation, torque or speed
fluctuation. Instability may grow indefinitely in time, or
become bounded by new constraints.

Half-frequency whirl - Instability of dynamically rigid rotor

in hydrodynamic fluid-film bearings, arising from rotor fluid
interaction. This instability commences near the first
system critical speed and occurs with a frequency somewhat
less than half the rotational speed. Associated with rigid
rotor systems where bearing properties predominate.

Resonant whip - System instability involving rotor flexure,

occurring in fluid-film bearing systems. It arises from
flexible rotor fluid-film interaction. It is similar in
nature to half-frequency whirl, except for presence of rotor
flexibility effect.

Fractional frequency whirl - Instability of dynamically rigid

rotor in externally pressurized fluid-film bearings. Other-
wise similar to half-frequency whirl, except that whirl
frequency is usually much less than half rotational speed,
e.g., 0.2Sw to 0.4Sw whirl frequency range.

1.3 Nature of Whirl Motions

A rotor is said to whirl when the e.g. of any cross section

traces out an orbit in time, instead of remaining at a fixed
point. If identical whirl orbits are traced out with
successive shaft rotations, the whirl is said to be stable.
If the orbit increases in size with successive rotations, the
Introduction 7

whirl is then unstable and may subsequently grow until the

orbit becomes bounded either by the internal forces of the
system, or by some external constraint, e.g., bearing rub,
guard ring, shutdown, etc. Smooth machine operation is
characterized by small, stable rotor whirl orbits, and by the
absence of any instabilities within the machine operating
range.

Some typical whirl orbits are shown in Figure 1. The circular

orbit in Figure 1 represents the synchronous whirling of a
rotor in isotropic radial supports. The frequency of the
excitation is synchronous as shown by the absence of loops
within the orbit. Excitation in such a case usually arises
from rotor unbalance.

An elliptical orbit, Figure 2, may arise from dissimilar

bearing stiffnesses or support stiffnesses in the horizontal
and vertical directions. These stiffnesses give different
displacements for the same rotor centrifugal force. Where
complex (i.e., cross-coupled) stiffness and damping proper-
ties exist as in fluid-film bearings, the major and minor
axes of the orbit then occur at some angle with respect to
the x and y coordinate directions, as shown. \'/here
cross-coupled effects are absent, the axes of the ellipse
coincide with the x and y coordinate directions.

If the whirl is non-synchronous, i.e., the rotor whirls at a

frequencyV, other than the rotational frequency w, the orbit
will contain a loop, frequently like that shown in Figure 3
for half-frequency whirl. The loop is internal indicating
that the whirl is in the direction of rotation.

Other non-synchronous excitations may occur at several times

rotational frequency, e.g., multi-pole electrical stimuli
give rise to multi-lobe whirl orbits such as that shown in
Figure 4.

Instabilities such as half-frequency whirl are frequently

bounded. In such a case the whirl would be initiated by
crossing the instability threshold speed. The whirl then
becomes a growing transient whose radius increases until a
new equilibrium position is found. Frequently this is
established by the fluid forces themselves and no damage is
done. The whirl then continues at a larger radius. Otherwise
the whirl increases until some structural constraint such as
the bearing surface or the machine casing is encountered. A
bounded instability whirl is shown in Figure S, including the
transient motion from the first condition of equilibrium to
the second equilibrium condition.
8 N.F. Rieger

Another type of transient condition is shown in Figure 6.

The rotor is initially operating in a small stable unbalance
whirl condition. .The rotor system then receives a transverse
shock, and the journal displaces abruptly in a radial
direct ion within the bearing clearance, but without con-
tacting the bearing surface. Following the shock, the rotor
motion is a damped decaying transient, as it returns to its
original small unbalance whirl condition.

Many other interesting types of whirl orbits have been

observed, such as those associated with system non-linear-
ities. Further comments on whirl motions are given in detail
in the chapters dealing with this and related topics.

1.4 Rotordynamics Information Sources

The rapid development of rotor-bearing dynamics has resulted

in a very extensive published literature and several books.
The best known books on this subject are listed in Table 1.1:
see references [1.1] through [1.13] of this Table. Details
are given in Section 1.6 at the end of this chapter. r.lost
books deal with some selected aspect of the subject, e.g.,
stability or balancing, rather than a comprehensive treatment.

The great practical relevance of rotordynamics studies has

also led to several major conferences, at which the current
technical developments have been presented. Vibrations
conferences featuring rotordynamics sessions were initiated
by the American Society of Mechanical Engineers in 1967.
Since then conferences have been held in the u.s.A. every two
years, and papers are published in the AS~lli Transactions. An
international conference on Vibrations in Rotating Machinery
has been held every four years since 1972 by the Institution
of Mechanical Engineers of London. Bound volumes of proceed-
ings are available from 1972 through 1984. An international
Symposium on Rotordynamics was also held by the International
Union of Theoretical and Applied Mechanics (IUT~I) in Lyngby,
Denmark in 1974. A book of proceedings is available from
this meeting. In recent years symposia on rotordynamics have
been organized by the International Federation for Theory of
Machines and Mechanisms ( IFToMM), in 1979, 1983, and 1987.
Published proceedings from these symposia containing state-
of-the-art papers are also available. An international
Rotordynamics Conference has been organized by IFToMJol every
four years, in 1982 (Rome), and in 1986 (Tokyo).

Beginning in 1976 the Vibration Institute in the U.S.A. has

organizud annual seminar meetings on vibration problems, with
an emphas~s on problems of rotating machinery. Other
conferences held by the Vibration Institute which emphasized
instrumentation and test measurement and balancing of rotors,
are also listed. Proceedings are available for these
Introduction 9

meetings. In each instance these proceedings consist of a

collection of original papers.

1.5 The Classical Literature of Rotor-Bearing Dynamics

The critical speeds of a uniform elastic shaft were first

investigated by Rankine [1.14] in 1869, who devised the term
'critical speed.' This phenomenon arose in factory overhead
pulley shafting at that time. It was incorrectly thought by
certain investigators to be an unstable condition. Dunkerley
[1.15] in 1894 presented an excellent collection of critical
speed studies related to pulley shafting, and gave his well-
known method with its experimental verification. Kerr [1.16]
in 1916 precipitated an extended discussion of turbine
critical speed problems and the mechanics of whirling.
Jeffcott [1.17] in 1919 resolved the evident confusion
surrounding the mechanics of rotor unbalance whirl in his
classical analysis of this problem. Stodola [1.18] proposed
the linear coefficient representation of fluid film bearing
properties in 1917. Hysteretic whirl was first investigated
by Newkirk [1.19] in 1924 during studies of blast furnace
compressor vibrations, and explained by Kimball [1.20].
Newkirk and Taylor [1.21] observed oil film whirl and resonant
whippina for the first time in 1925. An explicit review of
basic rotordynamics problems was presented in 1933 by Smith
[1.22]. Robertson [1.23], [1.24], [1.25], [1.26], [1.27]
presented a series of important rotordynamics papers between
1932 and 1935 on the subjects of bearing whirl, rotor
transient whirl, and hysteretic whirl.

Demands for larger and faster rotors, together with develop-

ment and application of the computer, had led to sustained
development of rotor-bearing technology since World War II.
Ha11 [1.28] conducted experiments on half-frequency whirl in
1946. The mechanism of bearing instability was first
explained by Poritsky [1.29] in 1952, and experimentally
confirmed by Boeker and Sternlicht [1.30] in 1955. The
linear theory of rotor instability was established by Lund
[1.31] in 1963. Prohl [1.32] introduced discrete mass
numerical calculation of rotors in 1946. Hagg [1.33] studied
unbalance response of rotors in bearings in 1948. Hagg and
Sankey [1.34], Sternlicht [1.35], Lund [1.36] and others
developed numerical hydrodynamic&! analysis procedures for
bearinas between 1955 and 1965. Procedures for flexible
rotor balancing were suggested by Linn [1.37] in 1928 and by
Thearle [1.38] in 1935. Their application was demonstrated
by Grobe! [1.39] and developed into practical procedures by
Goodman [1.40], Bishop and Gladwell [1.41], Lund [1.42] and
others since 1952.
10 N.F. Rieger

1.6 References

1.1 Dimentberg, F. M., Flexural Vibrations of Rotating

Shafts, Butterworth and Company, Ltd., London,
England, 1961.

1.2 Tondl, A., Some Problems of Rotor Dynamics, Pub-

lishing House, Czechoslovakian Academy of Sciences,
Prague, 1965.

1.3 Gunter, E. J ., Jr., 'Dynamic Stability of Rotor-

Bearing Systems,' NASA Report SP-112, 1966.

1.4 Rieger, N. F., Poritsky, H., Lund, J. W., et. al.,

'Rotor-Bearing Dynamics Design Technology,' Volumes
1-9, Wright-Patterson Air Force Base Aero Propulsion
Laboratory Reports, 1965-1968.

1.5 Wilcox, J. B., Dynamic Balancing of Rotating

Machinery, Sir J~aac Pitman and Sons, Ltd., London,
England, 1967.

1.6 Smith, D. M., Journal Bearings in Turbomachinery,

Chapman and Hall Ltd., London, England, 1969.

1.7 Eshleman, R., Shapiro, W., Rumbarger, J ., Rieger,

N. F., 'Flexible Rotor-Bearing System Dynamics,'
Volume I: Critical Speeds, Volume II: Bearing
Properties, Volume III: Unbalance Response and
Balancing, ASME Design Division Monographs, 1973.

1.8 Wilson, D., Pan, C., Allaire, P., 'Rotor-Bearing

Dynamics Design Technology,' Second Edition, WPAFB
Fuels and Lubricants Division, 1977.

1.9 Federn, K., Auswuchttechni k, Bd. 1, Berlin,

Springer, 1977.

1.10 Rieger, N. F., Vibrations of Rotating Machinery,

Vibration Institute, Clarendon Hills, Illinois, 2nd
Edition, 1982.

1.11 Rao, J. s., Dynamics of Rotors, Wiley-Eastern

Publishers, New Delhi, India, 1983.

1.12 Rieger, N. F., Balancing of Rigid and Flexible

Rotors, Shock and Vibration Information Center,
Naval Research Laboratory, Washington, DC, 1987.

1.13 Kellenberger, W., Elastisches Wuchten, Springer-

Verlag, 1987.
Introduction 11

1.14 Rankine, W. J. lolcQ., 'On the Centrifugal Force of

Rotating Shafts,' Engineer London, Volume 27, p.
249, 1869.

1.15 Dunkerley, s., 'On the Whirling and Vibration of

Shafts,' Phil. Transactions of the Royal Society,
Series A, Vol. 185, p. 229, 1895.

1.16 Kerr, W., 'On the Whirling Speed of Loaded Shafts,'

Engineering, pp. 150, 296, 386, 410, and 420.
February 18, 1916. Discussions by C. Chree, A.
Morley, A. Stodola, H. Naylor, H. Jeffcott. J.
Danus. W. Kerr.

1.17 Jeffcott, H. B., 'The Lateral Vibration of Loaded

Shafts in the Neighborhood of a Whirling Speed-
The Effect of Want of Balance,' Phil. lofagazine.
Series 6. Vol. 37. p. 304, 1919.

1.18 Stodola, A., Steam and Gas Turbines, Vols. 1 and 2.

Translated by L. C. Loewenstein. McGraw-Hill Book
Company, Inc., New York, 1927.

1.19 Newkirk, B. l., 'Shaft Whipping,' General Electric

Review. Vol. 27, p. 169, 1924.

1.20 Kimball, A. L., Lovell. J., 'Internal Friction as a

Cause of Shaft Whirling,' Trans. ASME. Vol. 48.
1926.

1.21 Newkirk, B. L •• Taylor, H. D., 'Oil Film Whirl -An

Investigation of Disturbances on Oil Films in
Journal Bearings.' General Electric Review, Vol.
28, 1925.

1.22 Smith. D. M.. 'The Motion of a Rotor Carried by a

Flexible Shaft in Flexible Bearings.' Proceedings
of the Royal Society, Series A•• Vol. 142. p. 92.
1933.

1.23 Robertson, D•• 'The Vibrations of Revolving Shafts.'

Phil. Magazine, Series 7. Vol. 13. p. 862. 1932.

1.24 Robertson. D.. 'The Whirling of Shafts.' The

Engineer. Vol. 158. p. 216, 1934.

1.25 Robertson. D.. 'Whirling of a Journal in a Sleeve

Bearing.' Phil Magazine. Series 7. Vol. 15. p. 113.
1933.

1.26 Robertson, D.. 'Transient Whirling of a Rotor.'

Phil. Magazine. Series 7, Vol. 20; p. 793. 1935.
12 N.F. Rieger

1.27 Robertson, D., 'Hysteretic Influences on the

Whirling of Rotors,' Proceedings !MechE, Vol. 131,
p. 513, 1935.

1.28 Hagg, A. C., 'Some Vibration Aspects of Lubri-

cation,' Lubrication Engineering, pp. 166-169,
August 1948.

1.29 Pori tsky, H., 'Contribution to the Theory of Oil

Whip,' Trans. ASME, Vol. 75, pp. 1153-1161, 1953.

1.30 Boeker, G. F., Sternlicht, B., 'Investigation of

Trans 1a tory Fluid Whir 1 in Vertical Machines,'
Trans. ASME, Vol. 78, 1956.

1.31 Lund, J. W., 'The Stability of an Elastic Rotor in

Journal Bearing with Flexible Damped Supports,'
Trans. ASME, Journal of Basic Engineering, Vol. 87,
Series E, 1965.

1.32 Prohl, M. A., 'A General Method for Calculating

Critical Speeds of Flexible Rotors,' Trans. ASME,
Vol. 67, Journal of Applied Mechanics, Vol. 12, p.
A-142, 1946.

1.33 Hagg, A. C., 'Some Vibration Aspects of Lubri-

cation, ' Lubrication Engineering, pp. 166-169,
August 1948.

1.34 Hagg, A. C., Sankey, G. 0., 'Elastic and Damping

Properties of Oil Film Journal Bearings for Appli-
cation to Unbalance Vibration Calculations,' Trans.
ASME, Journal of Applied Mechanics, Vol. 25, p.
141. 1958.

1.35 Sternlicht, B., 'Elastic and Damping Properties of

Cylindrical Journal Bearings,' Trans. ASam, Journal
of Applied Mechanics, Series D., Vol. 81, p. 101,
1959.

1.36 Lund, J. W., Sternlicht, B., 'Rotor-Bearing Dynamics

with Emphasis on Attenuation,' Trans. ASME, Journal
of Basic Engineering, Series D, 1962.

1.37 Linn, F. C., 'Method of and Means for Balancing

Rotors,' u.s. Patent No. 1,776,125, September 16,
1930.

1.38 Thearle, E. L., 'Dynamic Balancing of Rotating

Machinery in the Field,' General Electric Company,
APM-56-19, Schenectady, NY, 1935.
Introductioq 13

1.39 Grobe!, L. P., 'Balancing Turbine-Generator Rotors,'

General Electric Review, Vol. 56, No. 4, P. 22,
1953.

1.40 Goodman, T. P., 'A Least-Squares .Method for

Computing Balancing Corrections,' ASME Publication
No. 63-WA-295, September 1964.

1.41 Bishop, R. E. D., Gladwell, G• .M. L., 'The Vibration

·and Balancing of an Unbalanced Flexible Rotor,'
Journal of Mechanical Engineering Science, Vol. 1,
No. 1, p. 66, 1959.

1.42 Lund, J. W., Computer Program, See: Rieger, N. F.,

'Computer Program for Balancing of Flexible Rotors,'
Mechanical Technology Incorporated Report 67TR68,
September 1967.
 Table 1.1 Books on Rotordynamics
I~
REFERENCE AUTHOR l'UBLISBER 'l'OPIC_S EMPBASIZED EDITION ~

1.1 Dimentberg Per gammon Classical shaft dynamics for turbine 1 1961
generator applications.

1.2 Tondl Publishing Instability problems of rotors in bearings. 1 1965

House CSSR

1.3 WP/MTI WPAFB/AFAPL Rotor-bearing dynamics technology. 1 1965

1.4 Wilcox Pitman Balancing of rotors. 1 1966

1.5 Gunter NASA Instability of simple rotors. 1 1966

1.6 Smith Pitman Bearing analysis. 1 1970

1.7 Eshleman. Shapiro. ASME Rotor-bearing dynamics literature. 1 1973

Rumbarger. Rieger

1.8 WP/SRI WPAFB/APAFL Rotor-bearing dynamics technology update. 1 1973

1.9 Federn Springer Balancing of rigid rotors. 1 1978

1.10 Rieger V.I. Vibrations of rotating machinery. 2 1982

1.11 Rao Wiley- Dynamics of rotors. 1 1983

Eastern •z
>-r:t
1.12 Rieger NRL Balancing of rigid and flexible rotors. 1 1987 I~
....
(1)

1.13 Kellenberger Springer Rotordynamics. 1 1987 I~

Introduction 15

Whirl
Motion
Figure 1 Circular Orbit

Figure 2 Elliptical Orbit

Figure 3 Non-Synchronous Whirl Half-

Frequency Bearing Excitation
16 N.F. Rieger

Figure 4 Non-Synchronous Whirl Bearing

Rub or Electrical Excitation
Determines Number of Cusps

New Orbit

Original
Orbit
Figure 5 Transient Growth to New Stable Orbit

Impulse Orbi't
Growth

Figure 6 Transient Whirl Decay to Stable

Orbit Following Impulse
 CHAPTER 2

BEARING PROPERTIES

E. Krimer
Technische Hochschule Darmstadt, Darmstadt, FRG

ABSTRACT

The fundamental analysis of hydrodynamic lubricated

journal bearings is presented. It is shown how displace-
ments at static load, stiffness- and damping coefficients
may be computed. The formulas of the short circular
bearing are presented. Finally some remarks on special
bearing types are made.

Symbols
F Journal load
F Static journal load
s
L Bearing length
R Bore radius
s Duty parameter
So Sommerfeld number
dik Uamping coefficient
e lccentricity
h Film thickness
h0 Film thickness at centric journal
18 E. Kramer

kik Stiffness coefficient

p Film pressure
p Mean bearing pressure
r Journal radius
t Tir.1e
x1, x2 Displacements
~ Angular velocity of journal
~ Attitude angle
8 Width ratio
Bik Uamping coefficient, dimensionless
y Angle between eccentricity and X-axis
yik Stiffness coefficient, dimensionless
6 Radial clearance
£ Eccentricity ratio
n Dynamic viscosity
n Mean dynamic viscosity
~ Angle of resulting film force

Rotors are supported by two or more radial bearings~ These

are mounted in bushings carried by pedestals. Two types of
bearings are used, journal bearings and ball bearings. Journal
bearings are lubricated by liquid or gas, normally oil is used.
The flexural vibrations of rotors essentially are influenced
by the dynamic properties of the bearin~s. This chapter deals
with the main results of the theory of oil lubricated journal
bearings.
A review for gas bearings is given in /1/. Sor.e results for
ball bearings are published in /2, ... ,8/.

2.1 Reynolds Equation and Resu1tin~ Pressure Force

We consider the syster.1 journal, lul::ricant filr.1, bushing. The
journal rotates and it is loaded by static and dynamic forces,
leading to appropriate motion. This behaviour may be studied by
Bearing Properties 19

measurements or by cor.1putations. In the follO\:ing the basic ana-

lytical solution is presented. More details are given in /9/,
/10/, /11/ among others.
The system is shown in figure 2.1. The bore of the bushing is
assumed cylindrical with arbitrary, but nearly circular cross-
section with radius R. ~e assume a fixed coordinate system X, Y,
Z with the origin c8. The journal is cylindrical (radius r,
center CJ) rotating with an~ular velocity r.. It is assumed, that
it moves only translatory in X- and Y-direction, that means the
axis remains parallel to Z-direction. The position of the journal
y y

/ ' /

~L__j

Fig. 2.1 Geometry of a journal bearin9.

c8 Bearing center, CJ Journal center
center CJ is determined by the eccentricity e and the angle y. In
case of centric position (e=O) the film thickness is denoted by
h0 which is generally h 0 =h 0 (~). With e<<r in case of arbitrary
journal position the film thickness is approximately

(2.1)

If the journal moves transversally, e=e(t) and y=y(t), (2.1) be-

comes
h(lp,t) = h0 (\4>)-e(t) cosl\p-y(t)], (2.2)
which is called "gap function".
20 E. Kramer

Furthermore the following additional assu~ptions are made:

1. The lubricant is massless, incompressible and sticking at
the walls.
2. The lubricant is Newtonian.
3. Flow is laminar.
4. The lubricant pressure is constant in radial direction.
5. Flow velocities in radial direction are neglected.
6. Velocity gradients in radial direction are lar~e compared
to those in tangential and axial direction.
7. The film thickness is small compared to journal radius.
8. The curvature of lubrication gap is neglected, i.e. R=~.
9. The surfaces of journal and bore are smooth and rigid.

According to assumption 8 the lubricant gap now is shaped as

shown in figure 2.2, where
x, y, z coordinates
u, v, w flow velocities
U=Qr circumferential velocity of journal.

Fig. 2.2 Cross-section as

basis for theory

z
w

For a small fluid element of sides dx, dy, dz the condition of

force equilibrium yields
-
a-rxy = 2£ and a-rzy = 1e (2.3)
ay ax ay az
Bearing Properties 21

\\'i th p pressure of fluid

Txy' Tzy shearing stresses.

According to assumption 2 the shearing stresses are

and (2.4)

where n denotes the dynamic viscosity.

Equations (2.3) and (2.4) result

(2.5)

Integration of (2.5) yields

(2.6)

and finally with the boundary conditions

u=O, w=O at y=O (fixed bushing)

and u=U, w=O at y=h:

1
u = -2n -(y
ap 2 u
ax -hy) +-hy
(2. 7)
1ap 2-hy)
w =- -(y
2n az

Fig. 2.3 Continuity of flow

in a column
22 E. Kramer

Furthermore the condition for continuity of flow is needed.

For this we consider a column of fluid of height h and base dx·dz
(figure 2.3).
With the flow rates Q~, Q~ in x- and z-direction the condi-
tion of continuity is
aQ• aQ•
axxdxdzdt + azzdzdxdt + ~tdzdx = 0

respectively
aQ~ aQ~ ah
-+-+-
ax az at = 0 (2.8)

Uith (2.7) the flow rates are

h -t~ 3 a 1
Q1 = Judy = ---- ~+-Uh
x 0 12n ax 2
(2.9)
h -h3 a
Q• = Jwdy = ~ iP
z 0 !~n az

and (2.8) leads to

a h3 ~ + a h3 ap = l}.ah + E!!
ax(n ax) az(n a-z) 6 ax 12 at (2.10)

which is called Reynolds equation.

Now we return to the original cross-section of figure 2.1.
With x=fR and (2.2) the needed derivatives on the right hand side
of (2.10) are

With U=nR,assuming constant viscosity n=n and neglecting

the small difference between rand R the following form of
the Reynolds equation is obtained
Bearing Properties 23

The solution of (2.11) is the film pressure p(,, z, t). It de-

pends on one hand upon the viscosity n and the angular velocity n,
on the other hand upon the position e,) and the velocities e and
y of the journal center. In case of noncircular bore the term
ah/cnp+ 0, thus it generates filrr. pressure.
To obtain the resulting force firstly the pressure is inte-
grated in axial direction:
+L/2
dF = [ f P('f),Z, t)dz]Rd\1) , (2.12)
-L/2
where dF is the force on a small area LRd\1) at arbitrary angle
with the components
dF 1 = dF COS\1) dF 2 = dF s i mp (2.13)

A further integration in circumferential direction yields to the

components of the resultant force F:
21T 21T
F1 = f dF 1 F2 = f dF 2 l2.14}
IP=O IP=O
The resultant is
(2.15}

Its angle to X-coordinate (figure 2.4} is given by

F2
tan.S= F (2.16}
1
y

X
Fig. 2.4 Film forces dF, F and
its components
24 E. Kramer

In the following we consider the journal load \'~hich is equal

to the film force by amount and angle.

2.2 Static Load

Consider a horizontal rotor which is only loaded by its

weight. The components of the journal load are

where F5 is the weight force. Here the journal center is located

in the fourth quadrant of the X, Y sys tern (figure 2. 5). ~!i th F1,
F11 as components of~ in the x•, v• system the attitude angle a
is given by
-FII
tan a= -F- (2.17)
1

Fig. 2.5 Location of journal

center in case of
static load

Conventionally the Sommerfeld number is used as a dimension-

less form of the static journal force, i.e.

So= pi
nn
(2.18)

where o = R-r radial clearance

ljJ = 0/r clearance ratio (2.19)
e: = e/o eccentricity ratio
F
- s mean pressure .
P =M
Bearing Properties 25

The Sommerfeld nuober and the attitude angle are functions of

eccentricity ratio: So(£), a(£). For example see (2.45), (2.43) in
chapter 2.4.2. Hence the location £,a of journal center is a func-
tion of Sommerfeld number. The example given by figure 2.8 is
typical: the plot is nearly semicircular. Sometimes it is called
displacement orbit, since it describes the journal position at
different angular speeds (C 0 at n=O, c8 at n=oo).
The Sommerfeld number sometimes is substituted by its reci-
procal value. D.M. Smith introduced in /9/ the duty parameter,
which is defined under using the above symbols by

2.3 Stiffness and Damping

Initially static load as considered in chapter 2.2 is assumed.

In case of small displacements x1, x2 and small velocities x1, x2
(figure 2.6) the com~onents of the journal force are in first
approximation

(2.21)

Fig. 2.6 Displacements x1, x2

of journal center from
X static position
x,
26 E. Kramer

The additional forces are stiffness forces

Fs1 = kllx1 +k12x2

}(2.22)
Fs2 = k21 x1 + k22x2
aF.
with stiffness coefficients k = -1 (2.24)
i k axk
and damping forces

Fo1 = dnx1 + d12><2

}(2.25)
Fo2 = d21x.1 + d22><2
aF.1
with damping coefficients dik = aS<k (2.26)

The coefficients may be expressed by

and ( 2. 27)

where the dimensionless terms yik' Sik depend on the bearing para-
meters and are functions of eccentricity ratio £. For the short
circular bearing they are derived in chapter 2.4.3.

2.4 Short Bearing Theory

Many papers are published on the solution of the Reynolds

equation and the estimation of stiffness and damping coefficients.
Exemplary see /12, 13/, a review is given by J.W. Lund in /14/.
The computations are somewhat extensive and normally only numeri-
cal solutions are possible. For bearings with circular bore G.B.
Dubois and F.W. Ocvirk showed in /15/, that in case of short bea-
rings, L<<R, the derivate of film pressure in circumferential
direction may be neglected. This means that in (L.11) the first
term may be set to zero since the term h3a2p;az 2 dominates. The ~am­
putation of stiffness and damping coefficients by this theory is
published in /15, 9, 10/ and is summarized in the following.
Bearing Properties 27

2.4.1 Equation of Film Pressure and of Resulting Force

With circular bore the gap h0 is constant and equal to the

clearance
h0 = R-r = S,
whereby (2.2) becomes

h('P,t) = .5-e(t)cos('P-y(t)] (2.28)

Neglecting the first term in (2.11) leads with 3h 0 /3~=0 to

2 -
dp 2= ~[e( r.-2·() sin {\p-y) -2e cos (~y )1 (2.29)
dz h
by integration to

and by the conditions 3p/3z=O at z=O and p=O at z=~L/2 to the

film pressure function

(2.30)
Integration within the limits -L/2~z;+L/2 gives with (2.28) the
film load per unit length
. .
-L3_., e:(1-2*) sin {\p-y)-2Ticos{\p-y) - 3
q ('P· t) =- Tl I' _ ___;•;;....•- - - - - - , - - - =-~ q'('P,t). (2.31}
~ [ 1-e:cos {\p-y) 1 26

In static case, ~=0, s=O, the load becomes

= - 30
nl :::sin(y)--:)
q('P} - 252 [1-e:cos{\p-y)] 3 ( 2' 32 )
This theoretical load (2.32) is plotted in figure 2.7, showing
regions with positive and negative loads. In reality negative
loads are impossible and therefore they will 'be assumed as zero.
28 E. Kramer

Uith this assumption and with (2.12), (2.13}, (2.14) and (2.31)
the components of the resulting force are
'( y
F1 = -K : q'(4),t)costp14), F2 = -K f q'(4),t}sirA4kf'P (2.33)
--:-+y -!+'(
- 3
with
-;-r
K - ~L R~ (2.34)

Fig. 2.7 Theoretical film

load of short
bearing

The equations (2.33) may be integrated in closed form as shown in

/11/. Uith the dimensionless force
F. 2 1
- i --
F m 1 Ln~ l .· 1
. = 1 • 2 • I • II • (2.35)
where
L
B= ~ is the width ratio (2.36)
the components are

F1 = FIco s·r - FI I s i n·r

F2 = Fisiny+FIIcosy
• 2£ 2 E 1+2£ 2
where FI = ( 1-2~) 2 2 +T ~ 2 S/2
(1-E) -~ (1-E)
(2.38)

are the components with respect to the X', Y' system. The latter
may be splitted in a static part which depends on the eccentrici-
ty ratio E and in a dynamic part which is a function of the velo-
cities ~.~ and also of E:
 Bearing Properties 29

FI ( e: '~ ,() = Fi (e:) + Fl (e:' ~ 'y)

(2.39)
FII(e:,~,y) =FiJ(e:)+FII(e:,~,y).

2.4.2 Static Load

The static case is determined by

2
F - F' - 2e: F F' TI E (2.40)
I- I- ( 1-e:2)2 II= II =-"2" ( 1-e:2)3/2
Assuming the same force as in chapter 2.2 with the components

f.' 1 = o, F' 2 = -Fs

leads with (2.35), (2.18) and (2.36) to
- So (2.41)
Fs=8"2"
The angle y is determined with the first equation (2.37) and F1=0
by
0 = Fi cosy - Fi 1s i ny
and with (2.40) by
4 e: (2.42)
tany = - 1T 2 1/2
(1-e: )
Finally the attitude angle a results with (2.17) and (2.40) from
TI ( 1-e:2)1/2
tana = 4 e: (2.43)
In our case the second equation (2.37) is
- = -rs
F ~ So = r~. s1ny+
=--::-z- . F-'IIcosy , (2.44)
2 1
B
whereby siny and cosy may be expressed by (2.42). This leads to the
following expression for the Sommerfeld number

So = ~62 e: 2 2J1+[ (~) 2-1]e:2' (2.45)

(1-e: )
Notice that the Sommerfeld number is proportional to the squared
width ratio.
30 E. Kramer

0.1
0 goo

0.2
E
0,4
j 0,6

0.8

Fig. 2.8 Displacement orbit of short bearing

The displacement orbit of journal center is determined by (2.45),

(2.43) and plotted in figure 2.8. For numerical evaluations
So/S 2(E) and a(E) are plotted in figure 2.9.

100

lgo· 10
ex So
60" pr1

30" 0.1

o· 0.01

0 0,2 0,4 0,6 0,8

E--

Fig. 2.9 Sommerfeld number and attitude angle of short circular

bearing
Bearing Properties 31

2.4.3 Stiffness and Da~ping Coefficients_

Stiffness coefficients are derivatives of the force components

with respect to the displacements x1 and x2, (2.24). In static
case the components are given by (2.37)

F1 = K(Ficosy- Fbsiny)
} (2.46)
F2 = K(Fisiny+ Fbcosy) ,

considering (2.34), (2.35) and (2.40). F1 and F2 are functions of

£ and y, hence

(2.47)

Fig. 2.10 Coordinates of

sma 11 displace-
ment dx of journal center

Referring to figure 2.10

.2£
ax = -kos
o Y
1
(2.48)
ay = -s1ny
-1 . ay _ 1
-ax
1 e , ax- ecosy
2

leads finally to
82 Fs Fs
kik = 'SOfik(£) 6 = Y;kT
with
82
Yi k = So f i k (£ ) (2.49)
32 E. Kramer

and
2s
--2""'2 [1 +a ( s) 1
( 1- s )

(2.50)
1 . 4 2 2,
f 21 (s) =--z 2 S/2"1+[(-::-) a(s)+21E;
(1-s )

2E , 4 2
f 22 (s) = - -,..--,.;
2 3 ·_1+[(-=) a(c.)+2]t. 2}
(1-E )
where
1
a ( s) = --4---=-2-~2
1+[(;:) -1]t:
Figure 2.11 shows the dimensionless stiffness coefficients yik as
a function of So;e 2 or s.

t~o~~----------~----­
l;k
~+---~~~----~~~~

0.01 0.1 10 50 100

-;;z-
~
0.05 0.2 0 L 0.6 0.8 0.9
E-

Fig. 2.11 Dimensionless stiffness coefficients of short bearing

Damping coefficients are derivatives of the force components

with respect to the velocities x1, x2 (2.26). The derivatives of
the static terms Fi, Fir are zero, therefore the components F1 ,
F2 are given by (2.46) if Fi, Fir are replaced by the dynamic
terms FI, Fir (see (2.39), (2.38)). The derivatives are given by
Bearing Properties 33

aF .
_1 =
aF. _oe:_
_l
~• aF. ~ •
+ _ l 2L
axk a£ axk ay axk
which leads in similar way as before to the damping coefficients

with
(2.51)
and
TI 4 2 2
2 3/2{1-[2(n) -3]e: a(e:)}
(1-e: )

4e: 4 2 2
g12(e:) = g21 (e:) =- 2 2{1-[2(:rr) -3]e: a(e:)} (2.52)
(1-e: )

TI 4 2 2+e: 2 2
g22(e:) = 2 3/2[ 1+(1T) :-ze: a(e:)]
(1-e: ) 1-e:

with a(e:) from (2.50).

The dimensionless damping coefficients are shown in figure 2.12.

t~o+-~~--~---+--~~
~ik
IDf---~-~~~~--J--

- ~12 = -f3z, ~,
0.1+---~--~---+-----1-

1 I iililhj I ilillllj I iliiiiij ; iliidlj I I

0.01 0.1 1 ~So 100

jj!-
0.05 0.2 0.1. 0.6 OJ 0.9
E-

Fig. 2.12 Di~ensionless damping coefficients of short bearing

34 E. Kramer

2.4.4 Journal Orbit as a Result of Rotatina Force

We consider a static load Fs and an additional load Fa which

rotates slowly in XY-plane. The angular velocity of Fa is assumed
to be small enough to neglect dynamic forces. For Fa<<Fs the
displacement of journal center is determined by inversion of (2.22)

x1 = hll FS 1 + h12 FS 2
} ( 2. 53 I
x2 = h21 FS1+h22 FS2
-1
with [hik 1 = [kik 1

The components of the additional force are

hence the orbit of CJ is elliptical.

As mentioned before equation (2.53) is limited to small forces
Fa. Also for arbitrary large forces Fa a simple determination of
the journal orbit is possible as explained in the following. The
relation between static force and eccentricity is given by (2.45),
and the attitude angle by (2.43). Therefore the position of CJ is
given in case of Fa=O by figure 2.13a and in case of Fa>O,wt>O by
figure 2.13b. The orbit is determined by O~wt<2n.
y y
a b

X X

Fig. 2.13 Determination of orbit for arbitrary Fa/Fs

Figure 2.14 shows orbits for Fa=0,5 Fs at different eccentri-
cities (e=0,3; 0,6; 0,9). The solid lines show the exact solu-
Bearing Properties 35

tions, the broken lines show the linearized one. The arrows indi-
cate the direction of Fa. It may be stated that up to F~0,5 Fs
the linear solution is rather good.

Ca X

Limit Circle
Co

Fig. 2.14 Journal orbits in case of slowly rotating additional

force Fa= 0,5 Fs.
- - exact, ----- 1inearized

The figure shows that the orbit is nearly circular for small
eccentricities and shaped like a banana for large eccentricities.
For small eccentricities the bearing behaves nearly isotropic,
but the displacements are nearly vertical to the additional force.
At high eccentricities the bearing is stiff in radial direction
and rather elastic in tangential direction. Roughly the stiffness
is in the order of Fs/o.

2.5 Properties of Special Bearing Types

The theory given in chapters 2.1 to 2.3 is not only valid for
circular bore but also for other cross-sections. Furthermore par-
tial fil~ areas are possible. Special bearing types with different
geometry and also with flexible pads are used with respect to
36 E. Kramer

stability, load capacity, fricton losses and other demands. Some

types are plotted schematically in figures 2.15 a ... g.

a ~

Fig. 2.15 Some special bearing types

a Short arc bearing
b Lemon bearing
c Pocket bearing
d Three-lobed bearing
e Special three-lobed bearing
f Tilting-pad bearing
g Tilting-pad bearing for vertical shaft
Bearing Properties 37

For bearings with arbitrary cross section and cylindrical

bore T. Someya gives in 1131 a numerical solution for the static
displacement orbit. J. Glienicke describes in 116.17 I results of
comprehensive measurements for evaluation of stiffness and dam-
ping coefficients for several bearing types. The displacement
orbit of a special three-pad-bearing is published by H.H. Ott 1181.
Tilting-pad-bearings are investigated by Z.E. Varga 119/, H.J.
~1erker 120/ and H. Springer 1211. In 1221 E. Pollmann considers

the influence of variable viscosity. These references are only

some of the large number of published papers. More references are
given in I 141.

References

111 Chandra, M., M. Malik and R. Sinhasan: Gas Bearings. Part 1:

Dynamic Analysis and Solution Method. Wear 88 (3) pp 255-268.

121 Brandlein, J.: Die radiale Federsteifigkeit von Walzlagern:

Linearisierung der Kennlinien. Masch.-Markt 83 (1977) Nr. 61,
s. 1182/84.

131 Klumpers, K.: Untersuchung des Dampfungsverhaltens von Walz-

lagern. FVA-Informationstagung Bad Homburg, Nov. 1977, 32 S.
Forsch.vereinig. Antriebstechnik e.V., Cornelfusstr. 4,
6000 Frankfurt/M.

141 Klepzig, W.: Experimentelle Ermittlung der Dampfungskonstante

eines doppelreihigen Zylinderrollenlagers. Masch.-Bau-Techn.
26 ( 1977) Nr. 3, S. 112/15.

151 Voll, H.: Walzgelagerte Spindeleinheiten: Einflu8 der Lager-

gro8e auf die statische Steifigkeit; Masch.-Markt Bd. 86 (1980)
Nr. 29, S. 572/73.

161 Tamura, H., Tsuda, Y.: On the Spring Characteristics of a Ball

Bearing. Bulletin of JSME, Vol. 23, No. 180, 1980.
38 E. Kramer

/7/ Week, M. u. L. Ophey: Experimentelle Ermittlung der Steifig-

keit und D~mpfung radial belasteter W~lzlager. Ind. Anz.
Bd. 103 (1981} Nr. 79, S. 32/35.

/8/ Fukata, S.,.E.H. Gad, T. Kondou, T. Ayage: On the Radial

Vibration of Ball Bearings (Computer Simulation}. Bull. JSME
28 (239), pp 899-904 (1985).

/9/ Smith, D.M.: Journal Bearings in Turbomachinery, Great Britain:

Chapman and Hall, 1969.

/10/ Cameron, A.: Basic Lubrication Theory, London: Longman, 1971.

/11/ Lang, O.R., Steinhilp~r, W.: Gleitlager. Berlin, Heidelberg,

New York: Springer 1978.

/12/ Someya, T.: Stabilit~t einer in zylindrischen Gleitlagern

laufenden, unwuchtfreien Welle. Diss. TU Karlsruhe, 1962.

/13/ Someya, T.: Das dynamisch belastete Radial-Gleitlager belie-

bigen Querschnitts. Ing.-Arch. Band 34, Heft 1, S. 7-16,
Berlin, Heidelberg, New York: Springer 1965.

/14/ Lund, J.W.: Evaluation of Stiffness and Damping Coefficients

for Fluid-Film Bearings, Shock and Vibration Digest, 1979.

/15/ Dubois, G.B., Ocvirk, F.W.: Analytical Derivation and Experi-

mental Evaluation of Short-Bearing Approximation for Full
Journal Bearings. Report 1157, Cornell University, 1953

/16/ Glienicke, J.: Feder- und D~mpfungskonstanten von Gleitlagern

fUr Turbomaschinen und deren EinfluB auf das Schwingungsver-
halten eines einfachen Rotors. Diss. TU Karlsruhe, 1966.

/17/ Glienicke, J.: Experimental Investigation of the Stiffness and

Damping Coefficients of Turbine Bearings and Their Application
to Instability Prediction. Symposium in Nottingham. London:
The Institution of Mechanical Engineers, 1966.
 Bearing Properties 39

/18/ Ott, H.H.: Berechnung von Wellenlage und Reibung im Dreikeil-

Traglager. Brown, Boveri Mitt. Bd. 46 (1959) Nr. 7, S. 395-406.

/19/ Varga, Z.E.: Wellenbewegung, Reibung und Uldurchsatz beim seg-

mentierten Radialgleitlager von beliebiger Spaltform unter kon-
stanter und zeitlich veranderlicher Belastung. Diss. ETH ZUrich,
1971.

/20/ ~1erker,H.-J.: Ober den nichtlinearen Einflul3 von Gleitlagern

auf die Schwingungen von Rotoren, Fortschrittsber. VOl Z,
Reihe 11, Nr. 40 (1981).

/21/ Springer, H.: Stiffness and Damping Characteristics of Tilting-

Pad Journal Bearings. CISM Courses and Lectures Nr. 273. Wien,
New York: Springer Verlag, 1984.

/22/ Pollmann, E.: Das Mehrgleitflachenlager unter BerUcksichtigung

der veranderlichen Ulviskositat. Konstruktion 21 (1969) H. 3,
s. 85-97.
 CHAPTER 3

ANALYSIS OF ROTORS IN BEARINGS

E. Krimer
Technische Hochschule Darmstadt, Darmstadt, FRG

ABSTRACT

Thi~ part begins with the main facts about the

5il .• ~.Jlt>st rotor model. the Jeffcott rotor. A short
cot.!l.ent follows about the approximate computation
of \.lie first critical speed. Then the start-up and
shut-down behaviour of a rotor is described. The
influence cf bearings also is studied using the
Jeffcott model. A simple model shows the influence
of uearing dar.,ping. In the last section the main
iiS!Jects of unbalance behaviour of multi-bearing
syste1.:s are discussed.

In general two kinds of rotors occur in practice. rotors

for turbomachines an~ those for reciprocating machines. Both
tyves vibrate in a rotatory and translatory manner. In this
cl;"pter translatory vibrations of turbomachine rotors are
considered.

The a i111 of the designer is to develop rotors which

operate safely during their whole lifetime (several decades).
and do so with moderate vibrations. Many investigations
about this problem have been published since the beginning of
the century. A short review of several results is presented
in the followin~.
42 E. Kramer

3.1 Principles of Rotordynamic Analysis

Real rotors are rathe;- complicated systems : their geometry is

not simple, mostly- they are supported static undetermined on more
than two bearings. The oil film, the pedestals and the foundation
may influence their dynamic behaviour. In spite of this many pheno
mena may be illustrated by the simple rotor as described in the
following chapter.

3. 1. 1 Jeffcott Rotor

The cross-section of a real rotor on two bearings is given in

figure 3.1. The vibrations of such a rotor may be studied with
good approximation by the mod~1 of figure 3. 2. Investigations abou
such a model, called Jeffcott rotor or Laval shaft, are given in
/1, .. . , 6/ and many other pu~lications.

Fig. 3.1 Example of a

real rotor

Fig. 3. 2 Jeffcott Rotor

Laval Shaft

The properties of this rotor are :

- massless shaft, symmetric to mid-span, stiffness k,
rigid circular disc, mass m, polar mass moment o,
rigid bearings,
viscous damping d of translatory disc displacement.
Analysis of Rotors in Bearings 43

W shaft center, displacements y1, y2

S mass center, displacements z1, z2 •
With eccentricity e and angle~

z1 = y1 + e cos~
z2 =y 2 +esin'9

hence the system has three degrees of freedom.

By Newtons law the equations of motion are

mz 1 = -ky 1 - dy 1
mi 2 =- ky 2 - dy 2 - mg (3.2}

04) = -(ky 1 + dy 1}e sin~- (ky 2 + dy 2 }e cos'Pt" MT

where MT denotes external torque.

Normally the radius of inertia i=y0/m is large compared to

eccentricity e, therefore
'P = nt (3.3}

in case of external torque MT=O is a good approximation. With

these assumptions and (3.1) the two equations

my 1 + dy 1 + ky 1 = men 2cosnt
my2 + dy 2 + ky 2 = men 2sinnt- mg

for the displacements of shaft center remain. They are uncoupled,

being the equations of a simple oscillator.
The homogenous equation

m y.1 + dy 1. + ky 1. = 0, i =1' 2 (3.5}

has the solution

(3.6}
44 E. Kramer

where ~d =vt.;;-o 2' (3.7) damped natural frequency

(3.8) undamped natural frequency

:. d quantity of damping
= 2m
constants, determined by the
initial conditions

By (3.6) the natural motion of the shaft is determined. This

means that the shaft center describes a decreasing elliptical orbit
(figure 3.3).

Fig. 3.3 Natural motion of

shaft center

The inhomogenous equations (3.4) are valid for unbalance exci-

tation of the weight-loaded rotor. Further external loads F1(t),
F2(t) acting on the disc in directions 1 and 2 may be considered
by adding them on the right hand side of (3.4). For the given
case the steady state solution of (3.4) is

y1 = yCOS (rlt-E)
y2 = ysin (rlt-E)-y 0
where y = en 2£(1-n 2 ) 2 + (2Dn) 2 ( 112 = eV

tan E = ~ (3.10)
1-n

static displacement
Analysis of Rotors in Bearings 45

n =-
n
wk
(3.11)
D = _d_ damping factor.
2{km'
According to (3.9) for constant angular velocity r. the shaft
center describes a circular orbit with center (0, -y 0 ) and with
constant angle E (figure 3.4). The relative amplitude y/e is given
by the wellknown amplification function V [(3.10), figure 3.5)].
It shows that the amplitude y is maximum at 0-::::wk. The correspon-
ding speed
n
cr = ncr /2rr::::wk/2rr (3.12)

is called critical speed.

t
Y;
Fig. 3.4 Orbit of shaft
center in case
of unbalance excitation

y, ....

Further information is given by the polar plot 3.6, which shows

the position of shaft and mass center W, S at different relative
velocities n=r2/wk and at t=O. In case of an undamped rotor two configu-
rations are possible (figure 3.7), the mass center is outside (D<wk)
or it is inside (r2>wk) of the orbit. At n=oo the mass center coincides
with W0 , which is the static position of shaft center W (self-
centering).
46 E. Kramer

-H+-H--- 0: -----
-- 0
-0.1
,0,15
0.2
- 0.25 ---------
0.3
Fig. 3.5
Amplification function

o1-Ld::::::::=±=~r
0 2 3
..,

Fig. 3.6
Position of Wand S for
different angular velo-
cities at t=O

Yz Yz Yz

y,

w Fig. 3.7
Undamped rotor.
Qcwk Q>wk Q:oo Possible configurations
 Analysis of Rotors in Bearings 47

3.1.2 Real Rotors with two Bearings

The first critical speed of real rotors, as shown for instance in

figure 3.1, may be computed without using a computer accurately
enough in the following manner.
As first approximation the critical speed for the Jeffcott rotor
is according to (3.12)

and by (3.10)

therefore ::::J_.r£
n
cr 2n v-y: (3.13)

where g is the constant of gravity and y0 the static displacement

at mid-span, caused by rotor weight mg. The first natural frequen-
cy of a uniform bar supported in the same manner as the Jeffcott
rotor is
1 . {ff'
f 1 = 1,57 2 v~
£, p

and its static displacement is

_ 5 pA£. 4g
Yo - 384 EI
which leads with ncr :::: f 1 finally to

1,13. ~
n
cr
:::
2n v-y: (3.14)

The geometry of real rotors lies between these two models. There-
fore their first critical speed is limited by

0 159'/I" < ncr 0 179'/I" (3.15)

' Vy~0 <
' Vy~
0

If y0 is known, ncr may be estimated accurately enough by (3.15).

 48 E. Kramer

Another approximation is given by the equation T+U=constant for

undamped systems, which leads to

U - T (3.16)
max max
Assuming the·first natural mode to be approximately equal to
the static mode, given by displacements yi, then the maximum
potential energy

where mi are masses of the rotor concentrated at points i. The

kinetic energy
Tmax = -21~m.y.
•2 = ~
1 1
12 2
2 n~m.y.
1 1

where wn is the unknown natur3l frequency. Equation (3.16) leads to

2
2 ~m.y.
w = ~ , where y• = ~ (3.17)
n y0 o miyi

and with wn::::2Tincr and (3.13) finally to

ncr :::: '2\ {f- ·

0
(3.18)

Usage of this equation needs only the static displacements yi

caused by the loads mig at several points along the rotor axis.
This procedure is also relevant to rotors with overhang, if the
loads at overhang are assumed opposite to those between the bea-
rings.

3.1.3 Acceleration through Critical Speed

In chapter 3.1.1 constant angular velocity of the rotor is

assumed. During start-up and shut-down the angular velocity increa-
ses or decreases with time. Investigations about this problem are
published in /7, ... , 12/ and in further papers. In the following
the main results for simple models are described.
Analysis of Rotors in Bearings 49

The typical behaviour during acceleration is shown in figure

3.8, which was computed ace. to /8/ by J. Greb in /13/. It shows
the displacement as function of time for a damped single degree of
freedom oscillator which is excited by the force
,.. 1 2 ,..
F( t) = F sin Cz o.t ) = F s i n [w( t ) t ] (3.19)

where w(t) =io.t, which means that the exciting frequency grows
linearly. The figure shows that the envelope of amplitudes is
slightly undulated.

1 o=o.o1; t=o.o161

t 10
.!.
Jr"'

xs / 1' 1 1\1"~

4f(
5 ,'

v\11.01v
0 7'
v 0.5' 1. 2.0
t
~
\i T-
-5

-10

Fig. 3.8 uisplacements of an oscillator, exited harmonically

with linear increasing frequency.
-F
xs = k static displacement
o.t dimensionless time
1 =-
wk
D damping factor
Lt
I,
- wk2 relative acceleration

ulk ~Vf natural frequency

A Jeffcott rotor behaves during start-up similar to an accele-

rated oscillator. Figure 3.9 shows for example the envelope of am-
plitudes of shaft center for different values of acceleration.
50 E. Kramer

With increasing acceleration the peaks decrease and are shifted to

later times. During shut-down (figure 3.10) the peaks decrease
with higher decelerations, they are also shifted to later times
(shut-down:T=10~0).

The figures show that the effect begins to be appreciable at

lsi> 5·10- 4 . To make this limit clear we define a starting time
as follows. With T=o.t/wk and s =o./w~ the real time is

-e I D= o.nl
I
10 , .. ·- j

1.5 2

Fig. 3.9 Amplitudes of Jeffcott rotor during start-up

120
y
e
1 o=o.o2 1

10

0 0,5 15 2
T---

Fig. 3.10 Amplitudes of Jeffcott rotor during shut-down

Analysis of Rotors in Bearings 51

For small accelerations the main peak is reached at T~1, as the

figures show. Therefore T=1 may determine the starting time

ts = t(T=1) = --1-- (3.20)

~;;wk

By this definition the acceleration

which is plotted in figure 3.11, where ncr=wk/2n is the critical

speed. The figure shows that lsi> 5·10- 4 means extremely short
starting or stopping times.

I w-'
~ 10- 3 Fig. 3.11
5·10- 4 Relative acceleration as
function of starting time
and critical speed

0,1 10 100 1000 5

ts---

3.2 Theory of Rotor Bearing Systems

In chapter 3.1 only simple rotors supported on two rigid bea-

rings are considered. In reality rotors are more complicated. They
have more than two bearings, pedestals and a foundation. Sometimes
the base also should be mentioned. This means that the two rigid
bearings ~ust be replaced by a more or less complicated dynamic
system. This will be done step by step in the following chapters.

3.2.1 Jeffcott Rotor with Flexible Bearings

Firstly the model of figure 3.2 will be replaced by that of

figure 3.12 which has two flexible bearings with stiffness kB 1 ' kB 2
in horizontal and vertical direction. Denoting the stiffness of
52 E. Kramer

Fig. 3.12 Jeffcott Rotor with flexible bearings

rigidly supported rotor by kR the new system has the stiffnesses
2kRkBi
ki = . -kR.. .;_+;.,.;2k;;.. ;B-i i = 1, 2 (3.21)

The equations of motion may be obtained by substituting k1 for

kin the first and k2 fork in the second equation of (3.4). The
system has two undamped natural frequencies

and two damped natural frequencies

wdi = Vw~-o2 = 1, 2 (3.22)

where o=d/2m ace. to (3.7).
The particular solutions of the inhomogenous equations are
y1 = y1 cos (nt-£ 1) }
(3.23)
Y2 = Y2 sin (nt-£2)-yo2

where
- = en.[(1-n.)
y. 2 2 2 + (2D.n.) 21-1/2
1 1 1 1 1

2D.n.
1 1
tan £i = -:----2
1-n·1

Yo2 = i:2 (3.24)

n D.= Nd , i = 1, 2 .
ni = W.
1 1 2 k .m
1
Analysis of Rotors in Bearings 53

2
Fig. 3.13 Orbit of shaft center
in case of unbalance
excitation.
Rotor with unequal
bearing stiffnesses

l"kwdl
forward whirl {
n=wdl (~ first critical speed)

threshold speed

backward whirl

threshold speed

r _f)_
1~
Q•wd2 ,. second critical speed)

forward whirl
'+'- n"wd2

Fig. 3.14 Orbits of a rotor with unequal bearing stiffnesses at

different angular velocities
54 E. Kramer

The equations (3.23) describe the orbit of shaft center at un-

balance excitation (figure 3.13). It is an ellipse (circle) if
k82 4 k81 (k 82 =k 81 ). The shaft center may whirl in the same direc-
tion as the rotor rotates (forward whirl) or in opposite direction
(backward whirl). rigure 3.14 shows the typical behaviour at diffe-
rent angular velocities. The rotor whirls forward at low and at
high velocities and it whirls backward in a region between wd 1 and
w
d2.

3.2.2 Re~arks about Bearing Damping

In the previous chapter external damping at the disc is assumed.

However rotors usually are damped at their bearings. To study this
we consider a system shown in figure 3.15. Only displacements in
one direction and damping acting only at the bearings are assumed.

f y. f(t)
Fig. 3.15 Rotormodel with
bearing damping

With the symbols of the figure the equations of motion are

mY+ kR(y-x) = f(t)

} (3.24)
kR(x-y) + 2(d 8x+k 8 x) = o
The homogenous system leads to the characteristic equation

(3.25)
Analysis of Rotors in Bearings

,\
where X --
WR

WR =fi
dB
dB - ykRm,

kB
K
= kR stiffness ratio.

Normally (3.25) has the roots (eigenvalues)

The third eigenvalue has no practical importance. It yields a monotonous

decreasing natural motion which disappears very quickly. The first
two eigenvalues stand for a damped natural vibration with related fre-
quency wi=wi/wR. This frequency is plotted in figure 3.16 as a func-
tion of bearing stiffness and with bearing damping as parameter. The
figure shows that at small bearing stiffnesses the influence of bea-
ring damping on the natural frequency is considerable. With increasing
bearing stiffness the curves approach to 1,0 (wi ~wR) and_ the damping
becomes unimportant.

1
0.5 0.75
0.5
0.' de
0

Fig. 3.16 Natural frequency of model figure 3.15

 56 E. Kramer

The negative real part of a complex eigenvalue ) k indicates how

much the corresponding natural vibration is damped . Therefore it may
be called natural damping. Accordingly the expression

(3.27)

is called related natural damping.

For our model the related natural damping is plotted in figure
3.17 as a function of bearing damping with parameter bearing stiffness .

o·
~ 05 ~
/ , ·' 075 ko
0.5-

0~~~~======~~
0 0.5 1.5 2 s 10 -
ds-

Fig. 3.17 Natural damring of model figure 3.15

The curves begin at zero, reach a maximum and end again towards
zero. The maximum indicates optimum bearing damping. However the
question is whether the optimum can be realized in practice. The in-
fluence of stiffness is important: large bearing stiffness compared
to rotor stiffness lowers the natural damping.
For unbalance excitation
f(t) = Re(men 2eint)
the steady state solution of (3.24) yields the harmonic vibration for
the rotor
Analysis of Rotors in Bearings 57

with the amplitude

(3.28)
[K(1-n 2 )-0,5n 212+[nd- 8 (1-n 2)1 2

where n=Q/wR is the related exciting frequency .

The amplitude-frequency plot is the well-known resonance curve
with maximum amplitude i n the neighbourhood of Q=wi · Figure 3.18
shows the related maximum Ymax/e as a function of bearing damping
with parameter bearing stiffness. For the assumed values and for
small damping
-
Ymax C
-e-::- (3.29)
dB
where C depends very significantly on the related bearing stiffness
(figure 3.19).

1000

e
10

''"'"I
0,01 ''"""0,1 '"""I1

de

Fig. 3. 18 Maximum amplitude of model 3.15

Remember that with a small bearing damping the natural damping

decreases with increasing bearing stiffness . Therefore the maximum
amplitude of the rotor may also be evaluated approximately by the
natural damping.
58 E. Kramer

100
c
10

0,1

0,1 10

fig. 3.19 Factor C [Eq. (3.29)] for small related bearing damping dB

3.3 Unbalance Response of Rotors in Bearings

In practice the most important problem in rotordynamics is unba-

lance response. In the previous chapters the principal behaviour
of the Jeffcott rotor for this kind of excitation is described. As
a further step the Jeffcott rotor with journal bearings may be
studied. This has already been done in many published papers. The
main results are shown in figure 3.20.

ie 3
i
e
,___
·:
:
. t
vertoco }"
honzonto t
1. Rotor
e ·
Fig. 3.20 Unbalance res-
1 ponse of Jeff-
cott rotor with journal
bearings {161)

1 3
Iw.l . . . -
Analysis of Rotors in Bearings 59

Theoretically the rotor has two critical speeds, because the oil
film behaves anisotropically. But usually the critical speeds differ
so little that only one resonant peak appears. The example in figure
3.20 shows two peaks only for the horizontal rotor vibration. This
problem is treated rather comprehensively by J. Glienicke in /14/.
Knowledge of unbalance behaviour of Jeffcott rotor is doubtless
very important. For real rotors however more complicated models are
necessary, which are characterized mainly by:
different cross-sections of rotor along its axis,
more than two bearings,
consideration of additional effects, above all the gyroscopic
effect,
realistic models for the pedestals and the foundation.

The computation of such models is rather complicated and to date

many efforts have been made to obtain a good and economical procedure.
The development began in 1944 with the publication of the fundamental
papers /15, 16/ of N.O. Myklestad and M.A. Prohl. Their method, called
transfer method, was first applied to undamped, rigidly supported
multi-bearing rotors. In the following years many researchers extended
it for more complicated models. Even today most industrial computer
programs are based on this method. For an introduction see the book
of E.C. Pestel and F.A. Leckie /17/. The user must know, that nume-
rical difficulties are possible, which are discussed in /18/.
By the transfer method displacements and forces of one end of
the system are transfered by matrices to the other end. The solution
is determined by the boundary conditions of the two ends. All inter-
mediate conditions are considered during the transfer procedure.
This method has the advantage that the number of unknowns is only
the sum of the considered displacements and forces at both ends of
the system.
Another way of computing was to extend the known procedures for
the statics of large systems to dynamic problems. This leads to a
large number of coupled equations, whose economical solution was
60 E. Kramer

possible only after fast computers were available. As unknown quan-

tities either the displacements or the forces could be chosen. Both
possibilities are investigated and compared by K. Marguerre and
H. Wolfel in /19/. Nowadays the first possibility, called displace-
ment method, is almost exclusively used in modern programs. The
main advantage of this method is its flexibility. The system matri-
ces are obtained in a simple manner from the matrices of the ele-
ments and of substructures. Therefore extensions or variations of
systems may easily be introduced.
For the displacement method the equation of motion in matrix
notation is
[MJO<l+[B]{)d+[S]{x} = {f{t)} (3.30}

where {x} is the vector of the unknown displacements (n displacements

and rotations: n-DOF system) and {f{t)} is the vector of the known
exciting forces or moments. The (n,n) system matrices [M], [B], [S]
represent the masses and mass moments, damping and gyroscopic pro-
perties and conservative or nonconservative stiffnesses. The solution
of (3.30) for most practical cases is well-known.
Let us look at some results concerning unbalance behaviour of
multi-bearing systems, computed by the displacement method. Figure
3.21 shows a schematic picture of a multibearing rotor with its pe-
destals and foundation. For such a system usually models are assumed

// / / /, •· 1 .' , I I I I

Fig. 3.21 Schematic picture of a turbo machine

Analysis of Rotors in Bearings 61

with several hundred degrees of freedom with the same number of

natural frequencies. Often 20 to 50 natural frequencies lie in the
range of operating speed. For example figure 3.22 shows the natural
frequencies of a five-bearing rotor with rigidly supported journal
bearings. Aside from the twelve natural frequencies of the rotor
the natural frequencies of foundation sometimes must be considered.
The natural frequencies of the rotor depend on the rotational speed,

5000
cpm
,,2
f 4000 ,,,
f.
3000 ,'()
,9
ta
2000

"'
f6 t
t~, 5
t3
1000
f ,2
1

0
0 2000 4000 rpm
n-
Fig. 3.22 Natural frequencies of a five-bearing rotor /20/

which is caused by the bearing properties. The influence is however

mostly moderate. The curves marked by f 1,f 3, ... belong to natural
vibrations whirling backward and those marked by f 2 ,f 4 ... belong to
natural vibrations whirling forward. The critical speeds are deter-
mined by the intersection points of the line f=n.
The natural modes generally are not plane and they vary perio-
dically with time. These special properties are however not very
significant and it is sufficient to consider the mode shapes
as plane and independent of time. Figure 3.23 shows the natural
modes of the example, simplified as stated.
For computations of unbalance vibrations the distribution of un-
balance along the rotor axis has to be assumed. In practice any kind
of distribution is possible. For cases similar to that of our example
62 E. Kramer

Order

9 ·~ ~ I I Ll1
Fig. 3.23 Natural modes of a

7.8 A I five-bearing rotor

5,6 rPI
3,4 1-L¥l
1·2 ~-I+-1-+--oofC-A----Jt

it may be adequate to assume only two kinds of unbalance distribu-

tions: unidirected and opposite directed between two bearings. The
unbalance may be constant distributed or concentrated to one or
two single unbalances. Figure 3.24 shows for example the amplitudes
of the shaft at bearing 5. In case G) only one important peak
appears at ::::660 rpm, while in case @ two rather high peaks are
produced in the neighbourhood of 2000 rpm.

I
y"
100

Jlm
@
6 zs zs ~
5

50 zs 1-!

0
0 1000 2000 3000
-- rpm
n-

Fig. 3.24 Shaft amplitude at bearing 5 in case of unbalances

in the generator rotor
Analysis of Rotors in Bearings 63

Finally some remarks about the influence of foundation on un-

balance response: Nowadays the rotor and the foundation are usually
computed separately. Then the behaviour of the combined system is
estimated either on the basis of experience with earlier machines
or by computational approximations. The computation of the combined
rotor-foundation system is time consuming, above all in the prepa-
ration of data. A result of combined computation is shown in figure
3.25. For comparison the behaviour of the rotor without foundation
(that means with rigidly supported journal bearings) is also plotted.
On the basis of such comparisons the influence of foundation may be
studied, making it possible for future similar cases to avoid the
time consuming combined computation.

1
A
40
1\
I 1 11,.
11 o or,r1g1
I I I~ bearings
R t
. "d

y pm I VI
I I
I I
20 I ~

0
0 1000 2000 3000 rpm
n-

Fig. 3.25 The influence of foundation on the result of figure 3.24 ~

References

/1/ Foppl, A.: Das Problem der Lavalschen Turbinenwelle, Der

Civiling. 1895, S. 335.

121 Foppl, A.: Ober den ruhigen Gang von. schnell umlaufenden Hange-
spindeln, Der Civiling. 1895, S. 335.
64 E. Kramer

/31 Jeffcott, H.H.: The Lateral Vibration of Loaded Shaft in the

Neighborhood of a Whirling Speed - the Effect of Want of
Balance, Philosophical Magazine, Series 6, Vol. 37, 1919.

/4/ Biezeno, C.B., Grammel, R.: Technische Dynamik, 2. Band,

Berlin, Gottingen, Heidelberg: Springer 1953.

/5/ Rao, J.S.: Rotordynamics, Wiley Eastern Ltd., 1983.

/6/ Kramer, E.: Maschinendynamik, Berlin, Heidelberg, New York:

Springer 1984.

/71 Poschl, Th.: Das Anlaufen eines einfachen Schwingers. lng.Arch.

IV. Bd. (1933) S. 98-102.

/8/ Henning, G., Schmidt, B., Wedlich, Th.: Erzwungene Schwingungen

beim Resonanzdurchgang. VDI-Berichte Nr. 113, 1967, S. 41-46.

/9/ Capello, A.: On the Acceleration of Rotors through their

Critical Speed. Meccanica, Journal of the ltalien Association
of Theoretical and Applied Mechanics, Vol. 2, Nr. 3, 1967.

/10/ Markert, R.: Resonanzdurchfahrt unwuchtiger biegeelastischer

Rotoren. VDI-Z Fortschrittsberichte Reihe 11, Nr. 34, 1980.

/11/ Hassenpflug, H.L., Flack, R.D., Gunter, E.J.: Influence of

Acceleration on the Critical Speed of a Jeffcott Rotor. Journal
of Engineering for Power, Vol. 103, S. 108-113, 1985.

/12/ Tsuchiya, K.: Passage of a Rotor through a Critical Speed.

Journal of Mechanical Design, Trans. of ASME, 104 (2),
s. 370-374, 1982.

/13/ Greb, J.: Berechnung des Hochlaufs und Ablaufs von Rotoren.
Studienarbeit am FG Maschinendynamik, TH Darmstadt, 1985.
 .
Analysis of Rotors in Bearings 65

/14/ Glienicke, J.: Experimental Investigation of the Stiffness and

Damping Coefficients of Turbine Bearings and Their Application
to Instability Prediction. Symposium in Nottingham. London:
The Institution of Mechanical Engineers, 1966.

/15/ Myklestad, N.O.: A new method of calculating natural modes of

uncoupled bending vibrations. J. Aeronaut. Sci. 11 (1944)
153-163.

/16/ Prahl, M.A.: A general method for calculating critical speeds

of flexible rotors. J. Appl. Mech., Trans. ASME, Series E 67
(1945) 142-146.

/17/ Pestel, E.C., Leckie, F.A.: Matrix methods in elastomechanics.

New York: McGraw-Hill 1963.

/18/ Marguerre, K., Uhrig, R.: Das Obertragungsverfahren und seine

Grenzen. Z. angew. Math. Mech. 44 (1964) 1-21.

/19/ Marguerre, K., H. Wolfel: Mechanics of structural systems-

Mechanics of vibration. Sijthoff &Noordhoff, Alphen an den
Rijn, The Netherlands, 1979.

/20/ Kramer, E.: Models for Computation of Turbomachine Vibrations.

ASME 85-DET-138, 1985.
 CHAPTER 4.1

PRINCIPLES OF BALANCING AND OF BALANCING MACHINES

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

The construction and operating principles of

several types of modern balancing machines are
reviewed. Soft support. hard support. and resonant
machines are described. Plane separation and the
need for Wattmeter filtering are discussed in
relation to balancing requirements and signal
conditioning. Larger rotor facilities and auto-
mated production facilities are described.

4.1..1 Introduction

The objective of rotor balancing is to minimize the effects

of residual unbalance on the rotor system during normal
operation. The main effects of excessive rotor unbalance
are:

a) Undesirable vibratory forces applied at the rotor

journals to the supporting structure and foun-
dation.

b) Non-concentric rotor operation (rotor run-out).

c) Excessive noise level.

A perfectly balanced rotor will transmit no unbalance

vibratory force or vibratory motion to its bearings or
68 N.F. Rieger

supports at any operating speed. Acceptable levels of

residual unbalance are described in ISO balancing documents
[1, 2]. A rotor is balanced when the e.g. of the rotor mass
distribution in all normal modes of the system lies on the
axis of rotation. The objective of the rotor balancing
process therefore, is to achieve this condition. Typical
rotor balancing involves:

a) Detect ion and measurement of the effect of

unbalance on the rotor at selected locations.

b) Modification of the rotor mass distribution at the

correction planes.

c) Repetition of (a) and (b) until the rotor residual

unbalance is reduced to an acceptable level.

This procedure may be followed in a general purpose balancing

machine, a special balancing machine, in a balancing
facility, or on-site. Rigid rotors may be corrected in two
planes in a general purpose balancing machine. Flexible
rotors may be balanced in similar equipment with suitable
high speed capability. Most rigid and flexible rotors are
also trim balanced on-site. In most cases particular needs
determine the type of balancing required.

4.1.2 Balancing Methods

Single Plane Rigid Rotor:

The simple single disk rotor consists of a thin uniform disk

mounted eccentrically on a uniform shaft of circular cross
section. The rotor unbalance then lies in the plane of the
disk, and its effect may be removed by adding a suitable
weight opposite the disk eccentricity. It is common practice
to determine the angular location of the rotor unbalance by
placing the shaft on two knife edges and allowing the rotor
to roll until its c .g. finds its lowest position. A known
trial weight is then added to the disk at a selected
location, and the disk is again allowed to come to rest on
the knife edges. The trail weight is then moved to another
location, say 120° away from the first trial location, and
the procedure is repeated. A third trial may be attempted
with the weight an additional 120° from the previous
location. The magnitude of the required correction weight
may then be obtained by solving the resulting vector force
problem: see Sommerville [3].

Two Plane Rigid Rotor:

Any rigid rotor may be balanced by the addition of correction

weights in any two correction planes. The location of these
Principles of Balancing and of Balancing Machines 69

correction planes is usually limited by available access to

the rotor in its casing. Two plane balancing of stiff rotors
of moderate length requires several measurements to be made
with the rotor spinning at low speed, typically between 100
and 600 rpm. Modern rigid rotor balancing involves the
following steps:

a) Spin the rotor at a suitable speed.

b) )Ieasure the transmitted unbalance force and

reference phase angle at the left bearing.

c) Measure the transmitted unbalance force and phase

angle at the right bearing.

d) Use a suitable calculation to determine the angular

location and correction weight required in each
balance plane. This procedure must allow for force
interaction between the correction planes, known as
'plane separation'.

e) Insert the weights at the required locations in

both correction planes.

f) Measure the new transmitted forces at both

bearings. Compare these forces with values
permitted by balance criterion.

g) Repeat the above sequence if needed.

Multi-Plane Balancing:

Flexible rotors may require correction in more than two

planes. Multi-plane balancing may be accomplished by a
number of procedures, the best known of which are:

a) J.lodal balancing, in which the modal components of

the unbalance are corrected mode by mode.

b) Influence coefficient balancing, in which the rotor

correction masses are determined using a computer
program to process the trial weight response data.

Multi plane balancing may be accomplished either in a

balancing facility or in-situ. Techniques range from
trial-and-error balancing to automated computer balancing.
)fult i plane balancing requires the following equipment:

a) Mechanical drive input for required balance speeds.

b) Vibration sensors for data acquisition (displace-

ment probes, pedestal trans4ucers, etc.).
70 N.F. Rieger

c) Signal processing and data reduction equipment,

e.g., tracking filters, wattmeter circuit, mini
computer, etc.

d) Trial weights, and access to rotor correction

planes.

Multi-plane balancing differs from two plane balancing in

that it may require additional modal corrections for high
speed operation. Appropriate two plane balancing of rigid
rotors leads to smooth operation at all speeds.

4.3.3 Tnes of Balancing Machines

Types of Balancers:

General-purpose balancers. These machines are used to

balance rotors of various types and sizes. They are usually
designed for two plane balancing at low speeds although
larger units have been designed for higher speeds. All
recent general-purpose balancers use a mini computer to
determine the correction weights.

Custom balancers. Special purpose balancers have been

developed for such items as small fans, gyros, automotive
wheels, satellites, etc.

Automated balancers. Special custom balancing facilities are

part of the manufacturing process for crankshafts (with metal
milling facilities), small electric motors, conveyor systems,
and for production balancing of connecting rods: see Section
4.1.6.

Calibration Types:

Two important characteristics of a balancing machine are its

calibration and readout capabilities. Balancing machines are
classified with respect to calibration procedure, as follows:

a) Trial-and-error balancing machines.

b) Calibratable machines requiring a balanced proto-

type rotor.

c) Permanently calibrated hard bearing machines.

Support Type Classification:

Balancing machines are classified as soft support, hard

support, or resonant machines. The influence of the rotor
supports on the behavior of the balancing machine is shown in
Figure 1. Figure 1 indicates the regions of operation of
Principles of Balancing and of Balancing Machines 71

soft support, hard support, and resonant machines in relation

to the dynamic support properties of the balancer. For a
hard support machine, the natural frequency of the support is
high. and the machine operates below the support natural
frequency. The unbalance force and the support displacement
are always in phase.

With a soft bearing machine. the balancing is always

performed above the natural frequency of the rotor in its
supports. Unbalance force and response are then 180° out of
phase. Resonant balancing machines operate by passing down
through the natural frequency of the support system, as rotor
speed decreases. The associated resonant amplitude build-up
is then used to amplify the unbalance signals.

Rotor Types:

Balancers are sometimes described as either rigid rotor, low

speed, or flexible rotor, high speed types. These terms are
imprecise. They omit other considerations in the balancing
process. For example, a two bearing, low speed balancer may
also be capable of multi-plane balancing. If such a balancer
were capable of operating at higher speed, then a flexible
multi-plane rotor balance may also be possible. It is
therefore the operating speed of the balancer which defines
its balancing capabilities.

4.1.4 Major Components of Balancing Machines

Rotor Supports:

The rotor support structure of a balancing machine includes:

a) Journal supports.

b) Journal pedestals.

c) Pedestal supports.

d) Foundation base.

There are two types of support in current use: soft and

hard. The soft support principle is shown in Figure 2. It
consists of a low stiffness horizontal spring support for
which the period of free vibration may be one to two
seconds. The vertical stiffness of this support is much
higher. The advantage of soft supports is that, compared to
hard supports, they give a larger signal strength for the
same level of unbalance due to the larger displacements which
these supports permit. Strong signals at rotational
frequency require less sophisticated electronic signal-
processing equipment. Soft support machines also tend to be
72 N.F. Rieger

simpler and less expensive than hard support machines. They

are well suited to most rigid rotor balancing applications
and are used for a variety of small to medium size universal
balancers for the balancing of armatures, crankshafts, fan
rotors, impellers, drive shafts, and so on.

A typical hard support balancing machine is shown in Figure

3. The hard support is moderately stiff in the horizontal
direction, and very stiff in the vertical direction. Modern
designs of such supports commonly make the journal pedestal,
the hard springs, and the movable foundation block from a
single piece of metal. Support motions may be sensed by
displacement probes, .strain gages, and other means. A
typicaf arrangement is shown in Figure 4. Hard support
machines may be more susceptible to extraneous vibrations.

The smaller unbalance signals of hard support machines are

compensated for by the electronic equipment which is normally
provided with such machines. This equipment may incorporate
refined filter and amplifier circuits which condition the
unbalance signals for amplitude and phase angle. Universal
hard bearing balancers for both rigid and flexible rotor
balancing are now available in a range of sizes. ~~ny large
special purpose balancers and facilities are now using hard
support equipment. Rigidity of construction is generally
desirable for high speeds. The electronic sophistication now
available _is also attractive because of its precision.

Rotor Drive:

A variety of techniques are used to impart rotary motion to

the component being balanced; the belt drive and the end
drive shaft are the most common. The drive selection is
determined by rotor size, power involved, influence of
bearing eccentricity, and system dynamics. Flat belt drives
are common with small bench-type balancers. These drives
allow for easy set-up, but additional vibration may be
imparted to the pedestal by the belt motions.

The cantilever belt drive, shown in Figure S is used in

resonant balancers. When lowered to contact the rotor upper
surface it imparts rotation. The belt is moved away from the
rotor when the desired speed is reached (somewhat above the
rotor support critical speed). The rotor then drifts down in
speed, passing through resonance. During measurement the
belt is not in contact with the rotor. This eliminates any
belt excitation effects.

For installations with large rotor inertia and drive power

requirements, it is often desirable to use an end drive. The
end drive may consist of a coupling shaft with a universal
joint at each end. The shaft reaches from the balancer drive
Principles of Balancing and of Balancing Machines 73

unit to the overhung end of the rotor as shown in Figure 6.

Such drives are widely used in medium and large general
purpose balancing machines, and in many custom machines
because they transmit the higher power necessary for
acceleration and regenerative braking. In very large instal-
lations, a specially designed drive shaft coupling may be
required to supply adequate drive power.

Foundations:

A massive foundation is needed to support the rotor being

balanced and to attenuate external vibrations. The
foundation must permit axial adjustment of the balancer
pedestals. For small balancing machines, e.g., Figure 7 the
foundation may support only the pedestals, with the
electronic equipment mounted elsewhere. Medium sized
balancing installations, figure 5, are often built as a unit
with the electronic equipment mounted for convenience at one
end, on the foundation. In large special purpose units~ the
foundation may merge with the protective equipment of the
spin pit.

Electronics:

Electronic equipment used to acquire the vibration signals

ane to process the signals into unbalance information is an
essential aspect of all modern balancing machines. The
following electronic equipment is used:

a) Vibration sensors: inductance probes, acceler-

ometers, strain gages.

b) Filtering circuits: wattmeter circuit, tracking

filters.

c) Operational amplifiers.

d) Plane separation circuit.

Depending on their application, vibration sensors detect

velocity, displacement, or acceleration. Though the sensor
output may be weak or contain high background noise, this can
usually be amplified and filtered out. Vibration sensors for
a hard support machine are shown in Figure 3 and for a soft
support machine in Figure 5.

Signal filtering can be accomplished by a number of special

circuits. The most widely used is the so-called Wattmeter
circuit, which acts as a filter to exclude all non-synchro-
nous a.c. components from the balance signal. The resulting
output signal is a clean harmonic waveform which may then be
used to define the rotor unbalance. Other circuits use
74 N.F. Rieger

tracking filters with an analog/digital converter coupled to

a micro-processor to achieve the same operation.

A balancing machine Wattmeter circuit is shown in Figure 8.

Alternating curre.nt from an unbalance sensor is supplied to
the Wattmeter field coil. Alternating voltage from an
a.c. generator coupled to the drive shaft is supplied to the
Wattmeter moving coil. The deflection of the moving coil is
then proportional to the Wattmeter power W = EI cos 9, where
E is the generator voltage, I is the unbalance sensor
current, and 9 is the phase angle between the voltage and
current signals. Figure 9 shows how a wattmeter can combine
waves of identical frequency, and how waves of differing
frequency fail to produce a voltage and current when these
components are in phase. Where the unbalance current signal
leads the a.c. voltage, the unbalance power signal is reduced
by cos 9. The unbalance Wattmeter also requires careful
measurement of sensor signal phasing to avoid incorrect
balance readings. Only voltages and currents having the same
frequency can be combined in this instrument.

Accurate balancing depends on obtaining accurate signals

which relate in a consistent known manner to the unbalance
force being imparted to the pedestal supports. The Wattmeter
method is a simple procedure for preventing unwanted signal
components from appearing in the unbalance signal, and also
for excluding unwanted signals arising from rotation such as
non-circular journal harmonics drive stick-slip effects,
small impacts, misalignment, and excitation from the external
environment.

Plane Separation:

Figure 10 shows a rigid rotor to be corrected for unbalance

in the two planes indicated. The rotor is supported by two
end bearings. It has two correction planes inboard of the
bearings, and the residual unbalance force is represented by
the two applied forces shown. If the left bearing support is
restrained so that no lateral motion is possible at that
location, and if the right bearing support is free to move,
then it is possible to select a balance weight for the right
correction plane such that the right end of the rotor will
run smoothly. The same rotor could also be corrected by
restraining the right bearing and inserting a suitable
correction weight in the left correction plane. The
criterion for acceptable balance is that the rotor shall
operate without transmitting any dynamic force to the
bearings. Having corrected the rotor in the left and right
planes independently, it might be thought that the rotor will
run smoothly if both bearing restraints are re1eased. This
is not the case. The rotor will run roughly because the
rotor force balances will not have been achieved indepen-
dently of the support forces.
Principles of Balancing and of Balancing Machines 75

This problem can be overcome using the principle of plane

separation, as follows, First, in Figure 11 instead of
restraining the rotor motion at the left bearing, assume that
lateral motion can be restrained at the left balance plane,
Moment and Force equilibrium then give:

b-a
u~

0: -F 1 + U - C2 = 0:

Suitable weights are then added in the right correction plane

until the rotor runs smoothly, The second step is to
restrain the rotor at the right correction plane. and then
balance it in the left correction plane until it again runs
smoothly, i.e.,

c-b
l M2 0: U(c-b) - C1 (c-a) ::: 0: c1 = u-
c-a

b-a
lF = 0: -c1 + u - F2 = 0: F2 = u~

Note from the above that c 1 = F1 and c 2 = F2 , This indicates

that taken together the balance correct1ons will automa-
tically cancel the bearing forces, and both moment and force
equilibrium for the rotor are thereby satisfied, If the
rotor is then operated with corrections c1 and c2 installed.
the transmitted forces at the bearings wfll be zero and the
rotor will operate in a smooth, balanced manner,

4.1.5 Modern General Purpose Balancing Machines

Soft Bearing Machines:

The Gishol t Model 31S shown in Figure 12 is a typical soft

bearing general purpose balancing machine, It is designed to
operate well above the resonant speed of the supported test
rotor at speeds between 500 rpm and 2000 rpm, It is able to
accommodate rotors weighing between 1 lb, and 500 lbs, Drive
power is supplied through a belt which wraps around the
rotor. in a tension pulley arrangement, Unbalance data is
obtained as velocity transducer readout from movements of the
76 N.F. Rieger

supporting pedestals. The electronic equipment includes

plane separation circuits and Wattmeter filtering. The
readout is calibrated in terms of the required balance
correction units. The 'heavy' spot may be located by either
stroboscope phase measurement or by contrasting surface
photocell readings against numbered strips placed at the
correction planes. The readout equipment is specified as
being able to read down into the ten micro-inch range. The
machine is not permanently calibrated, and pre-calibration
against a rotor of known unbalance is required periodically.
The soft pedestal construction is achieved with a pair of
vertical flat springs. One end of each spring is attached to
the machine frame while the other end is secured to the
adjacent pedestal support.

Resonant Machines:

The Stewart-Warner 2380S machine, Figure 13, has a unique

support which permits tuning so that the natural frequency of
the rotor will occur at the desired balancing speed of the
machine. In practice, the rotor to be balanced is loaded
into the pedestal supports, whose dashpot damping is manually
set to zero. The rotor is then bumped gently, causing it to
vibrate in its lowest mode on the supports. A vibration
meter indicates the frequency of this mode at the pedestal.
A suitable value of damping is then chosen.

Both dashpots are then adjusted to the selected damping

value. The required rotor speed is pre-set on the handle of
the cantilever drive arm and an overhead cantilever belt
drive brings the rotor up to speed. A common balancing speed
for this machine is 450 rpm. The natural frequency of the
rotor in its supports is adjusted to occur somewhat below
this speed, so that large mechanical amplitude magnification
occurs as the rotor passes through resonance when the drive
is removed. Rotor unbalance is read on a graduated scale (no
units) for the left and right correction planes indepen-
dently. Unbalance angular orientation is determined strobo-
scopically, as with soft bearing machines. The balancing
operations proceed by trial-and-error.

This machine has found widespread use in the automotive

industry.

Hard Support Machines:

The Schenck H30V, Figure 14, is a typical hard bearing

machine with plane separation circuits, Wattmeter filtering
and measurement, push button set-up of the balancing
operations and options for either direct or belt drive.
Unbalance signals are detected at the pedestals using
inductive force transducers. The balancing speed is usually
Principles of Balancing and of Balancing Machines 77

i thin the range 275 rpm through 1500 rpm - considerably

ow
below the resonant frequency of the rotor on its support
system. It accommodates a maximum rotor weight of 880 lbs,,
carried synm1etrically on the pedestals, as well as asymmetric
rotors of smaller weight.

Advantages are: easy set-up and operation, permanent

calibrations with no trial weight runs, simple dial-in
balance with vectormeters, precise balancing, and the ability
to handle a wide range of rotors. Disadvantages are: higher
installation cost and relative delicacy of the machine.

This machine is said to be capable of balancing down to

0,0035 oz. in. Such accuracy may be possible with light
rotors, e.g., 30 lbs., though this corresponds to a e.g.
eccentricity of 3.5 micro-inches. For 880 lbs. rotors, such
balance precision would mean 0.25 micro-inches e.g. run-out
in the balanced condition.

General Purpose Hard Support Balancing Machines:

A typical industrial bard support balancing machine is shown

in Figure 15, The hard support general purpose machine
provides an efficient procedure for two plane balancing.
Strain gages or displacement sensors are attached to the bard
support frame, as shown in Figure 4. These sensors transmit
pedestal motions as electrical signals to the measurement
console. Both pedestals are permanently factory calibrated
prior to shipping. Special calibration is needed prior to
balancing.

A synchronous phase reference voltage signal is taken from

the drive, which may be either a universal coupling or
quick-attach belt. Inductance transducer signals are then
processed using a Wattmeter circuit in conjunction with a
plane separation circuit. The vectormeter screen used in the
model of Figure 16 shows either the residual unbalance and
phase angle in both balancing planes or the required balance
weights and their orientation.

With such equipment, an immediate, precise two plane balance

is possible for a wide variety of ISO Class I and Class II
rotors. The basic proportions of the rotor to be balanced
(din1ensions a, b, c, r 1 , r 2 shown in Figure 16) are first
dialed into the balancer console. The rotor is then run at
the desired balancing speed, and readings of the magnitude
and location of the required balance weights and their phase
angles are displayed. The rotor is then stopped, the
required correction weights are inserted in the two
correction planes, and the rotor is re-run at the balancing
78 N.F. Rieger

speed to check the balance. The simplicity of this process

permits a relatively unskilled operator to balance rotors
rapidly and effectively.

4.1.6 Balancing Facilities

Turbine and Generator Balancing:

Turbine and generator balancing facilities are designed so

that high speed balancing operations and overspeed testing of
assembled rotors are both possible. S11ch facilities may
incorporate:

a) Concrete overspeed burst-proof tunnel.

b) Vacuum spin test chamber,

c) Variable speed drive with 20 percent overspeed

capability.

d) Provision for various fully bladed rotor assem-

blies.

e) Bearing pedestals designed to simulate machine

support properties.

f) Transporter,

g) Control room,

h) Overhead crane,

i) Rigid attachment of transporter to tunnel

foundation,

Figure 17 shows details of a transporter loaded with a rotor

assembly and internal details of the tunnels. Figure 18 is a
section schematic of a turbine-generator balance facility.

ISO Class I and Class II rotors will operate satisfactorily

after having been balanced in a low speed balancing machine.
Though such rotors may require load testing and overspeed
testing, these additional operations may be conducted in a
less fully instrumented facility, and probably without a
vacuum chamber.

Class III flexible rotors require high speed balancing

involving trial runs near one (or more) resonant speeds
within the operating range. It is convenient to have n
common high speed facilit) in which both balancing and
overspeed testing can be undertaken. Class III rotN~ are
usually large and long and may carry thousands of blades.
Principles of Balancing and of Balancing Machines 79

Modern spin pit balance facilities are complex installa-

tions. Costs may range from SS million to S10 million or
more for a complete, installed unit. Figure 19 (burst pit)
is a general view of a concrete balance spin pit facility
showing the rotor on its transporter, with special bearings
and pedestals and other details. The bearings are bolted to
the foundation during testing. A section through a balance
facility, with a genera tor rotor installed, is shown in
Figure 18. Details of the drive, drive coupling, lubrication
system, and other features of such facilities are evident.

A special bearing pedestal support for use in turbine

generator balance pits is shown in Figure 20. Such units
have been designed to provide a tuned pedestal support in
which the specific bearings of the rotor being balanced may
be installed. This permits the rotor to be balanced while
operating in its own bearings and it incorporates simulated
operating pedestal effects.

Rotor support stiffness properties may exert a significant

influence on the critical speeds and dynamic properties of a
rotor bearing system. If a rotor is balanced in hard
bearings, and then operated in softer bearing supports, the
rotor system modes will be different, and the balance
achieved in the balancing stand will not be fully realized
during operation. The rotor may then run 'rough' unless
corrected by further in-situ trim balancing. The support
design shown in Figure 20 allows the rotor to be balanced in
the bearings in which it will operate. Matched dynamic
properties are especially important for large Class II rotors
in which unbalance effects through the fourth flexural mode
may be critical. Stiffness properties of such supports may
be adjusted within the range 30,000 lbs./in. through 3.0 x
10 6 lbs./in., according to whether soft or hard support
balancing is required. This preserves the balance quality
achieved for operating conditions, and leads to less field
balancing. Figure 20 shows details of such adjustable
support bearings, including the lubrication inlet (fore-
ground) and the tangential force transducers from which the
transmitted unbalance force signals are read. Fluid film
supports avoid the problem of 'brinelling' which may occur in
rolling element bearing supports, due to rotor weight which
may range up to 250,000 lbs. per pedestal.

Automated Balancing Facilities:

High volume production industries depend on automated

equipment for large numbers of repeated operations.
Automated and semi-automated balancing is widely used in the
automotive industry for crankshafts, propeller shafts,
tire-wheel assemblies, etc., and in the electric motor
manufacturing industry for vacuum cleaner motors, blender
motors, fan motors, and so on.
80 N.F. Rieger

Figure 21 shows an example of an automated balance facility

for crankshaft-clutch housing assemblies. Thh facility
incorporates a production line for:

a) Inithl measurement of residual unbalance in two

planes.

b) Correct ion of the unbalance by drilling at the

required locations.

c) Checking of the corrected assembly to meet required

balance specifications.

The operation of this crankshaft balance facility is as

follows. Incoming batches of crankshafts are loaded in an
ordered fashion on to a conveyor with a gravity feed roller
conveyor. The process is automatic from this point. Two
displacement sensors record the unbalance as the crankshaft
is rotated at low speed (400 rpm). The signals are stored
and the crankshaft is autc•matically indexed to the required
angular position for each plane in turn, using the data
obtained during the previous measuring operation. Two
separate drills then remove the required depth of metal at
the desired orientation. The drilled assemblies are then
spun again in the balance checking unit for a second measure-
ment. Once the crankshaft is positioned, the two
pneumatically controlled halves clamp it for drilling. The
linkage and pneumatic cylinder of this equipment may be seen
in Figure 21.

After checking, crankshafts are classified 'within tolerance'

or 'outside tolerance' automatically. Accepted crankshafts
pass out on the conveyor, while rejected crankshafts are
removed for re-processing.

4.1.7 References

1) ISO Draft Document, 'Balance Quality of Rotating Rigid

Bodies,' ISO/TC 108 DR 1940, 1967.

2) ISO Draft Document, 'The Mechanical Balancing of

Flexible Rotors,' ISO/TC 108/SC1/WG2 (Secretariat-7)
Document 12, 1976.

3) Sommerville, I. J ., 'Balancing a Rotating Disc, Simple

Graphical Construction,' Engineering, February 19, 1954.
Principles of Balancing and of Balancing Machines 81

Resonant Machines

Amplitude
= 0.0

= 0.15
Hard Bearing
Machines
,....,r-- Soft Bearing H chines

Rotor Speed

Figure 1. Relation Between Rotor Support Stiffness

and Balancing Machine Type

Figure 2. Soft Support Balancing Machine Principle

82 N.F. Rieger

Figure 3. Hard Support Balancing Machine

(Courtes~Schenck-Trebel Corporation)
Principles of Balancing and of Balancing Machines 83

Bearing carriage
Inductance Pedes ta 1
transducer Location

ODD
Vertical
1110ti_ons of
carrtage
~re negligible
Flexible members
for horizontal
1110vement

Rigid foundation

Figure 4. Hard Support Balancing Machine Principle

Figun' 5 . Cantilever Belt Drive

(Courtesy: Schenck-Trebel Corporation)
84 N.F. Rieger

Figure 6. End Drive

(r.oiJrtPsy: !'rhenrk-TrPhel Corp.)

Figure 7. Massive Foundation, External Electronics

(Courtesy: Schenck-Trehel Corporation)
 "'C
::! .
:s
n
-o
Unba I anee ii"'
Wattmeter ,\n.llog "'0
Swi t c hi r. g C1 r :Ji ......,
Ana l og
~
Switching t~rcuit
:- -, e:...
$»
:s
I n
:;·
C/(l
$»
:s
I 0..
0
......,
~
e:...
$»
:s
!:!.
:s
C/(l
Impedance ~
Probe $»
n
;:r
U"ba lanced Rot or :;·
: ... :.~t.! Ct • ·• t' n
:.l r 'l~r
"'
Universal
Couol i ng

Support
Flexibility

Figure 8. Wattmeter Filtering Principle

00
V'l
86 N.F. Rieger

E,I

2n
l El
_a
sin wt sin wt d(wt) El!i

wt

sin wt sin{wt+e)d(wt) El~os e

wt

21(
L--+--+-+--r-- +-+- fEr sinwt sin Zwt d(wt) = o
wt 0

E,I

zr
f El sinwt sin(2wt+e) d(wt) = 0
L......<-~~+f--+-f-·1---\--'-----<l
wt

Figure 9. Wattmeter Elimination of

Differing Frequencies
Principles of Balancing and of Balancing Machines 87

left correction plane Right correction plane

I
.I.
b_±_j
Figure 10. Rigid Rotor With Two Unbalance Forces

correction plane 1 correction plane 2

Figure 11. Correction Planes and

Resultant Unbalance Force
88 N.F. Rieger

Figure 12. Gisholt Model 31S Soft Bearing Machine

(Courtesy: Schenck-Trebel Corporation)
Principles of Ba;ancin~ ,•nd of Bal.mcinb Machines 89

Figure 13. Stewart-Warner 2380S Resonant Support Machine

(Cnurt~sv: Schen,·k-TrPh~l Corporat i<>n)

Figure 14. Schenck H30V Hard Bearing Support Machine

(Courtesy: Sch~nck-Trebel Corporation)
90 N.F. Rieger

Figure 15. General Purpose Hard Support Machine

(Courtesy: Schenck-Trebel Corporation)

Figure 16. Vectormeter Two Plane Readout Console

(Courtesy: Schenck-Trebel Corporation)
Principles of Balancing and of Balancing Machines 91

Figure 17. Transp,•rter Supports With Generator Rotor

(Courtesy: Schenck-Trebel Corporation)

Figure 18 . Schematic of High Speed Balancer Spin Tunnel

92 N.F. Rieger

Figure 19 . Burst Pit With High Speed Rotor

(Courtesy: Schenck-Trebel Corpor a tion)
Principles of Balancin~ .111d of Balancin~ Machinl's 93

Figure 20. Tuned Dyn~mic Support Pedestal

f0r Artual Bearin~s
(Cou! t L'SV: Schenck-Trebel Corpor .rt ion)

Figure 21. Automated Cranksh~ft Balance Facility

(Courtesy: Schenck-Trebe1 Corporation)
 CHAPTER 4.2

FLEXIBLE ROTOR BALANCING

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

Developments which lee to modern flexible rotor

balancing techniques are described. The theory of
flexible rctor balancing is outlined, and the
theoretical basis for Modal balancing, and for
Influence Coefficient balancing, is presented. In
bot}) instances, the publications from which the
source material for this section was obtained are
identified, witr discussion. Certain other
balancing techniques are also discussed. Recent
developments toward the establishment of criteria
for appropriate levels of residual unbalance in
flexible rotors following balancing operations are
included. The state-of-art for flexible rotor
balancing is summarized, and the work which remains
is identified.

4.1.1 Development of Flexible Rotor Balancing

The earliest reference to flexible rotor balancing appears in

the patent awardee to Linn in 1928, reference [1], in which a
sequential process for correcting the whirl modes of flexible
rotors in bearings is described. Figure 1 shows the title
sheet of the Linn patent, which refers to a steam turbine
rotor in end bearings, to demonstrate the proposed process.
The next development appeared in a paper by Thearle [2]
concerning a vector method for two-plane balancing of the
rotors of a three-bearing turbine-generator set. The
application of a proposed method is described in detail, but
without mention of whether or not the operating speed is
beyond. any bending critical speed of the unit. The method
proposed is in effect a trim balancing procedure, intended to
obtain an improved final balance condition for the three-
bearing shaft assembly, over the two-plane balance results
96

achieved separately for the indhidual rc•tors. Two other

early papers, by Kroon [3] deal "itb He theor) of rigid
rotor balancing. These papers also include certain 1·ractical
details for field balancing of turline-generat~r~.

Grobe! [4] is a discussion of r•ractical flexible rctor

balancing for large turbine!. and generatt'rs. The iJtportance
o f a c cur a t e 111 an u fa c t u r i n g 1:1 e t hod s t o D'· in iD· i z e in he r e 1: t
unbalance is stressed. Special ~;robleD'Is arising f1or-• tle
generator 'ltindings, froD'I thermal instability of the tt.rbine
rotor, and from the buckets arou~d tte t~rbine circuD'Iference
are mentioned. AD empirical trial-weight balancing method is
described for the first three bendhg D1odes cf the rt'tcr,
indicating that since Thearle's earlier paper improved
techniques had become necessary to account for flexible rotl>r
effects in larger machines. Details of residual vibration i~
installed units are given. The paper contains an interesting
de script ion of the practical aspects of turbine-generator
balancing at that time (1953).

Moore and Dodd [S) is a detailed practical discussion of the

Modal balancing technique appliec! to tlirbiDe-generat.or
rotors. Only single-span rotors are considered, not entire
units. The vibrations arising fro111 each 1:\0de are first
described, and tben the Modal balancing technique and its
application is described in detail, ir. relati01: to the
suppression of these vibrations. An interesting exa~ple of a
pump rotor which responded strongly in its second anc third
modes simultaneously at operating speed is described. The
problem of se~arating and balancing D'lixed 1:1odes is discussed,
with detailed numerical values. Recognition 0f the cau~e of
this 11robleD'1 l"nci its solution Icsolve«! rersisteJ:t diffi-
culties with thts unit.

Another recent practical contribution is due to Lindsey [6}

who described an empirical 'one-shot' method for the belanc-
ing of large turbine-generator rotors. The 1:1etl1od is most
effective when the interaction f1·om adjacent spans is
~:~iniral, and where the whirl ellipse is circular. In
essence, the method is a procedure for balancing single-span
rotors in their first three DIOdes without regard to adjoining
spans. It relies on collected previous experiences with
similar units. The method could probably be adapted to core
general balancing of other types of supercritical rotors.
This paper preseDts an outline of the D'lethod, but gives no
theory nor supporting details of its perforD'Iance.

4.2.1. Theory of Flexible Rotor Balancing

The theory of flexible rotor balancing was begun by ~!e]daht

[7] who outlined the principles of modal balancing and their
application to the three-plane balancing of r11 end-beering
Fk·xibk Rotor B.1lancing 97

flexible: r.:·tcr. ~~c!dahl recognized the inadequacy of two-

rlnne balancing f0r flexible rotors, and observed that it was
possible t(: baJan~.:e a rc-tcr n,ode-by-mode on the condition
that the rlacer.crt cf corrections for one mode does not
influel'ce the rt:>J·On~oe in any c·tler mode. The justification
for this requiter,ent lies in the orthogonality relation 1
wtJch shows that tbe irteraction between any two modes will
be J:'liniuized when tte condition:

bttwcen tloc r;1odal diss:·Iace111ents Y. and YJ· is satisfied. This

. . I cnn d'JtJons,
111eans t h at f or non-trlVJa J

>0 i • j
· y • j · dz • 0 {
j
• 0 i • j

"Jere c::., c.•. are natural frequencies for the ith and jth
rctc•r n1o1des tespcctively, andy. andy. are the modal ampli-
tudes. It follows that a flexfble ro~or may be bal'anced by
~li~i~ating the effect of residual unbalance mode-by-mode. As
the d~flection of the rotor at any speed may be represented
as the su~:~ of nl'ious n,odal deflections multiplied by speed-
dependant constunts, a rc·tor which has been balanced for
sruoott operation at each of its critical speeds in accordance
wit}• the abo\·e condition will remain balanced at any other
SJ;et"d.

The remainder cf 1\':eJdahl 's paper outlines the procedure for

balancing a rctor in its first three modes. Measurement of
whirl amplitudes away from the bearing locations is implied
by the staten1ent that fles.ible rotor balancing in rigid
hearings is not possible. This statement can evidently be
relaxed if r.eas~rer.ents are made at other locations along the
rotor length. For overhung rotors, or with sufficient
r>easurelllent and balance planes located along the rotor,
rc>tors in rigid bearings a•ay be readily balanced. The
J1roceclure described in this paper neglects damping effects,
but this causes no problem, as damping merely renders the
modes comrle~. and may become of minor significance where the
system response is dominated by rotor fles.ibility.

1 For a more complete discussion see, for example,

Timoshenko [8] or Bishop and Gladwell [11].
98 N.F. Rieger

4.2.3 Moda 1 Ba 1a nc ing

The Modal balancing method Dlay be applied to an axisylllllletric

elastic rotor of arbitrary stiffness anu mass distribution in
rigid bearings without damping such as tlat shown in Figure 2
as follows: first assume that all unbalances act in the same
y-z plane. Let the rotor deflection share y (z) at an)' speed
be described by the series:

+ . . .

where the 0 1 are coefficients , and the Y. (z) are 1:1odal

displacement expressions fC'r each of the no:lmal modes. The
distribution of mass eccex r ici ty throughout the rotor n,ay
also be expressed in series form as:

e (z) + +

+ . . .

At any speed, the amplitude response resulting from the rotor

mass eccentricity and the elastic deflections may also be
expressed as an infinite series in the normal modes by the
expression:

U(z)

+ ... >...Y.(z).
1 1

The orthogonalit y relations between the modes may now be used

to determine the unknown coefficien ts in the above
expressions:
Flexible Rotor Balancing 99

{"0
£ 1 ,. j
S m(z) Y1 (z)Yu(z)dz:
0 • M
1 1 - j

~0 EI(z) \
II

(z)Yj' (z)dz {•0

• K
i ~ j

i -
i j

First let the rotor experience a virtual displacement y in

the y-z plane such that:

The work done by the centrifugal force during the virtual

displacement is:

L
Wcent • S m(z)ro2 [y(z) + e(z)](oQ 1)Y 1 (z)dz
0

Now utilize the first orthogonality expression by multiplying

both sides of the rotor deflection expression by m(z) Y. (z)
1
and integrating over the length of the rotor:

l
~ m(z)Y 1 (z)y(z)dz - (j i
c" 2 (z)dz
j m(z)Y 1
0 0

lienee:

The work done in the virtual displacement is stored as

elastic energy in the rotor. For simplicity, let the rotor
cross sect ion be uniform along its length. Then the stored
energy is:
100 N.F. Ricvcr
C>

v .. jo
d2
1
,.t 21 EI(z) [ ::t dz • jo
,.t 2
1
EI(z)
~
LQ i Y~ (z)J
, 2
dz

i•l
in keeping with the second orthogonality relation given
above. The virtual work of the elastic forces may be
expressed as:

By the Principle of Virtual Work W t + W 1 0• Pe r form in g

this equation leads to the l ··nclus~gg that~

and near the ith critical speed

Thus,
1
or
( w/ w . ) 2
1n

The unbalance coefficients Al may also be evaluated by using

the first orthogonality relationship. Multiply both sides of
the unbalance equation by m(z)Y.(z)dz and get:
1
Flexible Rotor Balancing 101

In o r d e r t o b a 1 a n c e t he r o t. o r td t h d i s c r e t e ma s s e s at
sr·ecifjc locations it is convenient to deterJtine the effec-
tive unbalance at these locations. The deflection from a
concentrated force F or frorr, a uniformly distributed load q
acting over a small length ~ distant c from z = 0 may be
expressed as a Dodal series, vix.,

Determine the coefficients a. using the first orthogonality

1
relationship:
t
S 0
t
S
Pm(z)Y 1 (z)dz = a 1Y1 (z)m(z)Y 1 (z)dz = a 1M1
0

c+~l2

= ~ q m(z)Y 1 (z)dz
c-~12

The coefficients a. u:ay be found for q constant as ~

becomes vanishingly s~all as:

P m(c)Y (c)
i

The condition for balance is that the ith modal component

pust vanish. To achieve this a correction weight must be
inserted at a location z ~ c on the rotor such t~at U(z) = P
or:

or
-
Thus,

if

U(z) ~ m(z)e(z).
102 N.F. Rieger

To balance a rotor in its ith mode it is therefore necessary

to know the mass distribution m(z), the natural frequency w1
of the mode to be balanced, the modal amplitude ~(z) and tle
measured mode shape y(z)"b. From these the it deflection
coefficient 0. and the it eccentricity coefficient e. are
calculated. he {equired unbalance correctjon Pat locition
z = c for the it mode may then be obtained using the above
formula.

4.2.4 Papers by Bishop and co-workers on Modal Balancing

Between 1959 and 1967 an important series of papers on

flexible rotor balancing was published by Bishop and his co-
workers at University College, London. These papers greatly
extended the theory and application of the modal balancing
method, and drew attention to the need for flexible rotor
balancing in general.

Bishop [9] discusses the vibrations of a rotating circular

shaft having distributed mass and elastic properties, in
which the displacement amplitudes are represented as a power
series in terms of the modes. The analysis describes the
modal concept, with examples. Gladwell and Bishop [10]
applies the analysis of (9] to an axisymmetric shaft of non-
uniform section in flexible bearings. It is shown how
natural frequencies and characteristic modal functions may be
found. A d-iscussion of _free and of forced mot ions is given.
The receptance method is generalized to achieve this. and the
modal equations are again obtained. Bishop and Gladwell [11]
presents the underlying theory of a general method for the
balancing of flexible rotors, mode by mode. The Jeffcott
whirl theory is presented for a distributed mass-elastic
rotor, and the series modal whirl amplitude solutions are
given. The implications of low speed balancing are first ex-
amined analytically, and then the balancing of flexible
rotors is considered. An example involving a uniform shaft
balanced in two modes is discussed. The effect of a bent
shaft and of shaft weight are also exa~t.ined. Bishop and
Parkinson [12] discusses a procedure whereby the whirl modes
may be isolated for balancing. This procedure requires the
rotor to be run near each critical speed in turn, to magnify
the modal distortions at the balancing speeds. Two disadvan-
tages of modal balancing are discussed, together with an
adaptation of the Kennedy-Pancu [13] method of resonance
testing to overcome these shortcomings. The practical
aspects of applying modal balan~ing methods are discussed in
great detail, and the results of balancing tests on three
experimental rotors are described.

Lindley and Bishop [14] discusses the balancing of large

turbine rotors. After reviewing the overall balancing
problem, certain laboratory experiments are mentioned,
Flexible Rotor Balancing 103

followed by a discussion of industrial turbine rotor

problems. A 200 mw generator rotor with 500 oz.in. unbalance
was measured for modal balancing at 90 percent of its lowest
critical speed. It was then trim balanced at full speed
(3000 rpm) when cold, and after hot overspeeding. Improve-
ments over the initial unbalanced condition of between 10 and
20 times are reported. This paper contains an informative
contributed discussion by several experienced balancing
engineers, and a closure by the authors.

Parkinson and Bishop [15] discussed the problem of residual

vibration in a rotor after the modal balancing meth"od has
been applied. The suitability of measurements made with
seismic pedestal-mounted transducers is considered, following
from a comment by .Morton in the discussion of [14]. Residual
vibration in rotors running below their (first) critical
speed may be corrected by the addition of a single mass
correct ion, and such is demonstrated for a boiler feed pump
rotor. Modal balancing is proposed for all modes through
which the shaft runs, and an averaging technique is
recommended for the remaining higher modes.

Bishop and Parkinson [16] is a study of second-order whirl in

flexible rotors caused by gravity sag, which remajns after
the modal synchronous unbalance has been removed. This
problem is prominent in two pole generators, due to dis-
similar lateral stiffness of the shaft. This paper contains
a modal analysis of the motion of such a rotor, in rigid
bearings for a uniform dissimilar-stiffness shaft, and a
study of the stability of such a rotor in free vibration and
forced vibration. Isolation of the second-order modes by the
Kennedy-Fancu method is examined, and a brief discussion of
industrial rotor problems is given which includes both
dissimilar shaft stiffness and the effect of coupling
stiffness fluctuation.

Bishop and Mahalingam [17] undertook experiments on a simple

rotor with dissimilar lateral stiffness and gravity sag,
considering both synchronous whirl and second-order whir 1
effects. First, an alternate method to proximity probes is
proposed for measuring the whirl motions, utilizing a slotted
disk shutter. This apparatus consists of a heavy disk on a
heavy shaft, operating in rolling element bearings. The
elastic and inelastic properties of the rotor-bearing system
are discussed, and also the balancing of the shaft. During
balancing, a vibration level was reached beyond which further
improvement was erratic, due to changes in shaft initial bend
and from the relaxation of strains incurred near the critical
speed during balancing. Balancing of an asymmetric shaft at
its critical speed is discussed, including the observed
104 N.F. Rieger

strain relaxation effect. The second order gravity vibra-

t ions are then exa~r<.ined, including the variation of gravity
sag with speed.

Parkinson [18] describes a technique for balancing shafts

which are not axially symmetric. In this, the codal balancing
method is applied to find the single plane of unbalance
corresponding to each pair of (close together) modes. This
condition results from the orientation of the unbalance tc
the principal inertia axes. A procedure for locating the
unbalance angle is described, using the Kennedy-Pancu polar
plot.

Parkinson [19] examines the effect of asymmetric bearing

properties in the x- and y- directions on the vibration and
balancing of a uniform shaft, again using the modal technique
to generalize the interpretation of the distributed mass-
elastic result. The equations of motion from earlier works
are utilized and the modal functions are developed in terms
of the receptances of the system. Two series solutions for
the modal displacements in the principal stiffness directions
are used to obtain general expressions for any displacement
and speed. The specific example of an undamped simple shaft
in asymmetric bearings is considered, for the cases of free
whirl stability and for unbalanced forced whirling. Bearings
with large stiffness asymmetry and with small stiffness
asymmetry are studied. Finally an extension of the modal
technique to balancing with asymmetric bearing stiffness is
presented.

Parkinson [20) summarizes the contributions made by this

group of workers to the unbalance response of shafts.
dissimilar stiffness whirl, and stability of rotors in
bearings, and to the vibration and balancing of shafts. The
basic principles of each aspect are presented and discussed
in terms of the modal approach.

4.2.5 Other Modal Balancing Contributions

A number of modal balancing problems were exandned

analytically by Kushul' and Shlyakhtin [21). The procedures
described apply to complex rotor shapes with both con-
centrated and distributed mass-elastic properties, without
gyroscopic or rotatory inertia effects, in two or more
radially-rigid bearings. A number of theorems for balancing
flexible rotors are given, and determination of the 111ost
effective distribution of balancing planes is discussed.
Three numerical examples are given to demonstrate use of
these theorems and procedures. Although complicated. the
mathematical procedures are expressed in sufficient detail to
make this a valuable paper. Generally speaking, the o~ission
of gyroscopic and rotatory inertia effects detracts fre>r. He
Flexible Rotor Balancing 105

prediction accuracy attainable, but should not represent a

serious problem except for single-span rotors with overhung
end disks. The restriction of the method to rigid bearing
rotors for which the critical speeds and mode shapes must be
calculated as eigenvalues in each instance appears to be a
serious shortcoming, especially as the authors' experience
indicates slow convergence of the modal series, even for the
simple examples chosen. The influence of damping is
neglected, and so the modal forms are plane curves rather
than space curves. Although the analysis is thereby made
tractable, the results achieved are more hypothetical and
less related to practical rotors. No correlation with
experiment results is included, and so the effectiveness of
the method cannot be assessed from the results given.

Kellenburger [22] made a systematic analysis of the balancing

requirements for a flexible rotor in two flexible end
bearings. Distributed, variable, mass-elastic rotor
properties are assumed. The unbalance is continuo_usly
distributed. The required correction masses are determined
for a specified number of balancing planes, and the errors
caused by neglecting the effect of higher modes are
determined. Rules to guide the location of balance planes,
and for selection of the required number of correction
planes, are specified. The author's intention is a precise
formulation of the requirements for mass balancing of
flexible shafts with an infinite number of degrees of
freedom. Damping effects are not considered, foundation
mass-elastic properties are assumed negligible, and bearing
stiffnesses are assumed identical in both coordinate
directions. The amount of residual bearing force is taken as
the criterion of the balance obtained. It is claimed that a
minimum of (r + 2) balancing planes suitably distributed are
required to achieve an effective balance, where r is the
number of cr it leal speeds within the speed range of the
machine. Instances where this formula may fail due to
unfavorable positioning of the balance planes are discussed.
Residual errors arising from insufficient planes, various
distributions of planes, and from unbalanced whirl modes are
studied using a sample problem. No experimental or test
confirmation of the procedure is included, and no commentary
on application of the method is provided.

4.2.6 Influence Coefficient Method

The practice of inserting a trial weight to determine the

effect of unbalance on a given whirl mode is fundamental to
all balancing methods. The influence coefficient method is a
formalization of this procedure whereby: (a) trial weights
are inserted at selected locations along a rotor in a
specified sequence, (b) rotor amplitudes and phase angles
are read at convenient locations along the rotor, and (c)
106 N.F. Rieger

the required correction masses are computed from the

amplitude and phase data. and installed. This method requires
no preknowledge of the system dynamic response character-
istics. although such :knowledge is helpful in selecting the
most effective balance planes and readout locatic:ms. The
earliest theoretical procedure to explain this method for
flexible rotors appears to have been due to Goodman [23]
around 1961.

The Influence Coefficient method may be applied to balance

the elastic rotor in damped flexible bearings shown in Figure
3 as follows. The rotor system is axisymmetric. and may have
any variation in its axial geometry. At speed it experiences
circular synchronous whirl under the influence of residual
unbalance and modal whirl displacements. The analysis
requires only that the whirl amplitudes and phase angles
should be available at the selected measurement locations. in
accordance with the procedure given below for determining the
speed-dependent influence coefficients a ij (w). These
coefficients relate whirl amplitudes w. to un:kno'WD. rotor
1
unbalances ui. i.e ••

•.u

-
or

w
p
a
pn

{w} = [a] {U}

Assuming the unbalance to be concentrated in n correction
planes. the balancing procedure is to first run the
unbalanced rotor at speed •w1 ' and to measure the whirl
amplitudes w and phase angles f at p locations along the
rotor. This gives:
Flexible Rotor Balancing 107

where 111 = 111 (cos 111t + i sin 111t) = 111 + i 111 as shown in
Figure 3. Now insert a trial weight T fh corre!tion plane 1.
and re-run the rotor at 111 1 • Again read w and f which
correspond to:

r-~~ ·~~
1
1
aA1 •A2 • • • u1 + T
1 1
I WB1 aB1 aB2 •Bn u2

,_

Now

r VAl-
I VAO
1
1
aAl
1
.A2
1
•An

1
u 1 + T- u 1

-u2
··o
1 1 1 1
aB1 .B2 •Bn u2
r~l-
-
1
•p1
1
•p2
-ull
108 N.F. Rieger

Thus

1 1
1 WA1 WAO
·~u • T

-
1
1 WB1 wBO
.,1 T

-
1
wp 1
T

The balance weight is then removed from correction plane 1,

and is then successively inserted in the remaining (u - 1)
correctioo planes. Successively re-running the rotor at w1
and measuring w and fl for each trial balance gives, .iJI
general:

1 < i < p
1 < q < n
Flexible Rotor Balancing 109

This procedure must be repeated q = (n/2) (n even) and q =

(n+l/2) (n odd), where n is the number of correctioo planes.
The resulting simultaneous equations define the rotor
residual unbalance, referred to the correction planes:

1 up
a
pn

82
An
a2
Bn

- \
\

wpn un
110 N.F. Rieger

from which {U} = [a]-1 {w}. The insertion of n correction

masses, each equal in magnitude but opposite in sense to the
corresponding resultant unbalance at these locations has the
effect of cancelling the original unbalance of the rotor.

The above procedure has been described by Rieger [24] who

made an analytical study of the effectiveness of the
influence coefficient method. Three practical rotor-bearing
systems were examined, (a) rigid rotor in gas bearings, (b)
supercritical flexible three-disk overhung rotor in fluid-
film bearings, and (c) supercritical three bearing rotor
with one disk overhung, in fluid-film bearings. The influ-
ence of measurement errors and correlation weight install-
ation errors on the resulting balance was studied, along with
balance improvement with two, three and four balance planes.
The number of bearing supports involved was shown to exert no
influence on the quality of balance attainable. Bearing
misalignment may affect the location of the system critical
speeds and shaft bending stresses, but exerts little effect
on the quality of balance attained unless the whirl ellipses
are excessively elongated, or unless the ellipse axes are
oriented at different angles by the misalignment and bearing
damping.

This method has been verified experimentally for a flexible

three-disk rotor operating through its bending critical speed
by Tes.sarzik [25] who performed the balance weight calcu-
lations with a computerized procedure. The rotor-bearing
system used was designed to contribute negligible damping to
restrain the rotor whirl amplitudes at the bending critical
speed. After balancing, maximum whirl amplitudes of 1.6 x
10- 3 inches peak to peak were observed at this critical
speed, thus demonstrating the effectiveness of the balancing
technique used. Tessarzik, Badgley, and Anderson [26]
discuss these results in detail. A least-squares influence
coefficient procedure also has been proposed by Goodman [27]
for obtaining a best-fit balance for a rotor which operates
over a speed range which may contain several critical speeds.

Experimental results using a least-squares balancing approach

have been obtained by Lund and Tonne son [28], which verify
the effectiveness of this method. A further development has
been given by. Little [29] in which linear programming was
used to optimize the balance of rotors operating through
several bending critical speeds. Baier and Mack [30] have
described the ba lane ing of long helicopter drive shafts
through six critical speeds to beyond 7000 rpm. These
authors achieved smooth shaft operation using influence coef-
ficient balancing after having previously tried several other
flexible rotor balancing methods.
Flexible Rotor Balancing 111

The influence coefficient method has the advantage of

simplicity in application, which makes it suitable for a wide
range of complex turbomachinery applications (helicopter
shafts, mul tis pool aircraft engines, ultra-centrifuges,
etc.), plus the convenience that the correction mass calcu-
lations may be computerized. Effectiveness of this method is
not influenced by the presence of damping in the system, nor
by vibratory motions of the locations at which readings are
taken. Initially-bent rotors may be balanced as readily as
straight rotors, and no assumptions concerning perfect
balancing conditions are involved to detract from the quality
of balance attained. It shares the shortcomings of certain
other methods, i.e., the number of readings required to
acquire the input data, and the accuracy with which these
amplitude and phase readings must be made. Existi~g versions
of the Influence Coefficient method assume circular whirl
orb its. The influence of elliptical orbits from certain
fluid-film bearing conditions, or from dissimilar pedestal
stiffness, is presently not precisely known, nor are effects
drive torque changes (causing shaft wind-up), changes in
bearing operating eccentricity, stiffness, etc., and system
non-linearities.

Goodman [27] also proposed a balancing procedure which uses a

least-squares technique to minimize the rms residual ampli-
tudes at selected locations on the rotor. A second appli-
cation of the technique then uses weighted least-squares to
reduce the maximum residual vibration. For a rotor which has
n balance planes, and on which m vibration readings (m > n)
have been obtained for k different conditions, at p different
locations. Then m = k x p. The least-squares balancing
procedure finds the optimum size and arrangement of the
required balance correction masses and angular orientations
in the n balance planes, by minimizing the sum of the squares
of the m vibration readings. Initially, the uncorrected
unbalance data w0 and the trial weight unbalance data wi at
the m measuring locations are obtained. Influence
coefficients ai. are then calculated as described previously.
However, the p~evious case is for m = n, and the required
values of the correction weights were computed directly.
This reduces the whirl amplitudes to zero at p locations at
the selected speeds, and generally reduces the amplitude of
rotor whirl throughout the speed range: see Figure 4. At
speeds other than the balancing speeds, a small residual
unbalance ei remains such that at the nth location and speed:

wem • wem
r + i wi
em • wmo + am1u1 + . . . + amNUN

CD

- wmo +
n•
~ a U
mn n
112 N.F. Rieger

where the response amplitudes w • w • the influence

ilm mo
c o e f f i c i en t s a (Ill ) • an d the un b a 1 an c e s U are complex
quantities. M..iftiplying out and dividing tfi'"is expression
in to its real and imaginary components gives:

wr • w r + a r ur - a 1 U 1+
em mo m1 1 m1 1

w 1 - w i + a 1 u 1 + a m1 1
r u1r + . . • - wmo
em mo m1 1

(a 1 u 1 + arur).
mnn mnn

Let:

M
s
-I +

The objective is now to select the balance weights for which

S is a minimum. To do this requires that:

-~- ..li. - 0
~w r ~w 1
n n
Flexible Rotor Balancing 113

This leads to 2n linear equations of the form:

\m
L Jl a mnr L- wonr + 'n') (• mnr urn - 8 1u1)]
mn n
+ a i L- w
mn on
1

I{- •nit[wo1n + L(a~UJ - •JnU~] + •nfn [ wJ-n

m

+I (•aitU~ + •r!'nU~J• 0
n

The unknowns in these equations are the components of the

balance weights Unr and Uni required to minimize the rotor
response. These terms may be found using standard equation-
solving routines when the above procedure is programmed.

Goodman described a first iteration using the above

procedure, and then a weighted least-squares procedure which
may be applied in several successive iterations to minimize
the residual unbalance following each iteration, until a
satisfactory final balance is achieved. This final set of
balance corrections and orientations may be obtained auto-
matically by continuing the iterations until a pre-specified
balance criterion is achieved. Further discussion of
computerized balancing is given in Goodman [24], Rieger [21],
Lund and Tonnesen [25], and Tessarzik, Badgley, and Anderson
[23].

4.2.7 Other Analytical Balancing Procedures

Several other flexible rotor balancing techniques have been

proposed, a number of which are variations of the modal
balancing procedure. In most instances the shortcomings of
these techniques lie not in the theoretical approach, but in
the practical acquisition of suitable measurement data
required to implement the method.

Church and Plunkett [31] applied the mobility method to the

balancing of a simple uniform flexible rotor. They gave an
an a lysis for determining the magnitude and orientation of
correction weights. Shaft deflection is represented in terms
of normal modes for a single span shaft in rigid bearings,
without damping. The proposed method requires that a set of
modal response curves should first be obtained, preferably by
analysis. The static deflection and the dynamic deflections
114 N.F. Rieger

of the shaft are then measured. The vector difference

between these measurements allows the modal equations to be
solved. The required balance corrections are then computed
from the forces calculated to be causing the dynamic
deflections. J!or the test example given, practical diffi-
culties caused the authors to resort to shake-test measure-
ments to obtain the desired mobility response. Recent
experiences with helicopter drive shaft balancing suggest
that this was probably due to the flexibility of the test
model used, and to the measuring teclmique employed. As.
presented, the method was restricted to a single span rotor,
but in principle the method should apply equally well multi-
span rotors in flexible bearings.

Den Hartog [32] presented an interesting discussion of

flexible shaft balancing in which the requirements for
perfect balance were reduced to a theorem. Assuming that
perfect rotor balance is attainable, the bearings will then
function as simple rigid supports, as no transverse whirl
motions will then occur to generate dynamic displacements
(and hence forces) at these locatioo.s. This assumption
allows the related conditions for perfect balance to be
deduced directly. Equations for balance equilibrium were
then established, first for a two-bearing rotor, and then for
multi-bearing rotors. Several numerical examples are
included. For clarity, the theory is developed with all
unbalance occurring in the same plane. The effects of
damping are ignored (no bearing velocity) though the author
recognizes that practical balancing may involve three-
dimensional deflections of the rotor. The theory also
assumes axial symmetry throughout the rotor-bearing system
and small initial unbalance. The validity of the balance
condition depends to some extent on the degree to which these
conditions can be met in a given situation. No means
(analytical, or through test examples) for assessing the
effect of practical departures from the stated requirements
of the method, are included.

More recently, Hundal and Harker [33] gave a modal balancing

analysis of a flexible rotor with noo.-uniform stiffness and
mass distribution, in rigid end bearings. This method also
requires that the critical speeds and normal modes should
first be calculated. The whirl amplitudes along the shaft
are then measured at a selected speed, to calibrate the modal
amplitudes. The rotor unbalance is then determined, and so
balancing corrections may be selected in the desired planes.
This method may also be extended to multi-bearing rotors in
flexible bearings, though this does not appear in the theory
presented. The theory has been developed by considering all
unbalance as located in the same plane, for simplicity.
However, it is noted that practical balancing involves
finding independent solutions in two mutually perpendicular
Flexible Rotor Balancing 115

planes, and subsequently combining these solutions to obtain

the required balance corrections.

A recent review paper by Findlay [34] contains comments on

the Modal balancing method, and on the analysis and experi-
ments of Hundal and Harker [33]. The fundamentals of the
Modal approach are first discussed, followed by a discussion
of the practical limitations of the Modal method. It is
noted that the normal modes and natural frequencies must be
determined accurately, and that the test measurements must be
taken mode-by-mode to achieve the required accuracy for the
balancing calculations. Questions unanswered by Hundal and
Harker are: How to select balance planes and correction
masses in combination to optimize balance: special problems
where two critical speeds exist close together: and details
of numbers of balance runs and measurements needed. Findlay
has not commented on effects arising from an asymmetrical
rotor, nor a multi-bearing rotor system (though this is
implied by the Hundal-Harker analysis), but several qther
balancing procedures are compared in general terms with the
Modal balancing method: in particular, the Influence
Coefficient method, and the least-squares procedure.
Findlay's conclusion is that modal balancing is not a
generally applicable balancing technique, because of serious
practical difficulties involved in acquiring useful response
data, and because of its reliance on calculated modal inputs.

Gusarov and Dimentberg [35] studied the dynamic effect of

distributed and concentrated unbalance forces on the
balancing of rotors. The balancing problem for flexible
rotors with a limited number of available correction planes
is studied. The absence of dynamic reactions at the supports
and of optimum reduction of rotor bending stress is the
objective of the balancing procedure. Damping effects on
balance quality are considered.

Gusarov [36] further investigated the problem of eliminating

the first two modes of vibration for an unbalanced rotor
operating at speeds below its second critical speed.
Balancing is to be achieved using two correction planes. It
is supposed that the influence of higher modes at low speed
can be neglected. The question of optimal placing of the
correction planes along the shaft length is also investi-
gated. A shaft of constant cross-section without disks is
considered. In cases where it is only necessary to remove
rigid rotor modes, two correction plans are evidently
sufficient to significantly impro,ve the rotor balance
quality. Similarly, a flexible rotor in rigid bearings may
be balanced using two correction planes for two modes. The
more general case of a flexible rotor in flexible bearings is
noted to require more planes.
116 N.F. Rieger

4.2.8 Criteria for Flexible Rotor Balancing

An International Standards organization document is now

available to guide the balancing of flexible rotors.
Flexible rotor balancing concepts are pres en ted in the I SO
Document ISO/TC108/SCI:: 'Mechanical Balancing of Flexible
Rotors' (1976) [37]. This document proposes a general
classification of rotor types into specific Classes I through
V, as shown in Table 4.2.1 from this ISO document. Specific
rotor types are classified in accordance with their operating
requirements, and guidance is provided on the selection of
the most suitable method of balancing the rotors in each
Class. Of the five recommended Classes, Class II rotors are
quasi-flexible rotors, which may be corrected for smooth
operation by the addition of masses in two balance planes.
Only Class III rotors are truly flexible rotors in the sense
that they require the use of some established flexible rotor
balancing method to provide smooth operation. Flexible
rotors are observed to require more than two correction
planes to achieve smooth operation throughout their speed
range. All other Classes of rotors may be adequately
balanced by either two-plane techniques, or by simple three-
plane balancing. Document ISO/TC108/SI1 (1976) is concerned
with the adaptation of rigid rotor balancing concepts,
equipment, and balancing machines to the balancing of high
speed, flexible, two-bearing rotors. Brief comments are
given to coupled rotor balancing. No detailed classification
of flexible rotor types is pres en ted (multi-bearing, multi-
spool, etc.). Both the Modal method and the Influence
Coefficient method are described in detail.

Criteria for flexible rotor balancing are presented in ISO

Document ISO/TC108/SCI: (1976), with a tabulation of typical
numerical values for bearing velocity for the various rotor
classes and measurement procedures used, to guide the
selection of acceptable support vibration response. These
criteria are reproduced in Table 4.2.2. This table accounts
for rotor class, machine type, measurement location, and
balance quality details. These criteria (readout velocity
mm/s) should be regarded at typical, and are consistent with
other machinery vibration criteria values for each rotor
class.

4.2.9 Flexible Rotor Balancins: State-of-the-Art

Modal Balancing:

The Modal balancing method is the aost developed and proven

flexible r9tor balancing method available in the open
literature. Modal balancing has been discussed by many
authors, but the most extensive contributions have been made
by Bishop and his co-workers. These investigators have
Flexible Rotor Balancing 117

successfully applied Modal balancing to turbine-generator

rotors, and their techniques are used in current practice.
Modal balancing methods have also been applied for problems
of dissimilar lateral stiffness, hysteretic whirl, and to
complex shaft-bearing problems, such as a shaft with
dis tr ibu ted dis similar stiffness and mass properties
operating in bearings with dissimilar x- and y- stiffnesses.
Effective balancing bas been achieved under laboratory
conditions, in these instances. Such developments are in
advance of the commercial applications of modal balancing at
present. They are confined primarily to symmetrical two-
bearing rotors.

Effects arising fr.om the bearing fluid film have received

only minor consideration in the modal balancing literature.
The fluid film is the major source of system damping, and it
also governs the rotor stability problem. The damping
contributed by the fluid film strongly influences the phase
angle and its rate of variation near the system critical
speeds. This further contributes to the practical
difficulties of taking effective balancing measurements. It
should also be recognized that rolling-element bearing
effects can be as important as shaft effects in lightly-
damped systems, such as aircraft gas turbines.

Concerning the modal balancing technique itself, the problems

which appear to confront its wider application to more
general i.e. complex, damped, multi-shaft, or multi-bearing
rotor systems are as follows:

o Calculation of modes and frequencies

To use modal methods, information is required on the mode

shapes and natural frequencies of the rotor-bear-ing system to
be balanced.3 The accuracy of the computed results depends
on the capabilities of the computer program used, and on the
input data (dimensions, coefficients, effectiveness of system
model) used in these calculations. In complex turbomachln-
ery, this may create problems, especially where system
damping is significant, such as with fluid-film bearings.

3Den Hartog, Kellenberger, X:ushul ', and Sblyaktin, and

others have required that rotor modes only need to be
obtained, using the criterion that rotor amplitudes at the
bearings are to be balanced to zero. In complex systems this
is frequently impossible, e.g., for an aircraft turbine
damper bearing, and in any case the effects of non-zero
bearing amplitudes on this approach remains unexplored. The
alternative of accepting the existence of bearing amplitudes,
as proposed by Bishop appears to be more realistic and moe
general, judging from practical experience to date.
118 N.F. Rieger

The mode shapes and resonant frequencies of heavily damped

systems often bear little resemblance to undamped mode shapes
and frequencies. The use of a damped response program is
therefore essential in such cases to obtain accurate resonant
frequencies au.d mode shapes. At present. the available
circular orbit response programs are usually adequate for
single-level systems. but few two-level response programs are
ava Hable. The reliance of modal balancing on predicted
modes and frequencies is. at least. an inconvenience. It
could constitute a significant disadvantage for the applica-
tion of modal balancing to complex machine structures
involving multi-spool. or shaft-structure system unless
adequate response programs are available to obtain the
analytical balancing inputs. It should further be noted that
damped rotor response also depends on unbalance distribution.
The influence of an unrepresentative distribution of
unbalance on the accuracy of modal balancing should be
further examined. Evidently a procedure which accurately
calibrates the unbalance distribution model would be a useful
contribution to modal balancing.

o Measurement of Modal Amplitudes and Frequencies

Exper fence has shown the difficulty of trying to extract

amplitude and phase data from broad-band measurements taken
where several modes are present simultaneously. It is well
known that the resonant amplification associated with any
given mode may be utilized to obtain the required modal data
(shape. frequency. phase) by running the rotor at or close to
resonance in that mode. Where sharp or severely resonant
modes are involved. approximate (near-resonant) data may
first be used to obtain an improved initial balance. before a
final balance is attempted. A further measurement problem
results from the rapid variation in phase angle which occurs
in a lightly damped system in the vicinity of resonance.
Even when practical modal balancing is performed on a mode-
by-mode basis. it is evident that a certain residual whirl
amplitude will still result at operating speed. due to the
effect of the higher modes. Recognizing this situation.
Parkin son and Bishop [15] proposed a correction for such
modes which uses data acquired at the rotor operating speed.

Multi-Mode Balancing:

The Modal theory suggests that several modes may be balanced

simultaneously. but this does not appear to have been
achieved in practice, Bishop and Gladwell [11] recommend
balancing mode-by-mode. Church and Plunkett [31] resorted to
shaking the rotor to ascertain the mode shapes. Bundal and
Harker [33] were unable to achieve low amplitudes in the
vicinity of the first critical speed. and appear to have been
unable to pass through the second critical speed.
Flexible Rotor Balancing 119

Influence Coefficient Uethod:

S u c c e s s f u 1 a p p 1 i c a t i on s o f t he I n f1 u en c e c o e f f i c i en t
technique appear to include a wider range of rotating
machinery types than those which have been reported for other
flexible rotor balancing methods. Known applications
include: a high speed (24000 rpm) pump simulator, a long
helicopter drive shaft (20 ft. unsupported span, six critical
speeds), certain ultracentrifuge applications, several small
steam turbines, and small aircraft gas turbine applications.
Notable features reported in these applications were:

o Convenience of Application

The method is widely recognized as being simple to apply, but

it usually requires the acquisition and careful analysis of a
large amount of readout data. 2n sets of amplitude and phase
angle data are required for the Exact-Point-Speed method, and
preferably several more than 2n sets for the least-squares
method. This data acquisition is fairly straightforward for
operation through one or two critical speeds, but the amount
of data required in cases where the speed range involves many
cr it icals, i.e., helicopter drive shafts, may become very
large. Some automated form of data-taking and recording on
magnetic tape is evidently advantageous, and preferably
arranged in a form sui table for direct use as input for the
balance mass and angle calculations which follow. With such
an arrangement, the Influence Coefficient method can provide
a rapid and efficient balancing procedure.

o Accuracy of Balance Attainable

The factors which limit the present form of the Influence

Coefficient method are (a) the precision to which the
measurements of amplitude and phase may be acquired, (b) the
assumption of a circular rotor whirl orbit in the present
procedures, and (c) the constancy of the test set-up, when
readings are being taken. These shortcomings are, of course,
inherent in every flexible rotor balancing method yet
proposed. To remove them will require: for (a) the
development of a precision electronic data-sampling system
which is capable of reading and storing amplitude and phase
data at all rotor locations simultaneously, (b) development
of advanced analytical procedures which include elliptical
orbit effects in the rotor balancing procedure, and for (c)
a programmed statistical teclmique for evaluation of the
sampled amplitudes, phase and speed data. Experience shows
that the acquired input data may vary substantially, even
over a very short period of time. <kle reason for this
variation is rotor speed fluctuations, which may be
especially serious in the vicinity of a critical speed. Each
data reading could be electronically samplea many times over
120 N.F. Rieger

a fairly short time period (1000 revs.= 1000 readings). The

statistically-analyzed results could theu be used to compute
the required balance weights and angles to give a refined
statistical balance. An indication of the balance accuracies
attainable with preseu t methods is giveu in the results of
Bishop et. al.. Hundal and Harker. and Church and Plunkett.
The effect of errors of measuremeu t and of balance weight
installation on the quality of balance obtained has been
studied by Rieger [24].

Other Methods:

Most of the other methods proposed have been variations of

the Modal balancing method. Where experimental verification
has been sought. tests have usually beeu made on long slender
shafts. The success achieved in most cases has been
marginal. This has usually been due either to inherent
limitations in the techniques themselves or to practical
difficulties of measuring the modes. including the skill of
the opera tors. Other practical problems of amplitude and
phase measurement. speed fluctuation. data variation. and of
shaft crookedness. etc •• have also detracted from quality of
correlation achieved.

Other techniques such as the empirical one-shot method are

restricted to the applications for which they were developedL
and s~em to offer little promise of further development.
Still other methods require that amplitudes should be reduced
to zero at the bearings. which may not always be possible due
to system restrictions in cases involving flexible bearings.
e.g.. with an overhung rotor. for multi-bearing rotors. and
with variation the operating drive loads. etc. The effect of
practical factors such as these do not yet appear to have
been investigated. Finally. suggestions such as the zero-
amplitude method which depends on accurately calculated whirl
amplitudes for selecting the balance weights and angles. do
not appear to offer any practical advantage.

4.2.10 Comment

It should be noted that the flexible rotor balancing

techniques discussed in this section are effective only
against synchronous rotor residual unbalance. In each
instance. some procedure for filtering out. and measuring the
synchronous unbalance component is an esseu tial part of the
balancing technique. Where a machine has becomes unstable in
its bearings. no amount of balancing will restore its
stability. Rotor instability is an entirely separate
phenomenon for which procedures other than balancing are
needed.
Flexible Rotor Balancing 121

4.2.11 References

1. Linn, F. C., 'Method of and Means for Balancing Flexible

Rotors', u.s. Patent 1,776,125, Filed September 17,
1928.

2. Thearle, E. L., 'Dynamic Balancing of Rotating Machinery

in the Field,' Trans. ASME, Vol. 56, pp. 745-753, 1934.

3. Kroon, R. P., 'Balancing of Rotating Apparatus,' Part I,

Journal of Applied Mechanics, Vol. 10, pp. •-225, 1943.
Part II, Journal of Applied Mechanics, Vol. 11, pp. A-
47. 1944.

4. Grobe!, L. P., 'Balancing Turbine-Generator Rotors,' GE

Review, Vol. 56, No. 4, pp. 22, 1953.

5. Moore, L. S., Dodd, E. G., 'Mass Balancing of Large

Flexible Rotors,' GEC Journal, Vol. 31, No.2, 1964.

6. Lindsey, J. R., 'Significant Developments in Methods for

Balancing High Speed Rotors,' ASME 69-Vibr-53, Vibra-
tions Conference, Philadelphia, PA, March 1969.

7. Meldahl, A., 'Auswuchten Elastischer Rotoren,' Z angew,

Math und Mech. 34, 1954.

8. Timoshenko, S. P., VibratiOD Problems in Bngineerinas, D

van Nostrand Company, Inc., Princeton, NJ, 3rd Edition,
1955.

9. Bishop, R. E. D., 'The Vibration of Rotating Shafts,'

Journal of Mechanics and Engineering Sciences, 1, 50,
1959.

10. Gladwell, G. M. L., Bishop, R. E. D., 'The Vibration of

Rotating Shafts Supported in Flexible Bearings,' Journal
of llechanical Science, 3, 195, 1959.

11. Bishop, R. E. D., Gladwell, G. M. L., 'The VibratiOD and

Balancing of an Unbalanced Flexible Rotor,' Journal of
Mechanics and Engineering Sciences, 1, 66, 1959.

12. Bishop, R. E. D., Parkinson, A. G., 'On the Isolation of

Modes in the Balancing of Flexible Shafts,' Proceedings
IMechE, 177, 16, 407, 1963.

13. Kennedy, C. C., Pancu, D. D. P.. 'Use of Vectors in

Vibration Measurements and Analysis,' Journal of
Aerooautical Science, 14, 1947.
122 N.F. Rieger

14. Lindley. A. G., Bishop, R. E. D., 'Some Recent Research

of the Balancing of Large Flexible Rotors,' Proceedings
IMechE, 177, 30, 811, 1963.

15. Parkinson, A. G., Bishop, R. E. D., 'Residual Vibration

in Modal Balancing,' Journal of Mechanics and Engineer-
ing Science, 1, 33, 1965.

16. Bishop, R. E. D., Parkinson, A. G., 'Second-order

Vibration of Flexible Shafts, ' Phil. Trans. Royal
Society of London, 259, Series A, 1, 1965.

17. Bishop, R. E. D., Mahalingam, S., 'Some Experiments in

the Vibration of a Rotating Shaft,' Proceedings Royal
Society of London, 292, Series A, 537, 1966.

18. Parkinson, A. G., 'OD the Balancing of Shafts with Axial

Asymmetry,' Proceedings Royal Society of London, 292,
Series A, 66 1966.

19. Parkinson, A. G., 'The Vibration and Balancing of Shafts

Rotating in Asymmetric Bearings' Journal of Sound and
Vibration, 4, 477, 1965.

20. Parkinson, A. G., 'An Introduction to the Vibration of

Rotating Flexible Shafts,' Bulletin Mechanical Engineer-
ing Education, 47, 1967.

21. X:ushul, M. Ya., Shlyaktin, A. V., 'Modal Approach to

Balancing with Additional Constraints,' Isvestiya
Akademii Nauk SSSR, Mekhanikai Mashinostroyeniye, No. 2,
1966.

22. !:ellenburger, W., 'Balancing Flexible Rotors on Two

Generally Flexible Bearinss,' Brown Boveri Review 54,
No. 9. 1967, pp. 603-617.

23. Goodman, T. P., 'A Least-Squares Method for Computing

Balance Corrections,' ASME 63-WA-295, September 1964.

24. Rieger, N. F., 'Computer Program for Balancing of

Flexible Rotors,' Mll Report 67T.R68, September 1967.

25. Tessarzik, J. M. 'Flexible Rotor Balancing by the Exact

Point-Speed Influence Coefficient Method,' NASA Tech-
nical Report CR-72774, Lewis Research Center, October
1970.

26. Tessarzik, J., Badsley, R. B., and Anderson, W. J.

'Flexible Rotor Balancins by the Exact Point-Speed
Influence Coefficient Method,' Paper Submitted to ASME
Third Vibrations Conference, Toronto, Canada, September
1971.
Flexible Rotor Balancing 123

27. Goodman, T. P., 'Correction of Unbalance by Force-

Canceling Bearing Pedestals,' GE Company Report No.
61GL110, May 15, 1961.

28. Lund, J. W., Tonneson, J. 'Analysis and Experiments on

Multi-Plane Balancing of a Flexible Rotor,' ASME Third
Vibrations Cconference, Toronto, 1971.

29. Little, R. M., 'Current State of the Art of Flexible

Rotor Balancing Technology,' Ph.D. Thesis Bibliography,
University of Virginia, Charlottesville, 1971.

30. Baier, R., Mack, J •• 'Design and Test Evaluation of a

Supercritical Speed Shaft,' The Boeing Company, Vertol
Divis ion, Morton, PA, USAAVLABS Technical Report 66-
49/R458, June 1966.

31. Church, A. B., Plunkett, R., 'Balancing Flexible

Rotors,' Trans. ASME, Journal of Engineering for
Industry, Vol. 83, Series B, No. 4, pp. 383-389,
November 1961.

32. Den Hartog, J. P., The Balancina of Flexible Rotors,

Air, Space and Instruments, Stark Draper Commemoration
Volume, McGraw-Bill, 1963, pp. 165.

33. Bunda!, M. S., Barker, R. J ., 'Balancing of Flexible

Rotors Having Arbitrary Mass and Stiffness Distri-
bution,' Journal of Basic Engineering, Trans. ASME,
Paper No. 65-MD-B, June 1965.

34. Findlay, J. A., Review of paper: 'Balancing of Flexible

Rotors,' by Hundal and Harker, ASME Paper 65-MD-8.

35. Gusarov, A. A., Dimentberg, F • .M., 'Balancing of

Flexible Rotors with Distributed and Concentrated Mass,'
Proceedings, Procnosti v Mashinistroenii (Problems of
Elastic! ty in Machinery) Publication Bouse Academy of
Sciences, 1960, Moscow No. 6, pp. 5-37.

36. Gusarov, A. A., 'On Placing of Balancing Planes on

Flexible Rotor,' X:olebania i Prochnost pri Peremanyh
Naprj en iah (Vibrations- and Elasticity for Variable
Stresses), Proceedings, National Research Institute of
Machinery, Soviet Academy of Sciences, PublicatiOD
House, Nauta Moscow, 1965, pp. 112-124.

37. ISO Draft Document, 'The Mechanical Balancina of

Flexible Rotors', I SO/TC 108/SC1/'IG2 (Secretariat-7)
Document 12, 1976.
124 N.F. Rieger

~II, 1930. rr. C. UNN

- W - II&US POl IAI.UCI . . - ·

ru .. ...,. 17, 1921

Inventor:
rrani! c linn
by~~
H.. Attorney

Figure 1 Title Sheet of Linn Patent

for Flexible Rotor Balancing
(1928)
Flexible Rotor Balancing 125

DISK MASS AND

INERTIA PROPERTIES,
' M, Ip• IT

BEARING
COEfFICIENTS
PEDESTAL
COEFFICIENTS

Figure 2 Y-Z Distribution of Mass, Elasticity, and

Residual Unbalance Within a Generalized
Rotor (X-Z Distribution is Similar)

BALANCE
PLANE
r.........
I .........
BALANCE I ""l
PLANE I }
~
PROXIMITY
PROBE

Figure 3 Representation of Axially-Distributed Rotor

Unbalance by n Discrete Unbalance Moments
126 N.F. Rieger

100
ORIGINAL UNBALANCE =
I \
10 I \
_, '\.
L

1.0
..,..,....... / '
/
"""' !'-...
~THREE-PLANE BALANCE.
~ READINGS AT OVERHUNG DISK._+- READINGS TAKEN AT
BEARING AMPLITUDE 'RESPONSE. BEARINGS. BEARING -A-
THREE-PLANE BAL.-
=
I I I ~ :. AMPLITUDE RESPONSE.
I -THREE-PLANE BALANCE. Iii I
=READINGS AT OVERHUNG DISK. - f h ·

---
~OVERHUNG DISK RESPONSE.

---
--- --- -- ---
."'-... _\
... \
""' K ==---
~
1\.. •·•

r- FOUR-PLANE BALANCE.
~READINGS AT OVERHUNG DISK. ' -.... ·. .. .•·
.·
r-- OVERHUNG DISK RESPONSE.
I I I '·'\. \ ..... /
', .. / /
10-3
4000 5000 6000 7000 8000 9000
.10,000
\ \ //
'
11,000
SPEED, RPM

Figure 4 Effect of Number of Planes on Balancing

Response at Bearings and Overhung Disk
Flexible Rotor Balancing 127

Table 1 Flexible Rotor Balance Criteria From ISO

Draft International Standard Document
5343 (1978)

Rotor Ranges of effective pedest•l vibration velocity Correction

Category at once-per-rev frequency Ft~ctor
v (mm /sec) nns
o. 2s I o. qs I o. 11 •I u l1. 1
l1. 12l1. al2. 11.2 11 21 45 71 c c, c,
1
A
I B I c I D II II
Small electric motors up to 20 HP.
Superchargers. ;~" ~. 0. 63
D. 63 2

A. I B I c I D r

Paper making machines 0.63

Medium size electric motors & generators,
II 2o-1 00 HP on normal foundllions. 0.63 4
Electric motors and generators up to

•'' •
400 HP on special foundations. 0.63 2t
Pumps and compressors. 0.63 15
Small turbines. 0.63

A I a I c 0
Ill Large electric motors. 0.63 5
Turbines and generators on rigid and heavy
foundations. 0.63 5 20
A I a I c I D
Large electric motors, turbines, and generators
IV on lightweight foundations. 0.63 3 10
Small jet engines. 0.63

A I B c D
v
Jet engines larger than category IV. 0.63 2 ID
 CHAPTER 4.3

CASE HISTORIES IN BALANCING OF HIGH SPEED ROTORS

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

Three case histories concerning rotor balancing

problems from practice are described in detail.
These case histories refer to problems in which the
cause was rotor unbalance and/or unsuitable rotor
dynamic conditions. In each instance. a diagnostic
phase preceded the eventual corrections made to the
system. These cases demonstrate both the diag-
nostic techniques used. and the successful correc-
tions made.

4.3.1 Introduction

The case histories described in this chapter demonstrate the

application of diagnostic techniques to balancing problems
and the successful application of balance correction proce-
dures to practical machine situations. In all instances it
is important that a correct diagnosis of the problem should
precede any attempt to apply a solution. The following
examples are drawn from actual experience. They demonstrate
both the procedure used for identification of the problem
cause. and the balancing procedure which was subsequently
used.
130 N.F. Rieger

4.3.2 Case History 1 - Thermal Unbalance Vibrations of a

Generator Rotor

Severe vibrations of the generator casing structure shown in

Figure 1 were observed. End plate amplitudes up to 11.0 mils
were felt adjacent to the generator bearings. It was
considered that the fabricated end plates would experience
fatigue failures of the welds if operation in this condition
was continued.

The generator had a maximum output of 4000 kw, and it was

operated at 3600 rpm to produce 60 Hz power for a paper mill.
The generator casing was of standard welded steel fabri-
cation, and the generator rotor was direct-driven by a steam
turbine unit. Both the generator rotor and the turbine rotor
were supported in two elliptical oil film journal bearings.
The generator rotor was driven through a Falk gear tooth type
coupling. The entire unit was located on the third floor of
a concrete frame building, mounted on a steel bedplate, and
supported upon massive reinforced concrete beams and columns.

A chart of axial vibration amplitude response vs. frequency

at the generator number 4 bearing is shown in Figure 2, after
the first stiffening fix had been applied to the generator
end plates. A frequency spectrum for the number 4 bearing in
the vertical direction for operation at 3600 rpm is shown in
Figure 3.

Initial operation had indicated that strong vibrations were

occurring at the generator bearing supports, in the
horizontal, axial, and vertical directions. The end plates
were subsequently stiffened. This reduced the vibration to
the levels shown in Figure 2, but did not eliminate the
vibration.

It was observed that the vibrations typically developed some

time after~unit startup. Initially, the unit ran smoothly.
Both the turbine rotor and the generator rotor had recently
been balanced in a local shop. Once started, the vibrations
persisted, and the unit eventually had to be shut down.

Characteristics of the Problem:

a) Axial vibrations of the generator end plates was

the worst aspect.

b) Vibrations were worst under high generator load

conditions. Low load vibration was typically low
amplitude.

c) Vibrations were worst when structure temperature

was high.
Case Histories in Balancing of High Speed Rotors 131

d) End plate axial mode natural frequency was around

68-71 Hz.

e) Rotor system natural frequency was around 34 Hz.

f) .Measured vibration frequency was mostly syn-

chronous, at 60 Hz.

g) Other harmonics were typically smaller than the 60

Hz component in the measured vibration frequency
spectra.

Possible Causes: The following possible causes were each

considered in detail during the problem diagnosis phase.

a) Residual rotor unbalance arising from:

Uneven mass distribution
Bent shaft
Thermal bow
Elliptical journals

b) Bearing fluid film instability due to:

Half frequency whirl
Resonant whip

c) .Magnetic structure interaction:

Waveform harmonics
Structure shell modes
Eccentric armature

d) .Misalignment:
Hot vs. cold alignment
Coupling forces
Thrust bearing force
Foundation distortion
Multi-bearing support alignment

e) 'Flat' Shaft effect:

1 /rev, 2/rev, possible 3/rev
4/rev

Observations on Possible Causes:

a) Mechanical residual unbalance can be discounted as

a serious cause because the startup vibration at no
load was small and within acceptable limits. Bent
shaft unbalance effect was likewise rejected as a
possible cause for the same reason.
132 N.F. Rieger

b) Sequential testing (removal of excitation,

increased generator cooling, decreased oil inlet
temperature) of the system indicated that:

Magnetic effects were not a major cause of

exc"itation, or their removal would have
minimized the vibration immediately.

Thermal effects could be likely cause. Under

cooler operating conditions the observed
vibration decreased in each instance.

c) Fluid film instability is discounted because:

No sustained half-speed frequency component

was consistently observed in the test
frequency results. Therefore there was no
half-frequency whirl occuring in the system.

No strong 34 Hz component was observed in the

bearing frequency spectrum, when operating at
3600 rpm.

The rotor operates below twice its bending

critical speed: therefore the cause is not
resonant whip.

d) No evidence of any strong foundation vibration was

found. Measurements showed that vibrations were
strongest at the generator bearing caps, and were
small elsewhere in the concrete foundation.

e) Flat Shaft effect: Exactly 2x, 3x, 4x, etc.,

harmonics observed in the vibration spectra suggest
that some flat shaft effect may be present in the
generator, most likely due to the two-pole rotor
cons'truction.

f) Hot vs. cold rotor alignment checks were made to

identify possible casing distortion. No
significant misalignment was found.

g) Coupling alignment was checked by sling test. Both

the drive coupling and the exciter coupling were
found to be well aligned.

h) Output current waveforms could also be checked.

Casing 'shell' modes are typically high, but they
have significant structural damping due to coupled
winding movements. No evidence of an eccentric
armature was found other than a strong 60 Hz
component, as expected.
Case Histories of Balancing of High Speed Rotors 133

i) Elliptical journal may cause 2/rev radial vibration

but not 1/rev.

Natural Frequencies of End Plates:

Vibration measurements were made on the generator pedestals

and on both sets of end plates. Finite element natural
frequency calculations were made for (a) the original
unstiffened end plates, (b) for the first stiffening of the
end plates, and (c) for a proposed second stiffening. The
originally stiffened construction is shown in Figure 1.
Details of the finite element model are shown in Figure 4.
The first mode of this structure is shown in Figure S. The
first mode is a d.iaphragm type vibration mainly involving
mot ions of the upper end shield. The second mode has a
vertical diameter with anti-symmetrical motions.

The original end plates had a measured natural frequency of

63.5 Hz. This agrees with the calculated natural freq~ency
of 64 Hz. Test results showed that the lowest natural
frequency of the initially stiffened end plate structure was
at 68 Hz at the turbine bearing end, and approximately 71 Hz
at the generator bearing end. The finite element result for
the initially stiffened case was 69 Hz, with reinforcing
members added to the original F.E. model. It also is noted
that the measured second natural frequency of the end plates
occurred at 86 Hz for the turbine end bearing, and at 87 Hz
for the generator end bearing. The calculated second natural
frequency of the end plates occurs at 84 Hz.

The correlation obtained between the test results and the

calculated results for natural frequencies is close enough to
validate the finite element models of the original structure
and of the existing stiffened structure.

Positive Indications:

a) Increased generator output power leads to increased

operating temperature. Some time lag occurs until
thermal equilibrium is reached. Vibration
intensity is directly related to higher temper-
atures, and vibration intensity follows temperature
growth.

b) Mechanical unbalance and bent shaft are discounted

as causes of the problem because the shaft runs
smoothly during no-load startup, and when it cools
down.

c) The strongest response component is always at 60 Hz

under synchronized operating conditions.
134 N.F. Rieger

d) Electrical excitation per se appears to have a

minor effect on the vibration, as indicated by the
sequential test results.

e) System. bending critical speed occurs at 34 Hz.

This was observed in the run-down test data. This
means shaft resonance is not aggravating the
problem.

f) Calculated and measured end plate first-mode

natural frequencies occur at 68 Hz and 71 Hz.
Increased end plate natural frequency values
resulted in decreased end plate axial vibration
amplitudes. This suggests that plate resonance is
contributing to the prpblem. It also suggests that
further detuning, i.e., further stiffening, should
lead to smaller end plate vibrations.

g) Thermal effects could cause the shaft to bow. The

relatively slow development of thermal bowing in
the generator rotor due to the slotted generator
construction is consistent with the observed slow
worsening vibration problem.

h) Shaft thermal bowing would also explain the

correlation between high vibration and high
generator power output (when steady state
conditions. exist), and the strong synchronous
component observed in the results.

i) Slow cooling of the rotor and a residual bow would

also explain the persistence of the vibration when
generator load is reduced, and the lack of cor-
relation between generator load and power level.

j) Th~ 2x, 3x, etc. components may be due to either

flat shaft effect (most probably), or to uneven
·electrical airgap effects. Flat shaft effect seems
likely because the multiple harmonics remain after
excitation is removed. Airgap effect could be
checked in detail by examining the electrical
waveform.

k) Bearing (or casing) misalignment (mechanical,

thermal) could cause some 1x effect.

1) Coupling may be occurring between radially

transmitted forces from shaft thermal unbalance
excitation at 60 Hz, and the axial end plate mode
at 68-71 Hz. This coupling could occur because:
Case Histories in Balancing of High Speed Rotors 135

The end plate mode is a zero diameter

(umbrella) mode. This is strongly evident
from measurements, and is confirmed by the
finite element calculations.

The thermally bowed shaft could be applying a

transverse rotating couple to the bearing
support as well as a radial rotating force.
This would tend to excite the end plate in the
observed phase relation of this mode.

The observed anti-phase relations between the

end plate vibration displacements agrees ~ith
the thermal bow/bearing support excitation
hypothesis.

4.3.3 Conclusions

o The observed end plate axial vibrations were caused by

synchronous vibrations transmitted from the generator
rotor.

o The transmitted vibrations were caused by mechanical

unbalance due to thermal bowing of the generator rotor.

o Rotor thermal bowing most likely results from uneven

expansion of the slotted armature, possibly due to coil
binding, local hot spots, etc.

o The end plates respond strongly to this excitation

because the bearings are overhung, and the lowest end
plate mode is close to the synchronous frequency of 60
Hz.

o The higher harmonic components in the response waveform

are due to dissimilar shaft moments of inertia, the
'flat' shaft effect.

o Damping is low throughout the system. This causes

sharply defined high amplitudes responses to occur at 34
Hz and at 68-71 Hz, as observed in the run-down test
results.

4.3.4 Corrections Made to System

o The shaft was mechanically balanced to compensate for

the thermal bow at full power and 3600 rpm. This caused
some increase in the transmitted cold vibration ampli-
tudes. The thermal balance correction was adjusted to
reduce the hot vibration amplitudes to minimal values
consistent with very smooth operation, and low axial
vibration of the end plates.
136 N.F. Rieger

o The end plates were further stiffened with members which

oppose the first mode shape. This further detuned the
end plate modes. This backup fix was recommended in
case the shaft becomes otherwise thermally unbalanced
sometime in the future.

4.3.5 Case History 2 - Unbalance Vibrations of a Boiler

Feed Pump Motor

A schematic of a boiler feed pump rotor is shown in Figure 6.

An unslotted forging of this rotor was successfully balanced
in a test stand. The balanced rotor forging ran smoothly.
close to its observed first and second critical speeds,
Figure 7. and at the intended operating speed of 2300 rpm.
Problems appeared when a final balance of the completed rotor
was attempted. The rotor was successfully balanced for
operation near its first critical speed, but severe
vibrations were encountered as the operating speed was
approached. It was assumed that these vibrations were caused
by the second mode of the rotor, which lay beyond the rotor
overspeed condition. An attempt was then made to correct the
second mode by inserting correction masses on opposite sides
of the rotor, near the points of maximum second mode
deflection, Figure 8. It was found that one arrangement of
correction masses could be selected to make one or the other
bearing run smoothly. but the remaining bearing would then
experience rough vibrations, as shown in Figure 9. With
these corrections, none of the operating speed vibrations
could be brought within acceptable limits. It was therefore
important to locate the root cause of these vibrations, to
balance the rotor.

No calculations were available for locations of the rotor

critical speeds. It was observed that near 2300 rpm the
vibrations were in-phase at the ends of the rotor. It was
therefore concluded that there was an error in the first mode
correction, due to mis-proportioning of the masses added for
the attempted second mode correction. It was eventually
deduced that the rotor was responding to unbalances in its
second and third modes simultaneously, even though it was
running well below it second critical speed. The schematic
in Figure 10 indicates how this 'mixed mode' problem could
arise.

Comments on Mixed Mode Correction Procedure:

When a rotor, running at full speed, is under the influence

of unbalance in both second and third modes, the vibration of
the bearing pedestals will normally contain an out-of-phase
component from the second mode and an in-phase component from
the third mode. These two components may be at any angle
with respect to each other, and the resultant vibrations of
Case Histories in Balancing of High Speed Rotors 137

the bearings will likely appear as two unequal vectors, not

in the same plane.

The solution lies in resolving the resultant vectors bact

into their respective components, as shown in Figure 9. In
addition, because either mode may have unequal effects on the
two bearings from rotor to rotor, the constant of propor-
tionality may vary widely. This information cannot easily be
deduced with sufficient accuracy from the calculated modal
shapes, since many relevant factors such as foundation
stiffness and bearing stiffnesses enter these calculations.
However, even if the unbalance could be positively resolved
into definite amounts in the second and the third modes at
specific angular positions, the information would not· yet
have practical value because it is virtually impossible to
add correction masses which will influence only one mode at a
time on a real rotor, given a limited number of available
balance correction planes.

Balancing of Mixed Modes:

First assume that the rotor has been balanced in its lowest
mode around 2300 rpm. At some higher speed, well below its
second critical speed, it now is deflecting in its second and
third modes simultaneously. The mode shapes are those shown
in Figure 6. The most suitable locations for balancing planes
are at 25 inches and 70 inches from bearing A for the second
mode, and 10 inches, 44 inches, and 90 inch~s from bearing A
for the third mode. Also assume that all correction masses
are to be added at the same radius in these planes.

The first requirement is that there shall be no disturbance

of the condition of balance already achieved near the first
critical speed. Considering the relative deflection curve
shown in Figure 7. if a calibrating mass were to be moved
along the length of the rotor, its effect on the rotor when
running near its first critical speed would be proportional
to the relative modal deflection at the location concerned.
Taking measurements from Figure 7. it can be deduced that,
for no disturbance to occur in the first mode:

a) If p ounces are added on one side of the rotor 25

inches from bearing A, then p x 5.4/3.9 ounces must
be added on the opposite side of the rotor 70
inches from bearing A.

b) If q ounces are added on one side of the rotor 10

inches from bearing A, and also at 90 inches from
bearing A. then q x (3.4/1.6)/5.9 ounces must be
added on the opposite side of the rotor 44 inches
from bearing A.
138 N.F. Rieger

These proportions must be maintained for both calibrating and

correction masses, in determining the true correction mass
proportions for the desired modal balance.

The calibrating masses added for either second or third mode

will influence the deflection of the rotor in both modes
simultaneously. The best understanding is achieved by
viewing the two effects separately and adding the results.

Figure 7a shows the change of vibration level at the bearing

pedestals, caused by deflection of the rotor in the second
mode, as a calibrating weight traverses the length of the
rotor at a given radius, in a given radial plane. It can be
seen that this pedestal response is similar to the second
mode shape, and takes account of the fact that the bearings
vibrate in antiphase, and that the amplitude of vibration at
pedestal A is roughly 1.75 times that of B. Figure 7b is the
equivalent diagram for third mode, using the same weight, at
the same radius, in the same radial plane.

Assuming that the rotor is running well below its second

critical speed, the response of the rotor will be in the same
plane for both diagrams, i.e., if positive readings occurred
at an angle of 900, negative readings would occur at 270o,
for both modes. It is therefore permissible to consider only
positive or negative responses rather than vectors.

When adding corrections mostly for the second mode, the total
mass added at any location must be divided between the two
planes. The proportions into which the mass is best divided
were decided by the reasoning given above. The overall
effect of these correction masses on the vibration of the
bearing pedestals is determined by summing their individual
effects in both modes. These sums are, in turn, deduced from
the corresponding magnitude of the response curves given in
Figure 7. The values involved ·are given in Table 1. A
further computation is then required to determine the overall
effect of corrections added mostly for the third mode: (see
Table 2).

The effects on both modes were found as follows. When mass

was added mostly for second mode, the resultant change in
vibration at the bearings was:

At A: (8.64x - 1.4b) = 7.23x units

At B: (- 4.95x -·2. 35x) = -7.3x units

where x is the desired correction in the second mode.

Case Histories of Balancing of High Speed Rotors 139

The ratio [Effect at end A/Effect at end B) = 0.99, numer-

ically, instead of the 1.75 value noted above. This latter
is the ratio for deflection only in second mode.

Similarly, when correction masses are added mostly for tho

third mode, tho resultant change in vibration at the bearings
was:

At A: (2.3ly + 4.S2y) = 6.83y units

At B: (-1.32y + 7.S3y) = 6.2ly units

where y is the desired correction in the third mode.

The ratio [Effect at end A/Effect at end B) = 1.1 numer-

ically, instead of the 0.6 value noted previously. This
ratio applies for deflection only in the third mode.

Although the original readings of vibration may have involved

an error in second mode and another in third, for balancing
purposes the vectors must be resolved into components of
modified proportions, to compensate for the influence of the
added masses on both modes. A suitable graphical construc-
tion for achieving this resolution is given in Figure 11. If
during application this construction gives an unattainable
answer it means that the unbalance condition cannot be
corrected by the addition of masses in the required
proportions, in the proposed correction planes. Other planes
must then be selected, and the procedure r~peatod.

Figure 12 gives the complete worked example from original

readings and calibrating runs, for the boiler feed pump
rotor. The original readinss are resolved into their
appropriate components,and the calibratins weishts are
adjusted to nullify these components.

4.3.6 Comments

The construction shown in Figure 11 was used to identify the

required correct masses in the second ·and third modes,
simultaneously.

These corrections were then added to the rotor. fi th some

further minor adjustments, the rotor was then able to run
smoothly.

4.3.7 References

1. Moore, L. S., Dodd, E. G., 'Mass Balancing of Large

Flexible Rotors,' G.E.C. Journal (BDsland) Vol. 31, No.
2, 1964.
140 N.F. Rieger

4.3.8 Case History No, 3 - Coffee -Roaster Rotor

Vibrations

A hot air fan for roasting coffee beans was experiencing

troublesome vibrations and excessive bearing wear. A
schematic of the fan installation is shown in Figure 13. The
fan impeller operated at 2SOOF when blowing hot air. Both
bearinss were sinsle row spherical roller bearings, drip fed
with lubricant from sisht glasses, installed in 1947. The
bearings were mounted in cast iron pedestal supports bolted
to a concrete machine foundation, on the second floor of the
coffee plant. The fan was driven by an AC induction motor
through a gear tooth type coupling.

Details of Problem:

a) The fan was observed to be vibrating while

operating at its normal running speed of 5,000 rpm.

b) Strong vibrations were observed in the impeller end

bearins, and in the fan hot air duct surrounding
the impeller.

c) Local noise level was high, but not excessive. No

noise readings had been taken.

d·) Shaft operating temperature at the impeller bearing

was about 2000F,

e) No spectral analysis of the vibrations had been

made.

f) The drive end bearing of the unit experienced only

low vibration levels and ran at 1400F.

g) Drive motor temperature was not excessive. The

motor casing was warm (100°F) but not hot.

h) Foundation vibrations at the impeller were notice-

able but not excessive.

Investigation:

a) The fan rotor had a calculated critical speed of

7000 rpm in its bearings. This was based on the
assumption that the impeller was a point mass
located at the overhung end. No computer critical
speed calculations had been made.

:b) Impeller bear ins wear rates had been high. The
impeller bearing had been replaced three times
during the previous 12 months.
Case Histories of Balancing of High Speed Rotors 141

c) The fan operating environment was dusty. It was

observed that the bearing seals were not effective
in preventing dust from entering the bearings and
contaminating the lubricating oil. Bearing failure
was frequent and appeared to involve lubrication
failure. There was evidence of cage wear and brass
race smearing over the bearing outer ring contact
surface.

d) Some relative axial sliding movement had apparently

taken place between the shaft and the impeller
bearing, probably due to thermal expansion. The
thrust bearing was located in the drive end face
which was an angular spherical contact bearing.

Calculations:

o The calculated critical speed value was confirmed to be

around 7000 rpm, assuming a uniform shaft. and point
mass for the disk.

o The critical speed for the same uniform shaft with a

massive overhung inertia disk was at 5060 rpm.

4.3.9 Conclusions

o The original critical speed calculation was found to be

incorrect because gyroscopic-inertia effects of the disk
had not been included in the calculation. Inclusion of
these factors showed that the rotor critical speed was
actually occurring around 5000 rpm.

o The true calculated critical speed of the rotor system

was found to lie within 60 rpm of the 5,000 rpm running
speed. During operation the rotor system was almost
resonant with its unbalance excitation.

o Residual unbalance and thermal bow could, for this case.

evidently cause high synchronous resonant vibrations at
the fan operating speed.

o It is concluded that rotor resonant operation at 5000

rpm caused the impeller end bearings to vibrate as
observed, and that this vibration together with lubri-
cant contamination caused the bearing wear and failures.
142 N.F. Rieger

4.3.10 Recommendations

o The critical speed of the rotor system was raised by

welding a steel stiffening sleeve on to the shaft
between its bearings. This construction is shown in
Figure 14. The calculated rise in critical speed for
this condition is from 5,000 rpm to 6300 rpm.

o Improved bearing seals were introduced to exclude coffee

dust from entering the bearing lubricant.

o Bearing operating temperature could also be reduced by

increasing the oil flow through the bearing.

4.3.11 Effect of Corrective Action

As a result of the shaft stiffening action, and the improved

dust sealing, the bearing failures were eliminated.
Case Histories of Balancing of High Speed Rotors 143

(a) Generator and Turbine Coupling

(b) Stiffened End Plate

Figure 1. Paper Mill Generator Casing Structure

144 N.F. Rieger

90

110

70

30 :.:
Q..
20 f
c:(
10

0 12.!i 25 37.5 50.0 62.5

Figure 2. Axial Acceleration at No. 4 Bearing

-Resonant Peak at 71.0 Hz

100

90

110

70

6(j
f.0. ~;!a
50

•o
IZOHe
30

20 1
~
10 '"'!)'

0
"""""-
40 60 80 100 20
0 20

Figure 3. Vertical Acceleration at No. 3 Bearing

- 2.8 Mw Load
 . o f Balan en~
Case Histones . ~~ ~tH gh Sp~eedR:.:::=otors_
L~i 145

resentation of
Figure 4 · Finite Element
Stiffene Rep End Plate
d Generator
146 N.F. Rieger

Sll' I I rtlt I Pllfl .• c.,

I___( r

''-l•t Ill WO: (;("'(Uf!M

Figure 5. Lowest Mode of Generator End Plate - 69 Hz

Case Histories of Balancing of High Speed Rotors 147

0
_.. _ - .
10
. __ . __
20
.
_ JO 'II 50 10 lO 10 to 100
-
-----\----
~ _a...~-~-1

A
r ~-
ot,.;,;;;; incht>' tro"' M••lnt A
--, 8
~ ' '
I / co..ecllon NQul.-4 to \
I ~~~
modal delect
\
I \
I '\
I' '·
·' (a) Unbalance Response in First Mode

I'
I r

\ Correcliort requl.-4 to
\ nuftlfy tha aeconcl
\ modal defect
I

' ' .... I

-------
_Jo --:-__-_:_ .
A , ,.
~-
''
_J_ ~-_...!!.-_!f.._.....!!...-
D•IIA"Ce "' inches from be•nftg A

I \
I \
'
''
R...li¥e deReolon
of rotor

.,1~
(b) Unbalance Response in Second Mode

R...,ll•• cleftaolon
of rotor

10 to 100
~-_.._,_11

(c) Unbalance Response in Third Mode

Figure 6. Undamped Mode Shapes of Rotors in

Bearings Near Critical Speeds
148 N.F. Rieger

~jt
.o

V•bt,ltO•• ot
.be111ng ,edt51.11 A

Vtbtlllhon of
M11n"o peclesllll 8

Figure 7. Influence Curves for Response to

Correction Weights Along Rotor Length

''' ~ I ' ''

'' '' ,
.
'
I
I
.. ., ,., ........ '
I
f ____ '
.
I -"
t-"""' ~----·~
•
I I

"
Figure 8. Explanations for Readings of Vibration
Vectors OA and OB:
a) Equal out-of-phase component, equal
in-phase component.
b) Equal out-of-phase component •. unequal
in-phase component.
c) Unequal out-of-phase component, equal
in-phase component.
d) Unequal out-of-phase component, unequal
in-phase component.
Case Histories of Balancing of High Speed Rotors 149

A A A
c A

lcl
1~~. c
----7ot-----=cO
Be (0!)
0 0 A
lac
\
'\ I '\ \
\ I
\
\ v '\
'\ \\I
'
a\ 8
8 8

Figure 9. Vibration Response of Vectors at Operating Speed

a) Original readings of vibration of rotor
pedestals.
b) Effect of out-of-phase pair of calibrati~g
weights at ends of rotor.
c) Effect of adjusting the calibrating weights
to eliminate vibration of Pedestal A.
d) Effect of adjusting the calibrating weights
to eliminate vibration of Pedestal B.
e) Effect of adjusting calibrating weights for
best compromise.

Running Speed
2300 rpm
0.015
1 Mode 2
Model
Mode 1
I
I
0.010
Vibr•tion
Mils p-p

0.005

0
0 1,000 2,000 3,000

Speed, rpm

Figure 10. Rotor Response Due to Second and Third Modes

150 N.F. Rieger

,.,.,.,,,_ ., 11'1114<"1' - ······- QA •"-4 Oft ..... I

i - - -.....: .... ~ - - - - • - ..
· · - ---'';;;, .,...,,,•• • .,• ..( , . ... ,_,....,.. e{ , . ...... ,., ...... . ....... , ,..
.,_..,., ... ,..,.. ·~"'' fl( , ...... ,.•.,.. ...... .
L•• ••H""c• .t •·~•••.-. .,_ ,,_.,.,,,..,, t, oA •
~ ......, "' oi .. ,. • .,., a
C•••" th&C ( oJ .... ,. telot.r.lt~~f _..,,tfltt ••r eHf"' ,.,
•••.-.4MNr.
lfftct.,. ,.,4 A/( flue .,. r~ I "'...,...,,,.,,,
, ... • .. , ••• ....., ...., ........
~w ~ ~.t •.
e#«l ,., .... ,
I tfut ... ""4 Aflfftcr- tttt4 II n -.,,..utf.t'r
To f•n<llc_,,......,.u e( oA -.t4 Oi ,, ••• pre;-t"•••JI""!
Cf')f"truflutft: Jettt AI ell4 (,v*.,. C 1•1~1 A(J(l
.. ,...,,.._ ,,.,,. .. ,,.,..,,,. ... elf41
I ,,..,.,ct
'" J.... oc..w~~
AO ,. "'HI ,., ,.tlrl llftrr rlwevr• I "" 0

o, - ---- - - --,--- --~- --

'c
:!:::.g:,;!".:.~": ,:·!:!.,:f~ riNt~VCfl A/,.~
Th.,ft : Alft ,.,., ,.,.,,~ 0 wtll "'ttl 1'- .. ,eHd t.IW1
/'., \ I tlllt..,p A eft41 .. A' eAtl I' NllwttOA "fOI •'"
I( 1"1 • r . ,,.._, ., afiiii.-CI- 01' ~ t~r/m
,'' \ ,' W M •"'"'•' ~~ OA'A. oa·o. OA'• ,..ol'
A.'A•"'I'O•"f· .·• A'A/I"I• .. rfr .. ,.
/ ,' I \ ..,,.().\·•
,..,.,._<>A...,oi·-~•'i-oi
/ \: -..,,of-- oi.. oi_ _.,.,..,..
·---,!·! ___ ____ _, _ _ . ,. t u - -- - --+ -- ,......,.. OA·. oi·. wtwttOA·,.,..,.I(oi· .. •-M-"
0 l I' r • C........ A~ i·~.. --~ A"A•AXI'I

Figure 11. Graphical Construction for Identifying

Correction Weights for Mixed Modes

'Jt.~.r..J".t61,.~
,,...,... ...... fU!t·· ·~·-· \1!""~'!....1"-~~

.
=~:-~
"~""'"'" ~.

..._
. .-. ....... .. ~ ....... -.uc-o.•r......,_,...... .,..
~-- ~· ·-
t~

...,
)JS
l.ao..t
, ••

....
.....
·~
~
'~•.rc .~ ···~-~
J
... ,,,. .... .--• . .......................
)S ,..,~ ·~~ t

_.. ..::.··
M...
· '·" .... ._...
• ...... .._ _ _ _ ,...._._. ...,,-._
..... ......,~... --:','- ~ -~
t.®l...,,.,,__,.._~-roo'·~......,....&-,-.,....,r:..,....--.,

.... ·- ...
~- ) ~·
-~. 1r ._.,__..
1\Ar•••. """"" ~·· ·• ._ .(\o.t•{._·~...t-.--1-"'

.
'1-f" ........ --a.
AA'- l, l'.. ~·l ..t••,_, t ... ....._, .....................

..:.:s·"'*\ib .................... ., ....y...,. '•

..,............. ..
t!s,·:t-.!..-'!.•&V........

.
~ ~

: .... --' · ............--. ............... . t.+.·•'f4o·

. :off~~~--
~S-::=-~·"" ·• ~·Cc4
/ ~-.
•- AI,_.,.. ......_.._ •.-.· _.._, • ....
/
&IlL-····-·- ... .,._.
~

~
/
. , ./ ......,
.1#• •

.
"-:6,.~ A _ _. b .
.......... At• ... _ ...... .#d .._ ...__ • .._,. h ~ 0

~. I\'-··· ~ ~-··..,.; ..
.··
J.- ~tt'u - • • · · • · ........... ,-~..._ · • - ..,.._,... ,....,"•
/ ~-
.... / ..... T.t.,...~"~• ........ ,.._ ..,._,....... o-'.ftb"•· .. ~~_l
• ................ .,.,.,.Yflf

-
.:::~::: .. t:r.vu~ t.:'..... ·~U'"'-"

....,...... .·· ·····.... ~ . ~~:_ • • \& .... ,Jo•

r·.......... ··· ·····:;.;. ..·. a .~ ~~~~~~. ~~~:!:=::·~·\:·~
\ / :,;;. ...,.,...... ./.. / ' .;
' ..... .
,....._. -\ ..................... -,....
e.... -.-..;; ...
.. r.e, . -
..
.....~~J~"!-~
\•,> ·.. . . _ ,/ r~ l... • •• •scr
II ' -....(
·~· !..!::..!'~-i.=~~,Lt-J.. "'-'" f!.•L1o)•""- -·-~-~
{it• l!~o) • llo ~-·6 ..,. ..._ _..,.,...,. lf.• C. 'J ·1~u • · a.. ~,......,,,
..........,. _.......... -.... 1·- ~ ,__,_4 -4 ..... ~"""

·~·"'

Figure 12. Application of Graphical Construction

to Balance Boiler Feed Pump Unit
Case Histories of Balancing of High Speed Rotors 151

Drive
Coupling
Bearing

Figure 13. Overhung Fan Impeller in Bearings

Figure 14. Stiffening Tube Over Shaft

Between Bearings
152 N.F. Rieger

Table 1. Comput ation of Overall Effect of Weights

Added Mostly for Second Mode

--
Un111 o( unl»..ncc Unu" .,( ua.MlitKc
.!:fht oclde<l, Di,uncC' in inches ift IC'C'OR4 mc-IC'
~ftect dl'ccl in "'iui~~MMtc
np 11 • multiple
o( dw alobncin' wciKht
uoe<l ,....,.oon fill.l. r~:"'ortK~:t:l :c::h, Bcorin~ A lkaunc 8 8C'ann1 A lcAtlftt' £

... ,.,. - I "'I• -0·6S• -lb

• :~
...
S-4 10 s~
.... ,.,5,. ro•
5·4
-Z·Zx-• -0 S5x n•
S·~
-Ohlt,
~

r.• (opposite sidC' of IOfOf) )·9

101'AL fOa IACII MOnf

.... ~. -~·95• -1·41•. -l.lSa

II B
'l"h.:rcforc, ac•ual \"ti'tnlion C'l' IM'ann~ •"'Uid be -+1·2l;f unus at A and - i·)x

Table 2. Comput ation of Overall Effect of Weights

Added Mostly for Third Mode

Units of unbalance t.:n••• '" uabalancc

nprc~l':.~:k.plc OiJcancc '" inchn
from ttc:aun,: A to
eft"''"' tn IC'COf'ld mode die« ia chird .....

c:!~-: fic~ht plane o( od.J.·d ...,.;~111 IIC'annc: A lk•rine 8 lkorin,A 8c'.1nn, I.

,..,,
o(.:::.

10 ..... -2·25.• +1·2y J4_r

••
.Y

s-o
f.9"
44
(arposil< oi.S.: or ....or) +0·9xf.i•
s-o -O·Sl< !:!!,
5·9
s-o
-ti·ISI<f.ir -,..,.;:;,
y 90 -2·4y +1·3Sy ~1·15y -tz+
-t-Z·lly -1·32,- ~4·S2.'" HIT
ft'TAL PM; IACH MC'tOC

Thereto«, actual wibration of bc'armcs wauW lac • 6·13.o• unico 11 A on4 6·21yocll
 CHAPTER 5

SEAL PROPERTIES

R. NordiiWID
University of Kallenlautem, Kailenlautem, FRG

ABSTRACT

Hydraulic forces in annular seals may have a large influ-

ence on the vibrations of high speed centrifugal pumps. To
have a better knowledge about such seal effects,theoreti~

cal as well as experimental studies are important in the

actual research of rotordynamics. In this chapter a theo-
retical model and an identification procedure are presen-
ted to determine the dynamic coefficients of seals by cal-
culation and measurements as well. The identified parame-
ters confirm the assumptions in modelling and point out,
that the stiffness and damping characteristics of seals
are significant for the stability behavior of pumps.

5.1 IDENTIFICATION OF STIFFNESS, DAMPING AND INERTIA

COEFFICIENTS OF ANNULAR TURBULENT SEALS

An important assumption for the reliability of high speed

centrifugal pumps 1s a good rotordynamic behavior. Connected to
this problem hydraulic forces acting on the rotor are of major
importance. It is well known that neck ring seals as well as in-
terstage seals (Fig. 1) may have a large influence on the bending
vibrations of a pump rotor. Besides their designed function of
reducing the leakage flow between the impeller outlet and inlet
or between two adjacent pump stages, the contactless seals have
the potential to develop significant forces. This type of forces,
created by lateral rotor vibrations can be described by a linear
154 N. Nordmann

TURBOPUMP TURBINE

SEAL LOCATIONS

Fig. 1 Seals of a High-Pressure-Turbopump Rotor

in Aerospace Engineering

model with stiffness, damping and inertia coefficients.

If contactless seal elements are used in a turbopump the

fluidmechanical interactions have to be considered when predic-
ting the vibration behavior of the pump rotor in the design pro-
cess. However, there is often a uncertainty, concerning the data
for the dynamic coefficients. Up to now the stiffness and damping
characteristics of seals are not very well known and there is a
need for additional research in this area. This is particularly
the case for grooved seals, which are very common in practice.
Different research projects have been started to investigate the
dynamics of ~eals by theoretical models as well as experimental
procedures. The following chapter presents a possible model, based
on a bulk flow theory and describes an experimental procedure to
identify the stiffness, damping and inertia coefficients.

Modelling of Annular Seals with Turbulent Flow Conditions

Seal Model. To explain the seal model, we consider a very

simple geometrical form, consisting of a cylindrical shaft with
circular cross section, surrounded by a cylindrical housing (Fig.
2). This annular seal seperates the two chambers with pressure p 1
Seal Properties 155

and pressure p2 , respectively. The pressure difference is

6p = p 1-p2 • Caused by this pressure difference there is a leakage
flow in axial direction, which is always almost a turbulent flow

_------r
__....- PRES SURE DROP ll p = P,- p2

PRESSURE p 2
PRESSURE P,

TURBULENT LEAKGAGE FLOW HOUSING

WITH AVERAGE VELOCITY V

Fig . 2 Modelling of an Annular Seal with Turbulent Flow

with average velocity V. A velocity ~n circumferential direction

is superimposed due to the rotation of the shaft with angular of
velocity Q. In order to obtain the governing equations for the
presented seal we assume pure translational movements of the shaft
~n radial direction.

To derive the pressure around the shaft and then the force-
motion-relationships for the vibrating rotor we are using a bulk
flow model, which was originally derived by Hirs /1/ . Childs /2/
introduced this bulk flow theory for seal elements. The first
basic idea of this theory is, that the fluid velocity distribu-
tions in radial direction are substituted by average velocities.
For a fluid element between the rotor and stator surface (Fig. 3),
located at the axial coordinate Z and the circumferential coor-
dinate 0, the average axial velocity is U (Z,0,t) and the aver~ge
z
circumferential velocity is u0 (z,0,t) . The corresponding pressure
for this locations is p(Z,0,t) and the seal radial clearance
H(0,t). The shaft circumferential velocity is U' = R O.
156 N. Nordmann

U-RQ
PRESSURE

Fig . 3 Velocities , Pressure

L and Radial Clearance
for a Fluid Element

The second basic assumption in Hirs theory is his empirical

finding, that the relationship between the shear stress at the
wall and the mean velocity of the bulk flow - relative to the
wall - can be expressed by well known formulas. With the bulk
flow velocity VR ={(U 9-Rn) 2 + uz 2 } relative to the rotor surface
we obtain the wall shear stress at the rotor
m
2H VR 0
T
R
= no (--)
v
.Q.y2
2 R
( 1)

n0 , m0 are empirical turbulence coefficients, p is the fluid den-

sity and v the kinematic viscosity of the fluid. In a similar
way Vs = {u9 2 + uz 2 } is the bulk flow velocity relative to the
stator surface and the corresponding stator shear stress is
m
2H V 0
T
s
= n0 (--s)
\)
.Q.y2
2 s
(2)

Formulas (1) and'(2) correlate the shear stresses TR, Ts with the
Reynolds numbers, which are defined in parantheses.

We can now derive the two momentum equations, expressing the

"equilibrium" for the fluid element in axial and circumferential
direction (Fig. 4). If we introduce the shear forces at. t.Le walls,
the pressure forces and the inertia forces we end up with the two
Seal Properties 157

't;Rd9dZ
pHRd9

Fig. 4 Pressure and Shear Forces acting on a Fluid Element

equilibrium equations (3) and (4), which are shown in Fig. 5 to-
gether with the continuity equation (5)

Axial Momentum Equation

1+m 1+m
0 0
2 }

(3)

Circumferential Momentum Equation

1+m_
_ o 1+m_
_ o
2

+ (4)

Continuity Equation

auz 1 a 1 oH
H F + i ae (H 0e) + Rn at = 0 (5)
158 N. Nordmann

Reynolds number in circumferential direction

R = HU = .Q...J:!__Q
e v n

Fig. 5 Governing Equations for a Fluid Element

Perturbation Analysis. With the assumption of small motions

of the shaft about a centered position we can expand the equa-
tions (3), (4), (5) by means of a perturbation analysis /2/

H ( e, t) = Ho + e:: H1

p (z,e,t) = Po + e:: p1
( 6)
uz(z,e,t) = 0 zo + e:: 0 z1

u 6(z,e,t) = 0 eo + e:: 0 e 1

e:: is a parameter. H0 , p0 , UZO' u80 are the quanti-

p~rturbation

ties for the zero eccentricity flow condition (centered position

of the shaft). H1 , p 1 , Uz 1 , u 81 correspond to the flow conditions
for small shaft motions.

If we introduce (6) into the governing equations (3), (4),

(5) we obtain zeroth-order and first order perturbation equations.
The substitution and the solution procedures of the remaining
equations are very extensive. They are described in more detail
in /2/ and /3/. In this presentation we discuss only the essen-
tial results.

The solution of the zeroth-order equations (e:: =0 , shaft

without radial motion) defines the steady state leakage or the
pressure drop 6p in axial direction (Fig. 6).

6p = p 1-p 2 = {1 + ~ + 2 o} Q
2
y2 (7)
Seal Properties 159

The pressure drop 6p is proportional to p and the squared velo-

city V and consists of three parts. The first one shows the
change of pressure energy to kinetic energy. The second part
points out the pressure loss at the seal entrance and the third
part expresses the pressure drop along the seal, caused by fric-
tion.
--r
_......- PRESSURE DROP llp- P. -p
..--- I 2

PRESSURE p 2

Fig. 6 Pressure Drop in Axial Direction

The seal behaves like a hydrostatic bearing. A static displace-

ment of the shaft in radial direction causes a restoring force
and the fluid acts like a spring. The second important result of
the zeroth-order equation is the development of the circumferen-
tial velocity of a fluid element proceeding axially along the
seal. This velocity influences the cross coupled stiffness coef-
ficients.

The first order equations describe the pressure and flow

quantities due to a small shaft motion H(S,t) about the centered
position. These equations can be solved numerically. When we in-
troduce a circular harmonic orbit as a special motion, we can ex-
press H(S,t) in terms of this motion. By the further ass~ion
of a harmonic pressure and velocity distribution in circumferen-
tial direction, the first order equations can be reduced to a
system of three coupled complex ordinary differential equations
160 N. Nordmann

for the unknowns UZ 1 ' u81 , p 1 /2/, /3/. From the pressure field
solution the reaction forces acting on the rotor due to the cir-
cular shaft motion have to be determined by integration of the
pressure along the seal and in circumferential direction. Finally
the force motion relationship is established. The dynamic system
of a seal can be modeled by a linear system with stiffness, damp-
ing and inertia terms, if small movements about the seal centre
are assumed (Fig. 7).

F m c c y k k y
y yy yy yz yy yz
= + + (8)
F m c k k z
z zz zz zy zz

In general a numerical procedure is needed to calculate the dy-

namic coefficients of equation (8). For the special case of a
short seal a solution in an analytical form is possible /4/, /5/.

Fig. 7 Dynamic Seal Forces

caused by Small Shaft
Motions about a Centered
Position

The main diagonal elements in each of the matrices are equal and
the cross-coupled terms are opposite in sign. The coefficients
are mainly dependent on the pressure drop, the average axial ve-
locity V, the rotational speed n of the shaft the seal geometry
(seal length L and Radius R) and on some quantities characteri-
zing the friction in a seal. It is important to note, that the
cross coupled stiffnesses kyz = - k zy are strongly infllh:uced by
Seal Properties 161

the rotational speed n and the fluid entry swirl, which is the
circumferential velocity of the fluid at the seal entrance. This
effect may cause serious instability problems in high speed rota-
ting machinery, when the cross coupled stiffness terms become
dominant.

In one of his first publications about seals, Black /4/ has

derived stiffness, damping and inertia coefficients for short
seals. Fig. 7 presents this coefficients in dependence of the
most important influence parameters 6p, V, n, T = 1/V and some
friction coefficients ~O' ~,, ~ 2 • In Black~ derivatives the entry

M • uR:p ·1-"-'T_' -I-~,Ti-1

PRESSURE DROP 4p
AVERAGE AXIAL VELOCITY v K
AVERAGE FLOW TIME T: L/V
ROTATIONAL SPEED g

Fig. 8 Dynamic Coefficients of a Short Seal /4/

swirl was assumed to be half of the circumferential velocity of

the shaft:Rn/2.

With the a priori knowledge about the seal dynamic coeffi-

cients we have valuable informations about the structure of the
model (linearity, order of the model, skewsymmetric etc.), which
will be used in the following parameter identification procedure.
162 N. Nordmann

Identification of the Dynamic Seal Coefficients

Most of the experimental methods to determine dynamic coef-

ficients of bearings or seals are working with test forces (input
signals) and are measuring the relative displacement between
shaft and housing (output signals). The unknown seal parameters
can then be calculated by means of input-output relations in the
time domain or in the frequency domain (Fig. 9).

SYSTEM : SEAL

INPUT OUTPUT

FORCES DISPLACEMENTS

PARAMETER k.,, k ..,••c,, c .....,m,. m..

IDENTIFICATION PROBLEM .
TO FIND THE PARAMETERS OF THE SEAL
FRCJ.t INPUT- OUTPUT- RELATIONS

Fig. 9 Identification of the Parameters of a Seal

The basic steps of the applied identification procedure are

pointed out in Fig. 10. Measurements are carried out at a test
rig, that consists of a very stiff rotating shaft and an elasti-
cally mounted rigid housing with two symmetric seals between shaft
and housing. Water flows axially across the two seals in opposite
directions, while the shaft is running with a fixed rotational
speed n. The housing 1S excited by a test force and the system
response is measured as a radial motion between the seal surfaces.
From measured input and output time signals frequency response
functions can be determined by signal processing.

Corresponding to the seal test rig a linear mechanical model

ex ists, consisting of the rigid mass (housing) and the stiffness
and damping coefficients of the seals. The analytical frequency
Seal Properties 163

HODEL MEASUREMENT

F,

ANALYTICAL

~;=F=R=E=OU=E=N=~==R=E=S~==N=SE========~~~~~======~,
1!1:NTlFCATION OF [J'(NAMIC SEAL COEFFICIENTS

CRITERION : IS CORRELATION BETWEEN MEASUREMENT

AND MODEL GOOD ?

RESULTS

Fig. 10 Basic Steps of the Identification Procedure

response functions of the model depend on the unknown seal para-

meters, which have to be determined.

The dynamic seal coefficients are estimated by a linear pro-

cedure and uses the measured mobility frequency response function
and the analytical functions of the model. The main steps of the
identification are following in more detail.

Mechanical model.Fig. 11 shows the mechanical model with a

rigidly supported very stiff shaft,the rigid mass m of the housing
and the stiffness and damping coefficient elements corresponding
to the seals and the flexible springs supporting the casing. If
test forces are acting in the center of the housing, the system
responds only with translatory motions in the two directions y
and z. The equations of motion for the model
(9)
m+2m c c y k k y F
yy YY yz yy yz y
+2 +2 =
m+2m c c k k z F
zz zy zz zy zz z
164 N. Nordmann

RIGID SHAFT IN
SEALS WITH
RIGID BEARINGS
k,)<,,,k;,.ku
c,.c,.c",c,,
m., . mlf

RIGID HOJSING MASS m

Fig. 11 Mechanical Model of the Seal Test Rig

describe the equilibrium of the inertia forces (housing), the

seal forces and exciter forces.

For the considered two degree of freedom test rig system a

total of four stiffness frequency functions as well as four flexi-
bility frequency functions can be derived. They depend on the
seal coefficients and'the exciter frequency w, as well. The ex-
citer frequency w is usually different from the frequency of ro-
tation n. Fig. 12 points out the mathematical expressions of the
* ..
two types of frequency response functions H(w) and K(w). By in-
• -
version of the frequency response g(w) we obtain the stiffness
-
...
response !(w). Both functions are used in the identification pro-
cedure.

Measurements of the Frequency Response Functions H(w).

• Up
to now we have considered frequency response functions of the mo-
del. For the determination of the frequency response functions
from measured input and output time data. we take advantage of
the fact that the ratio of the Fourier transformed signals is
equal to the frequency response. Due to this possibility, ex-
citation signals with broadband character in the frequency domain
(impact, random etc.) can also be applied. The force and response
signals are measured in the time domain, transformed to the fre-
Seal Properties 165

Complex Flexibility Frequency Response

~Y(w) l Hyz (w) k

zz
-w2 (~m
2 zz
)+iwc : -(k +iwc )
zz yz
1 yz
'" II
!!= ----T---- -----------~------------
1
1 I
H (w) 11 H
zy . zz
(w) -(k
zy
+iwc
zy
) 1k
1 yy
-w2 (~m
2 yy
)+iwc
yy
1

~ = (kyy-w2 (~
2 myy )+iwcyy )(k zz -w2 (~
2 mzz )+iwc zz )

- (k +iwc )(k +iwc )

yz yz zy zy

Complex Stiffness Frequency Response

I
-w2m
1
K (w) I K (w) k (=+m )+iwc I k +iwc
yy I yz yy 2 yy yyl yz yz
I I
----1-----
1
=2 -----------.---------- I

• I
K (w) I K {w) k
zy
+1WC
zy
1k -w2(m
-+m ) +1WC
•
zy 1 zz 1 zz 2 zz zz

Fig. 12 Frequency Response Functions ~(w) and K(w)

of the Mechanical Model

quency domain by means of Fast Fourier Transformation and the

quotient is calculated (Fig. 13). This procedure is executed by
efficient two channel Fourier Analyzers. Fig. 14 shows in prin-
cipal the measurement equipments. A hammer was used in this case
to excite the housing by an impulse force (see also Fig. 13). By
this excitation the signal contains energy in a desired frequency
range, which can be influenced by the hammer mass, the flexibi-
lity of the impact cap and the impact velocity. The relative dis-
placements between housing and shaft are measured with displace-
ment pick ups. The time signals are amplified, digitized by the
166 N. Nordmann

INPUT OUTPUT

FREQUENCY Fig. 13 Measurement of

DOMAIN
Frequency Response
Functions by Means
of Fast Fourier
Transformation

F~ER ANALYZER

AMPLIFIER FAST FREQUENCY MAGNET I

"~
.,, ,, ,, / FOURIER RESPONSE TAPE
DISPLACEMENT TRANS FUNCTIONC
FORMATION
PICK UP

Fig. 14 Measurement Equipments

Analyzer and the frequency response functions are calculated.

Seal Properties 167

Estimation of the Dynamic Seal Coefficients. Different

possibilities exist to determine the seal coefficients from
measured frequency response curves. One idea is to fit analytical
flexibility functions to the measured ones. When working with the
output error, this leads to nonlinear equations for the unknown
parameters. Another method will be presented here with the defi-
nition of an error for the input signals.

From a theoretical point of view the product of the complex

-
mobility matrix H•and the complex stiffness matrix K~should be
the unity matrix ~· By combining the measured matrix ~·with the
-
analytical matrix ~·the result will be ~ plus an additional error
matrix §, caused by measurement noise. Fig. 15 points out this
fact and shows either the complex equation or the two real equa-

IPARAMETER - ESTIMATION

rn r-r-.,
I
I s I
I
L_
I

1 --'
I

f]
E +

IT·;~o.l~ If = w~t!;-_-:__
H1 - ...ZM H'
..::--~----
+ wC H'
·-· __
=
~•
.
S'
_____.::._

Fig. 15 Parameter Estimation

tions for the unknown parameters, concentrated in ~' £, ~· The

two real equations belong to one exciter frequency. In the case
of broadband excitation (impulsive force) we have as much equa-
tions as frequency lines, generally much more, than unknown pa-
rameters. The overdetermined equation system is presented in Fig.
16. The rectangular matrix A contains all information about the
measured frequency response functions (H~, Hi real and imaginary
part of frequency response) and the related exciter frequencies w.
X contains the unknown matrices ~' £, ,! and !' is a modified unity
matrix. Applying the criterion to this equation, that ~· shall be-
168 N. Nordmann

EQUATIONS FOR THE DETERMINATION

OF UNKNOWN SEAL PARAMETER X

A· X • E' • S'

..
A
X
CONTAINS THE MEASURED FREQUENCY
CONTAINS THE UNKNOWN PARAMETERS
RESPONSE DATA
M, D, K
-
Hlwl

CRITERION s'-MINIMAL

NORMAL EQUATIONS

Fig. 16 Normal Equations for Unknown Parameters

come a minimum, we find the so called normal equations. This is

a determined system of equations for the twelve unknown seal pa-
rameters in the matrices ~, £, ~· Due to the definition of an
input error the solution procedure for the linear system may fail.
This can be avoided by introducing instrumental variables ./6/.
. case t h e matr1x
I n t h 1s . _AT 1s
. sub st1tuted
. . WT
by a matr1x .
_ , conta1-
ning the instrumental variables. In our case they have the mean-
ing of the frequency response data, determined with the estimated
dynamic coefficients from a previous step. Then the solution is
found by an iterative procedure.

Test Rig and Some Measurement Results

Seal Test Rig. The mechanical part of the test facility is

shown in Fig. 17. The main components are a very stiff rotating
shaft, driven by a speed controled motor and the stiff housing,
supported in flexible springs. The shaft is rigidly supported in
roller bearings. The fluid enters the housing in the center,
flows across the two seals in axial direction and is ex iting the
housing at the two ends. With removable stator parts of the seal
different geometries and roughnesses can be realized. The static
position (zero eccentricity) is adjusted by a speci.nl rr:•·r 11anism
and measured by eddy current pick ups, which are also used to
Seal Properties 169

HOUS ING SEAL

Fig . 17 Seal Test Rig

measure the shaft motions.

The range for the rotational speed is from 0 to 6000 rpm and
for the axial fluid velocity from 0 to 14 m/sec. With this we
achieve Reynolds numbers up to 15000, when the fluid temperature
. 0 .
~s about 30 Cels~us.

Dynamic measurements. In case of a dynamic measurement four

frequency response functions have to be determined for one meas-
urement set, defined by a working condition with constant axial
fluid velocity, rotational speed and fluid temperature. A computer
takes over the measured data and calculates the dynamic coeffi-
cients by means of the described estimation procedure. Both sets
of frequency response functions, the measured and the fitted one,
can then be displayed and plotted.

Some J.!easurement Results. The process of the dynamic meas-

urement is demonstrated for one working condition with V=12 m/sec,
n = 3450 rpm and a temperature of 30° Celsius. Fig. 18 shows the
corresponding measured and fitted response functions as magnitude
and phase characteristics in the frequency range 0 to 100 Hz. The
170 R. Nordmann

u = 345121 l/min, v = 12 m/sec,

.,
~

" Ill

Jl , ••
r ,.•q u•nt (Ht] Sl '"

Fig. 18 Measured and Fitted Frequency Response Functions

two direct functions H and H should be equal and the cross

zz
yy
coupled functions H and H should be equal in magnitude but
yz zy
opposite in phase. The correlation between fitted and measured
functions is more or less good.

Several measurements were carried out for different rota-

tional speeds with constant temperature and axial velocity. For
each set of functions the inertia, damping and stiffness coeffi-
cients were calculated. They are shown in Figures 19, 20 and 21
versus the rotational speed. The values for the complete system
with two seals and all known additional terms (mass of the hous-
ing, soft springs etc.) are presented. It was found, that the
direct coefficients are not equal, furthermore the expected skew-
symmetry could not be found exactly in the measured results. The
coefficients, which should be equal 1n magnitude are shown in one
diagram and treated as two values for the same operating condi-
tion. Besides the measured coefficients the corresponding values
of the above mentioned theoretical model are also shown.
Seal Properties 171

2000 -PREDICTED
k, ,k ..
(N/m)
PREDICTED C, ,Cu
(Nslm)
-----·
0 0
-----o -----
0 n 0 R n n
5 0 0 0 0 0""' 10 v-""'-
2·10 1000
MEASURED

0 0
I
0 0 2000 4000
ROTATIONAL SPEED n I RPMJ ROTATIONAL SPEED n (RPM)

/
c,. ,c.,
,,
k,. ik., PREDICTED
/
IN/m)
(Nslm)
2·10
5
," 500

,"
,"' MEASURED
l,.._...,_o
~/ " ""'v u I" - u
0 0~--r--+---.---r-~--,
0 2000 4obo 60)0 0 2000 -4000

SEAL STIFFNESS COEFFICIENTS SEAL DAMPING COEFFICIENTS

V= 12 m/s , II = 30°C V=12 mls 1 11=30°C;

Fig. 19 Stiffness Coefficients Fig. 20 Damping Coefficients

MEASURED

m,. m,.
(kg 1
10 -1----+---'-----jr-----1

0-6--r----+-...--+--.----1
0 2000 6000
ROTATIONAL SPEED n (RPM)

SEAL INERTIA COEFFICIENTS

v :12 mls, 0 :30" c

Fig. 21 Inertia Coefficients

The correlation is not that good for the stiffness terms,

which are presented in Fig. 19. The dependence on the rotational
speed ~s reasonable for both quantities, a parabolic decrease for
the main stiffness and a linear increase for the cross coupled
terms. However, the main stiffnesses are found out 40 to 50 %to
172 R. Nordmann

small and the cross coupled terms much more than this. Meanwhile
it is known, that the reason for the small measured cross coupled
terms is the low entry swirl in the test rig, which has a large
influence on k and k
yz zy

Fig. 20 shows the damping values. The theoretical model pre-

dicts nearly constant values for the main damping and an increase
with rotational speed for the cross coupled terms. The correlation
between measurements and predictions are good to fair.

Finally the total inertia terms are presented in Fig. 21.

The theory predicts constant inertia terms. The correlation of
measured and calculated values looks good. However, it has to be
noted, that the mass of the housing is approximately 15 kg. This
mass is included in the results. Therefore the relative error of
the seal inertia coefficients, related to the model prediction,
is much higher.

References

/1/ Hirs, G.G.: "Fundamentals of a bulk-flow theory for

turbulent lubricant films"; Diss. TH Delft, Niederlande,
1970.

/2/ Childs, D.W.: "Finite-Length solution for rotordynamic

coefficients of turbulent annular seals"; ASME 82-Lub-I.Q.

/3/ Nordmann, R. et. al: "Rotordynamic Coefficients and

Leakage Flow for Smooth and Grooved Seals in Turbo-
pumps"; IFTOMM-Conference Proceedings, Tokyo 1986.

/4/ Black, H.F.: "Effects of hydraulic forces in annular

pressure seals on the vibrations of centrifugal pump
rotors"; J. Mech. Eng. Sci., Vol 11, No 2, 1969,
s. 206-213.
Seal Properties 173

/5/ Childs, D.W.: "Dynamic analysis of turbulent annular

seals based on Hirs' lubrication equation";
ASME 82-Lub-4 1 •

!61 MaBmann, H.: "Ermittlung der dynamischen Parameter

turbulent durchstromter Ringspalte bei inkompressib-
len Medien"; Diss. Universitat Kaiserslautern 1986.
 CHAPTER6

STABILITY OF ROTORS

R.NordmaDD
Univenity of Kailenlautem, Kailenlautem FRG

ABSTRACT

Instability in rotating machinery may be caused by differ-

ent effects like oil film forces in journal bearings or
forces in contactless seals. A machine designer needs to
know, whether a rotor will run stable during operation and
what size the stability threshold speed will have. Further-
more he needs information about the parameters influencing
the stability. This chapter treats the special system of a
rigid rotor, running in two oil film bearings. The equa-
tions of motion are derived and the stability behavior is
evaluated by means of the complex eigenvalues. Analysis re-
sults show a good correlation with measurement results from
a test rig.

6.1 RIGID.ROTOR STABILITY

Why do we investigate the stability behavior of a rigid ro-

tor? Is this useful from a practical point of view or is such an
investigation of more academic character? Every rotor is an elas-
tic system, of course. However, in special operating conditions,
a rotor can be regarded as a rigid shaft, approximately. To ex-
plain this, let us consider a 40 MW gasturbine-compressor shaft,
running in two lobe journal bearings with an operating speed of
4485 rpm (Fig. 1).

The shaft has a mass of 16000 kg and a length of 4,4 m. If we

consider the dynamic characteristic of this rotor in the oper-
176 N. Nordmann

GAS COMPRESSOR
TURBINE

POWER 40 MW
OPERATING SPEED 4485 RPM
SHAFT MASS 16000 KG
JOURNAL CLEARANCE JOURNAL SHAFT LENGTH 4.4 m
BEARING EXCITATION BEARING JOURNAL BEARINGS: TWO LOBE d/0"' 0,8

Fig. 1 Gasturbine Compressor Shaft

ating range, its behavior is dominated by the first eigenfre-

quency combined with a very low damping constant. The change in
sign for the damping value points out the onset of instability
at a speed near the operating speed. The corresponding mode shape
is characterized by nearly same displacements for all points
along the shaft, showing that the bearings are relatively weak
compared·to the shaft (Fig. 2). We recognize, that this Gastur-
bine Compressor Shaft can be modeled by a rigid rotor and two
elastic journal bearings with stiffness and damping coefficients,
to study the stability behavior in the interesting speed range.

The rigid rotor- oil film bearing model has only a few de-
grees of freedom and is relatively ~asy to handle. However, it
is a good approximation for many practical cases. It is also
useful to study the dynamic behavior of different bearing types.

Mechanical Model

The mechanical model in our investigation consists of a

symmetrical rigid rotor with mass m, equatorial moment of 1ner-
tia eE and polar moment of inertia ep (Fig. 3). The rotor is run-
ning in two equal oil film bearings, characterized by stiffness
coefficients kik and damping coefficients cik (i,k = 1.
Stability of Rotors 177

200
3

>-
LJ
s
:z
....,
::J
8100
a::

Q I t I]OJ
u..

s
-' EIGENVECTOR

~
1000 ?000 ~000 400'1 5000
ROTATIONAL SP£EO RPH
Fig. 2 First Natural Frequen-
10
1 cy and Damping Constant
·r
~ o~---,----~---.---.~r-r-­ of the Gasturbine Com-
sooo
pressor Shaft
OPERATIONAL
SPEED
\:> -20 4485 RPM
:z
a:
~ ·30

OIL FILM BEARINGS WITH

STIFFNESS COEFFICIENTS

'ik DAMPING COEFFICIENTS

CENTER

OF GRAVITY

COOROINA TES
RIGID ROTOR WITH

MASS DISPLACEMENTS

EQUATORIAL HOHENT OF INERTIA ANGLES ABOUT AXIS 1 AND 2

ep POLAR MOMENT OF INERTIA

Q ROTATIONAL SPEED

Fig. 3 Rigid Rotor with Fig. 4 Coordinates for the

Oil Film Bearings Rigid Rotor
178 R. Nordmann

To describe the motion we introduce coordinates (Fig . 4).

A rigid body has a total of six degrees of freedom, three trans-
lations and three rotations. For the rotatior.. about the x 3-axis
we suppose, that the angular velocity Q is constant. Therefore
this is no real degree of freedom. Furthermore we do not con-
sider movements in axial directions. The remaining four degrees
of freedom are the displacements u2 of the center of gravity
11 1 ,

and the angles ~1 , ~ 2 about the axis 1 and 2.

From lubrication theory it is well known, that the pressure

distribution in the oil film of a cylindrical bearing with a
time dependent journal motion can be described by the Reynolds
equation (Fig. 5). The solution of this differential equation
leads to the pressure in derendence of the journal displacement
and velocity. By integrating the oil pressure in circumferential

OIL VISCOSITY t- e /ll r

PRESSURE P

SI •
~ 12
801)11- SI I t, a, t...a I
llr•R-r CLEARANCE

llr ·llr lr REL. CLEARANCE

sl -eo
6tllrl s I .
l)·n- l t' a . t
.
a I
I 13 • BID WIDTH - DIAMETER- RATIO

Fig . 5 Pressure Distribution and Forces ~n a Journal Bearing

as well as in axial direction, we obtain the resultant forces

F 11 in horizontal and F21 in vertical direction, respectively .
These forces can be written in dimensionless form: s 1 , s2 with
the relative clearance 6r, the width B, diameter D, oil visco-
Stability of Rotors 179

sity n* and the rotational frequency n. They are nonlinear func-

tions of the relative eccentricity £ = e/6r, the angle a and the
corresponding velocities E,a (Fig. 5).

s, = = s,(£, a,£, a)
( 1)

If only a static load Fstat is acting from the vertical

weight (no force in horizontal direction), the corresponding
equilibrium position of the journal is characterized by

£ =0 a =o , £ 10 , a 10

s, = 0

The remaining vertical dimensionless static force

= So (2)

1s the wellknown Sommerfeld number So.

We recognize, that the static equilibrium position of a rotating
journal in a bearing is determined by the So-number, or by the
average bearing pressure p, the relative clearance 6r, the oil
viscosity n* and the rotational speed n. Different So-numbers
correspond to different equilibrium positions, shown in the sta-
tic equilibrium curve (Fig. 6).
For n=O (So= 00 ) the journal center is at a low position and for
n="" ( So=O) the journal center moves up to the bearing center.
180 R. Nordmann

ML

a
Fig. 6 Static Equilibrium Posi-
E I tions for Journal

If small vibrations about a static position occur, additio-

nal dynamic forces are actin~ from the oil film to the journal.
They depend on the displacements u 11 , u21 and the corresponding
velocities u11 , ~ 1 , represented in an cartesian coordinate sys-
tem (Fig. 7).

E
Fig. 7 Dynamic Forces due to
Journal Vibrations

aF11 aF11 aF11 • aF11 •

t.F11 = -,..-- u 11 +a~
--u +--u +--u
ou11 1 21 au. 11 11 au.21 21
( 3)

aF21 aF21 aF21 • aF21 •

-- u
t.F21 =au + - - u21 + ..,..........- u + -=-r- u
11 11 au21 au 11 11 au21 21
Stability of Rotors 181

6F 11 , 6F 21 are first order expansions of the oil film reaction

forces (Taylor expansion). Equation (3) points out, that the dy-
namic system oil film bearing can be modeled by means of stiff-
ness and damping coefficients

aFiL
kik = - - - Stiffness coefficient of oil film
aukL (i,k = 1 ,2)
{4)
- aFiL
cik = • damping coefficient of oil film
aukL (i,k = 1 ,2)

for small motions about the equilibrium position.

The four stiffness and damping coefficients can be assembled

to the 2x2-matrices ~ and~. respectively. They show, that the
oil film behavior is anisotropic (k 11 ~ k22 , c 11 ~ c 22 ) and that
the coupling coefficients k 12 , k 21 generally are not equal. The
stiffness and damping coefficients are very often defined as non-
dimensional quantities, for example as

6r 6r It
=k -- c --- ( 5)
ik Fstat ik Fstat

For a given bearing type yik' Sik are dependent only on the Som-
merfeld number or on the static equilibrium position of the jour-
nal, respectively. Fig. 8 points out the dependence of this di-
mensionless coefficients on the Sommerfeld number for a circular
cross section bearing. The values are measured quantities, deter-
mined by Glienicke /1/. We recognize the anisotropic and nonsym-
metric behavior.
182 R. Nordmann

r:t · k F-~-F-1-:=P-jb~
+., .. '.
. .~ ~~:~~
1-'1 - .- ,,., . .

L. - :~~r
100
100 -= - ~ ~£==.:.:.~!!•_l_j

10
10

0.1 .!-- --+--!-+-~.?;.....--+-......-<.-.+Mr-~-

0,01 0,1 o,, So

Fig. 8 Dimensionless Stiffness and Damping Coefficients /1/

Equations of Motion

Based on the force-motion relationships for the components

of our mechanical system, we are now able to derive the equations
of motion. This can be achieved by application of Newton~ law for
the four coordinates u 1 , u2 , ~ 1 , ~ 2 • Figure 9 points out the bear-
ing forces in two planes acting on the rotor, when it is displaced
(u 1 , u2 ) and twisted (~ 1 , ~ 2 ) about the center of gravity.

The upper diagram in Figure 9 represents bearing forces ~n

the x 1-x 3 plane. With a displacement u 1 and an angle ~ 2 the left

bearing displacements is (u 1- a ~ 2 ) and the corresponding force
component k 11 (u 1- a ~ 2 ). Caused by the cross-coupled coefficient
k 12 , another force k 12 (u2 +a ~ 1 ) has to be added from movements
in the other plane. Furthermore the damping forces have to be
completed.

By taking into consideration all bearing forces the four dif-

ferential equations of motion are found, describing the dynamic
equilibrium of the rotor bearing system. Because of the special
Stability of Rotors 183

- I
knlu 1 •a'i121 • k12 1Uz-a'i111 •
u
t 11 l u 1 •a~ 2 1 • c12 l 2-aljr,1

k11 lu1 -a'i12 1• k,zlu2 •a'il,l

c 11 lu1 -a~2 l• c12 lil2 •a~,l

k 22 1uz•a'll,l•kz,lu 1 -a'll 2 1•
Czz lil 2 •alil,l • c 2,1il 1 -alil 21 k 22 1u 2 - a'll11 • k21 lu1 •a'll21 •
czzl ilz -alii,) • c2, I u, •aq,21

Fig. 9 Dynamic Bearing Forces acting on the Rotor

nature of the system (symmetrical rotor, same bearings on both

sides), the equations are partly independent. There are two coup-
led equations for the translatory motion and two coupled equa-
tions for the rotatory motion (Figures 10 and 11). The first one
contain inertia terms, bearing forces and in additio~ external
forces (Fig. 10). Concerning to the rotatory motion, we have
again inertia terms from the equatorial moment of inertia, addi-
tional gyroscopic effects with the polar moment of inertia (damp-
ing matrix), forces (moments) from the bearings and external mo-
ments (Fig. 11).
184 R. Nordmann

(6)

INERTIA FilmS Of OILFILM FORCES EXCITER

RIGID ROTOR !BEARING! FORCES

Fig. 10 Equations for Translatory Motion

(7)

INERTIA MOMENTS, CAUSED BY BEARIN:i FORCES, EXCITER

MOMENTS GYROSCOPIC EFFECTS MOMENTS

Fig. 11 Equations ~or Rotatory Motion

Stability Analysis

To analyze the stability behavior of the rigid rotor with

oil film bearings we neglect any external forces and moments,
respectively. Now the two sets of homogeneous equations without
the right hand side terms have to be solved. The solution proce-
dure for the translatory motion is explained in Fig. 12. The
equations of motion can be written in short form with mass matrix
~'damping matrix ~and stiffness matrix K. It has to be noticed
once more, that£ and ~are generally nonsymmetric and depend on
the operating conditions, particularly on the rotational speed n.
The natural motion of the rotor has the general form ~(t) =
u"' e A.t • subst1tut1ng
. . t h'1s 1nto
. t h e above equat1ons
. . l .l:
yv~ '3. qua-
dratic eigenvalue problem. Corresponding to the 2x2 matrices the
Stability of Rotors 185

EIGENVALUES AND
EIGENVECTORS

\ Ml:i • cu • Ku = 0

U(tl =

K l u = 0

EIGENVALUES
Fig. 12 Solution Procedure for
Eigenvalues and
EIGENVECTORS : Un= 5 +
0 I tn Eigenvectors

system has four eigenvalues and eigenvectors. They normally occur

in two conjugate complex pairs (Fig. 12). The part of the solu-
tion, which belongs to such a pair, can be written as

a t
u (t) = B e n (S sin(w t+y ) + t cos(w t+y ) (8)
n n n n n n n n

It consists of an exponential function and of harmonic functions.

w is the circular natural frequency, a 1s the damping consta~t.
n n
By means of this damping constant, which 1s the eigenvalue real:
part, one can evaluate whether the corresponding natural motion
u (t) increases (a > 0) or decreases (a < 0) (Fig. 13).
n n n
If all eigenvalue real pQrts are negative, the system is stable.
If only one real part is positve, we have an unstable system.

Fig. 14 shows the circular natural frequency and the damping

constant of the dominant first eigenvalue versus the running
speed. These results correspond to the translatory motion and are
presented in dimensionless form. w 1s a reference frequency, de-
o
186 R. Nordmann

fined by w2
' 0
= g/6r (g acceleration due to gravity). So
0
is the
Sommerfeld number defined with n = wo' which is constant in the
presentation of Figure 14.

I
NATURAL VIBRATION I
D a..t ( . )
u.ltl = u,e s.srniW.t+y.J + t. coslw.t + Y.l

VALUATION OF STABILITY BEHAVIOR

O.n < 0 : STABLE a.. > 0 : UNSTABLE

Fig. 13 Evaluation of Stability Behavior by Means

of the Eigenvalue Real Part

2.0 CIRCULAR NATURL

FREQUENCY

1,0

0
1,0 2,0 w=O lw., 3,0

~ DAMPING
Wo CONSTANT
1,0 3,0
0.

WCiR" 2,66

STABLE UNSTABLE
-1,0

Fig. 14 Natural Frequency and Damping Constant of the Dominant

Eigenvalue versus Running Speed
Stability of Rotors 187

Real and imaginary part of the eigenvalue are changing with

the running speed because of the changing dynamic coefficients of
the bearings. The real part becomes zero at the dimensionless
speed of rotation wGR = nGR/w0 = 2,66, pointing out the instabi-
lity threshold speed. The corresponding natural frequency is
about half of the instability onset speed. Therefore the whirl
motion of the journal in the bearing is often called "half fre-
quency whirl".

In Fig. 15 the previously calculated eigenvalue is shown in

the complex plane (imaginary part versus real part). The dominant
first eigenvalue for the rotatory motion is also included in this

>-
...
u
z
~
a
~
-2,0 ..!!-
-'
<(
a:
~
<(
z
V1
V1
1,0 -~
0

~
TRANSLATION
...~
::t:
a
Fig. 15 Eigenvalues presented
- 0,8 - 0,6 - 0,4 - 0,2 0 0,2 in the Complex Plane
DJI1ENSJONLESS DAMPING CONSTANT aJw.

diagram. Parameter in both curves is the dimensionless rotational

speed w = n/w • In both cases it can be seen again, that the dy-
o
namic behavior of the system changes with speed, caused by the
different bearing characteristics. Both natural frequencies in-
crease with increasing speed w, both damping constants decrease.
Once again the instability of the translatory motion occurs, when
the dimensionless speed reaches the value WGR = 2,66. At this
speed the eigenvalue for the rotatory motion 'is still well damped
188 R. Nordmann

and not dangerous at all. The remaining eigenvalues of the dif-

ferent motions are also very highly damped and not presented in
diagram 15.

The stability threshold speed WGR = 2,66 was found for the
translatory motion of the rigid rotor, characterized by a special
system parameter So • If the type of bearing is fixed, So 0 is the
0
only parameter defining the stability onset speed. For other pa-
rameters So 0 other limit speeds WGR have to be calculated. If
another bearing type is selected, the threshold speeds will also
change. Fig. 16 shows a stability diagram for the rigid rotor with
oil film bearings, pointing out the rotational speed limit WGR in
dependence of the Sommerfeld number So o • The different curves be-
long to different bearing types. Such diagrams are very useful in
the design process for rotating machinery.

i ~ :' ! ; l
~~ 1 :
1 ! I l l'
~.; -- · -- . s _ P·or2
.. Oo - ~ Wo

Wo = ~
_! :!''
.:;,·
stab le

-: J
! :

.,
0. 1 10

Fig. 16 Stability Diagram for Rigid Rotor in Journal Bearings

Comparison with Test Results

Figure 17 shows a rotor test rig, which was originally de-

signed to measure stiffness and damping coefficients of oil film
bearings /2/. The system consists mainly of a relative stiff ro-
Stability of Rotors 189

tor and two cylindrical journal bearings with circular cross sec-
tions. A d.c. electric motor with speed control drives the shaft
in a speed range up to 6000 rpm.

This test rig is also well suited to measure damping con-

stants of a rigid rotor-bearing system. To achieve this, the ro-
tating shaft can be excited by a special impulse hammer. The dis-
placements of the shaft, caused by the impulse are measured with
contactless inductive pick ups near the bearings.

HAMMER WITH DISPLACEMENT

ACCELEROMETER PICK UP

~
JOURNAL
BEARING

D.C. ELECTRIC MOTOR

WITH SPEED CONTROL
CONCRETE FOUNDATION
ON ELASTIC SPRINGS

Fig. 17 Test Rig with Rigid Rotor and Oil Film Bearings

The measured system response signals mainly contain one eigen-

frequency with the corresponding damping constant. The other
eigensolutions decay very rapidly, pointing out the high damping
of this eigenvalues. From the decay rate of the displacement sig-
nal the damping constant can be determined. There is another pos-
sibility to find the damping from frequency response functions,
calculated with the force input and the displacement output.

In Figure 18 measured damping constants a/w are presented

0
for different rotational speeds w = D/w0 • Exept the speed all
other bearing quantities are held constant during the tests.
There is a good correlation between measured results and the cal-
culated values from the above presented model. In both cases the
190 N. Nordmann

!
a,
i
w.
0.5 2.5

- - MEASUREMENTS

- - - CALCU.ATION

Fig. 18 Damping Constant of a Rigid Rotor with Oil Film

Bearings in Dependence of the Rotational Speed

damping constant decreases with the rotational speed. Measure-

ments were carried out up to w., 2.0. The extended curve shows,
that a zero crossing of the damping constant is expected at
WGR- 2,6, as predicted by the model.

References

/1/ Glienicke, J.: Experimental Investigation of the Stiff-

ness and Damping Coefficients of Turbine Bearings and
Their Application to Instability Prediction.
Symposium in Nottingham. The Institution of Mechanical
Engineers , 1966 •

/2/ Nordmann, R.: Identification of Stiffness and Damping

Coefficients of Journal Bearings by Means of the Im-
pact Method. Dynamics of Rotors, CISM Course.
Springer-Verlag Wien-New York, 1984.
 CHAPTER 7

COMPUTER ANALYSIS OF ROTOR BEARING SYSTEMS

P.A.L.L.A.: A PACKAGE TO ANALYZE THE DYNAMIC

BEHAVIOR OF A ROTOR-SUPPORTING STRUCTURE SYSTEM

G. Diana, A. Curami, B. Pizzigoni

Dipartimento di Meccanica, Politecnico di Milano, Milan, Italy

ABSTRACT
A ::.acKage of programs which can be used to study the static
and dynamic behaviour of a multi-supported shaft line is
illustrated. The analysis is carried out by Keeping account
of ~he interaction with a supporting structure of same
(foundatlon, casings, etc.>. Together with the structure of
the pacKage, this paper is aimed at illustrating the
~athema.tical mocel wsec ~o~ static and dynamic analysis.

7. 1. 1 Foreword
"P.A.L.L.A." is the name of a pacl<age of programs developed by a research
group worl<ing at the Sezione di Meccanica dei Sistemi of the Dipartimento
di Meccanica of the Politecnico of Milan. This package is the result of a
series of research worK, carried out over the last ten years. in
collaboration with ANSALDO, ENEL-CRTN, ENEL- DPT.SASt TECNOMASIO
ITALIANO BROWN BOVERI and 'WORTHINGTON NORD ITALIA. The pacKage
can be used to study the different problems that arise in rotor dynamics:
the analysis is carried out by l<eeping account of th• iteraction between
the shaft line and the supporting structure of same (foundation, c&se,
etc.>.
192 G. Diana et al.

7.1.2 Description of the Mathematical Model

Schematization of the system
The system analysed consists of a shaft-line composed of several rotors,
bearings, possible seals and a carrying structure: each single element is
schematized in turn as follows :
- ROTOR - The rotor is schematized with finite elements of the "beam"
type with four d.o.f. per node, as shown in fig. A.1. The choice of the
number of elements is conditioned both by the rotor geometry as well
as by the frequency range which one wants to analyse. In fig. A.2. the
schematization of a rotor is shown as an example.
- BEARINGS and SEALS - Bearings and seals are schematized with
equivalent springs and dampers. As will be explained in more detail
further on, this equivalent elements are calculated by evaluating, in the
neighbourhood of the rotor's equilibrium position, the force field due to
the bearings and the seals fluid film. The system composed of rotor and
bearings and/or seals is shown in fig. A.a.
- SUPPORTING STRUCTURE - By "supporting structure" we mean the
entire structure surrounding the rotor except the bearings and the
seals. The dynamic behaviour of the supporting structure is defined by
means of a modal approach. The modal parameters i.e. generalized mass,
stiffness and damping relative to the different eigenmodes of the
supporting structure are calculated by means of modal identification
techniques with one or more degrees of freedom ([1 J, [2J) starting from
the tramfer function of the foundation itself. The transfer function (or
the frequency response) can be defined both by means of a mathematical
model via F .E .M. ("beam" finite elements) of the structure itself, either
starting from the experimental measurements carried out on a real
structure or on a physical reduced scale model. The modal parameters
are used to define the mechanical impedances of the carrying structure
through which, in the rotor-fluid film- carrying structure overall model,
the dynamic behaviour of the latter is Kept account of.
The complete model and the fundamental eguatipns
An example of the complete model rotor-oil film-carrying structure is
shown in fig. A.4., where, as can be seen, as far as the carrying structure
is concerned, the single nodes connecting the rotor-structure, together
P.A.L.L.A. :A Package to Analyze ... 193

with the correspondent transmitted forces are considered. If X indicates

the d.o.f. vector relative both to the rotor .!R as well as the carrying
structure~F' the equations of motion of the entire system are:

<A.1>

where

-F* = {.fRJ
F* =
-F
<A.2>

where ~Rand fR are the subvectors relative to the degrees of freedom of

the single rotor, while ]F and !'~ are the subvectors relative to_ the d.o.f.
of the connection nodes. In eq, CA.U tM*J, tR*J and tK*J are respectively
the mass, stiffness and damping matrices of the system obtained by
assemblinQ the rotor matrices and where the equivalent stiffness and
damping coefficients of the bearings and seals are l<ept account of. F* is
the excitation forces vector composed, in turn, of forces ! effectively-
external, applied directly to the line rotors and/or to the connection
,
points and forces BF transmitted to the connection nodes by the carrying
structure. These vectors can be subdivided into subvectors:

f = {~:} <A.3>

By using the matrix partitioning technique eq. <A.U becomes:

In eq. <A.4> the forces .BF transmitted by the foundation are unl<nown
functions. In order to define these functions, as has been said, a modal
approach wu used. The vector g of the modal variables of the foundation
can be correlated to the displacements of the connection nodes .!F of the
foundation itself by means of the following equation:

~F=£~l9, <A.5)
194 G. Diana et al.

where t4 J is a matrix containing the eigenvectors of the foundation

examined. Ccpl is generally a rectangular matrix with as many columns as
modes considered and as many lines as degrees of freedom of the
connection nodes. It is possible to define the vector of the connection
forces between the rotor and the carrying structure BF as a function of the
variables 1F using equation (see e.g. c:n, [4J>:

<A.6>

where cm,J, CrFJ and [I<FJ are diagonal matrices containing the modal
parameters of the supporting structure. If as many modes as degrees of
freedom ASsociated to the connection nodes are assumed, CcpJ is a square
matrix and can be inverted thus allowing the transformation

-1
~ = [cf l ~F <A.7>

which substituted in eq.(A.6> gives:

<A.S>

where prime indicates the transposed matrix. Equation tA.S> t•> correlates
the displacements of the connection nodes ~F to the forces .BF transmitted
to the rotor through these nodes. By now substituting the expression of
the forces BF of (A.S> in equation tA.4> and by moving the terms containing
j,, iF and .!F to the left hand side, the fundamental equation <A.t> can be
rewritten as:

•• •
- -
[Hl X + [Rl X + [Kl X = F <A.9>

where CMJ, CRJ and CKJ repres!!nt the mass, damping and stiffness
matrices of the overall system that keep account not only of the

t•> If the eigenvectors are orthogonal in the rutrict sense, eq. A.7> can
I
also be written as g • ctJ !F·
P.A.L.L.A. :A Package to Analyze ... 195

inertia., damping a.nd stiffness of the rotor but also of the bearing/seals
fluid films and the effects of the carrying structure. Vector ..E in eq. <A.9)
therefore represents the vector of the single external forces applied to
the nodes of the rotor and to the connection nodes. Equation <A.9>
therefore permits, once E is Known a.nd different from zero, any problem of
forced motion to be solved. Always with eq. <A.9> on the other hand, by
assuming E= 0 it is possible to study the free motion a.nd to analyse
instability condition by solving the associated homogenous problem.
Methods of solution
As will be e:<pla.ined further on, the different problems that are
encountered when studying rotor dynamics <forced vibrations and
instability> are cast, by the P.A.L.L.A. pacKage programs, into the form of
the evaluation of the frequency response to a harmonic excitation force of
the entire rotor-fluid films-foundation system, thus adopting a more
stable algo_rithm of solution.
As has been said, all the problems are studied by analysing the frequency
response to a. harmonic e>:cita.tion force of the type:

~ = £0 exp( i~t) <A .1 0)

Both here and further on, we will indicate the excitation frequency by Oe:
and the rotating speed of the rotor by .DR' both expressed in r/s <radians
per second>. With this excitation force, the forces at the connection nodes
are also of the type :

B.F = BFo exp ( iJlet> <A .11>

In this case eq. <A.8> simply becomes :

<A.12)

Since ~Fo is the vector of the <complex> vibration amplitudes of the

connection nodes :

<A .13)

In the case of the harmonic excitation force considered, by, using the modal
approach it is easy to define the matrix of the mechanical impedances
CH.Ile;>J. In fact if we assume that a. unitary harmonic excitation force is
196 G. Diana et al.

applied to the carrying structure at one of the connection nodes "along" a

d.o.f. <fig. A.5>, the vibrations at all the other nodal points can be obtained
by using eq. <A.6> and by substituting vector .BF with vector .f 1 whose
components are all null except for the d.o.f. where the excitation force was
applied. Having obtained the components qi of vector 9. from eq. <A.6),
composed of all uncoupled equations, .!F is obtained from eq. <A.S>; beeing
! 1 =I t 0 exp<i0e;t> it is also l-F = 1F 0 exp<iQe;t>. Since the -flexib1litv ma.tn:
is defined by eauation:

~Flo= £H<Of>J fto

the solution obtained above represents the c::lurrr. of ~H<.Ogl J
corresponding to the d.o.f. where the unitary excitation force was applied.
By changing the application node of·the excitation force 1t 1s ooss1ble to
define the entire deformation matrix HC(Oe;lJ and from i~s ir.version

(A.15)

it is possible to obtain the matri:< of the mecha!"'ical l'T:::Jedances.

In this calculation it is only necessarv to checK '!~at the number of f'!atural
modes is equal or l'ligl'ler tl'lan tl'le number of d.o •.j:, relative to the
connection nodes. Tl'lis inversion is not possible otherwise. Let us now go
bacl< to ea. A.1>; since!*=! 0 *exp(i1le:t> and therefore ~ = ~c·e;;~(iQgt>, 1t
can be rewritten as:

(-A-2[M*l
Mt:
+ i''-tR*J
--E
+ lK*l> -o
X =-o
F * <A .16>

where ! 0* contains both the external forcn as well as the forces

transmitted from the foundation through the connection nodes, i.e.:

!a*= {fRo}+ { 0 l = !o + BFo (A .17)

fFo !Fo J
I'll
By substituting BFo with the expression <A.12> and by bringing the term in
lFo of <A.12) to the left hand side, in the same way as done for equation
<A.9>, eq. <A.16> can be rewritten as:

[ E<Je >l ~o =£o <A .l 8)

P.A.L.L.A. :A Package to Analyze ... 197

where [E<Oe:>J is "the elas"todynamic matrix of "the overall system and J 0

the vector of the single external forces applied to the rotor and to "the
rotor-carrying structure connection nodes. All the problems relative to
rotor dynamics can be solved by using the frequency response of thtt
system given by <A.1E:), which represents a complex non-homogenous
algebraic expression. By assuming that Jle:=O, eq. <A.lS> represents "the
response of "the system to a static excitation force. The program offers
the us•r th• simplified possibility of schematizing the foundation by
m•ans of uncoupl~Pd mass-spring-damper systems for each bearing or seal
as in fig. A.6. In "this case, as r•gards the foundation, i"t is n•cessary to
simply assign the parameters relative to mass, spring and viscous damper.

Fig. A.1 - 8 d.o.f. "beam" finite •lem~tnt

Fig. A.2 -Shaft line schematization

liig. A.9- Sh&ft lin• + b•&rings system

 198 G. Diana et al.

11ec.hanicaL Impe dal'1ce.s

_i j_'

...
, .. ·: ·. Fo{)ndation
... ..
~

Fiq. A.4 - ShAft linl'-fluid film-carrying structure overall system

Fig. A.S - Carrying structure

connection nodes

Fig. A.6 - A mass-spring-dampers simplified representation of the

carrying structure
P.A.L.L.A. : A Package to Analyze ... 199

7 .. 1. 3 Package Configuration
The P.A.L.L.A. pacKage is basically interactive both as regards the data.
input as well as the choice of the type of processing to be used.
Calculation can be carried out either using the "demand" and/or "batch"
modes. The handling' of the set of the different programs is carried out by
means of a "PacKage General Menu" which carries out the various functions
by transferring control to different main programs which in turn recall
specific subroutines. Data. transfer taKes place by means of
mterconnection logical units which support the different input and/or
output files of the various programs. As specified further on, some of
these files are created interactively by the user, others automatically a.s
the output of subprograms. The preparation and use of the above
mentioned files allows the calculation to be carried out even in the "batch"
mode. The logic chart of the pacKage that evidences the different functions
is shown in fig. B.t. with the numbers of the relative "Functional BlocKs".
BlocK 0 is the "PacKage General Menu" for the handling of the various
programs: this guides the user in the choice of the different functions
carried out in the other bloc~. BlocKs 1,2,3 and 4 are recalled for the
introduction, in colloquial form, of the data relative to the shaft line,
bearings, seals and the carrying structure, which can be schematized by
means of the mechanical impedances or by means of uncoupled
mass-spring-damping systems. These same blocl<s supply the plots of the
shaft, bearings and carrying structure in order to checl< the accuracy of the
data introduced: eventual correction are carried out interactively. Once the
accuracy of the data introduced has been verified, data. files corresponding
to the different components previously mentioned are created
automatically from BlocKs 1,2, 3 and 4. BlocKs S to 13 carry out
intermediate calculation functions and are activated by means of
sequences piloted by the "General Menu" in function of the final
calculation which the user intends to carry out. BlocK s, which calculates
the overall mass and stiffness matrices of the shaft line <without fluid
films>, is carried out automatically, in that it is an obligatory step,
whatever the final type of calculation expected. The same goes for BlocK 6
which carries out the static analysis of the shaft line, analysis that
always have to be done since it supplies the loads on supports once
200 G. Diana et al.

o·
1--

i: I s~~a-r ~-11 nt
~
t
H
I

I: I ;fa!":~;!
- )i~• fi 1 t! :rti~=~n

I,I :n:: H
- ~cm~c~•~~ ;~~~~~= c·sc·a~

H C.•r~!"ln; s t""''JC ~~r!'

~ - ?~ec:r:e,.~c
mt~·att
t·tcu:~-:.r.
cal:ulatron
c~ .,..~1!'"'­

- E;:ecuttcn oi ~tni.l '=aicl.Jlc.t • .:r:

- Pio~trng ~

[ on;~-~~::-:~~:~~~~~~~-~~~~~;,---]
~ull ana.' or s:< 1e
~eou~!d l
+
~hysr~a.~ I"''Odt~ 1

---------------1-------------
• !

[~~~~::~~~:~~~~:;~~~~:~~~~~~~~~~~~~~~~~~]---~-[~~~~~~~~~~~~]
~ ShaH-1 i nt mass and st: dn!H
atr1cts

6 St~ttc anal/Sts:
al without ilu•d f1lms
OJ w1th "lu1d -films

Gr a~n r c

:~rrvtng stru:turt :rans4•r

•unction 15 GraotoiC

Caiculat•on o-f mod<i parameter!

ana mtc~a~ICll lmPtcar.tt!

Eii.stoaynam•c mi.trtx E<Jl) of tht lo

outra ~ 1 svs t!m Stab1l1 ty analYSIS

!3 Non~synchronous txc•tat•on
- Journai ovei ::i.ttons
- Crac-< tf~t':t!

Figure B.l - Flow chart of program showing function blocks

P.A.L.L.A. : A Package to Analyze ... 201

alignment has been assigned, or the alignment, once the loads and the
pc•sition of some of the supports have been assigned. In this same BlocK it
is also possible to Kee~ ;::our;t of the - -~--.;;c;;, :; "':he lubricating film to
define the loads on s:. · · : ·"':s once the alignment has be~r- =::": :, ' :· :·r
vice-versa. For this reason it is necessary to Know the fluid film stiffness
of bearings or seals, characteristics which are calculated at a given
angular speed of the shaft line O.R in BlocKs 7 and 8, by assigning the first
attempt loads defined in BlocK 6 as input loads. To improve the calculation
an iterative procedure between BlocK 6 and BlocKs 7 and E: is followed.
BlocK 6, therefore, calculates the alignment conditions while the shaft is
static•nary <in the absence, therefore, of an oil film, or by Keeping account
of the oil film stiffness corresponding to an assigned angular speed). Once
the alignment conditions have been defined, BlocK 14, on request, gives the
stress calculation along the entire shaft line due both tO' the alignment
conditions as well as, for the horizontal shafts, to its own weight. BlocK 9
1s activated should the user wish to Keep account of the carrying structure
through the mechanical impedances. In this BlocK, the dynamic behaviour of
the carrying structure itself whose transfer functions are defined in
correspondence to the connection nodes with the shaft is simulated
through a Finite E:lement schematization. These transfer functions can
also be assigned by the user, as in the case where they are defined
starting from the e:<perimental data. BlocK 10 utilizes the transfer
functions derived from the calculation <BlocK 9), or assigned by the user to
define the modal parameters of the carrying structure. The matrix of the
mechanical impedances is also calculated from the modal parameters for an
assigned frequency .Oe;. The data processed in BlocKs 5, 6 1 7, 8, and 10 are
used in BlocK 11 to create the elastodynamic rna tri:< CE: <Ae;) J of the whole
system, matri>: defined by relation:

( B ,1)

This matri:{, besides the mass and stiffness matri:< of the shaft, contains
the rnatri:< of the mechanical impedances of the foundation and the
stiffness and damping matrices of the oil film which are a function of the
angular speed of the rotor ~ which coincides with the frequency of the
202 G. Diana et al.

excitation force OE in the event of synchronous excitation which is the

case of a rotating unbalance.
If the frequency Jls: of the excitation force is different from the angular
speed~ of the rotor, matrix [E J also becomes a function of 1lR, since the
stiffness and damping constants of the bearings and the seals are
dependent on the rotating speed of the rotor. BlocK 15, by using the
elastodynamic matrix assembled in BlocK 11, solves the forced problem
with sinusoidal excitation i.e. solves eq. (B.i>. Through BlocKs 12 and 13
the user can introduce different excitation forces to be applied to the
system for the calculation of the frequency response <BlocK 15). Possible
different excitation forms are:

Block Type of function

12 Synchronous excitation - Unbalance response

- "Crank" effect

13 Non-synchronous - 2 per rev. due to shaft

excitation variable stiffness

- 2 per rev. due to

journal ovalization

- transverse crack

BlocK 16, using matrix [E<flg>J defined in BlocK 11 and the stiffness and
damping defined in Blocl<s 7 and S carries out, at the user's request, the
stability analysis of the shaft line. Graphic outputs are also available for
Blocl<s 14 and 15. Bloc!< 3 relative to the introduction and creation of data
files for the seals is still in the implementation stage. So is Bloc!< 8
relative to the calculation of the equivalent stiffness and damping
coefficients of same.

7.1.4 Block 1: Definition of the Input Data of the Shaft Line

The introduction of data regarding the shaft line is carried out in this
Bloc!<. As has already been said, the rotor is represented with beam finite
elements. The subdivision into elements (see fig. A.2> must be done in such
a way as to respect the usual schematization techniques for the use of this
P.A.L.L.A. : A Package to Analyze ... 203

approach. The elements are thicKened in correspondence to sudden diameter

variations. Length/diameter ratios of the various elements higher than a
limit value ls and lower than a minimum value li are avoided. The value ls
is limited by the fact that the single element should have a natural
frequency sufficiently higher than the frequencies at which we are
interested in calculating the behaviour of the shaft line being examined.
On the other hand, value li is linKed to the requirement of not having an
element with too small a length-to-diameter ratio with respect to the
surrounding elements: the stiffness of this small element would negatively
influence the overall stiffness matrix of the system. For each element the
external diameter, internal diameter and length must be given, besides the
value of a fourth quantity, the added mass, which will be specified further
on. The shaft material is assumed to be steel <specific weight 7860 Kgtm3,
Young modulus 21000 Kg/mm2), The automatic calculation program starting
from this data (element external and internal diameter, length>, the weight
of the generic element <considered as uniformly distributed along the
element itself> and its bending stiffness are calculated. If the area of the
transverse section which is effectively resistent (from the stress point of
view> does not coincide with the real one <as in the case of bladed disKs,
etc. being present> it is necessary to input the values of the external and
internal diameter of the effectively resistent section and an extra weight
<added mass> given by the difference between the effec:tive weight of the
element and the one calculated by the program starting from the assigned
geometric data. The added mass is also considered by the program as
uniformly distributed along the element. The introduction of shaft line
data occurs element by element, added to which there is the possibility of
controlling the elements already introduced using graphic outputs (fig.1.1.>
and, consequently, to correct the input data.
7 .1.5 Block 2: Definition of the Bearing Geometry

The part of the program described in this BlocK deals with the introduction
of the geometric data of the bearings of the shaft line being examined. The
thr-ee types of bearings presently considered by the program and which can
therefore be chosen by the user are:
- multi-lobe
204 G. Diana et al.

- tilting pad <20 pads max.)

- roller bearings
The lobe of the bearing 1s the circumferential portion of the bearmg
bounded by lubricant inlet and outlet pocKets, in correspondence to wh1ch
Known <supply and discharge> pressures are assigned. The fundamental
parameters to be assigned are the journal radius R and the radial clearance
a, which represents the fluid film thicKness when the journal axis coinc1des
with the center of the lobe <or generic pad). For this reason it 1s assumed
that the lobe (or the generic pad) has a circular profile. Non circular
profile are assigned through the method described further on. In the event
of the different lobes and/or pads having different clearances, a. reference
clearance is assumed: the other values of the radial clearance will be
specified through a. parameter e 0 i, which defines the ratios between the
clearances of the different lobes and the reference one.

Multi-lobe bearings
With reference to the generic i-th lobe <see fig. 2.1), the geometr1c
parameters to be assigned are :
- xci, Yci: . coordinates of the center of the lobe i-nth in a reference
system with origin in the theonc center of the !Jea.rmg ana
oriented as in fig. 2.1;
- eoi : ratio between the clearance of the lobe i-nth <assumed
circular) and the reference clearance;
- 8'oi.: start angle, measured a.s in fig. 2.1, of the 1-th lobe;
-17'-~~ : angular extension of the i-th lobe;
- JD<i> : number of cylindrical parts into wich the i-th lobe 1s a>:ially
divided <max. 31.
Assuming that the JOurnal axis is displaced by xc1,vc1 with respect to the
center of the i-th lobe. the film thicKness is defined by a function h!~ <see
[5 J) which is dependent on the bearing characteristics. When the journal
center assumes the coordinates x ,v in the reference system of fig. 2.1, the
film thicKness htm assumes the value given by:

h <-&'> = h 0 i <-8-) + ( x - xc 1) sin& + ( y - y c 1 ) cos&

In the case of the circular lobe considered uo till now we have equatlon
P.A.L.L.A. :A Package to Analyze ... 205

hoi = const. = 8, <2.2)

which defmes clearance ai of the i-th lobe. This clearance is linKed to

clearance 8, assumed as a l'eference, by equation:

<2.3)

which specif1es the meaning of carameter eoi• For the cases in which the
bearing profile is not circular, function hEat assumes the expression:

<2.4)

w1th1n intervals of amplitude ~<in a maximum number specified in the

User Instructions Manual) such as eq. 2.4> approximates the effective
profile with the desired accuracy <see fig. 2.2>. &:q. <2.4> is used, with
a = b= o, for bearings equipped with "grooves" with a profile that varies
linearly with the angle (fig. 2.3>. In its most general form eq. <2.4> it is
used for cases when the effects of wear of the bearing has to be
cons1dered <fig. 2.2>. In the case in which, in the i-th lobe, there are
circumferential pocKets with a large clearance <sn fig. 2.1>, the single
cylindrical portions <which the program considers equal one with the other
and 1n a maximum number of 3> are considered active, as far as carrying
capacity is concerned. The width of each of these portions is represented
by pal'ametel' bi previously specified. The so called "elliptic Uemon>
standard" bearing is a two-lobe bearing with the following geometric
characteristics <fig. 2.4.> already fixed by the program :

radial clearance 8 = 2.66Y.. of journal radius R

xcl= .5 8 ; xc2 = -.5 ~
Yet = Yc2 = 0
eol = eo2 = 1
- 16'0 1 = 25 deg. ; t6'0 2 = 205 deg.
- ~2 i = 130 deg.
JD<l> = 1 ; JD<2> = 3

Therefore, only the widths of the single cylindrical portion of the lower
lobe <b 1> and that of the upper one <b2> <fig. 2.4) have to be assigned. The
cylindrical bearing without pockets or grooves shDuld be treated as a
one-lobe bearing with a 360 deg. extension.
206 G. Diana et al.

Tilting pad bearings

For this type of bearing the film thicKness of the generic pad assumes the
expression <see fig. 2.5)

<2.5)

in which r 0 and cp 0 are the polar coordinates of the hinge and 1\.jJ is the
generic small rotation of the pad from the position assumed to evaluate
h0 i<-fh. The quantities Xci and Yci determine what in literature is normally
indicated by preload, which in the program is assumed to be a. radial
displacement of the tilting pad center with respect to the theoretical
center of the bearing (generally the center of the circle passing through
pad hinges>. The geometric definition of a. tilting pad bearing besides the
quantities r 0 and cp 0 which appear in eq. <2.5> and the radial clearance
already defined at the beginning, (it is important to note that different
radial clearances are not expected for the var1ous pads and is therefore
always e 0 i = 1> requires, for each of the tilting pads, the following data :
- the preload of the tilting pad, defined as has already been said.
Non-dimensional ratio preload/radial clearance is used;
- the moment of inertia J 0 of the tilting pad with respect to the hinge,
assigned by means of an equivalent mass me= Jp/r 0 2,
- -\lctstart angle of the i-th tilting pad measured as in fig. 2.5;
-~~angular amplitude of the i-th tilting pad.
Moreover, the axial length b of the tilting pads must be assigned.

Roller bearings
For these type of bearings the stiffness matrix, i.e. the coefficients K>:>:t
Kxy' Kyx' Kyy in the previously defined reference system, must be directly
assigned. The introduction of the geometric data. of the bearings taKes
place bearing by bearing, with the possibility of controlllng the data
introduced by means of graphic outputs.

7 .1 , 6 Block 3: Seals Geometry

This BlocK has not yet been implemented in the pacKage.
7.1.7 Block 4: Carrying Structure
In this phase of the program all the data required to define the parameters
of the mathematical model of the carrying structure are orepared. The tyee
P.A.L.L.A. :A Package to Analyze ... 207

of mathematical model used is described in order to justify the necessary

input data.

Mathematical Model
The carrying structur·e is schematized via F .E .M. with 12 d.o.f. "beam"
elements ([6J, [7J,l. These elements (fig. 4.1l represent homogenous beams
w1th a constant section, undergoing a>:ial action, shear, twisting moment
and bending moment along the two a>:is of inertia of the transverse
section. In Tab. 4.1 the corresponding stiffness [~J and mass [mtJ
matrices evaluated in the local reference system are shown.
The corresponding structural matrices EKe J and CMt'J calculated in the
absolute reference assumed for the overall system are calculated by means
of the well Known relation:

£Kt J= £AJ T [ kt J £AJ

4 .1)
[Mt J=LAJ T [mt J rAJ

where [A J represents the coordinate transformation matri:<.The

mathematical model that reproduces the carrying structure in the
P.A.L.L.A. pacKage program can also simulate the presence of rigid bodies,
linKed to the carrying structure itself, whose inertia contribution is not
negligible. The equivalent corresponding mass matrix, in the absolute
reference system, is shown in Tab. 4.II, where Rx, Ry and Rz are the
coordinates of the center <!If gravity G of the body in the local reference
system x, y, z, <see fig. 4.2 ), m* is the total mass of the rigid body and
T i,i (i,j = 1 ,2,3> represent the components of a matrix ETJ defined as:

[ T] =[A~ T [ I G] [J\~] 4.2)

.
where CIGJ is the matri>: of the polar inertia moments of the rigid body
wit:1 respect to the three principal axis:

0 1 OJ0
IyG (4.3)
0 2G
208 G. Diana et al.

snd rAI()'J is the coordinate transformation matr1:·: which correlate

rotations in the local reference system to those m the absolute reference
system:

Once all the structural matrices of the elements that comoose the complete
model of the carrying structure have been defmed, it is coss1ble to
assemble the total stiffness matri>: [}{sJ and the mass matri:{ [MsJ using
the usual techniques. Having assigned the constraint conditions tc the
structure, after partitioning the equations of motion relative to the
constrained and unconstrained ([6J, [7J> d.o.f., the equations of motion of
the entire carrying structure therefore becomes:

[Msl~s+ [ Rsl~s+ [ Ksl2Ss=£ s <t)

where [Rsl represents the damping matr-i>: calculated as:

(4.6)

BY assuming! 5 = 0 in eq. <4.5) and [R 5 J = [OJ, the eouation is reduced to

the form:

. 4. 7.'

whose solution:

X =X eiwt
-s -s
enables us to obtain the natural oulsations of the svsterr. and relative
principal vibrating modes<*). On the other hand by assuming

(*) Frequency resconse calculation and modal 1denhficat1on tec~nioues a~e

used instead of the eigenvalues and eigenvectors calculation. Reaso!"'ls are

given further on.
P.A.L.L.A. : A Package to Analyze ... 209

r~,..
-- -so l''.D.t
t •=F ~4.9)

m 4.5> it 1s possible to calculate the freauencv response of the carrvina

s"tr:.Jctw:-e for the ass1qned harmonic e>:citation forces. This oossibilitv is
used to calculate the transfer functions of the carrying structure <Bloc!< 9>,
+unc-t1ons evaluated to calculate the modal parameters of the same in order
to define the corresconaing mechanical imcedances matrix <Bloc!< 10>.

7.1.8 Block 5: Shaft Line Mass and Stiffness Matrices

The rotors of the shaft line analvsed bv the P.A.L.L.A. pacl<age are
schematized. as alreadv mentioned in caragraph A, by means of homogenous
beam elements, with a constant section, that reproduce the different mass
and stiffness characteristics of the same rotors ([$J. Each bum element
has ~: d.o.f. isee fie. A.1> to reproduce the deformation shace assumed by
the rotor m scace: the vertical X and horizontal Y transverse
d1splacements and the rotations of the node section are therefore
represented. The axial and torsional displacements of the shaft are
neghgible in that the associated vibrations <axial and torsional) are
generally uncoupled from the bending motions considered herein. The
correspondmg st1Hness CKJ and mass [mJ matrices of the generic "beam"
element wh1ch recroduce the elast1c and inertia characteristics of an
element of the rotor are shown in Tab. S.I. For more details see
C4J,C8J,C9J and C10J.

7.1.9 Block 6: Statis Analysis

This phase of the crogram enables the user to carry out the static analysis
of the shaft line by also Keeping account of the bearing lubricating films.
In the hyperstat1c system which is the one constituted by a shaft llne on
several supports, the loads acting on the latter are also functions of the
bearmgs alignment, i.e. of the position that these hold in the reference
clanes orthogonal to the undeformed shaft line axis. The l<nowledge of
these loads is 1mcortant in order to establish the worl<ing conditions of
the machine. This determination is obtained, in the static field, using a
P.A.L.L.A. pacl<age program.The aims of this program are:
a) determmat1on of the ahgnment of the supports to obtain
ere-established loads on bearings.This is the method used to calculate
210 G. Diana et al.

the supports alignment on starting up the machine. For the rotors with
horizontal axis an alignment is normally imposed so that the loads on
the bearings correspond to the so called "single beam reactions", i.e.
reactions exerted by the uncoupled rotors. Each rotor is normallv
placed on two bearings: the constraint reactions determined in this wav
are also maintained when, by means of the end joints, the
aforementioned rotor is constrained, flexurallv and torsionally, to the
other rotors composing the shaft line.
b) determination of the loads once the alignment has been assi~ned. This
method is useful when we wish to evaluate the effects of the support
displacements: e.g. thermal distortions of the foundation structure can
cause, on one or mar• supports, load variations with respect to the
nominal with consequent phenomena such as oil film instabilitv.
.
As previously indicated, the shaft line is schematized with 8 d.o.f. "beam"
finite elements. To obtain the stiffness and mass matrices of the rotor-oil
film system it is necessary to suitably assemble the stiffness matrices
cln <alrudy described in Blocl< 5> of the single elements and the
equivalent oil film stiffness matrices <see Bloc!< 7>. The d.o.f. number of
the shaft nodes is gen~rally very large so that condensation becomes
necessary (fig. 6.1). In the static analysis carried out whether the
lubricating film is l<ept account of or not, the condensation adopted, of the
static type, proves to be rigorous. The use of this approach decomposes
the generic shaft line into a set of substructures <called "superelements">
each composed of several beam elements with connection nodes between
superelement and superelement as extremitY end nodes <called
"supernodes">. As a substructure the program assumes the portions of the
rotor connecting two supernodes. Extremity nodes and bearing nodes are
always assumed to be supernodes <see fig. 6.2>. For the generic
superelement the equation of static equilibrium can be expressed as:

( 6 .1)

where ~e is the vector of the displacements relative to the generic

superelement, [Ke J is the relative stiffness matrix assembled as shown in
fig. 6.3. Finally, in eq.(6.U, fe represents the vector of the generalized
forces due to the weight of the rotor portion schematized by means of the
P.A.L.L.A. : A Package to Analyze ... 211

sucerelement. Having called !ee the disclacement vector of the nodes

external to the condensat1on <nodes 1 and 10 of fig. 6.'3> and Zei the vector
of the remaining degrees of freedom. it 1s possible to express equation
\6,1) 1n cartit1oned form, by reordering the vector Xe as:

~e -_{~ee}
. ~e 1
(6.2)

thus obtaining:

(6.3)

or rather:

(Keel~ee+[Ke;l~e;=fee a)
(6.4)
£K,el;:$ee+[K; ;l!e;=fe, b)

Obtaming from <6.4):

(6,5)

and subst1tutmg (6.5) in <6.4a), we obtain:

(6.6)

BY defmmg the st1ffness matrix of the statically condensed generic

suoerelement with

(6.7)

and ·with

(6.8)

the vector of the ~eneralized condensed forces. it is oossible to rewrite

ecuat1on <6.6> as:
212 G. Diana et al.

[KcJX
-ee =P
-c
In this way it is possible, to solve the static equation m rigorous form. b'!
using as degrees of freedom those relative to the single suoernodes.
Having calculated the matrices of the single superelements and the
matrices of the equivalent stiffness of the 011 film <BlocKs 7 and :?.) in th1s
way, it is possible to proceed with the assembly of the complete rotor + oil
film system as shown in figure 6.4. As regards the only shaft line, the
equations system for the static situation considered is therefore given bv:

[Kl~=£ (6.10)

in which f is the vector of the nodal forces equivalent to the external

forces <natural weights> and ! is the vector containing all the shaft line
d.o.f. (4 d.o.f. per node times the number of shaft nodes>. Matri>: [I<J which,
for now, is assumed not to contain the stiffness matrices of bearings or
Mals, as sangula.r. illmlnation of I:KJ's singularity <static equilibrturr.
condition> is otjtained by imposing at least fc·· di=ola.cements <isostat1c
case> equally partitioned in the horizontal and vertical dlrect1ons. By
partitioning matrix I:KJ, the displacement vector ~ and the nodal forces
vector! we obtain:

~~H-~ t~} H~J =

(6.11)

From system 6.11> we can formally obtain the disolacements of the free
nodes !Land the constraint reactions .fv according to the system:

~L=[Klll-lfL-[Klll-l[Klvl~v (6.12)
fu=CKuLl~L+[Kuvl~v (6.13)

by solving system 6.12> first and by substituting the values found thus for
!L in eq. 6.13>. Coordinates !v are always, as is obvious, bearing node
coordinates. They are as has been said, in a minimum number of two per
plane, horizontal and vertical, and in a maximum number equal to twice the
number of bearings <at the bearing nodes rotations are never constrained>.
The problem, discussed in point b) of the previous paragraph, of
P.A.L.L.A. :A Package to Analyze ... 213

determmmg the support forces when the alignment is assigned can be

solved by means of equations 6.12) and 6.13). Obviously, the same system
enables us to determine l::v once forces .Ev have been assigned • This
s1tuation is described in point a> of the previous paragraph. However, some
caution is necessary, Vector! v cannot be completely arbitrarily assigned,
since it has to respect the equilibrium conditions between the external
forces and the constraint reactions. To avoid possible mistaKes in this
sense, it is preferable to impose, in this case too, the minimum number of
two displacements at any two supports, thus obtaining the isostaticity of
the structure; on the two supports thus constrained it will not be possible
to impose constraint reactions. It is now possible to assign any set of
constraint reactions to the n-2 remaining supports and to obtain their
alignment always by means of eqs. 6.12> and 6.13), cast for this purpose in
ar.other form. At the end of this calculation phase the loads acting on the
supports are Known. It is therefore possible to Keep account of the
presence of the film of the bearings and/or seals, once an angular
reference speed has been assigned, For this reason and for the two
situat1ons previously described the program operates in the following way:
1l the loads on the bearmgs in correspondence to a certain rotation speed
of tne shaft line are imposed.
The program, once the displacements have been calculated in absence of
the lubricatmg film as prev1ously described, determines the relative
JOurnal-bearing displacements for the assigned rotation speed. The
relat1ve coordinates thus obtained and added to the ones indicating the
pos1tions of the bearmg centers determine the alignment to be
ass1gned to the la'tter during the assembly phase.
2> the alignment is assigned.
Once the loads on supports have been calculated as previously
described, the relative d1splacements which are added to the assigned
alignment are therefore calculated. The deformation of the shaft line
thus obtained does not however correspond to the loads calculated an
the basis of the imtial alignment due to the relative displacements
determined by the lubricating film. This accounts for the need to use an
1terative process which, for the saKe of calculation economy, is limited
to only one iteration,
214 G. Diana et al.

The calculation method followed to define the characteristics of the

lubricating film is illustrated in the description of BlocK 7. A more
detailed explanation of the program is given in [8 J where the calculation
procedures followetl to avoid matrix partit1ons and mversions which
appear in the equations previously shown are also described.

7.1.10 Block 7: Determination of the Bearing Static and Dynamic

Characteristics
The part of the program relative to this BlocK defines the force field that
is set up between the journal and the bearing due to the lubricating film,
as a function of the relative journal-bearing position. This field is
linearized in the neighbourhood of the static equilibrium position, thus
defining the equivalent stiffness anp damping coefficients. The calculation
of the equilibrium position of the journal inside the relative bearing and
the corresponding values of the stiffness and damping constants of the oil
film is obtained by integrating Reynolds' equation for both a laminar and
turbulent regime, Keeping account of the temperature variability along the
film. Reynolds' equation is written, in cylindrical coordinates (fig. 7.1l in
the form:

7 .1)

having indicated as

b axial width of the bearing

h f i 1m thickness
p pressure
R j ourna 1 radius
~ non-dimensional cylindrical coordinates
p turbulent viscosity of the lubricant for axial
and circumferential flow
.n angular speed of the journal
v velocity of the journal center.

The program automatically verifies the type <laminar or turbulent) of the

fluid film flow.
P.A.L.L.A. :A Package to Analyze ... 215

C~l::Uia 'tlOr. C't the torce actmg on the JOUrnal

Botr: for the lammar as well as the turbulent reg1mes the program
mtegra -t:es eaua tion <7 .! J followmg an appro>:ima te method shown in ( 11 J,
whose apollcations are shown m [5 J. After integrat1on of eq. <7 .Ut the
values of the hor1zontal Fv and vert1cal F>: c:omoonents of the fluid film
forc:e acting on the JOUrnal (fig. 7.2) are calculated:

Fx= Fx~x.y,x,y,.!lR)
(7.2)
Fy= FyO:x ,Y,x ,Y, .Q.R)

as functions of the assumed values of the coordinates x ,y and velocities

x,y of the journal center and of the rotation speed .OR• To c:arry out this
c:alc:ulatlon it is necessary to also ass1gn, besides the geometric: quantities
that enable oil film thicKness htlto be c:alc:ulated. the values of density 1
of the spec:iflc: heat Cs and the viscosity )L of the lubricant considered,
together with the temperature <in °C> of same at the entrance. The values
of C5 and p are considered constant while the viscosity )A- is assumed
var1able ac:c:ording to the ecuation:

(7.3)

where A.B and C are dependent on the lubricant type considered. As

regards the temperature, the program assumes. as described in [UJ, that
the walls of the meatus are adiabatic:, and, as a first approximation
calculation. considers the pressure effect negligible thus obtaining
e>:oression:

C dT.Id& = 2 \)(T)UR.fh2 (7.4)

m which U represents the peripheral speed of the journal <.a.tR> and V the
Kinematic viscositv. From equation (7 .4>, having assigned a temperature T0
at the beginning of the meatust solution T~ and the local viscosity value
)J e> from eq. (7 .3> are obtained. In the c:ase of laminar flow regime the
visc:ositv is defined as:

( 7 .5)
216 G. Diana et al.

On the other hand, in the case of turbulent flow reg1me. tne orograr;.
extends the validity of Revnolds' ecuation bv mtroducing the e>:cress1ons
of a turbulent viscosity derived 1n [12J on the oas1s of ~acers [13J ana
[14J. According to the findings snown. 11: is assumed tnat there are t·,..,c,
different e>:pressions of the turbulent viscosity, i.e.:

0.6.1

for flow in the circumferential direction and:

in the axid direction. In [14J the values of C'O'a.nd C'1 a.s a. function of the
local Reynolds' number Rh a.re not directly given. For use mside the
program interpolating functions with expressions

C~=0.00327 Rh4/S -0.36

•7. 8)
c,=o.oot7s Rh 415 -0.24

were determined which g1ve results that are fa1rly close to those of the
paper cuoted above for values of Rh ranging from 900 up to 200000. Tne
experimental results obtained in a test campa1gn, earned out on elhpt1.:
bearings normally used real turbosets showed the necesslt'! to mod1fy
1n

these expressions of the turbuler.t viscos1ty. For this curpos: ·.:ee ~15J)
other modifications were also introduced to a.na.lvt1callv reproduce the
distribution of the oil film temperature measured e:·:oerimental!v. As
shown in [16J,[17J,[18J an e}:cellent agreement between the e:<perimenta.l
locus of journal center and the a.na.lvt1ca.l one was obta.med bv usmg,
instead of eq. 7 .8>, expressions:

G.e-=<0.00327 Rh4/5 -0.755)p + 0.395 { f Q I

· , I ' . .·

C"'=<0.00178 Rh4/S -0.41D,P + 0.171

with p = 2. The systematic comparisons earned out using the e:-:oerimenta.l

results of those test campaigns did however show that th1s correcbon
alone is not enough. Finally, for the turbulent v1scositv e:<presslons:
P.A.L.L.A. :A Package to Analyze ... 217

p~= f < ot c~+1 >

(7.10)
JAi = f < ot c"l+1>
were used to obtain a good agreement between the experimental values in
which C'O'' C'rt are given by eq. 7 .9> and ot. has the expression:

dh
( de-
)2 dh
d.. =a + b + c (7.11)

The mathematical model used to calculate the thermal field is modified in

the second iteration and in the case of turbulent flow regime, in order to
l<eep account of the influence of the pressure flow. The expression of the
fluid rate of flow in a circumferential direction was assumed to be:

h
Q'6' = u (7 .12)
2 12f R

Moreover, in the expression of the thermal gradient, account was l<ept of

the power Wd dissipated as result of the harmonic journal oscillations with
amplitude U0 and pulsation Ge;. When calculating the temperature, it is in
fact necessary to evaluate the power dissipated locally (in every small
fluid element> due to journal vibration. The calculation was carried out in
an approximate manner by evaluating the overall power dissipated by the
journal with the simplified expression:

Wd = 2~ n2u 2R
.KE 0 XX
s i n <D
r (7.13)

where Rxx represents the direct dissipation term of the oil film and ~ is a
corrective term. By dividing this expression by the carrying surface of
bearing rtR b, we obtain an averaged power per unit area.. Moreover, this
power wa.s "weighed" with the local pressure p<-lt> compared with the
a.vera.ge pressure Pmt thus obtaining expression:
218 G. Diana et al.

= b R rc
-- =
Pm
7 .14>

since Qx is the buring load. Finally, the temperature gradient, keeping

account of the pressure flow and the dissipation associated with the
vibration, becomes:

R
7 .15)
Q~

where Q,. should be evaluated through equation <7.12) a.nd Rxx can be given
the value rxx calculated by the program.

Determination of the static equilibrium position

The procedure to determine the static equilibrium position of the journal
inside the single bearing consists of an iterative procedure enabling the
determination of coordinates x and y of the journal center so that
equations:

Fy<x,y,O,O,Qi = Qy
7 .16)
Fx<x,y,o,o,~ = Qx

are satisfied and where Qx and Gy are the load components on the journal,
according to the vertical x and horizontal y directions, evaluated as
previously described in Blocl< 9.

Determination of the equivalent stiffness and damcing coefficients

The determination of the stiffness and damping constants, to obtain the
equilibrium position calculated in this way, is made by developing
equations 7. t· l in Taylor series, thus obtaining expressions:

Fx<x+~x,y+~y,~x,~y,~ = Qx+ Kxxx + KxyY + Rxxx + RxyY

(7.17)
Fy<x+~x,y+~y,~x,~y,~ = Qy+ Kyxx + KyyY + Ryxx + RyyY

where

Kxx= 0 Fx/() x Kxy=o Fxld y

( 7 .18)
Kyx= oFyi ox Kyy=o Fyl'a y
P.A.L.L.A. : A Package to Analyze ... 219

represent the stiffness constAnts And:

Rxx= oFx/'0 ic Rxy• o Fx/() y (7.19)

Ryx= () F yi'O x Ryy= '() Fyl () y

the dAmping constAnts oof the lubricAting -film. The numericAl values of the
derivAtives in expression 7 .18) and 7 .19) are calculated by assuming thAt
the functions F x and F y have a. parabolic trend in the neighbourhood oof the
equilibrium position and are defined by separately giving increments 8x,
8y, 8x and ~y.

7.1.11 Block 8: Seal Stiffness and Damping Coefficients

This blocK has not yet been implemented in the "P.A.LLA." pa.cl<a.ge
program.

------ --------- - -

fig. 1.1 - Graphic outpu1: example

Fig. 2.1 - Geometric characteristic s

of a general lobe
220 G. Diana et al.

·-Y

Fig. 2.2 - Stepwise representation of a worn bearing profile

Fig. 2.3 - Three lobe grooved bnring

'
y

JIJ(2}=3
~
......
- --
~
In ~..J

...

Fig. 2.4 - Elliptic bearing

 2 3 4 5 6 7 8 9 10 11 12 "1:1
>
~
fAll
~
12EJzll-s S"/Jt1ETRIC ;.
12EJy/1~ ..
GJ"'/1 I (t1tmtntsnot sh~ art ztro> >
£Ktl = -6EJy/l1 4EJv/l
6EJz/J2 · 4EJz/l if
-fAll
t--····--'- 1--- ~1 {
~2EJ;Ii -6EJziil 12EJ.,./1:5
-12EJ)'/~ 6EJv/J2 12EJw I a-·
8
-GJx/1 .. GJ.t!L ~
-6EJy/tl
-- r - - - - ·4£;;;/1... 1
2f;J111 6EJu/1 2 ~
6EJz/12 2EJ /1 -6E.Jzlt ~
------~
I4EJz/l

2 3 4 5 6 7 8 9 10 11 12
1/3
13/35+6J1/~1 2
.13/35+6JII'/~J2 SYtt1ETRIC
J..l~
-111/210-Jr/10AI JCZ/105 +2Jv/15'A <t h11tn tr not shDIIIII art ztro)
111/210+ Jz/10AJ )2/105 +2Jz/15'A
(t1t]•f'l· l/6 -<12_~_2) <1,1>
9fi0-6Jz/5'AJl! -(11,3) (2,2>
9/70-6Jy/5'A12 (3.3)
· - - - - -f- ··- -
J~/6A _J_~~l.
13J/420-Jv/10AJ -)1 /140-Jlt'/30A
-- r - - - - - - - -
-(5.3) (5.5) -,
-131/420+Jz/lOA1 -1 2/14D-Jz/30A -(6,2) I (,,,,
TAB. 4.1 - 12 d.o.f. beam tlemtnt stiffntss and mass matricts t-)
t-)
.....
222 G. Diana et al.

Fig. 2.5 - Tilting pad bearing

····a...j
!

Fig. 4.1 - 12 d.o.f. "beam" element

Fig. 4.2 - A beam element A B

+rigid body system

0 0 0 Rz -Ry

0 0 -Rz 0 Rx

0 0 Ry -Rx 0
m*
0 MRz Ry T 1l ( R/+ R/ )/m* T 121m•- RxRy T 13/m•- RxRz

Rz 0 -Rx T 21 ;m•- RxRy T22<Rx2~ Rz2>;m• T2 3;m•- RyRz

-Ry Rx 0 r 31 ;m•- RxRz r 321mf - RyRz T33<Rx2+ Rz2);m•

TAB 4.II- B:quivalent mass matrix of system shown in fig. 4.2

P.A.L.L.A. : A Package to Analyze ... 223

I I
I 12EJz112 I I
1----------1---------- I
I I 12EJy112 I SYtt1ETRIC I
1----------1----------1-------- I
I I -6EJy/12 I 4Eoly11 I <eltmtnts not shown art ztro> I
1----------1----------1--------1--------- I
I 6EJz112 I I I 4EJz/1 I I
1----------1----------1--------1---------1-------- I
I -<1,1> I I I -<8,1> I <1,1> I I
1----------1----------1--------1---------1--------1-------- I
I I -<2,2> I -<7 1 2) I I I <2,2> I I
1----------1----------1--------1---------1--------1--------1------- I
I I (3,2) I 2EJy/1 I I I -<3,2> I <3,3> I I
l----------l----------l--------l---------1--------l--------l-------l-------l
I <4,1> I I I 2EJz11 I -<B,I> I I I <4,4) I
I I I I I I 1___ 1_ _ _ 1

I I
I 1:56 I I
1------1------ SYit1ETRI C I
I 221 I 412 I I
1------1------1------ (t1tmtnts not shown art zero> I
I I I 1:56 I I
1------1------1------1------ I
I I 1-221 I 412 I I
[ml= 1------1------1------1------1------
I :54 I 131 I I I 1:56 I
I tltl<D,2-o.2>
I( --------!___ + 11
) I
t420
1------1------1------1------1------1------ I 4
I -131 I -312 I I 1-221 I 412 I I
1------1------1------1------1------1------1------ I
I I I :54 I -131 I I I 1:56 I I
l------l------l------1------l------l------l------l------l
I I I 131 I -312 I I I 221 I 412 I
1_ _ 1_ _ 1_ _ 1_ _ 1_ _ 1_ _ 1_ _ 1_ _ 1

Tab. S.I - 8 d.o.f. "beam" element stiffness and mass matrices

 N
N
~

Fig. 6.1 - Shaft line + oil film bearings system

~ t :0.. ~ Supornades
------.---
-n-~ ~~ ~n } , { i)
Supernodes

4 IS
10

Nodes internal to condensation p

II)
~
n
Fig. 6.2 - Subdivision into superelements "
~
P.A.L.L.A. :A Package to Analyze ... 225

4 t--
element
~
2
.3 .I matrix
4 ~~
s ~ ~-
6
7 f234S 618910
8 l
9
10
[l<e] .__ w
FiQ. 6.8 - SupeNlement matrix

4 2 3 4 5 b 7 8 9 10 11 12 13

Fig. 6.4 - Overall stiffness matrix of the system <on rigid .foundation)
226 G. Diana et al.

~X
I

Fig. 7.1 - Fluid film geometry

tx
I

Fig. 7.2 - Fluid film force components

7.1.12 Block 9: Transfer Functions of the Carrying Structure

In this part of the program the frequency response of the carrying
structure is calculated using the mathematical model already described in
BlocK 4: the equations of motion of the support structure can be expressed
as:

<9.1)

where EMsJ, ERsJ and EKsJ are respectively the matrices of mass, damping
and stiffness of the foundation. In equation <9.1> .!s represents the nod~t
displacements of the same supporting structure, while.! 5 is the vector of
the applied excitation forces.
P.A.L.L.A. : A Package to Analyze ... 227

In the modal approach used to define the mechanical impedances of the

foundation, eq. (9.1) is cast in the frequency domain, thus assuming a
harmonic excitation force of the type:

F =F ei~t (9.2)
-s -SO

to which a response corresponds (also harmonic):

(9.3)

In the frequency domain eq. (9.1), Keeping account of eqs. (9.2> and (9.3), can
be rewritten as:

(9.4)

In order t[j calculate the matrix of the mechanical impedances of the

foundation CI<Ia:>J defined by relation:

(9.5)

where !; soc is the displacement vector of the foundation relative to the

only nodes that connect the rotor and ! soc the vector of the forces
exchanged in the same places, it is necessary to evaluate the flexibility
matrix CH<Dg>J defined by:

~soc=[H<Ae>lfsoc (9.6)

fhis matri>: is evaluated for a generic pulsation ~' starting with the
modal parameters of the foundation, parameters that can be evaluated if
only one column of CH<Ae;lJ is Known, evaluated for a certain number of
pulsations De;.
The generic 1-th column of the flexibility matrix CH<O..e;>J
represents the <comple>:) vibration amplitudes of the d.o.f. of all the
connection nodes for a unitary harmonic e>:citation force applied to the
1-th d.o.f., therefore corresponding to the index of the column considered.
The generic term hKl of the column 1-th is, as has been said, complex and is
called a transfer function between the K-th d.o.f. and the 1-th d.o.f. of the
rotor-foundation ccmnection nodes. To obtain the 1-th column of the matrix
CH<Oe;>J it is enough to therefore impose a unitary harmonic excitation
228 G. Diana et al.

force in correspondence to the d.o.f. relative to the column considered (*),

thus solving equation (9.4). The frequency range in which it is necessary to

evaluate the generic column b1 and for which it is therefore necessary to
solve eq. <9.4) is defined by means of the following criteria:
1> it must start from a null frequency and extend itself at least above
twice the rotor regime frequency;
2> it must extend itself above the worKing frequency so that a number of
modes, respectively vertical and horizontal, equal or higher than the
number of supports for each of the two excitations <vertical and
horizontal) can be determined.
Equation (9 .4) is solved by means of a Gauss algorithm with double total
pivoting.
7.1.13 Block 10: Definition of MOdal Parameters and Mechanical
Impedances
This Blocl< of the program enables one to carry out the modal analysis of
the structure carrying the shaft line i.e. to identify the natural
frequencies of the system, the damping factors and the generalized masses
associated to them. Moreover, for each natural frequency, the
corresponding vibration modes of the structure are determined. The modal
parameters of the carrying structure will be used subsequently to
determine the matrix of the mechanical impedances of the foundation, in
relation to the d.o.f. corresponding to the support points of the axis line.
The modal parameters are determined by analysing the transfer functions,
evaluated at the shaft line-carrying structure connection nodes, and,if
necessary, in other significant points to identify the vibration modes.
These functions can be obtained both experimentally, by tal<ing
measurements on a real model, and analytically, by means of a
mathematical model of the system. In this case the frequency response due
to a unitary harmonic excitation force is calculated (see Blocl< 9).

<*> In effect, the vertical and horizontal modes are uncoupled; it is

therefore necessary to calculate two columns of the matrix [HJ, one
relative to a vertical d.o.f. and one relative to a horizontal d.o.f., having in
this way to solve equation (9.4) twice, once with a vertical and once with a
horizontal unitary excitation force.
P.A.L.L.A. : A Package to Analyze ..• 229

The definition of the modes dttpends, in the experimentAl cAse, on the

measurttment points used, which should be at least all the support points.
In the following, with "foundation d.o.f.'' we will mean either the degrees
of freedom of the mathematical model or those of the measurement points
of a real structure used . to define the transfer functions. The modal
pArameters should be determined by analysing separately the transfer
functions obtained by applying one harmonic excitation force at a time in a
suitable plAce of the cArrying structure. The modal analysis program is of
the iterative type and requires the user to have a basic knowledge of some
of the modal Analysis notions. The single modal parameters can be
determined by carrying out in succession a series of logic and calculation
operations described in a menu and organized according to a
pre-established procedure. In the context of eAch calculation phase of the
modal parameters it is possible to use different algorithms, characterized
by different applicability fields. ·The applicability of the methods depends
particularly on the level of independence of the nAtural vibration modes,
on the damping fActor entities that characterize the systttm and finally on
the frequency resolution with which the transfer. functions have been
evaluated. The user has the possibility of intervening in thtt results given
by the single methods, discarding the values considered unsatisfActory.
The modal parameters must therefore be determined by using the results
obtained with different algorithms.

Identification of the modal parAmeters

In matrix form, the equations of motion of any vibrating system can be
written as:

(10.1>

where [Ml CRl And [Kl respectively represent mass, damping and
stiffness matrices of the system, ! is the vector relative to "n" d.o.f. and
l is the vector of the excitation forces. The modal matrix r+J, generally
rectangulAr n x p <where pis the number of the vibrAtion modes considered)
is formed by p eigenvectors f (j), ordered in columns Cthese values are
obtained as eigenvectors of the matrix [M ]-1 [KJ and for a stable
conservative system are real), i.tt.:
230 G. Diana et al.

[~]=[¥ 1) ;f2>; .. ·i<pl>] (10.2)

in which every eigenvector is formed by n terms xkCj) Calso l<nown as

"residues"), so that:

(10.3)

X (j)
"
By means of the coordinate transformation:

(10.4)

it is possible to rewrite eq. UO.U in principal coordinates:

(10.5)

where r-m-J, t"-r...J and L-k-J are the mass, damping and stiffness matrices
in principal coordinates (square and diagonal matrices pxp> and g are the
principal coordinates themselves Cthe matrix t-r-J is diagonal if [RJ is of
the type: [RJ=a[MJ+b[KJ). The generic diagonal element mj of eq. U0.5)
represents the generalized mass of the j-th vibration mode r j is the
associated damping and kj the relative stiffness. If the vector of the
excitation forces! of C10.1> is harmonic:

(10.6)

the solution will also be harmonic:

(10.7>

and in principal coordinates:

(10.8)

Equation 10.5) could then be rewritten as:

 P.A.L.L.A. : A Package to Analyze .•. 231

(10.9)

or rather

(10.10)

Once the modal parameters mJ r j' and l<j and the corresponding vibration
modes have been defined1 with the algorithms that will be described
hereinafter and having applied the transformation <10.4) we obtain:

(10.11)

i.e. :

(10.12)

having indicated the flexibility matrix with CH<Qg:>J:

(10.13)

As can be seen, once the modal parameters are Known and once the
pulsation.O.e: of the excitation force has been assigned, the matrix CH<Ag:lJ
is numerically determined and therefore, from its inversion it is possible
to obtain <see first chapter) the matrix of the mechanical" impedances
CI<Ot!:lJ. The generic term h1<1<Ae:l of the matrix CH<Og:>J is also named as,
as has already been mentioned, transfer function and represents the
complex frequency response of the system evaluated at point K, having
applied a unitary harmonic excitation force of pulsation~ at the point 1.
From <10.13) the generic transfer function can therefore be expressed in
function of the only modal parameters as:

(10.14)

where
w-2
J = Kjm·J represents the j-th natural frequency of the system
(j)
XI< represents the generic 1<-th component of the j-th
vibration mode.
232 G. Diana et al.

is the 1-th mode component in the application point of the

excitation force
S'·J is the associated damping factor, linKed to the
nondimensional damping ~j = <r Jrc} through the following
relation:

r · r · r ·
6"· = ~·J· = •· -L = .,. ~ = :...L. <10.15)
J J J J r . J 2m·t~t· 2m·
CJ J J J

Determination of the natural frequencies

The na tura.l frequencies wj of the carrying structure can be determined by
using two different algorithms. The first enables us to calculate the
undamped natural frequencies and the second the damped natural
frequencies of the system.
Method n.1
The first mathematical method consists in defining the "spectral power"
function obtained by adding, for each sample frequency, the quadratic value
of the imaginary component of each transfer function. The "spectral power"
function thus defined will show very accentuated peaKs in correspondence
to the undamped natural frequencies of the structure. Furthermore, this
method enables us to evidence all the resonances included in the frequency
range analysed, even in the case where some transfer functions were
evaluated in proximity to the vibration nodes; the lesser the damping the
more accurate the results obtained with this method.
Method n.2
Once first approximation values are Known (defined with the first method),
the natural frequencies of the structure can be more rigourously
determined by means of an algorithm which simultaneously allows the
determination of the natural frequencies and the relative damping factors
(see the method in following paragraph).

Identification of the damcing parameters

In order to determine the damping factors Erj three different methods were
implemented. Some of these methods offer reliable results if the damping
is low: it has, however, been experimentally shown that these conditions
are satisfied in a large number of structures.
PoAoLoLoAo :A Package to Analyze o o o 233

Method n.1
If the coupling between the vibration modes is negligible, the transfer
0
function phase presents a variation near to 180 in a frequency range near
to a resonance. A method which enables us to evaluate the damping factors
is based on the analysis of the derivative of the phase evaluated in
correspondence to the natural frequencies of the structure C19J. In fact,
phase "(( of the transfer function in a system with one d.o.f. can be
expressed as follows:

(10.16)

where w represents the n-atural frequency of the system and ! the

non-dimensional damping r/rc. Ey deriving equation <10.16) with respect to
wwe have:

dl\!' 2
- = (10.17)

In correspondence to the natural frequency w this derivative assumes the

following expression:

1 1
2'!2 = w.S
(10.18)

Therefore, an extension ton d.o.f. systems, if the natural vibration modes

of the system are uncoupled, for the j-th natural frequency Wj gives (see
eq. <10.15)):

(10.19)
234 G. Diana et al.

Method n.2
This algorithm is based on the calculation of the frequencies wd and wt,
near to each resonance of the structure, in correspondence of which the
amplitudes of the transfer function is 11V2 of the amplitude of the
resonance peaK. The calculation of the frequencies wd and wb is performed
by determining the solutions of equation [2'0J:

(10.20)

assuming St<r Jrc.} << 1 and neglecting higher order terms we obtain:

= 1+2E!:·
- 5J (10.21)

Indicating the roots of this equation with wd and Wb we have:

(10.22)

or

(10.23)

The accuracy of the results depends on the acquisition step of the

frequencies.
Method n.3
If hK11.4J} is the transfer function evaluated in correspondence to the j-th
natural frequency and Aw the acquisition step <the step with which the
transfer functions a.re Known experimentally or analytically) we have:

Ahkl(j) = hKl<~j)- hKJ<WrAw)

(1 0. 24)
AYK 1<j) = hK 1 <Ct»j)(lllj - hK 1 <Qrflc..n<c.tr.Aw)

The dampin!l factor associated to the j-th vibration mode can be defined by
the imaginary part which from ratio Ay<j)Kl/Ah(j)Kl [2J is:
P.A.L.L.A. : A Package to Analyze ..• 235

AYI< 1 ( j)
bhl<l(j)
= .,d·+i6'·
J J
<10 .25)

where with wdj the j-th pulsation of the damping system is indicated.

Identification of the vibration modes

Method n.1
The evaluation of the imaginary part of the transfer function in
correspondence to the natural frequencies is the simplest identification
method of the vibration modes; in this case too the accuracy of the results
is strongly influenced by the value of the damping factors and the coupling
of the modes themselves.
Method n.2
A more sophisticated method, Known a.s the Nyquist method, consists in
determining the circumference arc which approximates the transfer
function in the neighbourhood of each resonance; the value of the diameter
of the circumference is proportional to the modal amplitude, while the
corresponding phase is determined by the coordinates of the center. If the
coupling between the modes is not negligible but contribution given by
modes near to the one considered can be approximated (in the
neighbourhood of the frequency range where the j-th vibration mode is
predominant> by a. constant, we have <C1 J):

(10.26)

where with U tiVj the residue associated Xt<(j) to the j-th vibration mode is
indicated. The terms Re<D.g> and Im <De:> represent respectively the real
part and the imaginary part of the transfer function hl<l<Oa:> evaluated at
frequency .Og. Equation <10.26) is the equation of a circumference whose
center has the coordinates given by:

Xcj=-U j/26j
(10.27)
y c j=-V j/2 6j

while the value of the radius is:

236 G. Diana et al.

(10.28)

from which the residue value relative to the generic j-th modtt is
proportional to the diameter of the circumference <eq. <10.26». The
accuracy of the results obtained with this method depends directly on the
solving method used to determine the transfer function and on the number
of points used to calculate the interpolating circumference.

Determination of the generalized mass

Method n.1
Let us now consider an equation of the system <10.9), relative to the
generic j-th vibration mode: by assuming that we have only one harmonic
e>:citation force applied to the 1-th degree of freedom we will have:

<-n2E mj+iQErj+Kj>Qjo=Xfj) Flo (10.29)

where x1<j> represents the amplitude of the j-th vibration mode evaluated
at the point 1, Flo is the 1-th term of vector f 0 and qjo the j-th term of
vector So of eq. <10.9). By assuming resonance condition <Qg=CAJ} we will
have:

jn-r·q·
'"'"!: J JO
=i~·r·q·
J J JO
=X1(j)F1 0 (10.30)

or, with suitable simplifications:

2mj~j
a)
2mj(A)j
<10.31)
2imj~j~j2qjo = Xl(j)Flo b)

from which:

xl<j>Flo
(10.32)
2i~jf.J/Qjo
P.A.L.L.A. : A Package to Analyze ... 237

In eq. <10.32> Flo is Known, ~ j' x1(j) can be determined by the methods
wj,
described in the previous paragraphs. For the evaluation of qjo the
relation:

(10.33)

is assumed, which is rigorously true only in the case of orthogonal

eigenvectors in a restricted sense <mass matrix CM J of equation 10.1
diagonal>. Using this hypothesis, equation <10.4> can be rewritten as:

(10.34)

or

n
q JO ~ XK ( j >xk o
. --~ (10.35)
K=1

where XKo is the generic K-th element of the vector 1o' evaluated
experimentally by applying to the system an excitation force
F = F 0 exp<:iDe;t> or
by analytically calculating the latter in the same way
<see BlocKS>. By Keeping account of equation <10.34) eq. <10.32> becomes:

(10 .36)

From the definition of transfer function:

(10.37)

and by Keeping account of equation <10.19):

(10.38)

It is possible to reach the same result in another way, by assuming that

the coupling between the different vibration modes is negligible.
238 G. Diana et al.

Using this hypothesis, the error between the given transfer function

hl<l*<Wj> and the approximating function hl(lWj), evaluated in the

neighbourhood of the j-th natural frequency, is given by:

<10 .39)

The sum of the quadratic errors evaluated for the transfer function
determined in N measurement points is given by:

n 1 )
Et<6tj > =~ E2 ('>j, -. <10 .40)
k =1 . mJ
It is possible to minimize this error by imposing the following condition:

'c) ( 1/m j)
=0 (10.41)

from which we obtain:

(10.42)

Method n.2
The second method adopted tries, for each 1<-th d.o.f., to simultaneously
calculate the generalized masses mj of all the vibration modes: the user is
thus able to accept, mediate or discard the generalized masses calculated
for the generic j-th vibration mode in correspondence to the different
degrees of freedom. To identify the generalized masses the method
minimizes. for all the Wj considered in correspondence to the generic 1<-th
d.o.f., the function hl<l<'-»} (j=1,2, ..... p) defined by equation <10.14), with the
same known functions <determined experimentally or analytically> hl<l*<w;.
Analytically this can be expressed as:

xk<1>x1<1> xk ( i >x 1( i ) xk ( n >x 1( n) 1

i 2G'1c.»1 Wj 2-(.012+2 i CS'iw1 ... 6Jn 2-6)12+2 i &"nul
=
XK ( 1 >x1 ( 1) xk (n >xl ( n) 1
c.12~n 2+2 i 6'1"'n i 26"nc.Jn

10.43)
P.A.L.L.A. : A Package to Analyze ... 239

or

l.l = [T] 10.44)

!!!

In the case in which the number of the sampling frequencies is higher than
the gen&ralized masses to be determined, it is possible to define the
following error function:

1
g = h*-h = h*-£Tl 10.45)
m
To minimize the quadratic error:

1 T 1
gTg = [H-[Tl - ] [ tf-[Tl - ] 10.46)
m m
it is necessary to impose th& condition:

10.47)

thus reaching:

- = <£TlT£Tl>-1 <£TlT[Hl> <10.48)

m
Synthesis
The frequency response in correspondence to a point I< of a structure
forced in a point 1 from a unitary harmonic excitation force depends on the
excit&tion fr&quency Oa: and on the single modal parameters of the
structure (&q. 10.14)). Once the modal parameters are l<nown, it is therefore
possible to calculate the transfer function in correspondence to the
frequencies included in the frequency range analysed. This operation is
commonly called "synthesis". By indicating with Yl<l and Zl<l the
contributions rel&tive to th& vibration modes having frequencies which are
respectively lower and higher than the frequencies 01 and o2 delimiting
the frequency range analysed, we have:
240 G. Diana et al.

(10.49)

The more accurate the evaluation of the single modal parameters, the more
the function hK1<tJ} approximates the starting function hKl*<Wj • The
comparison between the starting transfer function and the one obtained by
means of the synthesis operation therefore enables us to evaluate the
reliability of the modal parameters calculated. The modal parameters w.i' G' j
and mj of the carrying structure (natural frequencies, damping factors and
generalized masses) are memorized, together with the vibration modes (j),f
on a file to subsequently be used to calculate the matrix of the mechanical
impedances. As far as the vibration modes are concerned, only the mode
components corresponding to the connection points <i.e. between the
carrying structure and the shaft line> are memorized. In fact, these
components are the only ones that are actively involved in the definition
of the matrix of the mechanical impedances of the foundation. Nonetheless,
as has been mentioned, to better define the vibration modes of the
carrying structure it is convenient to also calculate the transfer function
<BlocK 9) in correspondence to foundation points different from the
connection nodes. It is important to underline that, in order to define the
matri>: of the mechanical impedances CI <.Oe;>J it is necessary, for each
single modal analysis (i.e. for each direction along which the e>:citation
force acts) to identify a number of natural frequencies at least equal to
the number of supports. Moreover, the natural vibration modes identified
must 'have significant components in correspondence to the connection
nodes between the carrying structure and the shaft line. Therefore, all the
vibration modes in which a considerable number of connection nodes
coincide with a vibration "node" (i.e. a non vibrating point) have to be
discarded. This fact might be verified e.g. for torsional vibration modes of
a foundation slab. (fig. 10.1>. The modal parameters are determined by
analysing the transfer function hK1<0e;> (with K=1 12 1... n> evaluated in n
points of the carrying structure whith a. unitary harmonic excitation force
applied at point 1. As already been said, the transfer functions can be
expressed as a function of the modal parameters only (see equation (10.14):
P.A.L.L.A. :A Package to Analyze ... 241

the transfer functions evaluated at node K, with an excitation applied at

node r, can be determined by using the modal parameters of the foundation,
without having to once again carry out the calculation of the forced
response of the structural model <see BlocK 9). The set of functions
hKr<OE:>, evaluated in the vertical and horizontal planes, ena.bles us to
define a square matrix CHlOE) J of order 2*n 1 Known as flexibility matri>:.
As is Known, the inverse of matrix [HJ represents the matri>: of the
mechanical impedances CI <Oe;>J.The terms of the matri>: [l(.QE:)J will be
added to the terms of the ela15todynamic matrix CE<OE:>J <rotor + oil film>
corresponding to the degrees of freedom associated with the shaft line
connection points <BlocK 11>.

undeformed slab
_.,,
- ,..,. ' ..
'
'
'
- -·
;""";.,
''

,--.
''
-·· - ·-·
I

: -- .. :
~-.
: ' ' .. I ,.

___ :. --
___ - ' .,..-· -""'--
'' ' I _ ......

'
supports axis line ' ' ',, - deformation shape

Fig. 10.1 -Torsional vibration mode of a foundation slab

7.1.14 Block 11: Creation of the Elastodynamic Matrix

To obtain the elastodynamic matrix of theoverall system rotor+ oil film, it
is necessary to suitably assemble the mass [mJ and stiffness [KJ matrices
of the single beam element together with the mechanical impedances matrix
of the foundation <already defined in BlocK 10> and the stiffness and
damping matrices of the fluid film of bearings and seals <BlocKs 7 and 8).
242 G. Diana et al.

The d.o.f. nu.mber associated with the shaft nodes, as already stated for
that which concerns the static analysis <BlocK 6>, is generally large and a
dynamic condensation becomes necessary. In the dynamic analysis carried
out in the frequency domain used to evaluate the response of the shaft
line to a harmonic e:<citation, this condensation is rigorous. Hereinafter
we will firstly describe the method adopted to reduce the shaft d.o.f. by
means of dynamic condensation and subsequently will show the assembly of
the condensed matrices of the shaft with the matrices of the mechanical
impedances of the foundation as well as the stiffness and damping ones of
the oil film and/or seals.
Dynamic condensation
We will now analyse, for the saKe of convenience, the shaft line shown in
fig. 11.1. Let us consider a set of beam elements corresponding to a
portion of the rotor which will be considered as a superelement. By
indicating with ~e the relative d.o.f. vector, the equations of motion of
this system can be written as:

(11.1)

where [Me J, CRe J and CKe J are respectively the mass, damping and
stiffness matrix of the generic superelement anal)tser. For the example
proposed, these matrices are asser.:tle.~ ,:..s shown in fig. 11.1. In eq. <11.1>,
.f e
represents the vector of the external excitation forces applied to the
nodes. In the P.A.L.L.A. pacKage the steady state response of the rotor to
harmonic excitation forces is analysed, so that we have:

F
-e = !-l?O
F eif:lEt (11.2)

In this case, the solution of <11.1> given by <11.2> is :

X
~
= -eo
X ei.O.Et (11.3)

By substituting equations <11.2) in <11.1> we obtain:

(11.4)
P.A.L.L.A. :A Package to Analyze ... 243

rSuperel emel)t I
7/ ~8
fL.= £r £~
f 2 3 4 5 6

:
E-• s~s
Itechdn/ca l impecld/Jces

2 3 4 5 6 7 8 9 S1 S2 S3 S4

1 A
2 A ITJ
3
4
5
6 A [i]
7 LBl = [E<.QE)l
8 A I.....L
9
Sl []] ';'k :: t·;.: ··: [I <.D ) ]·'''

~l}~~·Kttf:;~i
S2 [iJ_
S3 I B
S4 B I

n
Kxx <IIJl> + i.D£Rxx<.DR> 0 Kxy<~> + i.DeRxy<~>
[Al =[ 0 0 0
Kyx<~> + iDERyx<~> 0 Kyy<.OR> + iOERyy<llR>
0 0 0

CBl = -tAl

Fig. 11.1 - Overall assembly of shaft line - foundation system

244 G. Diana et al.

Having called Xee the vector of the "external" displacements, i.e. the
vector containing the d.o.f. of extremity nodes of the portion analysed
(nodes 1 and 10 of fig. 6.2> and kithe vector relative to the other nodes
("internal" nodes>. it is possible to renumber the equations a.s:

(11.5)

With this new order of the equations it is possible to rewrite eq. <11.4) in
oa.rti tioned form a.s:

(11.6)

or rather, by splitting eq. <11.6) into two subsystems:

<-~ 2 [Meel+inerReel+[Keel>~eeo+<-AE 2 [Meil+ine[Reil+[Keil>~eio=feeo

(11.7)
<-~ 2 [M i e J + i~[ Riel+( Ki e J>~e i o+ <-Jle 2[M i i l+ i.D.E[ Ri i l +[ Ki i ))~e i o=fe i o

For a. better understanding, by defining the complex matrices:

[Aee<QE)J=-~E 2 [Meel+iD£[Reel+[Keel
[Ae i <QE) J=-~E 2 [Me i l+ i.QE[Re i l+[Ke i l
<11.8)
[Aie<OE>l=-ne 2 rMiel+iQE[Riel+[Kiel
[A·II· <OE> l=-n-2rM
'¥t · · l + i.QE[ RII
II · · l+ [K II
· ·1

it is possible to rewrite equation <11.7) a.s:

[Aee<QE)l~eeo+[Aei<QE)l~eio=feeo a)
( 11 • 9)
[Aie<OE>l~eeo+[Ai i<QE)l~eio=feio b)

From <11.9b) it is possible to express ~eio as a. function of keo=

(11.10)

By substituting <11.10) in <11.9a.l we obtain:

P.A.L.L.A. :A Package to Analyze ... 245

(11.11)

By defining

<11.12)

as the condensed elastodynamic matrix of the generic superelement and

(11.13)

as;the vector of the condensed forces, it is possible to rewrite <11.11> as:

(11.14)

In other words, it is possible , without introducing any approximation, to

solve the equations of motion of the complete system using as degrees of
freedom only those ones relative to the supernodes (external nodes> of the
single superelements, thus reducing the computer memory occupation.
Should the user wish to Know the displacements of the internal nodes,
Known from (11.14), it is possible from <11.10) to obtain the value of ]:eio•
Assembly
Having dynamically condensed the shaft elements to obtain the
elastodynamic: matrix of the complete system shaft + oil film <or seals) +
foundation, we then proceed with the overall assembly as illustrated in the
case given in fig. 11.1 as an example. As can be seen from the figure, the
vector of the total d.o.f. number contains, in order, the d.o.f. relative to
the shaft and, subsequently, the d.o.f. relative to the shaft-foundation
connection nodes. The squared matrix [HOe;> J relative to the degrees of
freedom of the connection nodes (shaded area of fig. 11.1> is the matrix of
the mechanical impedances, to which matrices [AJ and [BJ, relative to the
oil film or seals, are added.

7 .1.15 Block 12: Synchronous Excitation

The synchronous excitation forces that can be assigned in this BlocK are
basically the ones caused by unbalance, both of the concentrated and
distributed type. In the first case we are dealing with concentrated mass,
246 G. Diana et al.

situated in a nodal point of the F .e:. schematization of the shaH line at a

distance R from the rotation axis and with a certain phase in the relative
reference svstem (fig. 12.1>. The distributed unbalances that can be
assigned within the generic element simulate the situation in which the
center of gravity locus of the generic section do not coincide with the
rotation axis. The eccentricity distribution considered by the program is a
function e(z) of the type:

e(z)=az2+bz+c (12.1)

lying in an assigned orientation plane (see fig. 12.2>. For the definition of
12.1 the eccentricities e 1 <z 1 >t e2 <z 2 > and e 3 <za> must be assigned in
correspondence to three distinct points z 1' z 2 and z3 ofthe element axis.
The origin of z a>:is is assumed in correspondence to the extreme left of
the element itself.

Fiq. 12.1 -Concentrated

unbalance
Fig. 12.2- Distributed
unbalance

7 .1.16 Block 13: Non-Synchronous Excitation

In this the mathematical model used to simulate the effects of
paragraph~

an excitation frequency twice the rotation frequency of the shaft line is

described. Double frequency excitation is considered here to be due to
transverse cracKs and to journal ovalization. Since the mathematical model
simulating the cracK effect is not at present implemented in the pacKage~
P.A.L.L.A. : A Package to Analyze ... 247

we will limit ourselves to describing the model relative to the ovalization

effect. The normal section of a journal of a generic bearing can exhibit
some deviations compared with the circular one, deviations indicated
hereinafter as "surface irregularities". These irregularities are detected
when the rotor is stationary. When the rotor is rotating the air gap in
correspodence to a proximitor fixed on a bearing and facing the surface of
the corresponding journal is determined not only by the journal-bearing
relative vibrations but also by the surface irregularities and by the
"run-out". The latter only appears when the journal is rotating and can be
considered as the rotation of a circular section around an axis different
fr.om the geometric one. As far as the steps to be tal<en to determine the
various quantities are concerned refer to [21J. With reference to fig. 13.1,
we assume that there are two picl<-ups placed in two "radial" directions
parallel to the axis of an absolute reference system X,Y. While the journal
is rotating at a certain angular speed .OR sufficiently low to consider the
displacements x and y of same as negligible, the signals of the two
picl<-ups are only representative of a surface irregularity and eccentricity.
These signals are periodic functions, with the fundamental period given by
2rt/DR, with a phase angle Tt/2 between them. By developing these
function in a Fourier series up to the second order term we have:

xj~Ilcos<DRt+ ~1)+12cos(2 ~t+ P2>

<13.1)
Yi~I1sin<~t+ p1>+I 2sin<2 ~t+ j!2>
The 2 per rev. component is indicated a.s "ovalization". The deviation of the
journal surface from circularity can be attributed to a harmonic motion of
the journal center itself. This motion has the same frequency as that of
the rotation or twice the same, depending on whether the first or second
terms of (13.1> are considered. For a generic rotation speed JlR and
assuming that the irregularity given by Xi and Yi of (13.1> is limited to the
ovalization alone, the journal-bearing relative displacement in the X,Y
directions is therfore given by:

<13.2)
248 G. Diana et al.

in which Xc and Ycare the journal displacements and X5 and Ys those of the
bearing. For small relative journal bearing displacements, in the
neighbourhood of the static equilibrium position, the components of the
force due :to the lubricating film are given by:

<13.3)

where Xr and Yr are the relative speeds. If second order effects are
neglected in a congruent manner with the linearization carried out by
<13.3>, one can, as has already been said, attribute the ovalization to a 2
per rev frequency motion of a journal with a circular section < fig. 13.2>.
Therefore, by substituting <13.2) and its first derivatives with respect to
time in <13.3), we obtain for F x and F y expressions that are both a function
of the independent variables Xc' Yc' Xs and Ys and the ovalization, which
should be considered assigned: the first are already included in the
stiffness and damping matrices of the system while the second are put in
the right hand side member. The following expressions are found with the
above mentioned substitution:

(13.4)

(13.5)

in which F x and F y represent the excitation forces to be assigned as a

known term.

7.1.17 Block 14: Stress Calculation

In this paragraph we will describe the method adopted to calculate the
maximum nominal stresses that are set up in a rotor on account of:
static load <own weight> and the assigned alignment of the axis line
dynamic load that can act on the same line due different vibration
phenomena.
For this reason the P.A.L.L.A. package programs exploit the results
P.A.L.L.A. :A Package to Analyze ... 249

obtained in terms of supernode displacements in Blocl<s 6 and 15,

respectively to calculate the rotating bending stresses due both to the
own weight and the alignment as well as due to the rotor dynamic loads.
Determination of rotating bending stresses
In Bloc!< 6 the P.A.L.L.A. pacl<age programs enable us, as previously seen,
to calculate the static deformation assumed by the rotor with or without a
lubricating film, determining in particular:
- the alignment to be assigned to the machine to obtain the desired
reactions with supports;
- the loads acting on the supports for a given imposed alignment.
In this analysis the stiffness matrices of several elements of the shaft
were statically condensed, thus decomposing the axis line in several
superelements with supernodes for extremity nodes. In this way the final
equations of the entire system are functions of the supernodes d.o.f. only.
Finally, to calculate the stresses, the displacements of the shaft nodes
~nternal with respect to the condensation made) are computed ("retracl<ing">,
as herein shown. For the sal<e of convenience we hereby give the equations
of static equilibrium <Bloc!< 6, eq. 6.4> relative to the single superelement
partitioned into external <supernodes) and internal nodes, with respect to
the condensed system:

<14.1)

The relation between !ei and Zee' as has already been seen in Blod< 6 (eq.
6.5) is given by:

(14.2>

By substituting <14.2) in (14.1a.l it ~s possible <Block 6) to obtain the static

condensation of the ~uperelement stiffness matrix. Since swpernode
displacements .Zee are l<nown from the solution of the static problem, from
<14.2> it is also possible to determine the displacements of the internal
nodes lei• By repeating this operation for all the superelements with
which the line is schematized it is now possible to determine the static
displacements of all the nodes of the rotor. For the generic finite "beam"
250 G. Diana et al.

element it is possible to evaluate the generalized forces at the extremity

nodes by means of r-elation:

F ·=[K J·lX
-J
·-P ·
-J -J
<14.3)

where !j are the generalized displacements of the extremity nodes of the

generic j-th elementt CKjJ is the stiffness matrix of samet .fj the "perfect
clamp" reaction vector or of the generalized forces applied to the nodes
themselves calculated to evaluate the static deformation of the overall
system. The vector.! j of equation <14.3> thus represents the value assumed
by shear and_ by bending moment Mf in the two deflection planes. Once the
value assumed by the bending moment Mf is Known in every node, bending
stresses for the generic finite element are calculated by means of the
Known relation:

D·
~fj = _l. (14.4)
2

where Jj represents the moment of inertia of the shaft section and Dj the
external diameter.

Determination of the stresses due to dynamic effects

In BlocKs 11 and 15 we described the methods adopted by the P.A.L.L.A.
pacKage programs to evaluate the frequency response of the rotor for
different dynamic load situations. In this paragraph, we will describe hOWt
starting from the Knowledge of the displacements of the dynamic
condensation supernodes, it is possible to evaluate the maximum nominal
stresses due to dynamic effects. As in the static case, to calculate these
quantities it is necessary first of all to determine, from the displacements
of the supernodes, the displacements of all the nodes of the shaft line. As
described in BlocK 11, the dynamic condensation of the elasto-dynamic
matrices of several elements has been performed in order to have the
displacements of the only supernodes as the only unKnowns of the problem.
We will now go on to describe the relationship that exists between the
displacements of the internal nodes and those of the supernodes already
expressed in BlocK 11 (eq. 11.10):
P.A.L.L.A. : A Package to Analyze ... 251

(14.5)

Known, from the solution of the forced problem, the response leeo of the
rotor at the supernodes for different frequencies <Bloc!< 1Sl, from eq.U4.5)
it is possible to obtain the response in all the points of the shaft line. To
obtain the maximum bending stresses we used the shape function of the
finite "beam" element which permits, once the displacements and the
rotations of the end nodes are Known, to obtain the bending moment in the
vertical Mx and horizontal My plane of the generic section <see fig. 14.1l,
by means of:

(14.6)

The program recognizes the bending neutral axis and calculates the
maximum value assumed by the bending moment in each section of the rotor
successively calculating the nominal stress using eq. (14.4). These
stresses will give rise to fatigue phenomena should the rotation speed of
the rotor differ from the pulsation of the dynamic forces acting on same or
should the orbit of the shaft not be circular.

X "average" circle
ovalization
bearing node

Fig. 13.1 -Detecting run-out Fig. 13.2 - Ovalization

and ovalization
252 G. Diana et al.

+ z

Fig. 14.1 -Reference system and s1gn convent1on5

7.1.18 Block 15: Frequency Respons~

Once the elasto-dynamic matrix of the entire system ~ rotor+ oil film +
foundation CE<De;>J~ has been calculated~ as described in BlocK 11~ and once
the dynamic forces F = F 0 exp<i0gtl applied to the rotor <see BlocKs 12 and
13l have been evaluated~ it is possible to calculate <see eq.< A. 18)) the
response of the complete system by solving equation:

(15.1)

The displacements in terms of module and phase, both of the d.o.f. of the
rotors of the line as well as of the connection nodes of the foundation, can
be printed or visulized by means of graphic output that show the
displacements along the generic d.o.f. as a function of the frequency or
displacements of all the nodes of the line <spatial deformation>, for a
certain pulsa1:ion of the excitation force.

7.1.19 Block 16: Stability Analysis

The study of the dynamic behaviour of large turbogenerator groups
includes, as a fundamental phase, the verification of the stability of the
dynamic equilibrium of the different ro1:ors that cons1:itute the machine~

commonly Known as the shaft line. This paragraph describes the method
adopted to evaluate~ by means of the frequency response of the rotor-oil
film-foundation system to a harmonic excitation force, the overall
stability of the entire line and the of single bearings. The method used in
the P.A.L.L.A. pacKage, called "forced method" is described in paper C23J.
This method is essentially based on the determination of the energy
P.A.L.L.A. :A Package to Analyze ... 253

introduced by an external excitation force with an assigned pulsation into

the overall system. As will be more fully described in the following
paragraph, the relationship between the energy introduced by the
excitation force and the total energy of the system enables us to calculate
the ratio r/rc between the damping of the entire system and the
corresponding critical value. The sign of this relationship permits the
evaluation of the instability threshold.

The "forced" method: one d.o.f. systems

To illustrate the method used it is convenient to start off by considering a
one d.o.f. system. Let us consider the system of fig. 16.1 in which Ks and rs
represent the stiffness and damping of the system while ri is an active
viscous element capable of introducing energy into the system i.e. capable
of reproducing the instability conditions into the one d.o.f. system <ri<O>.
The equation of motion of the system is:

(16.1)

which, by putting r =rs+ri' becomes:

(16.2)

The solution of equation U6.2> is, as is Known:

X = Xe~t (16.3)

where

(16.4)

Fig. 16.1 -One d.o.f. system

254 G. Diana et al.

If r<O, coefficient of eq. <16.4) becomes positive and solution x(t) grows
e>:ponentially in time: the system is unstable and the instability threshold
is defined by the sign of r. This sign can be determined by also analysing
the steady state response of the system to a harmonic: excitation force. In
fact, if the system of fig. 16.1 is forced by an excitation of the type
F = F 0 e>:p<i0e;tl the response will be of the type x = Xaexp<iJle:t>: the
instability conditions can be verified by the relationship between the
energy e:ic: introduced by the external force in a c:yc:le and the total energy
e:Kma>: of the system itself. For De; = Wn <excitation force frequency
coincident with the natural frequency of the system> the value of r/rc: is
given by:

r Eic
-=- ( 16 .5)
rc 41t EKmax

Since the relationship between the energy e:dc dissipated by the damping r
in a cycle and the maximum kinetic energy is:

Edc: 1t r.QE I X0 12 21t r

(16.6)
EKmax = l/2rnOt:21 Xo I 2- %2

bearing in mind the expression of the critical damping:

rc = 2rnc.On ( 16. 7)

and since e:dc=Eic' for Jle: = Cl)n we have equation (16.5>. The energy Eic
introduced by the excitation force is:

( 16 .8)

where Cf represents the phase lag between the excitation force vector F 0
and the displacement vector X0 :

2ar/r c ) (16.9)
1-a2

where a = D.e:l~aJn,If the damping of the system is positive, eq. <16.8)

assumes the positive value since sin <f <O. For r<O, i.e. for instability
P.A.L.L.A. : A Package to Analyze ... 255

conditions. we have positive values of sin f· To define r/rc: with the forced
method it is therefore possible to use equation <16.5) which in explicit
form is:

r F0 tr I X0 I s i n cp
=-- <16.10)
ll2mDe21Xo12

being. as we remember • ..Ce: =CA>n• From a pratic:al point of view the system
is e>:cited by an excitation force with a variable pulsation until we obtain
the maximum of the amplitudes: in these conditions <resonance conditions)
the ratio r/rc: is determined from equation (16.10> and from this the
characteristic: parameter that determines system stability ol/Wd which. as
is Known. is:

(16.11)

If r/rc:>o the system is stable. otherwise it is unstable.

Systems with several d.o.f.

In the sytems with several d.o.f. instability conditions can arise due both
to the effect of the active velocity coefficients, i.e. non dissipative, (due
for e.g. to damping matrices CRJ defined non positive> and to the
positional forces field. in the case of fluid film instability or of
hysteresis. The equations of free motion of a system with n d.o.f. in
matri>: form is:

(16.12)

where CM J. [R J and rKJ are the square matrices of order n and.! the vector
of n independent coordinates. The solution of eq. <16.12> is given by:

(16.13)

with n values of ~i equal to:

(i=1,2,3 •• ,n) (16.14)

256 G. Diana et al.

1fCR J is non symmetric or is either or non symmetric or defined non

positive. at least one value of o( i can become positive and the system
unstable. In systems with several d.o.f. the instability conditions are
indicated by all the solutions~i with oti>O: the vibration mode defined by
the comple>: eigenvector ii> and by its conjugate is an unstable mode. If
the system is forced in a non nodal point of the K-th vibration mode, the
amplitudes become ma:<imum for Og = ~K' the lower the value of 0( K the
higher the resonance peaK. Moreover, in resonance condition, the
contribution of the other modes can be negligible and this enables us to
consider, in these conditions, the system as if it had only one d.o.f. and
ther·efore apply the concepts already seen. In particular:

(16.15)

where E: ic is either the ener·gy introduced or dissipated by the excitation

force in a cycle and E: ma>: is the maximum value of the total energy relative
to the K-th vibration mode. By varying the frequency Jlg of the excitation
forces it is possible to define the natural pulsations of the system WK and,
from equation <16.15>, obtain the non-dimensional damping relative to the
generic K-th vibration mode. In this case the energy introduced by the
excitation force is:

(16.16)

where XCfand cp are respectively the vibration amplitude of the application

point of the e>:citation force and the phase angle between the vibration and
the excitation force evaluated at the same point. The so called "forced
method" therefore enables us to evaluate the stability conditions of the
system by evaluating the index (r/rc>k relative to the generic natural
frequency ~K and associated to a vibration mode ~<I<>, using the response of
the system to a harmonic excitation force in a frequency range in the
neighbourhood of the frequency whose corresponding mode can become
unstable. In the case of the rotors the instability manifests itself in
correspondence to the lowest flexural natural frequency of that rotor of
the shaft line placed on the two bearings which can cause instability to
P.A.L.L.A. : A Package to Analyze ... 257

arise. Normally, these frequencies are Known. By applying an excitation

force <for e.g. horizontally> to that rotor which could become unstable, by
suitably varying the forcing frequency it is possible to evaluate the
natural frequency of the system. In these conditions we calculate the
energy introduced by the excitation force and the total energy by assuming
that the deformation of the K-th vibration mode is coincident with the
deformation obtained by forcing the system in resonance conditions.
Having evaluated the two types of energy it is possible to obtain the value
of (r/rc>K and therefore evaluate the instability index. In order to apply
this method, it is necessary to repeatedly solve the system:

(16.17)

where, as has been said, CM J, CRJ and CKJ, are the mass, damping and
stiffness matrices of the rotor + oil film + foundation system already
defined in BlocK 5 and I= ,! 0 e>:p<iDEt> is the harmonic: excitation force. By
substituting solution:

x=X e i.O.Et (16.18)

- -o

equation (16.17) becomes <see BlocK 11>:

<-D£ 2 [MJ+iQE[RJ+[KJ>~ 0 =£ 0
(16.19)
[E(.QE) l~o=fo

The advantages of the method used lie in the fact that it is possible in
this way to Keep account of the foundation and of the carrying structure in
general, through its mechanical impedances: apropos of this it seems
opportune to underline the difficulties of reproducing the
rotor-foundation behaviour due to the difficulties of schematizing both the
foundation and the casings of the machine. This can be obviated by
experimentally defining, in correspondence to the supports, the response
of the case and the foundation to a harmonic excitation force and from this
to define the mechanical impedance matrices which can be used, if
introduced in <16.19>t both in the calculation of the frequency response as
well as the stability analysis of the shaft line considered. For further
details see ref. [22 J.
258 G. Diana et al.

7.1.20 References

E1 J Brown D.L. et al. - "Parameter Estimation Techniques for Modal

Analysis" - S.A.E. Paper n. 790221

[2 J Curam1 A., Vama A. - "An Application of Modal Analysis Techniques

" L'Energia Elettric:a, n.7-8, val LXII, 19E:5

~ 3J Bishop R .E .D., Johnson D.C. - "M echanic:s of Vibration" -Cambridge

University Press, New York, 1960.

~4J Diana G.- "Appunti dalle lezioni di Dinamica e Vibrazioni delle

Mac:c:hine"- Milano. 1983

~5 ~ Pizzigoni B.. Ruggieri G. "Sulla Determinazione delle

Caratteristic:he Statiche e Dinamiche di Cusc:inetti Lubrificati" -
L'Energia Elettrica, n.2, val. LV, 1978

E6J Przemieniec:Ki J.S.- "The Theory of Matrics Structural Analysis"- Me.

Graw-Hill, New YorK, 1968

(7J Bathe K.J., Wilson E.L. - "Numerical Methods in Finite Elements

Analysis"- Me:. Graw-Hill, New YorK, 1970

~::: J Curaml A •• Pizzigoni B.- "Un Programma di Calcolo Automatico per

1' Analisi Static:a di una Linea d' Alberi" - L'Energia Elettric:a, n. 12,
voi.. LVIII. 1981

C9 J Bachschmid N., Pizzigoni B., Di Pasquantonio F. - "A Method for

Investigating the Dynamic: Behaviour of a Turbomachinery shaft on a
Foundation" - Design Engineering Technical Conference, paper
77-DET-16, Chicago 1977

E10J Diana G., Bachschmid N.- "Influenza della Struttura Portante sulle
Veloc:ita' Critic:he Flessionali di A:beri Rotanti" - L'Energia
E:lettric:a, n.9, val. LV, 197S

E11J Ruggieri G.- "Un Metoda Approssimato per l'Integrazione

dell'Equazione di Reynolds" - L'Energia Elettrica, n. 2, vol. LIII,
1976
P.A.L.L.A. : A Package to Analyze ... 259

[12J Pizzigoni E., Ruggieri G. - "Caratteristiche di Funzionamento di

Cuscinetti Lubrificati in Regime Turbolento" - V Congresso AIME:TA,
Palermo, 1980.

[13: '"':.;ta"':'!:'~i?.;~-· '-/. ~:.-"On Turbulent lubrication" - Proc. Inst. Mech.

E:ngrs., !..::;-.::'~- .. ::. :73 n. 38, 1959.

[14J Chung Wan-Ng, C. H. T. Pan - "A linearized Turbulent lubrication

Theory" - Journ. of Basic E:ngr., Sept. 1965

C15J Diana G., Eorgese D., Dufour A. - "Experimental and Analytical

Results on a Full Scale Turbine Journal Bearing" - Proc. 2nd Conf."
Vibrations in Rotating Machinery", Cambridge, 1980.

[16J Frigeri c., Gasparetto M., Vacca M. - "Cuscinetto Lubrificato in

Regime laminare e Turbolento : Parte 1" - Analisi Statica" -
l'E:NERGIA ElETTRICA, vol. l VII, 1980.

[17J Eiraghi E., Falco M., Pascolo P., Solari A. - "Cuscinetto lubrificato
in Regime laminare e Turbolento : Parte 2 - Lubrificazione Mista
Idrostatica-Idrodinamica" - l'E: NE:RGIA E:LETTRICA, vol. l VII, 1980.

[18J Falco M., Macchi A., Vallarino G.- "Cuscinetto Lubrificato in Regime
laminare e Turbolento : Parte 3 - Analisi Dinamica"- l'E:NE:RGIA
ELETTRICA, vel. LVII, 1980,

[19J Pendered I.W., Bishop R.E.D.- "A Critical Introduction to Some

Industrial Resonance Testing Techniques" - Journal of Mechanical
E:ng. Science, Vol. s, n.4, 1963

[20] Thomson W.T.- "Vibrazioni Meccaniche" - Tamburini E:ditore, Milano,

1974

[21J Frigeri C., Gasparetto M., Pizzigoni E.-" Metodologia di Misura e di

Elaborazione delle Vibrazioni Assolute di un Albero Rotante" - Atti
V Congresso AIMETA, Palermo, 23-25 Ottobre 1980

[22 J Cheli F., Curami A., Diana G., Vania A. - "On the Use of Modal
Analysis to Define the Mechanical Impedances of a Foundation" -
260 G. Diana et al.

Technical Rept. of Dipa.rtimento di Mecca.nica. del Politecnico di

Mila.no. 1984

C23J Diana G., Ma.ssa E •• Pizzigoni B. - "A forced Vibration Method to

Calculate the Oil-film Instability Threshold of Rotor - Foundation
Systems" - Proc. IFToMM Conference "Rotordynamic Problems in
Power Plants" • Rome 1982
 CHAP'DRI.l

SENSOR TECHNOLOGY

.J.Toanesen
The Tedmlcal Unlvenlty of DIIIIIW'k, Lyaaby, DIIUIW'k

The basic concepts of sensors are described and discussed

for non-contacting displacement types and piezo-electric
type force and acceleration transducers.

8.1.1 Introduction
Sensors of some type must be used to measure vibrations when deal-
ing with rotordynamic problems of turbomachinery. It is important
to select the proper instrument transducer or even different types
of transducers in order to cover a particular situation. One
should consider the following questions: "What is to be measured?,
'What are your looking for?, How is the machine built?, What access
area(s} is available?" and also the follow-up questions like:
What kind of display is wanted of the phenomenon?, and what is the
time element?". That kind of question will almost eliminate the
so called universal sensor, because there is no one sensor for all
jobs.

In many instrumentation systems the sensor is still the limiting

factor of accuracy, though not universally so, for the time-hon-
oured barrier of 1 per cent is fast approaching 0,1 per cent. This
trend has of course been matched by more accurate electronic and
recording equipment and also more sophisticated calibration methods
and equipment. Another general trend in sensor development is to-
262 J. Tonnesen

wards wider extremes of environmental conditions. Higher and low-

er ambient temperatures and more severe conditions of accelera-
tion, vibration and shock have to be considered.

A final requirement of a sensor is invariably the necessity for

small size and weight, for exceptional reliability, and for low
cost. In the light of these exacting and often contradictary re-
quirements it is difficult to understand why the effort extended
on sensor selection is minor compared, for instance, with that
allocated to the electronic sections of modern instrumentation
systems. Data processing and data reduction always seem to carry
greater weight and be of greater attraction than data acquisition,
whether it be in the planning, the development or the production
stage of an instrumentation system. After all, what is common
knowledge in the computer age: "Garbage in, garbage out". This
experience may be partly due to the fact that sensors, regardless
of their high performance, are small and comparatively cheap items
which people tend to overlook, but it may also be due to one of
Parkison's laws that sensor being, as they are, on the borderline
of mechanical and electrical technology, often do not get any sup-
port from either the mechanical or the electrical expert, because
of their two-sided character. Sensor design and development as-
sumes indeed as much expertise in the mechanical as in the elec-
trical field, the former extending deeply into many branches of
physics and the latter not excluding the art and science of elec-
tronics.

In the following it is attempted to demonstrate the diversity and

interdepence of the various brances of science and technology
which are employed in the design and the application of these un-
assuming and therefore often neglected "gadgets", sensors.

8.1.2 Displacement sensors

This type of sensor is also often oausd proximity transducer and

the physical quantity to be measured can be made to vary the in-
ductance, capacitance or eddy-current. The first two groups re-
present the sensor types used in the laboratory and the last group
is used both in the laboratory and in the field (industrial app-
lication}. A common feature of the three sensors is that a car-
rier signal (typically 8 kHz to 2 MHz} must be fed to them in or-
der to operate, and this signal is modulated in proportion to the
gap size. A demodulator then converts the transducer output into
a voltage that many or may not be linearly proportional to the
gap size. Thus both static and dynamic displacements can be mea-
sured and since it is basically a gap variance that causes the
effect the sensors are also termed the non-contacting type.
Sensor Technology 263

8.1.3 Inductance sensors

As the variation of inductance is the property of such a sensor a

coil arrangement is required which gives the largest possible in-
ductance change for a given input quantity, and an ironcored coil
will basically yield larger effects than an aircored coil. The
variable air gap type with ferro-magnetic core is probably the
most sensitive type if the fractional change of inductance for a
given air gap change is considered. Inductances with ferro-mag-
netic cores have also the additional advantage, that in a closed
magnetic circuit, as provided by a ferro-magnetic core with only
a small air gap, it is the least affected by external magnetic
fields.

The inductance of a coil is inversely proportional to the small

air gap and hence the characteristic is non-linear. This is an
undesirable feature of a displacement sensor and several correc-
tive means must be taken. The simplest one is to arrange two
coils in a push-pull arrangement (e.g. in a bridge circuit of the
carrier amplifier). In such a set-up the non-linearity will pro-
bably be 1 to 3 per cent at full scale deflexion, depending on
the individual design. A further advantage gained by this method
is a higher sensitivity (or resultion). Another solution is to
use the assosiated electronics to convert a non-linear input sig-
nal in the feed-back path to provide a linear output.

The physical dimensions are small of both inductance and eddycur-

rent sensors, the smaller industrial types going down to 3 mm in
diameter and then standard connestors (plugs and sockets) are usu-
ally much too large and their contact properties are not reliable
in the presence of shock and vibration. A convenient type of cab-
le attachment is a permanently soldered and cleated cable, say 1
meter long with a connector socket attached at the free cable end,
which can be protected from shock and vibrations. The connectinq
cable, by virtue of its low impedance, can be used in great lengths
and a shielded cable is preferrable, but not necessary, however
ground loops, caused by more than one grounding point along the
line, must be avoided. The shunting capacity provided by the cab-
le must be considered when the sensor is calibrated with a cable
different in length from that required for the actual measurement.
At high carrier frequencies the permissible cable length is lim-
ited by the cable capacitance.

The inductance and eddy-current displacement sensors are widely

accepted due to their small size, high sensitivity and rugged con-
struction. When used to measure rotating machinery they can be
installed with a relatively large air gap, say 1 to 1.5 mm and
are unaffected by lubricating oil and most gasses. On the other
hand, one will have to accept that the sensors are susceptible to
inhomogenities of the shaft like electrical and magnetical proper-
ties, coatings of different conductivity and shaft finish.
264 ]. Tonnesen

8.1.4 Capacitance sensors

By changing the distance between two parallel electrodes a varia-

tion in capacitance is obtained. The dielectric may be simply an
air or gas gap (or even fluid filled gap}. Most variable capaci-
tance sensors can be presented by a pure capacitance and the com-
putation of the "geometrical" capacitance of a given condenser
configuration, neglecting fringing, can be found in many textbooks
of physics or electrical engineering. Even at very high carrier
frequencies of the order of several megacyles per second and with
very small capacitance values, losses are normally negligible in
capacitance sensors, possibly with the exception of some high-tem-
perature sensors.

In practice one of the electrodes constitutes the moving part and

the actual displacement sensor is the second electrode. The mov-
ing electrode is typically connected capacitatively to the rest
of the electronic equipment.

The capacitance of a pair of electrodes is inversely proportional

to the small air gap changes and just like previously mentioned
under the inductance type sensor the same type of corrective mea-
surements must be applied in order to have a linear output. Si-
milarly, the connecting cable must be screened right down to the
sensor housing without leaving an unscreened gap. For this reason
a convenient length of cable is usually made an integral part of
the sensor and when it is calibrated the entire cable must be in-
cluded.

When discussing the construction of variable capacitances used in

sensors it seems more difficult to divorce the essential variable
capacitance unit from the specific sensor design than in the pre-
vious mentioned variable inductance sensors where coil and core
were distinct elements. Essentially, a conductive electrode is re-
quired. The electrode must be well insulated from the mounting
parts and still connected solid mechanically. The choice of a
suitable insulation material is of great importance. It must be
of sufficient mechanical strength and, even more important of an
extremely high form of stability. Its thermal coefficient of ex-
pansion should be as low and as predictable as possible, because
it will most likely affect the stability of the active air gap.
In some cases its coefficient of expansion should match the coef-
ficient of expansion of the other structural parts of the sensor,
usually made of metal, so as to achieve improved zero stability
by way of compensation. Ceramic insulation materials are general-
ly a better choice than plastics or organic materials. The face
of the electrode cannot normally be cleaned in use so it should
be protected from humidity and condensation as well as from cor-
rosion and rust. Metals and alloy combinations used in the design
of the sensor must be chosen to avoid electrolytic corrosion, es-
pecially if the sensor cannot be effectively sealed against am-
bient influence.
Sensor Technology 265

The variable capacitance displacement sensor is mostly used in

the laboratory because it can easily be designed to match a
given test set-up. The major disadvantage is contamination of
the air gap or a variable dielectric condition. A common disad-
vantage to all types of displacement sensors is their inherent
characteristic of sensing geometric changes in a moving surface
commonly termed the mechanical out-of-roundness.

8.1.5 Acceleration sensors

The acceleration sensor most used, both in the industry and in

the laboratory is the piezoelectric type. Piezoelectric sensors
are energy converters and are therefore generator-type qr,active
type sensors and the generated quantity is an electrical charge.
The indicated voltage thus depends on the capacitance of the sen-
sor and indicator circuit. This fact, together with the high im-
pedances normally prevailing in these circuits determines the ap-
proach in the design of piezoelectric sensors and their associat-
ed electronic equipment. Owing to the finite insulation resis-
tance of the sensor circuit and the shunting effect of the load
resistance, the generated charge gradually leaks away and there
is therefore no steady-state response.

Piezoelectric sensors are force sensitive devices and are there-

fore employed for the measurement of physical properties which
can be reduced to forces, such as pressure, stress or accelera-
tion. Piezoelectricity occurs in crystals of certain configura-
tions when exposed to compression or tension. Pyre-electricity
is a closely related effect observed in some crystals when heat-
ed. In the design and use of piezoelectric sensors the pyro-e-
lectric effect is unwanted and in some modern piezoelectric ce-
ramics the high voltages generated by temperature changes may in
some applications become a serious source of errors.

The main attraction of piezoelectric sensors for the measurement

of pressure or acceleration is their high mechanical input im-
pedance (stiffness). They are genuine force-measuring instru-
ments of negligible deformation under load. In the lanquage of
sensor performance they are instruments with high natural fre-
quencies. The output side normally represents a very high elec-
trical impedance which often requires special precautions and
slightly more complex circuits than sensors of low output imped-
ance.

The other advantage of piezoelectric sensors is the possibility

of designing instruments of exceptionally small dimensions.
There are piezoelectric acceleration sensors weighing less than
one gram. From the point of small size and weight these sensors
are rivalled only by resistance strain-gauges, thermocouples,
and thermistors.
266 J. Tonnesen

The main disadvantages of piezoelectric sensors are their lack of

steady-state response and their high electrical output impedance
coupled with the need for low-noise cables of low capacitance
value. The methods of mounting need also special care and atten-
tion.

The maximum working temperatures of polarized piezoelectric ce-

ramics arenow in the neighbourhood of 500°C and this is a re-
markable achievement considering that no less that four sensor
parameters: charge sensitivity, permitivity, dissipation factor,
and leakage resistance are variable with temperature.

Humidity may affect piezoelectric materials such that they suffer

loss of insulation resistance when water vapour condenses on
their faces. This effect applies equally well to other parts of
high-resistance circuits of piezoelectric sensors, such as in-
ternal connections, terminals, transducer housing etc. It is
thus important that condensation of water vapour be prevented
from entering, or affecting, critical parts of piezoelectric sen-
sors. Waterproof coatings should be applied to these areas and
silicone grease filled into cavities of sensors and connectors.
Cold curing silicone rubber has proven useful in places where so-
me mechanical durability is required.

When a linear-acceleration sensor is exposed to a sinusoidal lin-

ear acceleration in a radial direction perpendicular to the main
axis of polarization the directional characteristic of sensiti-
vity assumes a two lobed shape with a transverse sensitivity be-
tween 2 to 5 per cent of the longitudinal sensitivity. Manufac-
turers are constantly trying to improve this by paying attention
to the selection of homogenous piezoelectric ceramics and in care-
ful machining, polishing and lining-up of the parts in such a
fashion that the direction of the anticipated transverse accele-
ration run perpendicular to the main axis of the lobe. Similarly,
connecting leads must not exert radial forces on the sensor and
the optimum position of the leads may have to be found experi-
mentaly.

Piezoelectric acceleration sensors may also be sensitive to air-

borne high-intensity noise, but if a sensor is designed so as to
prevent admittance of sound pressure to the crystal faces, say
the cylindrical faces of a disk enclosed in a sealed housing, a
net pressure sensitivity may be experienced.

Electrical cable noise is primarily an environmental effect since

it is caused by whipping and twisting of the cable during cali-
bration and mesurement. The mechanism of generation of cable
noise in piezoelectric circuits is frictional separation of con-
ductors and insulation which cause an electrostactic induction
at the point of separation and can generate appreciable voltages
on the load. The remedy is to ~ away the changes by a conduc-
tive film on the surface of the insulator. Piezoelectric sen-
Sensor Technology 267

sors are now normally supplied by the manufacturers with a suit-

able length of low-noise cable.

The frequency response of a piezoelectric acceleration sensor is

best when it is mounted with a steel stud into a clean flat sur-
face. However less rigid mounting methods works just as well de-
pending on the application, enviromental conditions and frequency
range of interest. Several possibilities are available like mag-
netholder, wax, double sided adhesive tape or disc, dental cement,
epoxy, and quick setting cyanoacrylate cement. By using some of
these methods, and also with an insulated stud, ground loops are
prevented which may otherwise interfere with measurement of low
acceleration levels.

8.1.6 References:

1) H.K.P.Neubert: "Instrument Transducers", Oxford at the Clarendon

Press, 1963.
2) C.Rohrbach: "Handbuch fur elektrisches Messen mechanischer
Grossen", VDI-Verlag, Dusseldorf, 1967.
3) C.Jackson: "The Practical Vibration Primer", Gulf Publishing
Company Book Division, 1979.
4) M.Neales and Associates: "A Guide to Condition Monitoring of
Machinery", Her Majesty's Stationery Office, London, 1979.
5) E.B.Magrab: "Vibration Testing-Instrumentation and Data
Analysis, A.S.M.E., AMD-Vol.12, 1975.
 CHAPTERU

EXPERIMENTAL TECHNIQUES FOR ROTORDYNAMICS ANALYSIS

J. Tonnesen
The Technical Univenity of Denmark, Lyngby, Denmark

Experimental test rigs are described for:

1: identification of bearing parameters.

2: the influence of mass unbalance on a rotor's

instability threshold speed and the effect
of adding external damping to the rotor-
bearing system.

8.2.1 Introduction
Experiments will always be needed, first at all in the laboratory
or at the manufacturer's facilities, in order to validate the theo-
retical model or to provide further physical insight and just as
often to give vital feed-back for new and improved theoretical
models.

In bearing parameter identification, and for our purpose here is

referred only to fluid-film bearings, many manufacturers are now
conducting experiments on full-size bearings. It is first of all
the static properties like load carrying capacity versus film
thickness that is being investigated but the equally important
dynamic properties like stiffness, damping and stability are get-
ting close attention. Although large discrepancies are found be-
tween theory and test the overall tendency is towards a much bet-
ter understanding of identifying these parameters and especially
with the aid of modal experimental techniques.
270 J. Tonnesen

8.2.2 Bearing Identification Test Rig

A typical bearing test stand is shown in fig.1 and the main data
are as follows:

Bearing diameter: 100 mm, D

Bearing length: 50 mm, L
L/D range: 0.3 to 0.6
Speed range: 0 - 20,000 RPM
Load range: 0 - 10,000 N, by applying force
and/or mass to the bearing housing.

Fig.1 Bearing Test Rig.

The journal's position in the test bearing is measured by capaci-

tance and eddy current displacement sensors. The capacitance dis-
placement sensors are installed on both sides of the bearing hou-
se, and at each location there are four sensors, two for the ver-
tical direction and two for the horizontal direction. The sensors
are arranged in a push-pull connection, where the difference be-
tween the signals measures the relative motion of the parts and
the sum of the signals gives the combined expansion of the parts,
thus making it possible to partly eliminate the error caused by
differential expansion of the parts. The measurement planes are
located 75 mm from the bearing midplane and the position of the
centre of the journal is obtained as the average of the two signals.
Experimental Techniques for Rotordynamics Analysis 271

The eddy current sensors are installed in the test shaft in two
planes each located 18.5 mm from the bearing midplane . There are
two sensors in each plane and they are arranged to give both in-
line and 180° apart readings. Figure 2 shows some details of the
installation and in addition to the displacement sensors is also
seen an oil pressure sensor and a thermocouple sensor, all mounted
flush with the journal surface. The signals from the sensors are
transmitted to the stationary parts via a mercury-cell slip-ring
unit and thus simultaneous information is available about the ab-
solute eccentricity, attitude angle, oil pressure profile and
journal midplane surface temperature.

Fig.2 Cross-section of Test Shaft.

A broad range of tests must be conducted to provide reasonable

data on the effect of the variable encountered in bearing para-
meter identification work. The static properties have three main
variables: speed, load and oil temperature (viscosity},but design
variables such as geometry and oil inlet conditions, •like flow and
supply pressure, may also have a noticeable influence on the be-
haviour. The test technique is straight forward: at a given speed
the load is varied in increments large enough to give distinct
changes in the displacement, oil pressure and temperature data.
The test series are then repeated with a different oil viscosity
or if the bearing geometry is of interest, a different bearing
type is used.
272 J. Tonnesen

l.~-~~H~""""~-----------------------;
5000 N

1510

J
mPa·s Do
m 4.25 12.1""""
l( 12.9 5000N
6 27.8 I

27.........
5000 N

Fig.3 Locus CUrves. Fig.4 Oil Pressure Profiles.

8.2.3 Static Properties of Bearings
Figures 3 - 6 show typical results obtained from the test rig.
Figure 3 shows the locus curve as a function of speed and oil vis-
cosity and interconnecting the individual test points will produce
a smooth curve. For comparison is the theoretical locus curve
shown as a dashed curve. Figure 4 shows the oil pressure as a
function of speed and oil viscosity for a constant load. Integrat-
ing the areas under the curves give a check on the accuracy of the
test method and in this case is the deviation less then 4 per cent
from the applied load.

Figure 5 shows the measured and calculated journal surface tem-

perature as a function of load and speed and as can be seen from
the curve, the temperature is largely independent of load at any
given speed, except for a rising trend at low loads caused by
the reduction in the side flow of the oil. Figure 6 shows the
Experimental Techniques for Rotordynamics Analysis 273

95~------------------------------~
--CALCULATED

----------------__--_
90 - - - - MEASURED
', ~OOR~
as •
'-.....
~
_.
6S00 RPM -

_
'----------------.sooo RPH
sOOO RP~-

;t 10
z
a:
::I '---~----
•
.____ __, __..... -...
,._ ~

~65

60

ss ...--- .
a:------~--
_.. - --- J!j!!i_ ~'i

~~~2=00~0._~~~~0~0~~~~~0~0~_.~~~00~~~~~00

LOAD, N

Fig.S Journal Surface Temperatures.

8Sr-------------------------------~
SPEED• 5000 RPM
--CALC. LOAO•S600 N
80 - ...-MEAS.

60

Fig.6 Bearing Bush Wall Temperature.

274 J. Tonnesen

results for the maximum bearing bush wall temperature which occurs
in the loaded portion of the bearing a little past the minimum oil
film thickness. The agreement between calculation and measurement
is good for the region where full oil film exists, however in the
ruptured film zone there are large discrepancies. A possible ex-
planation is that there may be a backflow of cooler supply oil
from the qrooves in the unfilled space in the striated film.
8.2.4 ~amic bearing properties
The dynamic bearing parameters are obtained by applying a force
function to the system. If only the stability threshold is wanted
this is aaxnp~ relatively simple by reducing the load or for a
given load increasing the speed until whirl occurs. Under these
condition is the whirl-frequency between 40 to 50 per cent of the
running frequency. The force function used in order to determine
the stiffness and damping coefficients of the bearing is either a
sinusoidal, step or impulse force each of which have their own
advantages.
8.2.5 Rotor-Bearing Tests
Once the bearing parameters have been identified in a test rig
where the rotor dynamics influence is minimized as much as pos-
sible the next step is to investigate the combined rotor-bearing
system. As the stiffness and damping coefficients of the bearings
vary with speed, the natural frequencies of the system will also
change and a speed may be reached where the damping becomes nega-
tive. At that speed, the threshold speed, the rotor becomes un-
stable in a self-exited whirl. Hence, from this point of view, it
is important to be able to determine the damped natural frequencies
of the system at the design stage. The information is, further-
more, of considerable value for monitoring purposes and in diag-
nosting vibration measurements, especially in connection with a
FFT-analyzer. The present trend in experiments is towards measur-
ing the onset of instability, methods of stabilization and the in-
fluence of mass unbalance as well as to determine which critical
speeds must be avoided, and which of them may be safely ignored.

Figure 7 shows a test rig that effectively may be used to identify

some of the before mentioned parameters. The main data of the
test rotor is:

Rotor mass. 188 kg, 6 disks shrink fitted

on a common shaft.
Rotor length: 1,20 m

Operating range: 0 - 20,000 RPM

The rotoris-supported in fluid-film bearings with a bearing dia-

meter of 62.7 mm and the L/D ratio is 0.3. The bearing arrange-
ments are shown in figure 8 and it should observed that in the
left hand bearing arrangement external damping can be added.
Experimental Techniques for Rotordynamics Analysis 275

Fig.7 High Speed Rotor Test Rig.

The shaft whirl orbit is measured at ten locations along the shaft,
namely at each of the six disks, at the two bearing pedestals on
the outboard side and in the middle of the two bearings. At each
location are capacitance displacement sensors for measuring in the
vertical and horizontal directions. The sensors in the bearings
are flooded by the supply oil, and to insure constant dielectric
properties in the gap, they are additional flooded through separate
supply holes around the tip of the sensor. From separate calibra-
tion tests the dielectric constant is found to be a factor of 2.3
less than that of air (at 50°6
with a variation of ±3 per cent in
the temperature range 20 to 80 C. It is found to be insensitive to
pressures up to 6 MPa.

In addition to the displacement sensors there are also pressure

tabs in the bearings' midplane for measuring the static and dyn-
amic oil film pressures at two places in the bottom half, ±15 de-
grees relative to the vertical. 1 mm diameter holes drilled at
these locations are connected by 600 mm long tubes to manometers
and piezo-electric pressure sensors.
276 J. Tonnesen

Fig.8 Bearing Arrangements.

The circumferential temperature distribution of the bearing bush

wall temperature, which is an indication of the oil film tempera-
ture, is measured by eight thermocouples: two in the upper half
of the bearing at ±45 degrees from vertical and six in the bottom
half with 30 degrees spacing. The thermocouples are located in
the centerplane of the bearing and are within 0.5-1 mm from the
surface.
8.2.6 Squeeze Film Damper Performance
The vertical and horizontal motion of the squeeze film dampers.
which contain the journal bearings, is measured by capacitance
displacement sensors which are flooded by the squeeze film oil
supply. The transmitted bearing forces are measured by force
gauges made up of a solid ring connected by eight flex bars to
the bearing housing. The force is measured using strain gauges
attached to the flex bars and by calibrating the force gauge up
to 2.000 N it is found that the non-linearity and cross-coupling
does not exceed 3 per cent. Taking simultaneous measurements of
the transmitted force and the squeeze film ~ makes it possible
to determine the system's external damping and stiffness.
Experimental Techniques for Rotordynamics Analysis 277

,_ ROT_Cl ~SPEED 16800 R,..

ISTAl LEI

I
2

~
2 3f
l.
~,If ~\ ~ 5
II II
ii!
"
IIU lJ IIIW
~~~QUENC~~Hz
:311' I
f 3
! f ROTO ~ SPEEDol7160
1
(UNS BLEI
R~
ii:
6Kt 2
2f
5

J ~ \} I Ill II
Fig.9 Test Rotor Frequency Response,
with no External Damping.
8.2.7 Stability Test Data
Some typical test results from a stability test run are shown in
figure 9. The test rotor is well balanced and should theoreti-
cally be stable up to the max. speed of 20.000 RPM,when the rotor-
bearing system consists of a rotor without the overhung disk and
with rigid bearing supports (no squeeze film damper). However the
calculations did also indicate that above 12.000 RPM the damping·
is (log.decrement) small, typically below 0.08. As may be seen
from figure 9 the rotor became· unstable at 17.160 RPM. There-
sidual unbalance in the rotor results in a bearing whirl orbit
which occupies approximately 20 per cent of the clearance just
before the instability sets in, say at 16.800 RPM. At the thres-
hold speed this whirl orbit grows rapidly, occupying about 50 per
cent of the clearance circle when the speed has been raised 50-
100 RPM.

Inserting a weight of 25 g•cm in the disk next to the rotor's free

end bearing brings the threshold speed down to 14.235 RPM and in-
creasing the weight 3 times more results in a further decline
278 J. Tonnesen

to 11.300 RPM. This rotor-bearing system is thus sensitive to un-

balance which lowers the instability threshold speed and the char-
acter of the instability onset is the same, a small speed increase
results in a very rapid growth of the whirl orbit.

The same behavior is observed when the overhung disk is installed.

Here the theory predicted the instability threshold speed to be
at 18.400 RPM, however, with residual unbalance, the measured speed
was 12.600 RPM, and this is caused by a small log.decrement at the
higher speeds.

The rotor-bearing system is now stabilized by adding external damp-

ing at the bearing supports and it can now be operated safely up
to 20.00 RPM even with a substantial residual unbalance resulting
in a shaft orbit covering 40 per cent of the bearing clearance.
However it is interesting to notice that a frequency analysis of
the oil film dynamic pressure signal, in comparison with the dis-
placement, did reveal most of the predicted damped natural frequen-
cies, an example of this is shown in figure 10.

10~
0 RPTOR SP ED= 200 PO RPM
0.. (~TABLE) "-..t
w I
a:
~J(t
1/)
w
a: 8
0..
..J 2 7
0 10'

)
u 6
3
~ s
""
J
z
~ ~4 .ftJlfi \l1500
uf0 1>0 200 300 400
FREQUENCY, Hz

Fig.10 Test Rotor Frequency Response

with External Damping.

From the foregoing test it is evident that the bearings exert a

controlling influence on the dynamics of the system. This influ-
ence is governed by the dynamic properties of the bearings, which
again depends on the operating eccentricity ratio of the journal.
A comparison between theory and tests is shown in figure 11 where
the measured position of the journal center at various speeds is
plotted against the calculated journal center locus. The overall
curve fit is reasonable, but a point by point comparison shows an
increasing descrepancy as the speed increases. It would require
an unrealistic increase in ~ 3ssumed oil viscosity value from
Experimental Techniques for Rotordynamics Analysis 279

• THEORY
• EXPERIMENT

Fig.ll Locus Curve, Theory and Experiments.

which the theoretical points are obtained, in order to bring agree-

ment and this would conflict with the measured temperatures in the
bearings. The possibility of some systematic error in the instru-
mentation or the measurement technique cannot be ruled out entire-
ly as it is notoriously difficult to insure accuracy of static
measurements with displacement sensors. The measurements, however,
repeat themselves whether the rotor is brought up to full speed
rapidly or slowly, and reversing the direction of rotation yields
a mirror image.
8.2.8 Recent Development
In recent years the technique of experimental modal analysis has
been used to measure the dynamic performance of rotor-bearing sys-
tem and with variable succes. It has so far been limited to
simple test rigs, where the system damping was relatively small
and where the rotor speed was below 6000 RPM.

In contrast to the previous example, where only synchronous force

excitation was used, we now use several other techniques like
swept sine, impact or random forces. The sinusoidal excitation
is long and tedious but it has one subtle advantage which is often
overlooked, namely it is a rather straight forward approach which
allow you to get a good "physical feel" for the situation as you
go through the analysis. For example, when the forcing frequency
sweeps through a natural frequency the response is often quite
visual or audible and as the response transducer is moved from
point to point it is possible to feel (or easily see on the dis-
280 J. Tonnesen

play media) the points of maximum response. Even an inexperienced

person could basically "see" what is happening and i f something
did not look quite right one will find it fairly easy to use ones
judgement to find the cause of the problem.

The advent of the mini-computer based digital analysis systems

with its' tremendous capabilities of Fast Fourier Transforms (FFT),
high speed computations, and data storage has drastically changed
the methods used to determine the dynamic characteristics of struc-
tures. The excitation that is applied is often a transient (such
as a pulse or step relaxation) or a force signal with energy at
many different frequencies simultaneously (such as a random signal).
The frequency response data is computed and can be automatically
put into a disc storage. The mode shape analysis techniques and
modal parameter extraction methods can be completely automated so
that one never sees, handles or evaluates any of the data nor under-
stands any of the preprogrammed functions that are being perform-
ed for you.

These techniques have revolutionised experimental modal analysis

and it is no wonder, therefore, that mini-computer based analysis
systems and these new analysis techniques have enjoyed such wide
utilisation in a relatively short time. However, simply owning
one of the new analysing machines does not gurantee good or useful
results. It should be regarded as a tool andits characteristics
must be specified properly by the manufacturer and you must under-
stand it. One must always scrutinize ones results and have a good
basic understanding of the theory that has been programmed and al-
so if compromises have been made. If this is not done the results
will often be worthless.

8.2.9 References:

1) Lund, J.W., "Stability and Damped Critical Speeds of a Flexible

Rotor in Fluid-Film Bearings", Journ. of Eng. for Industry.
Trans. ASME, Series B, Vol.96, No.2, May 1974, pp.509-517.
2) Thomsen, K.K., "Theoretical and Experimental Investigation of
the Stability of Hydrodynamic Radial Bearings", PhD. thesis,
Tech.Univ.of Denmark, Lyngby, Denmark, Sept.1975.
3) Christensen, E., Tonnesen, J., and Lund, J.W., "Dynamic Film
Pressure Measurements in Journal Bearings for Use in Rotor Bal-
ancing", Journ.of Eng. for Industry, Trans.ASME, Series B,
Vol.98, No.1, Feb. 1976, pp.92-100.
4) Glienicke, J., "Schwingungs- und Stabilitatsuntersuchunge n an
G-leitgel.cgerten Rotoren", Motortechnische Zeitschrift, Vol. 33,
No.4, April 1972, pp.135-139.
5) Morton, P.G., "The Derivation of Bearing Charachteristics by
Means of Transient Excitation Applied Directly to a Rotating
Shaft", IUTAM-Symposium, Dynamics of Rotors, Lyngby, Denmark,
Springer Verlag 1975.
Experimental Techniques for Rotordynamics Analysis 281

6) Nordmann, R., "Identification of Modal Parameters of an Elastic

Rotor with Oil Film Bearings", Journ. of Vibration, Acoustics,
Stress and Reliability in Design, Trans. ASME, V01.106, No.1,
Jan. 1984, pp.107-112.
7) Tonnesen, J., and Lund, J.W., "Some Experiments on Instability
of Rotors Supported in Fluid-Film Bearings", Journ. of Mechani-
cal Design, Trans. ASME, Vol.100, No.1, Jan 1978, pp.147-155.
8) Tonnesen, J., and Hansen, P.K., "Some Experiment on the Steady-
State Characteristic of a Cylindrical Fluid-Film Bearing Con-
sidering Thermal Effects", Journ. of Lub. Technology, Trans.
ASME, Vol.63, No.1, Jan 1981, pp.107-114.
9) Lund, J.W., and Tonnesen, J. "An Approximate Analysis of the
Temperature Conditions in a Journal Bearing. Part II:
Application", Journ. of Tribology, Trans. ASME, Vol.106, No.2,
April 1984, pp.237-245.
 CHAPTER 9

INTERACTION BETWEEN A ROTOR SYSTEM AND ITS FOUNDATION

L.Gaul
Unlvenlty of tbe Federal German Armed Forces, Hambura, FRG

ABSTRACT: A theoretical approach is developed and pro-

grammed to analyze the three-dimensional dynamic response
of machines on foundations interacting with soil. Struc-
tures and soil are coupled by means of a substructure
technique. The substructure behaviour of soil is treated
for rigid and flexible foundation slabs of arbitrary
shape by superposition of semianalytical solutions of
viscoelastic halfspace field equations. The interaction
between a single turbomachinery frame foundation and
soil as well as the interaction through the underlying
soil between adjacent block foundations are considered.
The assumptions of perfectly smooth and perfectly welded
contact at the interface between soil and bases bound
the influence of shear stresses. The impact of founda-
tion flexibility with respect to rotor vibrations is
discussed. Experimental studies describe the measured
sine sweep response and vibration modes of a small
scale frame foundation and a rigid circular block foun-
dation on compressed sand.

9.1 Introduction
The prediction of machine vibrations by theoretical approaches
as well as the modification of response after construction often
require taking the interaction between machine, foundation-struc-
ture and subsoil into account. Three examples are given. Fig. 1
shows a discretized model of a drilling machine with long founda-
tion slab on soil. The impact of static soil-structure interaction
was calculated and measured by Thurat [22]. A base for the dynamic
analysis is given by the soil model in the present paper.
A multi body model of a forging .hamner (Fig. 2) is covpled with
a viscoelastic truncated cone model of soil (Knobloch and Gaul
[17]). Thurat [22] calculated and measured the transient response.
Novak [19] treats a hammer foundation as a system of two masses
on a viscoelastic halfspace including embedment effects.
The global response of turbomachinery frame foundations e.g. the
low-tuned steel foundation with concrete raft (Dietz [7]) of Fig.
3, or the response of block foundations are calculated and studied
experimentally by small scale models in the present paper. Dynamic
284 L. Gaul

folXldation

Fig. 1 Model of drilling machine

foundation slab and soil

Oil,
Piston

Pad

Joint {

Pad

Fig. 2 Model of hammer, Fig. 3 Low-tuned steel founda-

foundation and soil tion with concrete raft
Interaction Between a Rotor System ... 285

response results from active excitations by rotor unbalances,

short circuit moments and shaft misalignement or by passive seis-
mic excitation. The function of the foundation is not only to sup-
port the weight of the expensive equipment; the light upper steel
plate on flexible columns (Fig. 3) minimizes the amplitudes of
shaft whirling relative to the bearings. Although the tendency
often prevails to treat the rotor, the frame and the foundation
as if they were independent, actually all these substructures in-
teract. This interaction was treated by Gasch and Sarfe'ld· [8] for
a Laval shaft on a block foundation and by Aboul-Ella and Novak
[1] for a turbogenerator on a pile-supported frame foundation. The
horizontal soil stiffness in the first paper was calculated by
Gaul [10], the vertical soil stiffness matrix of the second paper
by Gaul [9].
Methods for simulation of soil-structure interaction often take
advantage of substructure techniques by coupling the model of
structure and base plate with the model of the substructure soil.
Structures are usually discretized by finite elements or can be
treated in special cases by analytical dynamic stiffness matrices
as in the present paper. Besides simplified soil models (Gaul and
Plenge [14]) the substructure soil is usually described by finite
elements (Waas [23]), halfspace theory (Holzlohner [15], Gaul [11]
or boundary elements (Ottenstreuer [20]).
Finite elements do also allow for a simultaneous discretization of
structure and soil. The method is equally applicable to embedded
foundations and inhomogeneous soil. It has however serious disad-
vantages when applied to three-dimensional problems since it re-
quires extensive, complicated and expensive data mangement. Energy
radiation travelling out to infinity by waves (geometri~~l damping)
can be represented approximately by semi-infinite elements, which
do only simulate the infinite extension in the horizontal direc-
tion (Waas [23]).
The halfspace theory presented here treats the substructure soil
separately. The soil is assumed to be an elastic (Holzlohner [15])
or viscoelastic (Gaul [11]) homogeneous halfspace. Dynamic stiff-
ness matrices of the discretized soil surface can be coupled with
rigid or flexible base plates of arbitrary shape (Sarfeld and
Frohlich [21], Gaul [11]). Three-dimensional motion of structures
can be described even in the high frequency range. Soil inhomoge-
nity has to be approximated by one or two layers or the concept
of equivalent moduli. Embedment has to be approximated as well.
As another tool the boundary element method proved to be well
suited to handle soil dynamics problems. It is possible to calcu-
late embedded structures (Dominguez [24], Huh, Schmid and Otten-
streuer [16] as well as layered media.
Viscoelastic material properties and coupling effects of neigh-
bouring foundations can be described by all three methods.
286 L. Gaul

9.2 Coupling of Substructures

The neighbouring structures (Fig. 4) interact with soil. The trans-
fer behaviour of the three substructures can be described in the
frequency domain of Fourier transform by dynamic stiffness matrices
[K(i ~ )] including inertia, damping and stiffness properties.

The substructure ma-

S
trices of soil [K] and
I II
both structures [K], [K)
are coupled by compati-
bility requirements of
generalized displace-
ments {Uc} and forces
{Fe} at the contact
nodes of the interfaces
I and II. Seismic exci-
tation requires the in-
put of generalized dis-
placements {Vel at the
unloaded interfaces ge-
nerated by incoming
waves. With the genera-
lized forces of active
excitation {P} the sub-
Fig. 4 Substructures of soil-structure structure equations are
interaction given by

I {U}
[K(iw)] I
I 1=
t{uc} .i
I
{P}

I
-{Fc }
II
[K(iw)]
l
r
{Uc)
II
{U}
II II
-{Fc }
= II
{P}
(1)

and
I I I
s {Uc} - { vc
} {F c}
II = II
[ K( iw}) I I (2)

{Uc} {Vc} {F c}

where the generalized displacements of structures {U} are separa-

ted from those at the interfaces {U,}. With given excitation data
Eqs. (1) and (2) lead to the generalized displacement response of
the coupled system by solving
Interaction Between a Rotor System . .. 287

r11 l f{~} r{~}

s r {~c}
. I
I

t Kr Js {Uc} {0}

[ld
. K ( 3)
l II = + [K]l II
{U} {0} {Vc}
II II
{Uc} {P}

The solution of Eq. (3) leads to complex amplitudes {U} = {UR} +

i{UI} corresponding to real displacements {u(t)} = {UR} cos wt -
{U } sin wt for time harmonic excitation or to Fourier transformed
di~placements, corresponding to transient excitation. Calculation
of transient response requires the inverse transformation which
can be computed efficiently by the fast Fourier transform algorithm.
9.3 Substructure Soil
9.3.1 Interaction of soil with rigid and flexible base plates.
The substructure behaviour of soil is calculated by the halfspace
approach for idealized rigid base plates and for flexible plates.
The plane interfaces of soil (Fig. 5) are loaded by forces Fi and
moments Mi generated by the structures. Solutions of the field
equations of soil have to fulfil mixed boundary values:
- rigid bases require plane displacement fields at the inter-
faces,
- the soil surface is stressfre~ elsewhere.
While rigorous formulations by dual integral equations (Gaul [12])
lead to approximate solutions only for simple base geometries, the
presented superposition method provides solutions for arbitrary
shapes and allows for taking flexible base plates into account.
Arbitrary shapes are
modelled by subdivi-
ding the interfaces
into rectangular sur-
face elements. The
continuous stress dis-
tributions in the
interfaces are dis-
cretized by locally
constant pressures in
each element, acting
harmonically in time.
Each loaded surface
element in Fig. 5
Fig. 5 Mixed boundary value problem of defines a stress
soi 1 • boundary value problem
Stress boundary value problem of of the halfspace. The
one interface soil element. assumption of decoup-
288 L. Gaul

led horizontal and vertical displacement fields simplifies the

analysis. Only vertical displacements generated by the load compo-
nents in Fig. 5 are calculated. The horizontal displacements (Gaul
[9]) generated by the missing load components are superimposed. To
bound the influence of shear stresses at the vertically moving in-
terfaces
- smooth contact with vanishing shear stresses,
- welded contact with vanishing horizontal displacements
are compared. Semianalytical solutions of both boundary value pro-
blems lead to flexibility influence matrices. One complex frequen-
£Y dependent matrix element hij relates the compl~x displacement
Wij in the middle of element 1 to the amplitude {pA)j of the time
harmonic force acting at element j. '
Displacement superposition leads to
w1. = fi .. ( pA)
1J J
. (4)

written with the flexibility matrix [h)

{w} = [h) {F}
s
or with the inverse dynamic stiffness matrix [K) of soil halfspace
H
{F} = [K){w}.

For rigid bases the corresponding interface stress distribution is

determined by requiring
- the displacement boundary conditions of the plane interface
displacement fields to be fulfilled in the center of each
element,
- the resultants of the interface stresses to be equivalent to
the halfspace load resultants.
The interaction of soil with flexible base plates requires a plate
discretization by finite elements compatible to the soil element
discretization (Fig. 6).
The equations of motion of the discretized base plate, which is
loaded by nodal external forces {P}, moments {T} and half-space
reactions {F}, are partitioned with respect to the translational
{w} and rotational {q>} degrees of freedom. Coupling of mass matrix
[M] and viscoelastic stiffness matrix [K] of the base plate with
soil is achieved by displacement compatibility at the plate nodes
and soil element centers. Expressing the unknown displacements {w}
by Eq. (4) avoids the inversion of the flexibility matrix [h) and
leads to a linear set of complex equations

f ~~~~~~~:-~~~~~~~~:-~~~-~~~~~~~:-~~~~]
(-<~l[M'AP]T+ [hJ:-w
[Kw>]) J
{~~~} = {~~~}
2 [Mtl>) + [Ktl>) {q>} fT}
(5)
Interaction Between a Rotor System .. . 289

Finite elements from which the halfspace

reactions {F}~ determin-
ing the soil pressure
distribution~ and the
translational and rota-
tional degrees of free-
dom {~} and with Eq. (4)
{w} follow . The results
have to be interpreted
as complex amplitudes
for time harmonic exci-
tations or as Fourier
transforms of transient
response.

Soil elements

Fig. 6 Base plate interacting with soil .

Coupled finite elements and soil
elements.
9.3.2 Flexibility matrix of soil halfspace~ stress broundary
value problems for smooth and welded contact.
The solutions of the stress boundary value problems of halfspace
loaded vertically on one surface element (Fig. 5) are given semi-
analytically by the Fourier integral for smooth and welded contact.
Compared with elastic halfspace theories, a better approximation
of the rheological properties of soil is given here by using visco-
elastic constitutive equations. It turns out that the energy dissi-
pation by viscoelastic damping is of considerable influence when the
geometrical damping by wave radiation is of same order of magnitude.
The equations of motion of the viscoelastic continuum in terms of
displacements ui(xj,t)
t au. . . t aum,R.j
f E0(t-T) ~~Jl dT - eijk ekR.m JG(t-T) 01 OT =
-co -co
()2u.
1
= P at 2 (6)

are decomposed by
(7)

in two wave equations for the scalar and the vector potential ~
and ~k respectively
290 L. Gaul

(8)

This representation is complete (Gaul [11]) if the constraint con-

dition ~k,k = 0 is satisfied. Steady state harmonic motion
ui(xj,t) = Re{ui(xj) exp(iwt)} (9)
leads to reduced wave equations for dilatation £ = u1 1 = ~ 11 and
rotation 2wk = ekij uj,i = - ~k, 11 with relaxation function~ of
plane E0(t) and shear G(t) replaced by complex moduli
E*0(i ) =A *(iw) + 2G*(iw) and G*(iw)
-,U + oc--
£ E* - ,11 + oc--
0' wk
£ - G* wk - 0. (10)
D
Excluding reflextions, solutions are given by
E: =A exp[-a0z + i(Sx+yy)],
( 11 )
wk = Bk exp[-asz + i(Sx+yy)],
where Re(aD,S) ~ 0, wk,k = 0. Displacement field and stress field
are superimposed by these solutions

-
CJ ••
lJ

( 12)
with A, Bk determined by the boundary conditions. The stress bound-
ary value problem is solved by superposition of harmonic vertical
displacement waves at the halfspace surface z = 0
-s '(S,y,w)
w( x,y,O,t) = H w -
p(S,y) exp.[i(Sx+yy+wt}] (13)
generated by the stress wave
-o 2 z(x,y,O,t) = p(x,y,t) = p(S,y) exp[i(Sx+yy~t)](14)

propagating with phase velocity v = w/~, where~= (~ 2 +y 2 )!. Real

and imaginary part of the complex wave compliances A ,w correspond-
ing to smooth (s) and welded (w) contact are plotted in Fig. 7
versus the velocity ratio v/vs.
The compliances show the features of a single-degree-of-freedom
system. Welded contact leads to a resonant condition at the shear
 Interaction Between a Rotor System ... 291

wave ve 1oc i ty v = 1G1p ,

smooth contact feads to
Compliance
resonance associated with
1. 0
Ii=i•Hw Smooth contact
Welded contact
the slightly slower
Rayleigh wave speed vR.
Fig. 8 shows an experi-
mental set-up (Crandall
et al. [4]) by which the
Viscoelastic halfspace
Const. hysteretic solid damping factors no(w)'
- 1.0
11:0.2 '1\111. = 1 n~(w) of the complex
v: 0 .4 m~duli E~ = Eo(1+ino),
G = G(1+ins) and Poisson~
- 2· • .~...:-
.•-___._--':,'-:: .•-...l.-.--::3'-::.o:--M-='-:v:---!•. o rat i o can be mea sured • The
. •-_._--=z'-=
v, damping factors are rela-
Fig. 7 Wave compliances of viscoelastic ted by
ha lfspace.
n0 = nA + (n 5-nA)( 1-2v)/( 1--J,
with damping factor nA of Lame modulus A*= A(1+inA).
Taking advantage of avail-

1
able measured data, re-
Shear Dilatation sults are presented for
G*liW) the constant hysteretic
~·-· Eoliwl solid and Kelvin-Voigt-
solid leadin9 t~ damping
head factors ns ,A (w) = ns ,A
and nS,A(w) = a0 ~S,A
respectively, where
Fig. 8 Clay sandwich for alternating ao = wa/vs is a frequency
dilatation and shear tests. parameter with character-
istic lenth a.
The obtained harmonic solutions are now superimposed by integrating
with respect to the wave numbers B, y in Fourier's integral theorem
1 oo -s w -
w(x,y,O,t) = 2n ff H ' (B,y,w)p(B,y) x exp [i(Bx+yy+wt)]dB dy (15)
-oo

with
- - 2pj(sin saj sin ybj\
p(S,y) - -n-\ By )
being the two-dimensional Fourier transform of the exciting stress
field (Fig. 5) at one surface element of area 4aj b·. The elastic
halfspace leads to improper integrals due to poles ~f the compli-
ance in Eq. (15) at the shear wave and the Rayleigh wave speed.
Solutions can be obtained by choosing Cauchy's principal value and
performing a contour integration in the complex plane. Here a diffe-
rent technique is used. Because the viscoelastic halfspace yields
finite resonance amplifications instead of poles (Fig. 7), the in-
tegral of Eq. (15) is no longer improper with respect to the inte-
grand and can be integrated directly.
292 L. Gaul

By pointwise evaluation of the complex surface displacement field,

the soil flexibility matrix [h) of Eq. (4) is obtained with ele-
ments
hij = w(xi,yi)/[pA{xj,yj)].
9.3.3 F~exibility matrix of excavated soil halfspace for coupling
w1th embedded base plates by substructure deletion.
The dynamic stiffness matrix of substructure soil calculated ana-
lytically by solving the field equations of the halfspace can be
applied not only for describing surface foundations but also for
embedded foundations. The effect of embedment can be taken into
account by substructure deletion utilizing the available conti-
nuum and discrete solution techniques.
Finite element discretizations of soil require to introduce arti-
ficial boundaries at some distance from the base plate although
the foundation medium is geometrically unbounded. When applying
the boundary element method, surface discretization too has to be
truncated at some distance from the base implying a discretization
error.
These difficulties are circumvented in the present approach by
employing the substructure deletion technique (Fig. 9).
H
The dynamic stiffness matrix of the excavated halfspace [K 1 is
calculated from the known dynamic stiffness matrix of the ~alfspace
H
obtained Eby the continuum
.
approach [K] and the dynamic stiffness
matrix [K] of the excavated domain.
Finite element discretization of the excavated domain (Dasgupta
[6]) requires to c~ndense the internal nodal degrees of freedom
out and leads to [K] = [K] -w 2 [M] by static condensation.
The author applies surface discretization by the boundary element
method, where condensation drops out.
The boundary nodes are divided in those at the ground level surface
{5s} and those at the excavation surface {5e}· Interface conditions
at the ground level surface require identity of nodal forces and
displacements at the halfspace surface and excavated domain surface:
H E H E
{Fs} = {Fs}, {us} = {us}.
E H
At the excavated surface equil ibriumEof nodal forces {Fe}+ {Fe} =
{0} and displacement compatibility {ue} = {~e} has to be fulfilled.
As simultaneous prescription of nodal forces and displacements on
the common boundary points is required, Dasgupta [5] demonstrated
that well posedness can be guaranteed if and only if the discrete
models can reproduce those results which are the counterparts of
Almansi•s triviality theorem (Almansi [3]).
Interaction Between a Rotor System ... 293

These requirements and the known dynamic stiffness matrices of

halfspace and excavated domain lead to the dynamic stiffness matrix
H
of the excavated [K ] (Fig. 9). Interaction with an embedded base
plate can be calculated by coupling the dynamic stiffness matrices.

HALFSPACE
H H H
{Fs} =[Kitus}

=-
8
H H

H H
{Fe).(ue}

EXCAVATED HALFSPACE
H H H
{Fe!= [KeJiue}
E E H E -1 E }
{
-[Kee]-[KesH [K]-[KssD [KseJ

Fig. 9 Flexibility of excavated soil

halfspace by substructure de-
letion.
9.4 Substructure Frame Foundation and Shaft.
As an example of a sensitive structure interacting with soil a
frame foundation (Fig. 3) is considered. Only the global vibration
behaviour in the low frequency range is treated on the bases of a
simplified model (Fig. 9) which is suited for comparisons with
experimental results from small scale frame foundations.
The rigid upper plate is excited harmonically by the force Fi and
torque T~ generated e.g. by the unbalances of a Laval rotor with
excentri~ity e. Upper plate and base plate are connected by flex-
ible columns, rigid bearings and rotating disk are connected by
the flexible shaft. The halfspace reactions are reduced to point
B in the interface. The three-dimensional motion of upper plate,
base plate and disk are described by displacement coordinates v.,
u;. w; and the angles of small rotations$;. ~i· a; respectively.
294 L. Gaul

The geometrically linearized Newton-Euler equations of motion

yield with symbols, coordinates and constraints from Fig. 10
-for the upper plate with mass M, inertia tensor I~.
1J

Base
plate

Fig. 10 Rotor on frame foundation

interacting with soil.
D •• c 8 ( e .. k ar . Pk F+M
a . F) + T . 0
I .. <P. + M e " k rJ. vk = E 1
lJ J 1J a=1 J J 1 1

(16)

A
-for the base plate with mass m, inertia tensor J 1J
..
Interaction Between a Rotor System ... 295

A q>.
··
J .. + m e. "k X. uk
c ·· = S ( o. a: s a:s)· ( B o
L e. "k X. Pk +f~. - M. +e. "k X. Fk
B)
1J J 1J J o.= 1 1J J 1 1 1J J
mf·· c .. ) S o.p s F B ( 17)
"\ui-eijk xj ~k = - o.~l i - i '

-for d
the rotating disk with mass md , inertia tensor 0ij' angu-
lar velocity w
d .. d . . d
0ij o.j + 0 11 w(oi1 o.1- 0i3 o.3) =- Ti
..
md wi =- Kid + md e w2 ( coswt o1i-s1nwt
•
o3i ) • ( 18)

Flexible columns and flexible shaft are treated as space beams

wit~ distributed mass. The effects of shear, rotary inertia, static
axial force and viscoelastic material properties are considered in
dynamic, complex valued stiffness matrices (Aboul-Ella, Novak [2],
Gaul [11]).
- column o.
M.F 0. <P~
1
0.
P.F 0. v~-e~mn rm <Pn
1 [KH] ( 19)
M.s IV£
1
0.
P.s
1 U
£-e £mn Xm IVn
- shaft
M.d
1 0.1:,

K.d w£
1 = [KH] d (20)
M.o <P£
1
K.o v£
·1
In the frequency domain Eqs. (16) to (20) lead to a substructure
equation like Eq. (1)

[ K( iw)] (21)
 296 L. Gaul

9.5 Calculated Results of Soil-Structure Interaction

9.5.1 Lumped parameters of substructure soil.
The solution of the mixed boundary value problem, describing the
interaction between one rigid base and soil leads to the complex
s
elements of the soil stiffness matrix (Eq. 2) K1J .. =c .. (a 0) +
1J
i a 0 d .. (a 0), which can be modelled as lumped parameters of soil.
1 ,1
The spring and damping coefficients cz(a 0 ) and dz(a 0) corresponding
to vertical vibration of rigid square base are plotted in Fig. 10
versus the frequency parameter a 0 = wa/vs. The spring coefficient,
describing elastic restoring forces and inertia forces, is slight-
ly higher in the low frequency range for welded contact than for
smooth contact. This is due to the displacement constraint at the
halfspace surface. The
same reason causes a
higher geometrical
damping primarily
associated with the
Rayleigh wave for
smooth contact. Thus
6 the damping coefficient
- - Smooth contact d~ exceeds d~. In the
---- Welded contact present paper all
lumped parameters ad-
' v • 0.4 ditionally depend on
the energy dissipation
Visco•lastic halfspac•
I<Mvin -Voigt modM
of soil governed by
2 viscoelasticity. One
&.• 0.1 llll·· 2
important result of
the analysis with re-
• 1 tc•···ia 01
1!:•iii• d"·•lw
spect to the uncer-
tainties of the contact
0 OS 10 15 2 0 boundary conditions is
that it makes little
difference whether the
Fig. 11 Spring and damping coefficient of contact at the inter-
soil for vertical motion of a face is smooth or
rigid base. welded.
9.5.2 Interaction between two structures.
Real and imaginary parts of the complex interface pressure distri-
bution are given in Fig. 12 corresponding to the interaction
through the underlying soil between two rigid structures, which
are excited by forces P!, P! 1 and torques T!, T~ 1 due to rotating
unbalanced masses. acting with a phase shift.
Interaction Between a Rotor System . . . 297

9.5.3 Dynamic response of machine foundations.

The dynamic response of a frame foundation (Fig. 3) with eight con-
crete columns is evaluated as an application of the presented sub-
structure technique. The system is excited by an unbalanced rigid
rotor. The magnification functions in Fig. 12 describe the ampli-
tudes of horizontal displacements v1, u1 and vertical displacements
v3, u3. According to Fig . 9 Vi belongs to the upper plate and Ui
to the base plate. The coupled rocking and horizontal motion gives
rise to two resonant amplifications indicated by the horizontal
displacement amplitudes within the regarded frequency range. The
resonant amplifications are affected by material damping of soil
because rocking motion causes only small geometrical damping by
wave radiation (Gaul [11] ) . This indicates vertical motion being
associated with strong geometrical damping. Frequency independent
static lumped parameters of soil lead to the compared deviations
of response. The vibration modes in Fig. 14 show the coupling bet-
ween a rocking and sliding motion.

a 0 = 1.0 v =0.4
lis = 0 .2 11/lls=l
mr.n I p(a 1) 3 =10
JI.n
IC•,YY
/p(ai)L20
-
0,54

Fig. 12 Interface pressure distribution for adjacent excited bases.

298 L. Gaul

Rigid columns of the frame

t c • u1to1 o •u,ta. A • ii 1to 1 foundation simplify Eqs.
8,6
(16) to (21) and lead to
x10" 1 •I
I
+•v1to1 X•v,to1 O•v1to1
the description of a Laval
I
7,5
rotor on a block foundation
3
I
- - . Frequency dependent (Gasch and Sarfeld [8],
----- Static parameters Kramer [ 18]). If only a
of substructure soil
plane motion is considered,
the system has 5 degrees of
freedom. Fig. 15 compares
..
I::I

:a
the response of the rotor
on a rigid foundation with
the response corresponding
..I
2
to a flexible foundation .

.
i Analogous to the frame
! foundation (Fig. 13) the
a
a.!I first three resonant ampli-
c fications are predominately
due to the foundation,
while the last two are
governed by the rotor. The
foundation influence splits
one resonant frequency of
the rotor in two of the
combined structure with
lower amplitudes.
1 2
Frequency parameter a,
Fig. 13 Response of frame foundation
on soil.

Deflection
scale
111t•11t2 '151J.m

Fig. 14 Vibration modes of frame

foundation on soil.
Interaction Between a Rotor System ... 299

9.5.4 Turbomachinery frame foundation supported by pil~s.

Aboul-Ella and Novak [1] analyzed the dynamic response of turbo-
machinery frame foundations supported by piles or a foundation
slab. Their study investigates interaction of all components of
the system, i.e. flexible rotors, viscoelastic oil film, space
frame, flexible mat, piles and soil (Fig. 16). The mat is composed
of rectangular finite plate elements. The pile and soil resistance
is included into mat element stiffness matrix. The dynamic complex
soil stiffness matrix is obtained from Gaul [9].
In the study of Aboul-Ella and Novak [1] special attention is paid
to the effects of soil structure interaction. It was found that
this interaction markedly affects the response of the frame as
well as the rotors in the lowest resonant regions. The interaction
reduces rotor and frame amplitudes. This results from the increase
in damping due to energy radiation in the soil and viscoelastic
behaviour of soil and mat. The interaction reduces the frame vibra-
tion more than shaft vibration.
E.g. Fig. 17 compares the vertical response of frame under bearing
pedestal corresponding to a rotor on elastic frame and rigid foun-
dation with a rotor on elastic frame and elastic foundation.

--Rotor on flexible
----Rotor on rigid
w, foundation
/e

2 3

Fig. 15 Response of Laval rotor on

rigid and flexible block
foundation.
300 L. Gaul

9.6 Experimental Investigation of Soil-Structure Interaction.

9. 6. 1 Measured response of a model frame foundation.
The response of a f~ame foundation according to the model of Fig.
10 was simulated by a small scale model (Fig. 18). Four rotating
unbalances driven via a control gear allow for coupled and uncoup-
led excitation by horizontal and vertical forces and by torsion
and rocking moments. The stiffness of coupling between upper plate
and base plate can be varied by interchangeable columns. Rubber
springs simulate the soil.
Turbine

Shaft~ D~

F:lr.±r:r-¥-==~-:!=-~~rr:F======~~ Oil film- ~~~~~~~~t!~:!:~

in journal
bearings

~~(IIi I i iiI iii i I i i iii

Piles

Fig. 16 Turbomachinery frame foundation and its

model .

.60 .0153rnm

iii
~ . ~8
....
N
.36
...
QJ
"0
::1

-a.
- - ROTOR ON EL ASTIC FRAME
AND RIGID FOUNDATION
E . 2~
< - - - - ROTOR ON ELASTIC FRAME
c AND ELAST IC FOUNDATION
~
..... .12
QJ

>
00
"' -
100 200 300 900 1000
w rod/sec

Fig. 17 Effect of foundation (piles and soil)

flexibility on vertical response of frame
11nrlPr hPilrino nPrlPc;till.
Interaction Between a Rotor System ... 301

The sine sweep response

of 12 degrees of free-
dom where measured by
velocity pick-ups. The
results are in good
agreement with those
calculated by Eqs. (16)
to (21) (Gaul, Mahren-
holtz [13]). The coupled
horizontal and rocking
modes in Fig. 19 corre-
spond to the calculated
modes in Fig. 14.

Fig. 18 Small scale model of frame

foundation with measuring
set-up.
9.6.2 Laval shaft on a model frame foundation.
For simulating the interaction between rotating shaft, frame foun-
dation and subsoil on a model scale, the lab model of Fig. 20 has
been built. A laval shaft with adjustable disc position and excen-
tricity driven by a variable speed motor is supported by two ball-
bearings on the upper foundation plate.
The experimental work in
progress simulates the
interaction effects of
the three substructures
where the base plate is
bedded on rubber springs,
on a foam layer or on a
model sand foundation
which is explained in the
next chapter.

Fig. 19 Vibration modes of frame

foundation on rubber springs.
302 L. Gaul

Fig. 20 Lab model for measuring inter-

action effects between rotating
shaft, frame foundation and
subsoil.

Fig. 21 Experimental set-up of shaker-

driven model footing.
Interaction Between a Rotor System ... 303

9.6.3 Steady-state vibrations of model footings.

The substructure behaviour of soil was measured on a model scale
by shaker-driven footings at the surface of homogeneous or layered
sand mixed with gravel (Fig. 21). The response of acceleration and
phase angle versus frequency of the sine sweep (Fig. 22) as well
as the deduced frequency dependent lumped parameters of soil are
found in satisfactory agreement with calculated results (Gaul,
Mahrenholtz [13]).
TF LIN B/A
2E- I -- ----·NB---G-- 3009
- -. -- WTG H
F-''- __ ,
A I 0 B 0 05 V
,~- r-="-

--- -- - -- --- --

E/E
r-- -- r-- --- --~· - - -
~
1\
v IW ?
I
A._ -.,..,_
0
...,..,....v -
..........
00 LIN X HZ 400
X. 107 HZ TF 1.44E-1 EU/EU AVG N 1203

00 LIN X HZ 400
Xo 107 HZ + 79.6 DEG AVG N 1203

Fig. 22 Sine sweep response of vertical

vibration for a circular model
footing.
9.7 Summary and Conclusions.
On the basis of substructuring a theoretical approach has been
formulated and programmed to analyze t~e three-dimensional dyna-
mic response of machine foundations considering the interaction
of the system components; namely, viscoelastic soil, frame and
rotor, as well as the interaction through the underlying soil
with an adjacent structure. Geometrical as well as material damp-
ing of soil are considered. Material damping is found to be of
304 L. Gaul

considerable influence for rocking motion. The influence of shear

stresses at the interface between base and soil is limited. It
makes little difference whether the contact at the interface is
smooth or welded. The dynamic response of a model frame founda-
tion and a model footing on compressed sand are measured. Both,
theoretical approach and experiments, provide a good understand-
ing of the basic interaction effects.
9.8 Acknowledgements
The experimental research with model footings performed by Dipl.-
Ing. M. Plenge, University of Hanover, is gratefully acknowledged.
This study was supported by the Deutsche Forschungsgemeinschaft
(German research council).
9.9 References
[1] Aboul-Ella, F.A. &M. Novak, 1978. Dynamic analysis of tur-
bine-generator foundations. Presented at the 1978 Fall
Convention, Housten, Tex., Oct.
[2] Aboul-Ella, F.A. & M. Novak, 1980. Dynamic response of pile-
supported frame foundations. Journal of the Engineering
Mechanics Division, Proc. ASCE, Vol. 106, No. EM6, Dec.,
pp. 1215-1232.
[3] Almansi, E., 1907. Un teorema Sulle Deformazioni Elastiche
.dei Solidi Isotropi, Atti della reale accademia dei
nazionale Lincei, Vol. 16, pp. 865-868.
(4] Crandall, S.H., Kurzweil, L.G. & A.K. Nigam, 1971. On the
measurement of Poisson's ratio for modelling clay.
Experimental Mechanics, 11, pp. 402-407.
[5] Dasgupta, G., 1979. Wellposedness of Substructure Deletion
Formulations. Proceedings, Sixteenth Midwestern Mechanics
Conference, Vol. 10, Manhattan, Kansas.
[6] Dasgupta, G., 1980. Foundation Impedance Matrices by Sub-
structure Deletion. Journal of the Engineering Mechanics
Division, American Society of Civil Engineers, Vol. 106,
No. EM3, pp. 517-523.
(7] Dietz, H., 1972. Stahlfundamente fUr Turbomaschinen. Beratungs-
stelle fUr Stahlverwendung, DUsseldorf, Merkblatt 146,3.
(8] Gasch, R & W. Sarfeld, 1980. Unwuchterzwungene Schwingungen
des Systems Lavallaufer-Blockfundament - elastischer Halb-
raum, VDI-Berichte Nr. 381, pp. 129-138.
[9] Gaul, L., 1977. Dynamische Wechselwirkung eines Fundamentes
mit dem viskoelastischen Halbraum. Ing.-Archiv, 46,
pp. 401-422.
[10] Gaul, L., 1979. Dynamisches Halbraumverhalten infolge eines
schuberregten Fundamentes. ZAMM 59, pp. 180-183.
Interaction Between a Rotor System ... 305

[11] Gaul, L., 1980. Zur Dynamik der Wechselwirkung von Strukturen
mit dem Baugrund. Habil. Univ. Hannover, June.
[12] Gaul, L., 1980a. Dynamics of frame foundations interacting
with soil. Journal of Mechanical Design, Vol. 102,
pp. 303-310.
[13] Gaul, L. & 0. Mahrenholtz, 1981, 1984. Dynamische Wechsel-
wirkung zwischen Maschine, Fundament und Baugrund. Ar-
beitsbericht zum DFG-Schwerpunktprogramm Betriebsverhalten
dynamisch belasteter Maschinen (not published).
[14] Gaul, L. &M. Plenge, 1983. Ein Baugrundmodell fUr geschich-
tete BaugrUnde mit viskoelastischem Stoffverhalten~
ZAMM 63, pp. T50 - T53.
[15] Holzlohner, U., 1969. Schwingungen des elastischen Halbraumes
bei Erregung auf einer Rechteckflache. Ing.-Archiv, 18,
pp. 370-379.
[16] Huh, Y., Schmid, G. &M. Ottenstreuer, 1983. Evaluation of
kinematic interaction of soil-foundation systems by
boundary element method. SMIRT 7 Chicago, August, Paper
K814.
[17] Knobloch, W. &Gaul, 1975. Dampfungs- und Federverhalten
elastischer und viskoelastischer GrUndungen bei harmoni-
scher Erregung. Kolloq. Viskoelastische Systeme, VDI-GKE,
TU Berlin, pp. 412-451.
[18] Kramer, E., 1984. Maschinendynamik. Springer Verlag, Berlin.
[19] Novak, M., 1982. Response of hammer foundations. Proc. soil
dynamics and earthquake engineering conference Southampton,
July, pp. 783-797.
[20] Ottenstreuer, M., 1982. Frequency dependent dynamic response
of footings. Proc. soil dynamics and earthquake engineer-
ing conference, Southampton, July, pp. 799-809.
[21] Sarfeld, W. &C. Frohlich, 1980. Dynamische Wechselwirkung
von Gebauden und Fundamenten auf dem elastisch-isotropen
Halbraum, Bauingenieur 55, pp. 419-426.
[22] Thurat, B., 1978. Machine-Fundament-Baugrund.
Diss. RWTH-Aachen.
[23] Waas, G., 1972. Linear two-dimensional analysis of soil
dynamics problems in semi-infinite layered media.
Ph.D. Thesis, University of California, Berkeley.
[24] Dominguez, J., 1978. Dynamic stiffness of rectangular foun-
dations. Dep. of Civ. Eng. MIT R78-20.
 CHAPTER 10.1

PROBLEMS OF TURBINE GENERATOR SHAFT DYNAMICS

D.W. King*, N.F. Rieger**

*Rochester Gas and Electric Corporation, Rochester New York, USA
**Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

The nature of faults and disturbances which can

occur in an electrical power system are reviewed.
The sequence of events to which a turbine-generator
unit may be exposed from such events is described.
Procedures for unit response analysis are mentioned.
and current problems in predicting such response are
discussed. The need for on-line shaft monitoring
and cumulative damage analysis is mentioned.
Several types of continuous monitoring and measure-
ment equipment now in use on turbine-generators are
described, with details of their performance.
Conclusions are presented concerning the state-of-
the-art in this area.

10.1.1 Introduction

Electrical faults and system disturbances can create severe

transient torques in turbine generator (TG) shafts. TG shaft
failures have been attributed to such transient torques in
1970 and 1971 by Jackson and Umans [1]. The IEEE Screening
Guide for Planned Steady State Switching Operations to
Minimize Harmful Effects on Steam Turbine Generators [2]
indicates that the following electrical disturbances may
create damaging transient torques in TG shafts.

a) Transmission line switching.

b) High speed rec1osing of circuit breakers following

fault-clearing on lines leaving power stations.
308 D.W. King- N.F. Rieger

c) Single phase operation, such as caused by single

pole operation of circuit breakers. This produces
alternating torques at twice line frequency. While
this frequency is generally above those of the
lower.modes of the TG set, there are many complex
higher modes of vibration which involve internal
deflections of the generator and turbine rotors.
Further, individual discs and blade groups can
respond to such pulsations.

d) Sub-synchronous resonance (SSR) in series-capacitor

compensated systems.

e) Out-of-phase synchronization.

f) Line faults reaching the generator terminals.

g) Full load trips.

It is known that when series capacitors are used in trans-

mission lines situated electrically close to the generator,
steady state and transient currents may be generated at
frequencies below the normal power system frequency. The
existence of such currents causes alternating torques in the
generator which can excite the natural vibration modes of the
shaft, and cause significant dynamic torques at or near the
shaft couplings.

From numerical calculations of such conditions, it has been

determined that the above mentioned fault conditions typically
occur in sequence:

The first impact on the generator might be a sudden short

circuit in a transmission line close to the generator, which
in e 1 ec t r ical engineering is frequently called fault
application.

The second impact results from electrical disconnection of

the faulty transmission line, which is called fault clearing
or fault removal. Usually this is all that is needed to
extinguish the electrical arc, which is initiated by the
fault, e.g., lightning, or by short circuits on the power
line.

After a brief disconnection, the third impact is then applied

when the line is electrically re-connected. This is also
done automatically. The technical term for this operation is
high speed re-closing.

High speed re-closing can be successful if the fault is

effectively cleared by the disconnection of the line, or
unsuccessful if the fault persists after re-closing. In this
Problems of Turbine Generator Shaft Dynamics 309

latter case a fourth impact is applied by a second clearing

attempt.

These shocks are applied consecutively. TG shaft oscil-

lations result from the initial shock. and the unit is re-
shocked from the second. and then the third impacts.
Depending strongly on the relative phasing of these shocks.
the resulting dynamic strains can result in very high total
dynamic shaft torques. These torques can be much higher than
in the terminal short circuit case. which had been considered
the worst design case until just a few years ago.

An evaluation of the effect of an electrical disturbance on

the amount of fatigue damage per incident on the TG shaft was
presented by Joyce and Lambrecht [3]. and is shown here in
Table 1. This chart was based on a large number of computer
calculations for the fatigue damage incurred per transient on
a four-flow 3600 rpm turbine with a 926 Mva two pole gener-
ator. and for a six-flow 1800 rpm turbine with a 1525 Mva
four pole generator. These transients resulted from the
successful and unsuccessful reclosure of system faults. The
black sections of the chart represent the results of studies
conducted on 45 machines of various design and sizes. From
calculations it appears that the amount of TG shaft fatigue
damage resulting from an electrical disturbance can vary by
more than an order of magnitude. This is in part due to a
lack of knowledge in certain critical areas of TG shaft
vibrations and stress analysis. Some of these areas are:

a) Data on electrical disturbances with regard to their

duration and transient amplitudes.

b) The effect of system size. impedance. load. and

climate conditions.

c) The conversion of an electrical line transient to a

mechanical torque transient within the generator is
not satisfactorily understood.

d) The effect of turbine and generator construction.

of blade-shaft interaction. and of modal damping on
torsional vibration requires further clarification.

It is the purpose of this chapter to explore some of these

areas. assessing what has been done in the past and what
appears to be needed for the future.
310 D.W. King- N.F. Rieger

10.1.2 Notation

c Damping Coefficient
Hz Frequency, Cycles/Second
I Inertia
K Spring Constant
T(t) Time Varying Torque
e• Angle of Twist
9 Rate of Angle of Twist Angular Velocity
~ Angular Acceleration
(I) shaft Angular Velocity

10.1.3 TG Shaft Vibration Analysis

In order to assess the fatigue damage incurred during a given

transient, the resulting stress history within the shaft
system must be determined from the imposed transient torques.
Such transient torques can produce stress in two ways. The
first is through an abrupt change in torque level which leads
to torque magnification followed by a slow decay in stress
amplitude which is inversely proportional to the damping
magnitude. In Figure 1, M is the mechanical torque imposed
on the TG shaft while M i~ the mechanical torque imposed on
the TG shaft while M it the measured air gap torque imposed
on the generator asea result of an electrical disturbance.
The second manner in which transient stress is produced in a
TG shaft is by sub-synchronous resonance (SSR). In this case
the ratio of M to M may reach as high as 15 in the lower
modes of the T6 shad system. SSR is typically of greater
concern than step changes in torques, because of the greater
rate at which significant damage can be acquired by the shaft.

TG shaft natural frequencies may be computed using a mass-

elastic model of the shaft system. A simple example of such
a model is shown in Figure 2. Actual analytical models
commonly contain 200-500 inertia stations, and include several
rows of LP blades for each rotor flow direction. Under
transient loading conditions, the matrix equation of motion
for such a system is:

., ~

[I] {9} + [C) {9} + [K] {9} [T(t)] (1)

where the notation used is specified above. By integrating

this equation using a time-marching technique in a finite
element code, such as ANSYS or NASTRAN, the time-varying re-
sponse of such a system may be readily obtained. Such
programming may also be used to obtain mode shapes for the
different natural frequencies of the TG system. These mode
shapes are normalized displacement amplitudes (degree of
Problems of Turbine Generator Shaft Dynamics 311

twist in the shaft) and are helpful in predicting those shaft

sections which are most susceptible to fatigue damage under
transient conditions, or SSR. In order to realistically
determine the TG shaft stresses by finite element calcu-
lations, inforcation on the following must be available:

a) Transient time history of the disturbance reaching

the generator rotor,

b) Transient load distribution within the generator,

c) TG shaft material strength,

d) Distribution of mechanical load on the TG shaft,

e) Shaft damping properties (magnitude, distribution

for each mode).

Further information is needed in each of these areas, and

this is of concern in the evaluation of stresses due to
generator transients and SSR.

10.1.4 Continuous Monitoring and Measurement Equipment

In order to acquire data such as the time-variation and

transient amplitudes during an electrical disturbance,
continuous monitoring equipment has been developed.
Continuous Monitoring Systems (CMS) are designed to start
recording both the electrical disturbance and the mechanical
transient torques when a threshold electrical disturbance is
detected. Presently there is no common threshold value due
to differences in the transmission line makeup and machine
variances. Once triggered, the CMS records the electrical
disturbance usually by measuring the three phase current and
voltage at the generator terminals. In certain systems,
generator currents are measured by use of a so-called Rogowski
coil, as illustrated in Figure 3. Voltage is measured with
potential transformers. The current and voltage values are
then transmitted to a recording device and stored on magnetic
tape (analog signals). The product of such voltages and
torques gives instantaneous values for the generator air gap
torque. In addition, devices such as shaft mounted toothed
wheels with accompanying magnetic pickups are used to measure
small variations in the shaft rotational speed. These are
the two primary measurements that are used on TG systems to
acquire basic data on the amount of fatigue incurred per
transient incident.

Two such CMS units currently in use are the Torsional Stress
Analyzer (TSA) developed by Kraftwerk Union AG, as described
in reference [7], and the Torsional Vibration Monitor System
(TVMS) developed at General Electric Large Steam Turbine
312 D.W. King- N.F. Rieger

Generator Division, as described in reference [8]. Block

diagrams for each unit are shown in Figure 4 and Figure 5.
The first TSA was implemented in September 1977 (no location
given) and several are now in service. The TSA uses the
Rogowski coil to measure three phase current. An ironless
core is used to eliminate inaccuracies due to core saturation
in DC offsetting of fault currents. Current values are
combined with three phase voltages by an analog computer to
produce an instantaneous value for the generator air gap
torque. Shaft angular velocity variations are measured as
described previously, and these values along with the air gap
torque are used as input signals to an analog shaft torsional
model. Figure 6 shows this analog model. The turbine torques
and electrical torques along with the shaft angular velocity
enter the computer, are transformed into analog work, and are
then transformed back into shaft torque exerted by all
contributing modes. Measured shaft angular velocity is used
as input to compensate for non-linear effects that machine
damping could cause. The shaft torque figure derived from
the model is used along with the properties of the shaft
section to determine fatigue damage by use of the Rainflow
Cycle Counting method. This method uses closed loop stress-
strain cycles, which are determined from the succession of
torque maxima and minima to obtain a cumulative figure for the
fatigue damage resulting from each transient event. A typical
stress-strain hysteresis damage loop is shown in Figure 7.

Sample results from the TSA are shown in Figure 8 for a TG set
with four couplings, where the shaft motions are monitored.
A code, shown in Figure 9, is used to describe the severity
of the incident and recommended act ion, if any. The second
column shows a figure for the shaft life expended in the
incident, based on a 10()4K, shaft life expenditure at first
crack initiation. The third column shows the cumulative
fatigue incurred since installation of the TSA. The accuracy
of this model is evaluated in Figure 10 for a 970 Mw, 3000
rpm, six-flow TG undergoing a sudden three-phase short circuit
(A), and for partial load rejection (B). The shaft torque
between the last low pressure turbine and the generator is
compared using the figures from the TSA and figures obtained
throughout the application of strain gages to the subject
shaft area. The results appear to endorse the TSA fatigue
life model.

The GE TVMS was first installed in early 1981. The TVMS uses
both air gap torque and angular velocity for its inputs and
produces figures for fatigue life expended in much the same
way as the TSA. The novel concept of the TVMS is the use of
Data Acquisition Systems (DAS) which sends in information
from the TG unit to a central computer to be analyzed. This
permits centralized monitoring of many remote TG units, and
tends to minimize the need for specialized test personnel at
the TG locations.
Problems of Turbine Generator Shaft Dynamics 313

10.1.5 Current Problem Areas

Due to the complex nature of this problem, many areas of

uncertainty remain. One important concern is the prediction
of remaining shaft fatigue life. This is a function of the
complex state of stress that the shaft is exposed to, and it
is also influenced by differences between actual shaft
material properties and those derived from simple test, e.g.,
size and loading effects. Jackson and Umana [1] conclude
that 'much more effort in this area (fatigue calculations) is
required before a general consensus can be achieved.
Predominant is the need for research in the area of torsional
failure and fatigue modeling in order to resolve a major
source of controversy which has arisen as to the loss of life
predictions for torsional shaft transients.' Since this
paper, programs to address these problems have been initiated
under the sponsorship of the Electric Power Research
Institute, Palo Alto, California.

Rusche and llitsche in the Discussion of [1] state that 'Even

its (fatigue and damage prediction) most basic aspects remain
under continuing investigation by specialists.' The General
Electric Company is presently working on a project under EPRI
sponsorship to develop a torsional fatigue methodology for
use by industry sources in fatigue calculations [8]. Similar
developments concerning the fatigue properties of typical
shaft materials are being undertaken by the Westinghouse
Electric Corporation, again under EPRI sponsorship.

As can be seen in Figure 11, there exists many different

types of mass-elastic models for the purpose of finding the
modal properties of TG shaft systems. Both Joyce and
Lambrecht [3] in Table 2, and Ramey and Kung [4] in Figure
12, show the difference between simple mass-elastic models
(less than 10 masses) and complex models (over 200 masses).
Both conclude that for basic calculations, adequate simple
models can be developed. However, differences between the
simple and complex models do not appear to have been explained
in the literature, particularly in relation to the effect of
such models on fatigue calculations. Adequacy in such cases
means 'an appropriately calibrated modal model of the system,'
and such a model must be defined using some proven system
identification procedure.

Figure 1 shows that many TG shaft natural frequencies result

from the use of such models. This is in accordance with
practical observations on TG units. Correlation between
observed and predicted torsional frequencies is typically
close. Both Bizume [5] and Joyce and Lambrecht [3] have
shown, in Figure 13 and Figure 14, that the torsional modes
which contribute most significantly to transient torques are
314 D.W. King- N.F. Rieger

typically the first four modes. These four are therefore the
primary modes which must be considered in such calculations.

The effect of machine damping on torque calculations has been

examined by Joyce and Lambrecht in Figure 14. This figure
shows the decaying torque oscillations for a 700 .r.tw TG unit
over a period of ten seconds after clearing a close-in th·ree-
phase fault. Figure 15 shows the effect of damping on fatigue
calculations for an undamped, a low damped, and an 'expected'
damped shaft. The fatigue incurred by the shaft was inversely
related to the damping value. Furthermore, the damping value
does not appear to be linear (with vibration amplitude).
Both CMS units described earlier use feedback of actual shaft
angular velocity to compensate for such non-linear damping
effects in their evaluations of shaft life.

Shaft torsional damping varies with the vibration modes

excited, and with the amplitude of vibration in that mode.
This has been shown by Joyce ·and Lambrecht [3] in Figure 16,
and by Walker, Adams, and Placek [6] in Figure 17. Figure 16
shows the measured mechanical damping of the first three
modes of torsional vibration of a 970 Mw turbine with a two
pole generator after a terminal to terminal short circuit, in
terms of reciprocal decay rate (seconds). Figure 17 uses a
log. dec. vs. load graph, but it clearly shows the variation
in damping values, as both a function of load and of model.
Details of how this graph was obtained are given in the
appendix to the paper. It is evident that the damping
increases as a function of generator load, and the rate of
increase depends on the mode of vibration. The need here
appears to be for a calibratable computer procedure which
will allow damping to be calculated for the relevant modes of
a given TG system.

Current problems associated with the use of CMS units are the
lack of a means for retrofitting the past fatigue history of
the shaft into the cumulative fatigue calculations, plus a
means for accounting for the effects of modal damping. A
means for specifying an appropriate threshold strain value
that will provide optimum use of these diagnostic tools is
also needed.

10.1.6 Conclusions and Recommendations

Since the first reported shaft failure due to transient

torques was disclosed, advances have been made as follows:

o Knowledge of these electrical disturbances which can

cause transient torques in TG shafts is for the most
part complete.
Problems of Turbine Generator Shaft Dynamics 315

o Vibration analysis can now be efficiently accomplished

by use of tools such as the finite element method. A
remaining question is: how accurate a calculation does
one need?

o Development of basic continuous monitoring systems to

gather much needed data such as the frequency of an
electrical disturbance, its duration, and its transient
amplitudes have been accomplished.

o Increased awareness exists concerning how system plan-

ning can reduce damage to the TG shaft from shaft, i.e.,
system disturbances.

o Analytical refinements such as Rain Flow Cycle Counting

technique, and of the Rogowski coil that increase the
accuracy of calculations, and of data acquired during
monitoring.

o Increased knowledge of what direction further research

should take.

Several areas exist in which further study of this problem is

needed, i.e.:

o Knowledge of shaft fatigue criteria must be increased in

order to better predict shaft torque crack initiation
time.

o Effect of damping on shaft torque should be explored in

greater detail.

o A method to predict modal damping should be developed.

o Continuous monitoring systems should be improved which

can incorporate past damage into the life algorithm.

10.1. 7 References

1) Jackson, M. C., Umans, S. D., 'Turbine Generator Shaft

Torques and Fatigue: Part III-Refinements to Fatigue
Model and Test Results,' IEEE Transactions Vol. Pas-99,
No. 3., pp. 1259-1269, May/June 1980.

2) 'IEEE Screening Guide for Planned Steady-State Switching

Operators to Minimize Harmful Effects on Steam Turbine
Generators,' IEEE Transactions Vol. Pas-99, No. 4,
pp. 1519-1521, July/August 1980.
316 D.W. King- N.F. Rieger

3) Joyce, J., Lambrecht, D., 'Status of Evaluating the

Fatigue of Large Steam Turbine Generators Caused by
Electrical Disturbances,' IEEE Transactions Vol. Pas-99,
No. 1, pp. 111-119, January/February 1980.

4) Ramey, D. G., Kung, G. C., 'Important Parameters in

Considering Transient Torques on Turbine Generator Shaft
Systems,' IEEE Transactions, 1978 IEEE ASME/ASLE Joint
Power Generation Conference, Paper No. 7, November 1978.

5) Hizume, A., 'Transient Torsional Vibration of Steam

Turbine Generator Shafts Due to High Speed Reclosure of
Electric Power Lines,' ASME Transactions Paper No.
75-QCT-71, pp. 968-979.

6) Hammons, T. J., 'Stressing of Large Turbine Generator

Shaft Couplings and LP Turbine Final Stage Blade Roots
Following Clearance of Grid Systems Faults and Faulty
Synchronization,' IEEE Transactions 1978 IEEE ASME/ASLE
Joint Power Generation Conference, Paper No. 15, November
1978.

7) Fick, H., Stein, J., 'The Torsional Stress Analyzer for

Continuously Monitoring Turbine Generators,' IEEE Trans-
actions Vol. PAS-99, No. 2, pp. 703-708, March/April
1980.

8) Gibbs, E. E., Walker, D. N., 'Torsional Vibration

Monitoring System,' General Electric Publication
Presented at the Pacific Coast Electrical Association
Engineering and Operating Conference, March 13-14, 1980.
Problems of Turbine Generator Shaft Dynamics 317

Figure 1 Response Characteristics

of TG Shaft Sections [3]

...
t' ...
.•. L
............, r.ntt
•r•.... ..._"' unollf

Figure 2 Simple Mass-Elastic System [3]

318 D.W. King- N.F. Rieger

Generator Terminal Box

I I

Current
Transformer

. !

b
I • Rogowski Coil
Bushing

Figure 3 Rogowski Coil [7]

Problems of Turbine Generator Shaft Dynamics 319

~~~-~
r-" : ______:---,
I : : :=-:-:-rr- ---=---, 1 I

I
I
I
I
I
.,.~,., .
Et.c••onc Modl'l of 1tw 51\ah System

I
I
I
I
I
I
I
I
O .. «t ~· M.f'nOiy l'roc•M Coonpu1C"
I
s..... f•ltQwl'
Pl'f lnc:•dent o.p .. of o.s·- "'
I
F,
A«t.tmula1ecl
t'.owph,. S.t•tor W•nd•na
I
I
Dtforrn•tiO,. S1rftt'"l
SNh F•t•Qvr

L_ ].F
__j

,-~~
L:::J
Figure 4 Block Diagram of Torsional
Stress Analyzer [7]
320 D.W. King- N.F. Rieger

.
w,,('

n
Wrs
l
HPT
JR
w
~ Ill
m GENERATOR
m
w
EXC

IA,EA
Ia. Ea
lc· Ec

- TORSIONAL
VIBRATION
- ELECTRICAL
FAULT
MONITOR MONITOR

J
DA.S.
CTVMDASI J D.A.S.

r-1
CEFMOASI
FAULT
I
l TORQUE
TRANSDUCER
SENSOR

l
rI I I I CON~RTER I
TRIGGER
~ CON.~TER J TRIGGER

MICRO COMPUTER
WITH
MEMORY
J MICROCOMPUT ER
WITH
MEMORY
I
I DATA
COMMUNICATIONS
I
l SYSTEM I

~ ~
_j TELEPHONE
R LINE

REMOTE COMPUTER

DATA ANALYSIS

DATA STORAGE

Figure 5 Block Diagram of Torsiona l

Vibratio n Monitor System [8]
Problems of Turbine Generator Shaft Dynamics 321

r --
1
I

Figure 6 TSA TG Shaft Analog Model [7]

Figure 7 Typical Stress-Strai n Diagram [1]

322 D.W. King- N.F. Rieger

DATE: 10.12.77 HOUR: ll,H

TORSIOSAL STRESS AMALYZEP.

Cl: IIAXIKUH TORQUI LEVt."l. 0 REACHED O.OOOOS 99.926~

C2: IIAXI!M'I TORQUE LEVEL 0 REACHED 0.246Sl 99. 32S8

C3: JWtin.'K TORQUE LEVEL I REACHED 0. 38194

C4: NO IKI'ACT 0.00000 99.763~

NO DIPACT OM STATOR Eliii WINDING

Figure 8 Typical TSA Results [7]

Four Levels of Mechanical Impact ol

Operating Incidents on Turbine-Generator ShaH
Couplings and Recommended Actions
0 No permanent deformation.

1 Locelized minor permanent deformations which beer no

consequence for the operation of the machine.

Localized permanent deformations. Inspection should

like piece as soon as poulble and convenienl If the run-
2 ning behavior deteriorates, Immediate Inspection Is re-
quired. The coupling bolts should be replaced and the
bolt holes reworked, II necessary.

MaJor deformation can be expected. Shut down for lm-

3 mediate Inspection and repair of the couplings.

Figure 9 Code for TSA Printout [7]

Problems of Turbine Generator Shaft Dynamics 323

.. ,.., ,., , .., ··-

"-
J

!g{:(J·txHXJ:q,.§
.
I'------_,....::;...-~~-::::.:.-_:-_-_-_-.,:-_-_-_-;.--:_-:- ___;

."
..
~
C"
0
'i
.c.
"'

,;
.e p.u. (B)
•rr o.J
1-
"
0
~'--------M-e
•
I
Q.

~
c I
p.u.
O.!;'

,;
.e
.,•
~

!·:!,..,. . . . . . .""~
~ p.u.

Timtt

Figure 10 Comparison of Modeled vs. Measured Shaft

Torque for (A) and Three-Phase Short and
(b) Partial Load Rejection [7]
324 D.W. King- N.F. Rieger

ClASSICAL I.UMPEO MASS MODEl

ADVANCED CONIIt<UUM MQO(l

Figure 11 Different Types of Mass-Elastic

Models [4] - Also See Figure 2

1.0 9.5 HZ
9.3 HZ

-1.0
--18.5 HZ
t.O - 18.3 HZ

-c
II) 0
~
.
Ell'F"::r:::::fu- '(...U
,~
....
·JIE5~-·· ···~
E ',
...,
II)
0

0. -1.0

..
0
~
1 ·0 ·-- 21.4 HZ
- 20.8HZ
iii
a;

..
a:
·;,
~
"ii
...0E -1.0
z 1.0
--· 40.4 HZ
-39. HZ

" 1 .0 EXC GEN lP2 LP1 IP HP

Ro~M Position

Figure 12 Comparison of Modes Between Simple Spring

Mass Model and Complex Continuum Model [4]
Problems of Turbine Generator Shaft Dynamics 325

•• ...
...
II':' LPI

Ju1

-···-· ... .... ....... ...... .-...,.

. . . . .4 ••

. . =-
44.1.:.,
.....
~
I liP
,...,~

Shall aratem of utllltr turbine and generator unH

.... ,. .
~~t
-l!!!!JII"f __.;.~
-~=~~jili"r",
1 . .1 T . . - •..._...._. eM••

....-.-r••c-I

---·-1 _ ..··-··--i. . . -,
1 •

I
.... u
-~-f~~-f=-----~------------4~··~
d ,•
,..............,......... ,.,..,
•IMit I

-~~~t• ·~,,uc
''"
--·...·vNJW·fri~~J!!;:
,1 ~-· ··"····
~l'' 1 i'~'AI"l"'''f'
:
..
~ ~ ·lilll!~A!\~: ~
1

mmrl1 VHVH:W •..

..., ""....... ,, ....... .
I£C
1
z

"'.........
!"In\"'"• .. " • ·• ...
4P

•Hu·•·•~····-,., ..
J J.. su·
1
.,.," .... " • -...,~.
~.... •iltmn~ii;,.~ ~~~ Four Modes
...
.. ~ __ .. 'r ........
ntn,Utlutnt IIIUU•t•tttUtff'f''''''' tt SLC'
IFI ttl t I tttl I I fIt to
I :

....

Figure 13 Effect of First Four Modes on Torques [5]

326 D.W. King- N.F. Rieger

HP LP1 LP2 LPJ GEN EXC

-cr{Xt{XJ--{):J-{J~
11_
0~~-----~--~~~------------
/
c=::=:::::..__ Mode 1 15 Hr

.. -1

~e 2
t_:r
1
~ ["""---~ 24.7 Hr

~
~ ~~/\ ~e3
35.6 Hz

~-~~
J -~pi 1 :
Mode 4 60.1 Hz

.
~ 0~<=:::>
;;;

i ~I /
Tors10nal v•brallon modes ol J 970 MW. 50 Hz nuclear-lueled
turbme-generator.

LPJ CFN IP l,., lPl CEN EJliC

..--..J'·...-':
lP~

{:rJ-0- ry. r - . / ; .. ...,

1 . - -
....
L •
---~.rv'....:v :: ;_ __ ...:~
, i I '

I
oo'- -- --· +- - - ·. - - ... --- - -
0 - - - - r - . - - --
";'tOO -· · - ~- --- ·- --·-·' : :
~
-
["
;;I()'!.. -·- ··-
• ~-
0
:I
~.,.. f-- - . ...
-

: o.:....
- 1·
' --'-------
. :t-
J .
w
;;oo
2
)

1- --

tI
Four Modes ~o:_ . -.,~~-.'.'.!:..:.2
.e 1 -
cr --
I ~ oo·-I
0~ l ~
; "'... 1-·
,,
1·
0· · ·· - - - -- - .... -- 4 h
123• \2l4 1234 1234 12'34.
0 ,.... - · - · - - - · --- -- 5 loiOCS.

Torsional modal response at different shaft locations

to a uni-directional electrical impulse. [3]
4

::> 3 ~- :-- · . !.A)

~
..
:::J
CT

~
;;
~
U)

Decaying torque oscillations [3]

Figure 14
Problems of Turbine Generator Shaft Dynamics 327

(A)

2.01
- 1.f!l
-
~
41
-~ 1.0
•.....

0 1 2 3 4 5 6 7 8 9 10
Time (sl

Figure 15 Relation Between Fatigue Calculations

and Shaft Damping Values [3]

0.15 i

!
~ C.1~
1
r I
t I
ioos~
i r--,-______
I Dm3

i 0
500
0 100
Time (c,cles)

Figure 16 Measured Mechanical Damping of Torsional

Oscillations in Three Different Vibration
Modes [3]
328 D.W. King- N.F. Rieger

0 60 eo 100 120
"X.OF FW. LOAD

Figure 17 Typical Modal Damping Characteristics

Problems of Turbine Generator Shaft Dynamics 329

Table Effect of Different Electrical Disturbances on the

Amount of Fatigue Incurred per Incident [3]

.... ... ...

...............
.-c.= J
.. ...
~ ~-~- ·~_"- ··:··:··_ ... -- --- -- -
., ..,.. -
I ,. ...._-.
;

,. _ , ,, of. ": •;"': _..,D o1 r '- ~

• J
-=:J
I
' •"l.···=-..- -.-
- :._ t • · " • - · I ... ._ "" · ;t · , I
~ --:l
I
~ ----=
..,_,,, (.o • ., ••• ! v •,.. ....... --=-:J I
i I •,
,,~ I, 1 ..,., , .,., , • .,

1
~ - ~:-· -:-· .
1 ~--~~- ~~--
: 'J.."'T<-"•' • -=-:;

I
0 .. , ....., ~

=·i
t •t ,
! !
y
. -=1
=~
:
I
!'

-
I ( ·- , ,
'I
ic:::b
-- - - --~--
• j .......... ... , ....
I ... .....,.""1 t-::;::,:ll"'t'U, ... ·-

*
iF:
:.-- ·.,,1 ....,.. !1-_'G-• ·.~)
IJn, · ~•C.cv••

,..,,""'....,.."' - - ·····
. ... ....,.. .,~ ._..:., .
s...,.f'S ... -.t....~
\JIOI· ..,....,. . c.or.,., rt":..::..;;,.~ :;,.,-s; ~r:.- t.~~

;** '- -

Table 2 Calculated Torsional Natural Frequencies

for Simple (10 Masses) and Complex Shaft
Models (200 Masses) [31

-
Catc.ulaled Fr<equency Meeauted Fr~uenc.Y
Uo\4 H'""t..t ViMtion (Hz) (Hz)
Rol;ng of Number or M•~o5ea tn by Double- by Frequency
I"'-Wl fo-e. a tOt System Fr•quenc, An•t,'liS from

I E•c•••••on
f'olu SuddenSho<t -
n <10 n >200
(S. C .Tut) ~rcu•t Test O.tt..
.
630 2 1
I 17.0 16 . 5 - 16 . 7

660
• 1 7. 8 8 .2 - 11 .1

1200
r--- • 1 6 .5 6 .5 - 6 .3

1 14 . 11 14 .!1 14.11 14.!1

1170 2 2 24 . 3 24 . 7 24 .7 24 . 5

3 34 . 9 34 .6 Ji. 3 34 . 3
 CHAPTD 11.2

TORSIONAL SYSTEMS: VIBRATION RESPONSE BY MEANS

OF MODAL ANALYSIS

P. Sclnrlllllla•, R. Nord....
UaiYenltJ of K.-..ue_.., K......._...., FRG

ABSTRACT

Large steam-turbine generators in operation may be stimu-

lated to torsional vibrations by dynamic moments at the ge-
nerator due to electrical system transients. To solve the
torsional vibration problem the turbogenerator shaft is mo-
delled by the finite element method. The equations of mo-
tion are solved for the transient vibrations with the modal
analysis technique- 'Time-History-Method'. If the designer
is interested in an approximation for the maximum response
the 'Response-Spectrum-Method' proves to work very effectively.
This paper discusses 'Time-History-' and 'Response-Spectrum-
Method' and presents the results for a 600 MW and a 722 MW
turboset.

10.2.1 Introduction

Large steam-turbine generators in operation may be stimu-

lated to torsional vibrations by dynamic moments at the genera-
tor due to electrical system transients. The induced torsional
stresses in the shaft have drawn growing attention over the past
few years /1-4/, Fig. 1.

For the solution of the torsional vibration problem it is

essential to find an appropriate torsional model for the turbine-
generator shaft. A common approach is to model the torsional sy-
stem by the finite element method. As a resultoneobtains the li-
near equations of motion for the rotor. The solutions of the ho-
332 P. Schwibinger- R. Nordmann

r-·-·-·-·-·,
I I
NETWORK ---~j TrQnsmossion
I Iones

,L....--..-1---1
I

I TrQn sformer
GENERATOR I
Peril" s EquQtions L·--·- ·--·_j

Turbine VoltQge
govenor regulQtor

r· .,
i
I
i
i
!
HP IP LP1
! Gener Qtor -Rotor , j
L._._.-.- · - . - . - . - . _.1.....,:-=.:·.-=-··=-···--' -· - · j
TORSIONAL VIBRATION SYSTEM
Finite Element or Doscrete 11Qss Model

Fig. 1 Subsystems for calculating the effects of

electrical system faults

mogeneous problem are the eigenfrequencies and modes to which

correspond modal internal moments.

For arbitrary transient excitation of the system due to

electrical moments the equations of motion are solved by modal
analysis:
The decoupled equations of motion are integrated numerically in
the time domain. But this so called 'Time-History-Method' needs
extensive computer time. Often the designer is only interested
Analysis of Torsional Systems 333

1n the max1mum of a displacement or a ~ension. In ~his case an

approximation is sufficient which calculates only the r.:axima and
works therefore more economically. In this raper ~he arplication
of the 'Response Srectru:.:.:odal Analysis' ~o calculate transient
torsional vibrations of turbogenerators is discussed.

10.2.2 Modelling of the Turbogenerator Shaft

The first step in the analytical solution of a vibration

problem in the field of machinery dynarr.ics is to find a mecha-
nical model for the built machine, which describes its dynamic
characteristics adequately. For the solution of our torsional
vibration problem it is a common arproach to model the shaft
finely by the finite element method. In this case the total shaft
is subdivided in N-1 finite 'torsional elements', Fie. 2.
If modelled in this way a 600 MW turbogenerator has about 250 DOF.

We consider small - that means linearized - torsional vibra-

tions about a stationary ro~ation of the shaft. By using the prin-
ciple of virtual work we can describe the dynamic behavior of our
model mathematically; we get the mathematical model.

N-1 N-1 N-1

8W = I: + I: oWTe + I: owFe = 0 ( 1)
e=1 e=1 e=1

virtual work virtual work virtual work

of elastic moments of inertia moments of external moments

If we approximate the unknown torsional displacement in an ele-

ment with static deflection functions, we obtain a 2x2 mass- and
stiffness-matrix. Each element has two nodes with one DOF each.
Superposition of the element matrices yields the mass-matrix M
and the stiffness-matrix K for the complete system, Fig. 3.
334 P. Schwibinger- R. Nordmann

TURBOGENERATOR
SET

Ip

HP

FINITE ELEMENT
MODEL ( 250 DOF)

TORSIONAL
ELEMENT

K GJ [1· 1 -11 ]
- T
-e- T

Hnss-Hntrix Stiffnus-Hntri x

Fig. 2 Finite element model for the turbogenerator shaft

In a similiar manner we can transform the distributed external

moments to the nodes and obtain the excitation vector E(t). The
damping matrix is assumed as proportional to ~ respectively K.

D =a M + 8! (2)

The result are the equations of motion for the turbogenerator

shaft: a linear, inhomogeneous system of differential equations
of order N.
Analysis of Torsional Systems 335

~. .
~~ ·.~[I
I •I
·I]
I

I· · I i•l··· I
CoiiKtdtnct
Tronstoraotion I: •lA ftH I
I;'· !1'·1:
r:' ..l'·I:

fiNITE ELE"ENT
!!1m!.

l1•p(t1'I: •.:. '

Stittntss "ot•i• .1• Elf· .1' ·I:

'
.,]
• .:.

Fig. 3 Superposition of element matrices

!:! Si + .Q .9. + 1i 9. = f(t) ( 3)

To solve (3) we'll use the modal analysis method; but before we
study the corresponding homogeneous problem <r(t) = Q).

10.2.3 Modal Parameters of the Torsional Problem

Eigenfrequencies and Modes. First the equations of motion

are solved for the undamped natural vibrations, we get from the
homogeneous equations.

(4)
336 P. Schwibinger- R. Nordmann

Ass~~:r.g a solution of the for~

).<:.
"' ( ...... )
~\ = tile ( 5)

·..-e :::>t"':ain the eiger.value pro'clerr.

-; -\-
(K - Y. :.:) cp
-
=0 with I = - ). 2 ( 6)

Eigen·1alues 'Xl. :~odes cp •

-l

Nith N eigenvalues, 1. respectively eigenfrequencies f.,

l l

f.=-
l
1
2':1'
£ l
(7)

and corresponding modes -cp l••

The lowest eigenvalue we obtain from (6) is zero and corresponds
to the rigid body mode of our unbounded torsional system. Fig. 4
shows the next five elastic modes of a 600 MW turboset. In the
first mode HP-, IP- and LP1-turbine oscillate with 18,19 Hz
&gainst LP2-turbine and generator. In the first, second and third
mode the whole turbogenerator vitratP.s, whereas in the higher
modes only parts oscillate, e.g. the slip ring shaft in the fifth
mode.

Modal elastic moments. The natural vibrations in one mode

cause elastic moments in the shaft - so called modal elastic mo-
ments f • If we apply the principle of virtual work to one ele-
----m
ment we get, Fig. 5,

f
-m
e
= [-f m1
= Me .. e
!l + -
Ke
.9.
e ( 8)

fm2
Analysis of Torsional Systems 337

EIGENVECTORS MODAL ELASTIC MOMENTS

i- · - -~ : ~

ll!mr:!: ~Ji:l :· · . ···-· · ---~

~3 j·--~~"'·-~-·11!1im<tll'',...,"""'mu""
I
J

57.22 Hz
I l
=ljt ,._,__.\,....,.
L.,-··11~\l : . •.
,_,,...,
7:
rt til /
.,c:.'"="""=----- - _fm4
'LwiiJ"

. ~; .-""ill'i'"'-
~s ~-- _ ....:;:.cJ
118,72Hz ' -
+ ------~~~~;,; {;·;,::l fms

Fig. 4 Modes of a 600 MW turbogenerator with the

corresponding modal elastic moments

f~, q~
-
19e(~),(jie(~)
® --
q: f~2

L~e
.IS;e .Me

Fig. 5 Modal elastic moments at element e

If the rotor vibrates 1n a mode ~- with the frequency w. it 1s

-1 1

e e
q. = cp •
-1 - 1

resp. (9)

..e 2 e
q. = -w. ~ .
-1 1-1
338 P. Schwibinger- R. Nordmann

and we obtain from (8)

( 10)

In the slender shaft regions (couplings, slip-ring shaft) the

contribution of the inertia term Me~e in (8) is small compared
to the elastic moments !eSe• In that case we can reduce (10) to

( 11 )

In Fig. 4 the modal elastic moments which correspond to the lower

modes are shown. The maximum amplitudes are normalizedtoequal 1.0.
This maximum occurs mainly in the couplings where the gradient of
the mode is high. Because the modes are normalized, we cannot in-
fer from the modal elastic moments to the absolute stresses in
the shaft. But they show clearly, where the maximum moments 1n the
shaft occur, when it is excited in one natural mode. In the next
point we'll expand the internal moments due to transients with
the modal analysis method and use the modal elastic moments as
ingredient.

10.2.4 Calculation of the Transient Vibrations with Modal

Analysis - Time History Method

The transient vibrations due to dynamic electrical moments

are calculated from the inhomogeneous equations of motion (3).
Let us start with the virtual work of the complete system.

ow= esT {-!!.9.- Qil.- ! s + f(t)} =o ( 12)

The numerical solution of the resulting system of differential

equations is very expensive, because the individual equations
are coupled. We can solve the equations of motion economically
with the modal analysis method. For this we expand the unknown
torsional displacements s(t) linearly with the known elastic mo-
Analysis of Torsional Systems 339

des , .•
-1.

N
g_(t) = i=l
r ,.n.(t)
-1. l.
=1 n<t) ( 13)

The modes are arranged in the modal matrix 1, 1 )

( 14)

expansion parameters are the modal displacements n.(t).

1

(15)

From the energy equation (12) we get with (13)

cS!!T{ -~T~ .! .!J·· - ! Tp ! !!· - ~ ~

T
.! n + ! T! ( t )}
= o ( 16)
•
diag [mi} diag [di] diag [ki] g. (t)
l.

Because the modes are orthogonal to mass-matrix ! and stiffness-

matrix K and on the condition that (2) is valid, we get N-1 de-
coupled equations of the type of a single DOF system, Fig. 6,

Fig. 6 Modal one degree of

freedom system

m.n.
l. l.
+ d.n
l.
+ k.n.
l. l.
= g.(t)
l.
(17)

1 )Without the rigid-body mode , •

-o
340 P. Schwibinger - R. Nordmann

with

T
m.l = cp . M cp . - Modal Mass
-l --1

'I
d. = cp. D cp •
l - l --1
- Modal Damping
( H!)
T
k. =cp. Kcp.
l -l- -l
- Modal Stiffness

g. ( t)
l
=-cp T•l -F( t) - Modal Force

Equation (17) can be solved for n. analytically or for arbitrary

l
modal forces numerically with suitable algorithms. The unknown
torsional displacements follow with (13) from a superposition of
the modal solutions. For technical applications it is mostly suf-
ficient to consider only the lowest 1, ••• ,n elastic modes, to ob-
tain accurate results.

n
s(t) = E cp. n.(t) ( 19)
i=1 - l l

Besides the decoupling of the equations throu~h this the comruter

time can be further reduced.

Solution of the single degree of freedom system (5DOF). The res-

ponse of a SDOF to an arbitrary excitation can be written with
the Duhamel-Integral

t -D.w. (t-t)
n.(t) =- I g. ( T ) e l 10 .
SlOW. ( t-t ) d
T (20)
l rn.w. l l
l l 0

with the initial conditions

n.l (t=O) = o n.(t=o>

l
=o (21)
Analysis of Torsional Systems 341

the r:am-:>ing factor

d.
D. (22)
1 = 2m.w.
1
1 10

and the eigcnfrequencies

w.
10
=~
1 1
w.1 = w.10 (23)

Tt>e numerical solution of (20) is very uneconomical. 'i'herefore

ol.her more t'fft>ctivt> algorithmus should be used. A linear inter-
··o}al.jc.·n of the excitation function proves to be most effective.

:.:n tl-is approxir1at.ion ir••~thod the given mo•ia.l fr:~,~·:: ;.(t.l is

l
replaced by a polygonal course, which is formed by the discrete
k
values g. at the times tk, Fig. 7.
1

Fig. 7 Linear Interpolation

of the modal force
g. (t)
1

For the approximated modal force the equation of motion can be

solved piecewise exactly. For the k-th interval is the equation
of motion for n.(t * )
1

m.n. + d.n. + k.n. = g.

k
+ s. k t * (24)
1 1 1 1 1 1 1 1

and
342 P. Schwibinger- R. Nordmann

k
s.
1

If we restrict the solution to the discrete times t k and assume

constant intervals ~tk, the solution of (24) is

n.1k+1 a,, a12 n.1k b 11 b12 g.

1
k

= + (25)
n.•1k+1 a21 a22
•
n.1k b21 b22 g.
1
k+1

whereas the a , b (m,n=1,2) are constants which must only

mn mn
calculated once /5/.

The only error of the method lies in the approximation of

the modal force. Our experience for a periodic excitation func-
tion is: If the period of the highest harmonic 1s approximated
by 30 intervals the error in amplitude and phase is smaller than
one percent.

As an example we take the response of the 600 MW turboset

to a short circuit excitation, Fig. 8. The torsional displacement
at a global DOF qk superimposes with (13) from the contribu-
tions of the modal SDOF n.1 multiplied with the corresponding
eigenvectorcomponent 'ik

n
qk(t) = E , 'k n. (t) (26)
i=1 1 1

In Fig. 8 the contributions of the individual modes to the angu-

lar displacement of the LP2-Generator coupling are shown. From
the first three modes only the two lower ones contribute a con-
siderable amount to the displacement qk.
 >
::s
II>
-<
~-
Modal Displacement qk "'0
Eig envectors --.
Displacements >-3
11 "shorl circuit .. ...0
(5'
"'
::s
e:..
til
-~ '<
N02 Genera tor
"'.....(1)
HP IP LPI LP2 Gen . SR
3
"'
30
Mode
mrad
I
® '11 0 0 q lk O ~ '<i:l:iili'100 "<b::cY =""Q 100 <D
19, I' "' · ··· ,. ' c:::C:: ......... ....... s Tirre tim:~
- 10

lO
.. , rn r ad 0
Alil\ Aool'\ D
@ 'lz ol ,,.ul ('{ , l'G'~' ',, .I F' '• · I ,,}u·, ~H P, \/\..1 -: · 0 0
1n 0 q 2• l.ll/ "" "loa W "" - 'zoo®
T:z · l .. :,,.,~, '·,• ..: · I
~CJ '•I'I'• i, ltt.i),

I - lo .. 30
I
F I
0

~
•n~
Y]
I
II @ 0 qlk 0
T. C> = ·~ ,... = ..... 200 Q)
c:::::::::::::> c==T">

- )0 - 30 •

10 @)
mrad

0
I & - -
q. r "' YiiJII 'i!roo v v .... '{zoo
-10

Fig. 8 Modal expansion of displacement \,.)

...,.
\,.)
344 P. Schwibinger · R. Nordmann

'!'l".e ar.-.;:E:~,;.:Oe ~! -:.he thir:i ::aocie at. the cout::ling ic ~~~.al1 9 i:c
:::::r::·.:.:br. :c :l".e :.:;;bce:::en~ 1k ~he1·efore indgni ficant.

Calc·..;:a:ic:-. c!' ;.:-.:e~:-.!.1 ~.o::-.er:~;. 7t.c 'JE:signer <~ill ask:

::.-;-.: are ~r.e ir.-:.er·ral r.".O~-~r.:.s an'! .;:hear forces .ira t!.e shaft due

For :l".e :ransiera: ·r:t.ra:ion ..-e can 1evelop the c.:la::tic for-
c~s at an ele=en~ e in a si~~liar ~ay as w~ have done it abov~

·.:itl". -:l".e princi~le -;f ·rir-:~,;al iiOr~ 9 Fig. 1.

- -.
Torsional
Moments DisRlacements
,:~,,

Lr.\. !L
~ '5' ~·
K'.t!
••
ltft nodt

Fig. 9 Exitation and torsional displacement& to calculatr. the

elastic moments

c
At a torsional element act the distrubuterJ moment r A9 the: lli~-
e
crete moments rEi• the inertia moments and the internal momentG
at the nodes. Assuming the distributed moment as constant in an
element. we get for the internal moment&

[
1
- ~i ]' (27)
(.
1
Analysis of Torsional Systems 345

As already mentioned at the modal elastic moments, the max~mum

stress occur~in the slender couplings. Here no external moments

act on the shaft, in addition is the contribution of the inertia
terms 1n (27) small compared to the elastic terms. For the coup-
lings we may therefore simplify (27) to

-f
e
e s1
f (t) [ (28)
-s
fs2

or written as a superposition of the individual modes, compare

(11), (13)

e n n
f ( t ) ~ E Ke cp ~ n . ( t ) = E f ~ n . ( t ) (29)
-s i=1 - -1 1 i=1 1111 1

For the 600 MW turboset Fig. 10 shows the calculated coup-

ling moment between LP2-turbine and generator.
At time t=O the shaft is tensioned by the constant turbine and
generator moments. At t=10 ms a two phase short circuit occurs
and excites the generator with net- and double net frequency
components. For the modal damping factors values were taken from
literature (D.~0,01-0,02) /6,7/. The highest amplitude of the
1
resulting coupling moment is four times the static moment. Be-
cause of the small damping the response decreases slowly.

As in the case of the displacements, Fig. 8, we discuss the con-

tributions of the modes to the maximum coupling moments, Fig. 11.
We see, e.g. for the LP2-Generator coupling, how the individual
modes superimpose to the coupling moment, from which we find the
maximum. In the diagram on the right hand side the contributions
for the different couplings are plotted. The lowest three up to
four modes have a determining influence on the maximum at the
couplings - except for the slip ring shaft, the elastic moment
of which is determined by the fifth mode. One compare with the
346 P. Schwibinger- R. Nordmann

2-Phase Terminal Short Circuit

[lpclrica l and Couplng Moment

HP IP LP1 LP2 lien era tor SR

-M.
e - -·- M11 •Mn•Mn•M14
...
z
i:
18 600

10 000
~ .,
e0
::1:: 0
8
~ -10 000
.:.:; -14 300

~00
I Time t/ms

Fig. 10 Excitation moments and corresponding coupling moment

at LP2-Generator

corresponding modal elastic moment ~n Fig. 4.

So when we use modal analysis we get some information on the

composition of the systems response. If we are only interested
in the maximum response we can also use a technique based on mo-
dal analysis to save further computer time - this technique is
called 'Response-Spectrum-Method' (RSM).
Analysis of Torsional Systems 347

COUPLING MOMENT

HO MD N01 N02 Gen. SR

ND2 Gen.

-
E
z
..ll::

~
E
0
l:
Cl
.!:
Mode c.
:l
0
u
E
.~
-4
r
I ~ ~ I
Mo 0·1
.t!lk
3
Jhv- Ah~o" ~yo..~oo ®
v
~
"8
I
I
I
I

-
l:

~-:~
I
~ I
I
I
v·ULP 1o't <uP' - - ""' 2oo (]) 15 I
:.;:
il E I
II
-~ 1
1
·;:: E I v
.&_k ~ § I I
I II
u
E I II
Mo -----------
Or·- - - - - - - - - -100 200 ® E I v
1000 E I v
E I I
~
-4
@ I
E ~~ I
I
~~
~ ~~ I ~~
t ~·
IE
@ ~ ~~ II

•
1:!JL3 II I
Mo 0 lmrm._,..\nnf"\nruonnn1"' ~~ II ~~
~~ 2
0 123412341234 1234 1-34
Mode

M0 = 2182 kNm

Fig. 11 Contribution of the modes to the maxlmum coupling

moment
348 P. Schwibinger- R. Nordmann

10.2.5 Calculation of the Transient Vibrations With Modal

Analysis - Response Spectrum Method (RSM)

In this point we apply the RSM to approximate the torsio-

nal response of a turbogenerator set to transient excitation.

Basic idea and theory. The response of our linear system

superimposes from the contributionsof modal SDOF, Fig. 6. Equa-
tion (20) gives the solution of a modal SDOF to an arbitrary ex-
citation. We consider the case that the excitation vector f(t)
~n (16) may be written as

f(t) = f·h(t) ( 30)

and therefore the modal forces (18)

"'
T F·h(t) ( 31 )
g.(t) =
~
<p.
-~ -
where h(t) is an arbitrary 'time function'.

If (31) is valid the response n.(t) from (20) can be written

. ~
as

tn. T "
F t -D.w. ( t-T)
:!~ -
n. (t) =---
~ m.
f
w. 0
h( T) e ~ ~ 0 sin w.(t--r) dT
~
~ ~

(g)
n. (t) ( 32)
~

or
T F" (g)
'P.
n. (t) =--- n. (t)
-~ -
~ m. ~
~

(g)
where n.(t) is the modal response of the i-th SDOF to the exci-
~

tation h(t), Fig. 12, and the term

Analysis of Torsional Systems 349

T
- <il· F
w =--- -].
(33)
i m.
l.

is called 'participation factor'.

Ill

"·-
J.
'1;-

~D•const
""~ I
I
I

hcitotion hltl Rnponse Spectru•

•odal SDOF Response of SDOF

Srstea with Syste11
W;,.D;

Fig. 12 Response of modal SDOF to excitation h(t)

Instead of the 'Time-History' that means the complete time func-

tion we work further on only with the maxima of (g())
n.l. t .

1 t -D.w. (t-T)
n. (g) = max{-- I g(~) e l. 10 sin w.(t-~)d~} Sd(w.l.O ,D.) (34)
1.max w.l. 0
l. 1
350 P. Schwibinger- R. Nordmann

This value we plot in a diagram as the maximum response of a

modal SDOF, characterized by w.10 and D., to the excitation h(t).
1
If we examine other modal SDOF with varied eigenfrequencies w.10
and modal damping D., and store always the max1mum 1n a 'Respon-
1
se-Spectrum', we obtain the following set of curves

~m,

Fig. 13 Response-Spectrum

Besides the 'Response-Spectrum' for the displacement also spec-

tras for the velocity resp. acceleration can be calculated.

We know the maximum of ni' which 1S

T ,. T A
cp. F cp. F
n~g) -1 - -1 -
nimax = 1max m = m. Sd(w.10 , D.)
1
(35)
i 1

and therefore the maximum of the i-th modal contribution to the

displacements (13)
(ji.T F
(jl.l
-1
-1
=---
m.
-
Sd(w. , D.)
10 1
( 36)
1

and to the coupling moments (29)

T A
cp. F
l-s f71
f711 max=ln.1max -m1 = 1.:1.:.....-::
m.
Sd(w. ,
10
D.) f
1 -m1
71 (37)
1
Analysis of Torsional Systems 351

An example

We apply the RSM to approximate the transient response of a 722

MW turbogenerator 1 ) to a 2-phase terminal short circuit excita-
tion. The approximations with the RSM are compared with the 'ex-
act' solutions of the THM.

First the 'Response-Spectrum' for a excitation function cor-

related to a 2-phase terminal short circuit, Fig. 9, was calcul-
ated. The corresponding electrical moment contains mainly the
single and double net frequency (= 60 Hz resp. 120 Hz in the
United States). We have assumed in (30) that only one excitation
function works at the system. But our shaft is excited with the
constant steam forces and the generator moment, which is constant
first and starts oscillating when the short circuit occurs. In
this special case of excitation, the static preload condition
can be transformed to an equivalent system with zero initial con-
ditions and an electrical moment to which the static part is ad-
ded /8/. Therefore we have a system which is excited by only one
excitation function and (30) is valid.

Fig. 14 shows the calculated corresponding response spectrum of

the displacements for different modal damping factors. It is im-
portant to note, that this response spectrum is independent of
the considered system!
After the response spectrum has calculated once, the modes -J.
•· and
the participation factors W. are calculated for the considered
l.
722 MW turboset - they are also marked in Fig. 14. The maximum
of the i-th modal contribution lq.l. Imax can therefore with (36)be
easily obtained from Fig. 14.

1 )built for the United States

352 P. Schwibinger- R. Nordmann

The question is now: How to superpose the contributions of

all individual modes to get the total system response? The simp-
lest approach is simply to add all the maxima of the individual
modes, e.g. for the displacements

n
I~~max., i~1 I ~i Imax (38)

But because the maxima in the different modes don't occur at the
same time this approximation overestimates the solution in ge-
neral. The experience shows that a mean sguare approximation
yields much better results. We use

IS Imax * I~.
1=1
I9i 1max
2 ( 39)

if the eigenfrequencies don't lie closely together, that means

w. -w.
JO 10>0, 1 (1 S i < j ~ n) •
w.10

For eigenfrequencies lying closely together the 10%-method is

used.

I9 Imax tv /1 9k 12max + 2 r
i~j I~i Imax I9j Imax (40)

whereas i,j are indices of the eigenfrequencies for which

w.,)0 -w.10
~ 0,1 (41)
w.10

is valid.
Analysis of Tonional Systems 353

5 100
...•
N
'o

4 80 o-o.oos • O•.O.OS
0·0•0,01 • 0•0,1
• 0•0,02S

3
-z 60
E

!40
E
!: 2
No
'3 ....
• '30
i 1 •
A

-
"' 20 0•2u60Hz
J. ,tl~ i"' ~ Wo
0 0
~o I 40 60 80 100
I
120 140
I 160
w,.wzo wlo '"'t.o '"'So w6o
,...... W./0
0 1 2 3 4

Fig. 14 Response-Spectrum for a two-phase short circuit

and participation factors of a 722 MW turboset

The results for the maximum displacements are presented in Fig.

15, calculated with

• Time-History-Method
• 10% Response-Spectrum-Method
• to check - the sum of maximum response of all modes

The sum of maximum response of all modes overestimates the solu-

tion all over the shaft. The 10% RSM using mean square averaging
354 P. Schwibinger- R. Nordmann

h(tl

'a
...E
Cl

.....
go 60

--Time-History-Method (THHJ
------ 100fo-Response-Spectrum-Hethod IRSHI
-·-·-·- RSH: Sum of max. response of all modes

Fig. 15 Maximum displacements of a 722 MW turboset calculated

with Time-History-Method, 10% Response-Spectrum-Method
(40) and sum of maximum response of all modes (38)

yields quite good results in all shaft regions except the gene-
rator. Here the approximation yields to small displacements com-
pared to the TH-Method. Probably in the response the maxima of
two or several modes supperpose here. It is important to note
that near the couplings, where the maximum twisting and the maxi-
mum stresses occur the RSM apprcximation is fairly good.

The intention of Fig. 16 is, to show the saving 1n computer

time by using the RSM. In the THM about 16% of solution time is
spent solving the eigenvalue problem and 84% calculating the res-
ponse of the modal SDOF-Systems and internal forces. In compari-
son to that and under the condition the adequate 'Response-Spec-
trum' is already calculated the RSM needs 16% time for the eigen-
value problem and only 1% calculating the maximum response for
the modal SDOF and the approximation of the internal forces. That
Analysis of Torsional Systems 355

II TIME -HISTORY- METHOD I USED TIME II RESPONSE-SPECTRUM-METHOD

RE~PONSE. SPECTRUM FOR

SPECIAL TIME FUNCTION
ONCE PARAMETERS:
- NATURAL FRECUENCY
-DAMPfNG

EIGENVALUEPROBLEM EIGENVALUEPROBLEM
-!l!;,w;,mi 16°/o 16°/o -!;,wi,mi

RESPONSE OF SDOF • MAXIMUM RESPONSE VALUE

E. G. POLYGON· PROCEDURE 0,5°/o FOR SDOF--SYSTEM
(RESPONSE SPECTRUM)
84°/o
CALCULATION OF INTERNAL APPROXIMATION FOR
FORCESISTRESSESI 0,5°/o INTERNAL FORCES

100°/o
-17°/o
Fig. 16 Comparision of computer time used by 'Time-History'-
and 'Response-Spectrum-Method'

means 83% time saved.

10.2.6 Conclusions

1. To calculate the torsional vibrations of turbine-generator

sets the dynamic behavior of the shaft system is modelled by the
finite element method. The result are the linear equations of mo-
tion. The solution for the natural vibrations yields the eigen-
frequencies and modes together with the corresponding modal elas-
tic moments.

2. For transient electrical excitation the equations of motion

are solved with the modal analysis technique. The equations of
motion are decoupled with the modes. As result we get n modal
single degree of freedom systems (SDOF). For the solution of the
SDOF an algorithm based on a linear interpolation of the excita-
tion function proves to be most effective.
356 P. Schwibinger- R. Nordmann

3. If the solutions of the SDOF are superposed in the time do-

main we obtain the total system response. We call this Time-
History-Method (THM). If we are only interested in the maximum
response we can use very effectively the Response-Spectrum-Method
(RSM) as an approximation. The THM and the RSM are applied to a
722 MW turboset. It is shown that the RSM saves 83% of computer
costs and yields good approximations in the coupling regions.

10.2.7 References

1. Berger, H., Kulig, T.S.: Simulation Models for Calcula-

ting the Torsional Vibrations of Large Turbine-Genera-
tor Units after Electrical System Faults, Siemens For-
schungs- und Entwicklungsberichte, Band 10, 1981, Nr. 4.

2. Gonzales, A.J., Kung, G.C., Raczkowski, C., Taylor, C.

W., Thomm, D.: 'Effects of Single- and Three-Pole Swit-
ching and High-Speed Reclosing on Turbine-Generator
Shafts and Blades', IEEE Trans., Vol. PAS-103, 1984.

3. Canay, M., Rohrer, H.J., Schnirel, K.E.: 'Effect of

Electrical Disturbances, Grid Recovery Voltage and Ge-
nerator Inertia on Maximization of Mechanical Torques
in Large Turbogenerator Sets', IEEE Trans., Vol. PAS-99,
No. 4, 1980.

4. Schwibinger, P.: 'Torsionsschwingungen von Turbogruppen

und ihre Kopplung mit den Biegeschwingungen bei Getrie-
bemaschinen', PhD-Thesis, Univ. Kaiserslautern, 1986.

5. Kramer, E.: 'Maschinendynamik', Springer-Verlag, Berlin,

1984.
Analysis of Torsional Systems 357

6. Lambrecht, D., Kulig, T.: 'Torsional Performance of

Turbine Generator Shafts Especially Under Resonant
Excitation', IEEE Trans., Vol. PAS-101, No. 10, 1982.

7. Hammons, T.: 'Electrical Damping and its Effect on Ac-

cumulative Fatigue Life Expenditure of Turbine-Genera-
tor Shafts Following Worst-Case Supply System Distur-
bances', IEEE-Trans., Vol. PAS-102, 1983.

8. Clough, R.W., Penzien, J.: 'Dynamics of Structures',

Me Graw-Hill, 1975.
 CHAPTER 10.3

TORSIONAL DYNAMICS OF POWER TRANSMISSION SYSTEMS

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

The sources of torsional vibration in geared power

transmission systems are reviewed. The modal
analysis of typical geared systems is described,
with useful formulas for natural frequency
analysis. Damper design principles are reviewed,
with criteria for suppression of troublesome modes.
Two Case Histories from practice are discussed,
indicating the diagnostic procedure used and the
vibration suppression methods in each case.

10.3.1 Sources of Torsional Vibrations

Torsional vibrations occur in many machine systems. Rotating

power transmission machinery is particularly susceptible to
torsional vibration problems because of the long shaft
sections and concentrated disk construction. Torsional
pulsations arise in practice from typical sources such as:
reciprocating mechanisms, e.g., internal combustion engines;
impulsive excitation of rotating equipment, e.s., punch
press, generator electrical transient, and in a wide range of
other machine systems. A general classification of torsional
excitation sources related to machine types is shown in Table
1.

Increased power transmission requirements are usually

accompanied by an increase in machine size and an increase in
operating speed. Increased power frequently means an
360 N.F. Rieger

increase in the magnitude of the applied fluctuating torque,

in addition to an increase in the steady torque. Increased
machine size means lower natural frequencies for the system
torsional modes, and increased speed means more modes are
possible within the machine operating range. Each factor
associated with increased power output therefore tends to
increase the vibration potential of the torsional system.

10.3.2 Jbeorx of Torsional Vibrations

The simplest torsional systems are shown in Figure 1. The

simple torsional oscillator, Figure 1 (a) obeys the same
vibration laws as the simple linear oscillator: the
analogous relations are shown in Table 2.

The two inertia semi-definite model shown in Figure 1 (b) is

more representative of torsional systems, which usually have
disks at their ends and are free to oscillate at these
locations. Machine drive trains are typical of such systems.

The natural frequencies of such a system are readily found.

Applying Newton's Law for free vibrations of the disks gives:

(1)

In matrix form:

+
[~ =

For free harmonic vibrations assume solutions:

9 1 = Ae i~t ; 9 2 = Be i~t (2)

Torsional Dynamics of Power Transmission Systems 361

On substitution these expressions lead to the frequency

equation:

4 2 =
I1I2(1) - q(I1+12)(1) 0 (3)

The roots are:

(1)1 = 0
(4)

(1)2 =
f•H1+I2l] 1/2

~ I1I2

The zero root is characteristic of semi-definite systems. It

arises from the free-end conditions, and is referred to as a
'rolling mode,' in which both disks move (roll) in phase.
This mode is only of significance in shock induced vibration
problems.

The second mode is important. It consists of the disks

vibrating against each other. Typically this mode is the
lowest observed mode in all free-end torsional systems.

Modes of torsional systems may be calculated as follows:

From equations 1 and 2 we have:

q - I (1)2
( 1 ) • 9 (5)
q 1

Knowing the natural frequency values and the system constants

allows the coefficient of 9 1 to be found. By allowing 9 1 =
1.0, the value of 9 2 may be Iound, for (1)1 and (1) 2 •

Example: A turbocharger consists of two steel disks 6.0

inches diameter by 2.0 inches thick, connected by a
uniform solid circular steel shaft 1.0 inch
diameter and 10 inches long, as shown in Figure 2.
Find the two lowest frequencies of this system.

o Disk inertia:

W = n/4 D2L·(I) = n/4 (36)(2)(0.283)

= 16.00 lb.
362 N.F. Rieger

(36)
= ----
g 8(386.4)
= 0.186 lb. sec. in. 2

o Shaft stiffness:

q = n/32 (12.10 6 ) 64 /10

= 152.68 x 10 6 lb. in./rad.

o Natural frequency:

7152~-7) (to 6 ) (0.372)

f-fi-~+y
w =
2 1112
=
v (0.186) 2

= 40.52 X 10 3 rad./sec.

o Angular relation:

I 2 I
'
llw I
92 = :t q i 91
!
,_ .i
r (0.186)w 2 .i
It- 91
!
L 152.7 X to 6 .

.Mode 1: w2 = 0 92 = 1.0 91 = 1.0 (rolling mode)

1

Mode 2: w22 = 1641.87 92 = [1-2.00]91 = -1.0 91

These modes are illustrated in Figure 3. Note that the

undamped mode analysis shown here is normalized on the dis-
placement 9 = 1.0 (selected arbitrarily). No knowledge of
applied excttation or damping is required for this result.
Torsional Dynamics of Power Transmission Systems 363

The analysis of other torsional systems proceeds in a similar

manner. The general torsional system shown in Figure 4 is
found frequently in practice. With 9 1 and 9 4 for the disk
displacements, 9 2 and 9 3 for the gear displacements, and q1
and q 2 for the shaft torsional stiffness, the equations of
motion for the end disks are:

Assuming the shafts have zero inertia (or that this has been
included in the disks), and neglecting the inertia of the
gears (close to node in mode 2), the equations of equilibrium
at the gears are:

where F is the gear contact force, and R , R2 are the gear

radii. From these equations the lowesl non-zero natural
frequency is:

1/2
L
2n ] hz

and:

when N is the shaft speed ratio.

Assuming that system excitation is caused by gear two only,

the following relationship applies between the gear cycles at
any time:
364 N.F. Rieger

and:

is the run-out angle for gear two, where a 2 is the eccen-

tricity of the pitch circle about the center of rotation, as
shown in Figure 4. For constant angular velocity ((1) 2 ) of
gear two:

Solving the above equations, the dynamic force F between the

gears at frequency (1) 2 is:

F =

The response amplitudes at this frequency are:

o.iKiK.4I4R2
91 = 2 2 2 2
11q1(q4- 14(1)2)R3 + 14q4(q1- 1 1(1)2)R2

o.2K.lK.4I4R2
94 = 2 2 2 2
11q1(q4- 1 4(1)2)R3 + 1 4q4(q1 - 1 1(1)2)R2

It is important to note that in geared system cases where the

shafts of a system rotate at m different speeds, the number
of resonant (or critical) speeds of the system is then the
number of shafts m times the number of natural frequencies n
of the system, i.e •• mn. Thus for the above case there are
two shaft speeds (m = 2) and one natural frequency (ignoring
(1) 1 = 0). Thus mn = 2, and the system has two resonant
speeds. The reason for this is shown in Figure 5. The two
per-rev excitation lines cross the natural frequency line at
f 1 and f 2 • the two resonant speeds.
Torsional Dynamics of Power Transmission Systems 365

10.3.3 Natural Frequencies of ~haft Systems

Formulas for calculating the natural frequencies of several

shaft systems are given in Table 3. Each of these systems is
semi-definite in nature. Thus, a three inertia system may be
solved for the two significant eigenvalues directly. The
frequency matrix for four order and higher systems must
usually be solved by some root-seeking procedure.

Table 3 gives frequency equations for branched systems and

for looped systems. Many cases frequently encountered i~
practice may be solved conveniently using these results.
There appears to be no information in the open literature on
looped systems, though branched systems are occasionally
discussed. The formulas relate to both harmonic oscillations
and to transient system dynamics, in which the first step in
a modal analysis is to determine the natural frequencies and
modal vectors of the system. Some experiences with branched
and looped systems are described in Case Histories 1 through
4 herein.

10.3.4 Viscous Damper Analysis and Design

The following general rules have been shown to be generally

effective for the suppression of torsional vibrations:

o Eliminate the source of vibration.

o Attenuate the vibration response.

o Provide additional damping in the responsive modes.

o Improve the fatigue resistance of affected components.

This section is concerned with rules (b) and (c). Consider

the undamped single degree of freedom system shown in Figure
6 (a) which has a single natural frequency ~ • If a second
inertia is added as shown in Figure 6 (b), ~he system then
has two natural frequencies ~ 1 and ~ 2 • These frequencies are
displaced about~ as shown rn Figure 6 (c). Evidently, the
addition of an aaditional inertia is a method for changing
the dynamic characteristics of a simple system. However,
without the addition of damping, the resonant peats will
still involve very large amplitudes of vibration.

Note that this de-tuning has been achieved at the potential

price of a second system resonant frequency.
366 N.F. Rieger

This above detuning procedure is a method for reducing the

vibration amplitudes of certain systems which receive
excitation at some constant frequency which is situated close
to a natural frequency of the system. In practice it is
usually only possible to add small inertias I 2 , e.g., the
inertia ratio (I 2 JI 1 ) frequently lies in the range 0.1 <
1 2 /1 1 < 0.3. Such attenuation is usually possible only where
tlie source of excitation is steady. Variable excitation
would evidently re-introduce resonance problems.

A vibration damper can also reduce resonant vibration ampli-

tudes. Again, consider an undamped single mass system,
Figure 7 (a), to which a viscous untuned damper is added, as
shown diagrammatically in Figure 7 (b). The damper has the
effect of: (a) reducing the vibration amplitude, and (b)
reducing the system natural frequency. (This may also detune
the resonance). The reason is readily seen by considering
two extreme cases:

Case 1: Damper coefficient c = 0. Zero damping, single

degree of freedom system is unaffected, amplitude
response as previously shown in Figure 7 (c).

Case 2: Damper coefficient c = m. Zero effective damping,

no relative motion. Single degree of freedom
syste' natural frequency reduced to w1 = (K/ (1 1 +
I 2 ) }1 2 , as in Figure 7 (d). Amplitude response
again high.

The practical case is the well-designed damper, which causes

a reduced system natural frequency, and the relative motion
(9 2 - 9L) between the inertias causes energy dissipating
slippage \damping) to occur.

It can be shown that the degree of dam_per effectiveness is

related to the inertia ratio V = I 11 1 • For optimum damper
performance the required damping rat~o z; D is given by:

1
z;D =
J2(2+V) (l+V)

where:

c q
z;D = and 11)0 = (::1)
21 2w0 11
Torsional Dynamics of Power Transmission Systems 367

The damped amplitude ratio is then given by:

This may be compared with the amplitude ratio without the

damper which is:

where ~ is the original system damping ratio. This factor

was ign8'red in the damped design because it is almost
negligible for many torsional systems, and ~ 0 » ~ u•

Fo~ a typical inertia ratio v = 0.2, the amplitude ratio is

(2.2/0.2) = 11.0, compared with an undamped amplitude ratio,
which for ~ = O.OOS is 100.0. The introduction of a
viscous dampel' in such a case would therefore lead to a
vibration amplitude reduction of 9:1 in the original system.

It has been found that optimum viscous damping occurs at the

frequency where the two undamped response curv~s intersect,
Figure 7 (b). This is called the 'tuning' point. The design
condition is that the damper responsive curve has its peak
amplitude at this tuning frequency. It is seen that the
larger the inertia ratio v, the greater the distance between
the undamped peaks, and the lower the value of the damped
peak amplitude.

From a practical standpoint, as the damper inertia 1 2 in-

creases,_ the inertia of the surrounding casing I~ also in-
creases. This in turn, increases I 1 to (I 1 + 1;1
attached to 1 1 • The actual inertia ratio is then:
as I; is

which being less than I 2 /I lessens the damper effective-

ness. Lightweight casings wiich minimize I;11 2 are therefore
another important aspect of damper design. Details of a
practical viscous damper are illustrated in Figure 8.
368 N.F. Rieger

10.3.5 Tuned Damper Analysis and Design

The effectiveness of the torsional dampers may be further

improved i f a suitable stiffness (or tuning) element is
inserted in parallel with the dashpot, as shown in Figure 9
(a). The effect of this element is shown in Figure 9 (b).
The original system response is shown, together with the two
mode response of the corresponding undamped two inertia
system. This response intersects the single inertia system
response at two tuning points, as shown. By a suitable
choice of damping constant, the damper response can be tuned
to have local maximum values at the tuning points. Two cases
must be considered when this is done, as follows:

Case 1: Response peaks equal at tuning points, Figure 9

(c).

Required inertia ratio:

Required damper ratio:

c
z; =

Required spring ratio:

Amplitude ratio:
Torsional Dynamics of Power Transmission Systems 369

Case 2: One response peak minimized, other disregarded,

Figure 9 (d).

Required inertia ratio:

\) =

Required damper ratio:

Required stiffness ratio:

Q =

These notations are defined in Figures 9 (a) through 9 (d).

Stiffness ratio calculation: Nestorides [1] discusses the

following procedure for calculation of the stiffness ratio
q 2 /q 1 • The required tuning condition reduces to the
expression:

or

The sum of the roots of this factorized2expression c~ 2 +~b 2 )

is equal to the coefficient of the ~ term in thea first
expression times (-1), i.e ••
370 N.F. Rieger

This expression. allows the required stiffness ratio to be

found by specifying the frequencies (1) 1 and (1) 2 at which the
system resonances are desired. These frequencies are sub-
stitut~d in the above biquadratic. which is solved for w 2
and (l)b • These terms then give the amplitude ratios at t&e
resonant peaks:

T
0
a = -------------2; b =
qq - (11+I2)(1)a q

The required damping ratio does not influence these peak

amplitudes. but it affects the off-peak amplitudes.
Nestorides [1] shows that the required damping ratio ~ can
be obtained from the following equation:

Examples of use of the above procedures are described by

Nestorides [11. Ker Wilson (2]. and others.
Torsional Dynamics of Power Transmission Systems 371

10.3.6 Case History No. 1

Methanol Compressor Drive Train Torsional Vibrations

Problem Details:

The methanol compressor drive train shown in Figure 10 was

exhibiting the following vibration symptoms:

o Excessive gearbox noise level, 103-105 dB.

o Rapid erosion of gear surface at pitch line (clearly

evident after 72 hour operation).

o Intermittent sharp metallic knocks from within gearbox.

o Noise and knocks increased in severity with increased

process loads.

o One gear pair removed from service due to pitch line

wear six months after startup.

Machine Specifications:

Motor type: AC squirrel cage rotor, 4 pole

Speed: 1780 rpm
Power: 1250 hp
Frequency: 60 Hz
Gearbox type: Bobbed double helical
Ratio: 1.685:1
Quality: AGMA Grade 8
Face Width: 10.875 inch
Pinion Teeth: 92
Gear Teeth: 155
Pitch: 9.148
Compressors: Methanol gas
Axial Flow: 10 stage
Inlet Temp: 220°F
Outlet Temp: 6500F

Inspection Details:

o Impact damage was clearly evident upon inspection as a

bright band at the pitch line after 72 hours of
operation with replacement gears.

o The sharp intermittent (impact) noise from the gearbox

was apparent, over the plant background noise (also
high, around 90 dB).

o Couplings were gear tooth type.

372 N.F. Rieger

o Compressors were rigidly secured to the poured concrete

foundation. No special provisions for thermal growth.

Diagnostic Procedure:

o Preliminary diagnosis suggested that gear tooth impacts

due to torsional vibration were the most likely cause.
The bright metal surface at the pitch line was though to
be due to plastic flow.

o The torsional natural frequencies and mode shapes shown

in Figure 12 were calculated to provide data on the
dynamic properties of the torsional system.

o It is evident from Figure 12 (b) that the second

torsional natural frequency (1728 cpm) and the motor
speed (1780 rpm) are very close so that a one-per-rev
gear excitation could strongly excite the second mode.

o Since the gearbox is nearly at a node of the system.

relatively small excitations result in large amplitude
response.

o The most probable source of excitation was one-per-rev

excitation from the low speed gear. This could arise
from machining tolerances or from non-concentric
mounting of the gear on its shaft.

o The motor is also a possible source of one-per-rev

torsional excitations. due to air gap and electrical
waveform non-uniformities.

Proposed Remedy:

o New gears were manufactured with a pitch of approxi-

mately 6 to AGMA 10 grade. This probably reduced the
excitation and increased the impact wear resistance.

0 An elastomer coupling was inserted between the drive

motor and the gearbox. The intention was to detune the
second torsional mode from resonance with rotational
speed. The elastomer also provides a small amount of
torsional damping.

o This remedy utilized off-the-shelf components. for which

the torsional stiffness was known (manufacturer's data).

o The plot of system natural frequencies vs. coupling

stiffness shown in Figure 13 was prepared to determine
the effectiveness of the proposed remedy. The specific
stiffness was selected from a range of available
coupling options.
Torsional Dynamics of Power Transmission Systems 373

o Other possible remedies were:

a) Replace gearbox rolling element bearings with fluid

film bearings. This would significantly increase
system torsional damping. Rejected because of (a)
possible center distance variation of gears, (b)
machinery costs, and (c) delay involved.

b) Fluid coupling. Rejected because of costs and

associated inconvenience (unnecessarily complex).

c) Torsional (Holset) damper. Rejected because

detuning is preferred to absorbtion where possible.

Effectiveness of Fix:

Trouble-free torsional performance was observed twelve months

after installation of the new coupling.

Note: System torsional measurements were made before and

after the fix was installed. A CEC torsiograph
(velocity transducer) was used to obtain waveforms
for frequency analysis. The rotor of this device
was conveniently attached to a stub shaft mounted
to the end of the motor shaft. The stator is free
in rolling element bearings. Signals are obtained
directly without slip rings or telemetry.

10.3.7 Case History No. 2

Branched Compressor Drive Vibrations

Problem Details:

The branched compressor drive train shown in Figure 14

consists of a hot gas expander turbine driving two axial flow
screw compressors. The drive input shaft carries through to
the LP compressor. The gearbox power take-off involves a
1 ine of five gears driving the HP compressor. The screw
compressor consists of two meshing helical groove rotors
inter-connected by timing gears.

The machine exhibited vibrations throughout its structure

from the time of startup. The problem became critical when
the HP compressor slipped its drive timing gears, allowing
the rotors to impact together torsionally during rotation.
This caused impact damage to the helical surfaces, and an
alarming noise. The timing gears were replaced and doweled
into place, and the machine operation was measured with an
IRD hand-held pickup. Unacceptable vibration levels remained
as follows:
374 N.F. Rieger

o Torsional vibrations of HP compressor rotor.

o Axial and lateral vibrations of the turbine.

o Many harmonics of rotational frequency present in the LP

compressor vibration signature, including half running
speed (0.5 x)

o Some indication of compressor surge was detected in the

HP output pressure trace.

o Vibration was also strong on the gearbox casing.

System Parameters:

Turbine: 4500 rpm, 3000 HP

1st Compressor: 4500 rpm, 1500 HP
2nd Compressor: 3000 rpm, 1500 HP
Gearbox: Bobbed double helical, ratio (a) 1:1.000,
(b) 1:0.57, 7.0 diametral pitch, plain
cylindrical sleeve bearings, 10,400 rpm
at 4500 rpm surface speed

Inspection Details:

Several possible sources of vibration existed:

o Unbalance vibrations of the turbine (frequency 75Hz).

o Unstable turbine bearing operation (frequency 38 Hz or

bending critical frequency, depending on nature of
system, rigid, or flexible rotor).

0 Torsional vibrations of system. Several possible

excitation causes exist.

o Coupling misalignment or foundation settlement.

o Observed gearbox casing frequency of 0.47 times turbine

speed, i.e., 35.3 Hz or 2120 rpm.

o Some 2x vibration evident on bearings of both shafts.

o Thermal distortion of the long gearbox.

Possible Causes:

o Unbalance vibrations of the turbine could cause lateral

vibrations of the turbine bearings and casing. Strong
vibrations could occur if the turbine was running close
to its critical speed.
Torsional Dynamics of Power Transmission Systems 375

o Unstable whirling of the turbine rotor in its bearings

could occur above twice the rotor critical speed. or
around Nu = SOO/ VC•
where c is the bearing radial
clearance.

o Torsional vibrations cou].d arise from drive gear run-

out. timing gear run-out. and from gas pressure
pulsations on the rotor helices.

o Thermal distortions of the casing could unload bearings

and make them unstable.

o Gas pulsations could cause the axial vibrations observed

at the turbine couplings.

Diagnostic Procedure:

The primary cause (or causes) of the timing gear slippage was
sought. Incidental possibilities were attended to as part of
the correction.

o Balance of the turbine rotor was checked. Bearing gap

accelerations were high but not excessive. Fix:
balance the rotor at the next maintenance shutdown.

o The turbine critical speed for the rotor in its bearings

was calculated to be 3100 rpm. Rotor critical speed
vibrations were rejected as the primary cause.

o Hot distortion of the gearbox was checked optically

(surveyors theodolite) and found to be small: good
foundations and boltins.

o Bearing instability was rejected as (a) the turbine

runs below 6200 rpm. and (b) the rigid rotor
instability speed is Nu = SOO/ Jc = SOO/ J0.002 = rpm.
0 Torsional natural frequencies for this system were
calculated to be 12.7. 17.7. 34.7. and 43.3. Hz.
Corresponding critical speeds are shown in the resonance
diagram. Figure 15.

o Gear run-out pin checks were available: see .Figures 16

and 17. All gears had substantial b. run-out. The
second idler gear had 3x run-out. This 3x corresponded
to the 0.47x turbine speed detected on the casing.

o The torsional forced response of this system was

calculated. Figure 18. Measured torsional amplitudes
agreed well with the predicted response.
376 N.F. Rieger

Primary Remedy:

A viscous shear damper was first proposed to suppress the

compressor peak amplitudes. but it was shown that this would
increase the non-resonant vibrations to unacceptable levels.

A quill shaft between the gearbox and the LP compressor was

then studied and chosen to detune the torsional vibrations.
The quill shaft detuned the compressor from the gearbox
excitation without increasing the non-resonant vibrations.

Secondary Remedies:

Several additional steps were taken to reduce the overall

machine vibrations. as follows:

o The turbine vibration problem was eliminated by re-

aligning the turbine casing and re-balancing the turbine
rotor with the coupling sleeve attached.

o All gears were machined to AGMA 9 standards and balanced

to reduce the torsional excitation.

10.3.8 References

1) Nestorides. E.. Torsional Vibration Anahsh. British

I~ternal Combustion Engine Research Institute Handbook.
Cambridge University Press. 1958.

2) Ker Wilson. W. J •• A Handbook of Torsional Vibration

Analysis. Volumes 1 through s. John Wiley and Sons. New
York. 1958. Second Edition.
Torsional Dynamics of Power Transmission Systems 377

(a) Simple torsional oscillator

(b) Two inertia semi-definite system

Figure 1 Simple Torsional Systems

378 N.F. Rieger

611 DIA 611 DIA

Figure 2 Dimensions of Turbocharger Rotor

~----------------------~1 (a)
91
Mode 1 2
w1 = 0

Mode 2 w22 = 1641.87

Figure 3 First Two Mode Shapes and Corresponding Natural

Frequencies for Turbocharger
Torsional Dynamics of Power Transmission Systems 379

Bz

(a) Simple torsional system

(b) Gear error geometry

Figure 4 Simple Geared Torsional System Showing

Geometry of Gear Runout Error
380 N.F. Rieger

/1 f - - - - 1-

lU Shaft 2

Il
I
Shaft 1
'
I

~System ~latural
• Frequency f 1·

n
l/
y
........
s aft 1 ........._

I
~ I

- -.,I
f Shaf~ 2 I
1--- I
I
.I
I
_Speed fi RPi:

Figure 5 Resonance Diagram and Corresponding

Mode for Geared Line Drive
Torsional Dynamics of Power Transmission Systems 381

(a} Simple torsional (b) Two Inertia Torsional

Oscillator Oscillator

Natural frequencies w 1 • w 2

Torsion" I
Amolit!Jde

Figure 6 Torsional Response Attenuation at Frequency

w0 by the Addition of a Second Inertia
382 N.F. Rieger

ew
(a) Undamped single degree of freedom torsional system

r, Iz

ql

e,
D-{J 62

(b) Damped untuned torsional system

(c) Response of single degree

System to T
0

(d) Effect of damping on response

Figure 7
Torsional Dynamics of Power Transmission Systems 383

screw

'>?~9='9='91:9!~~~7'""'~~i--N o t u ra I
rubbczr
joint ring

Plug

Natural
rubber
0~~~~~E;;;~lj oint ring

screw

Figure 8 Viscous Shear Damper Details

384 N.F. Rieger

c
(a) Simpl e torsi onal system with tuned damper

of damping
(b) Respo nse of simpl e system for vario us amounts
Figur e 9
Torsional Dynamics of Power Transmission Systems 385

v "'-../
1.0 ~
"" ~ t--

(c) Damper respo nse with both peaks tuned equal ly

V- r--

y I'-. JI -
.L.O

{d) Damper respo nse with one peak tuned to

low respo nse at selec ted frequ ency wa

Figur e 9 (cont ')

 ..,..
00
0\
COMPR(SSOR COIAPHf SSOR f()l.',l'lll !1!i01~

~--

11}11117111 1 11111}11 I ) ) ; I 1)111 I 7 I>> I I JI 7 I I}, I Ill I I I I I.,IT7"T"T., 7 ,-rTT7"7-, 7 7 • r r

Figure 10 Methanol Compressor Drive Train Schematic

MOTOR GEARBOX COMPRESSOR COMPRESSOR COMPRESSOR

SHAFT SHAFT SHAFT SHAFT

z
Figure 11 Block Diagram of Torsional System ~
:::0
~·
...
Torsional Dynamics of Power Transmission Systems 381

(a) First torsional mode. £1 960 cpm

(b) Second torsional mode. £2 1728 cpm

(c) Third torsional mode. £3 = 3792 cpm

Figure 12 Torsional Mode Shapes and Natural

Frequencies for Compressor Train
'388 N.F. Rieger

ELASlOMER ORIGINAL
COUPLING STiffNESS COUPLING Sllff NESS

FRE DUENCY•
CPf.

o~------~~----~~-----L--~----
105 4· .10 5 106 4 • 106 K
COUPLING STifFNESS LB IN./RAO.

Figure 13 Variation of Natural Frequency

vs. Coupling Stiffness

Figure 14 Schematic of Compressor Drive Train

Torsional Dynamics of Power Transmission Systems 389

OPERATING SPEED I TURSII;E t.ND CQI,IPRESSOR

I
5000 .----. -----, .---.- ---,-- r--:r, ----,

~ 4000~----~----r-----t---~~-r--r---~
u
>-
COMPRESSOR
u
~
::I
3000~---4-----+----~----t--r--~~~
0
w
a:
~ 2000i======~======~=====-+-~....,-;--;---,
Q
~
2S IOOO,f===J.~=s;~~==F====9f==i===r====1
~

1000 2000 3000 4000 5000 6000

TURBIN£ SPEED RPM

Figure 15 Resonance Diagram, Branched Compressor Drive

[111101': 011[ rtP. II[YOLUfiO N (1111011:

-
FIIIS1 IDLU

ONl r[A II(VOLU110N

Sf COIID I DLU

UROII: lHRU rU II[VOLIITIO K [RROII: ON( Plll II[V0LU110K

Figure 16 Gear Profile s, Branched Compressor Drive

 \N
"'D
0

SUN GEAR INSPECTION TRACE

FOURIER AMPLITUDE
-q -q
First 2. 35 x 10 in. Second 0.39x10 in.
-q -q
Third 0.27x10 in. Fourth ' 0. 03 x 10 in.

57 T

Figure 17 z
;,
:::0
;·
~
..,
Torsional Dynamics of Power Transmission Systems 391

10
1-
1- I I
I
1-
1-
r-----+-~--;------+--i
~l I
I
II'
' ~J,,~~
1-

1- I
lsi CRITIC t\L
I
I 2nd CRITI AL
b
I1
,!
J
~ I
j
z
<t
0 1.0 ~
I
<t
r-

\
a:

\
~
....J
....J r-
:E

\
-
I.U
0
:::;)
1-
\I
1\
....J
a..
:::!:

\
1\

\
<t

~
\
g
t-
0.1
~TORSIONAL AMPLITUDE
I \

1- RESPONSE AT FIRST
r- COMPRESSOR.
rEXCITATION: 2-PER-REV
r- SECOND COMPRESSOR \
GEARS.
r- ORIGJt:AL MACHINE WITH-
OUT QUILl SHAFT.

I
'
L
Turbine ~~peed, rpm J

3000 4f00 jSOOO

I
O.Ql 10 20 30 40 50 60 70
· CPS
EXCITATION FREQUENCY

Figure 18 Amplitude Response vs. Frequency and Speed

392 N.F. Rieger

Table 1 Typical Causes of Torsional Vibration

CAUSE

1. SIMPLE GEAR RUNOUT COI'IPRUSOR DRIVE REPLACE GEAR AND BALANCE.

HI.AMCNIC AVOID RESONANCE.
ELECTRO."-'GNETIC TuRIINE·GENERATOR SET EALANCE ELECTRICAL LOADS.
DE-TUNE SHAFT.
HYDRODYNAIIIC
INTERACTIONS
l'lARINE PROPELLER
VENTILATING FAN } DE·TUIIE SHAFT WITH FLEXIBLE
COUPLING. ADD VISCOUS DAIIPER.

2. COI'IPLEX CYLINDER GAS RECIPROCATING ENGINE

HUIUINIC FORCES REC IPROCA Tl NG CO,.,IIUSOR
GEAR INACCURACIES PRINTING ROLLERS STIFFEN SHAFTS. DoUBLE GEAI!IHIS.
HELICOPTER GEARBOX IsOLATE GEARIOX. fLUID•FILN
IEAIIINGS. DE·TUNE SHAFTS.
TuRIOCOIIPRESSOII. WIND TUNNEL DRIVE DE·TUNE SHAFTS WITH VISCOUS
ILADE DISCHARGE COUPLING,

3. TRANSIENT ELECTRICAL STARTER PAP(II !lACHINE 51100TH STARTER, STRENGTHEN

SHOCK ADJUSTI1ENT SHAFT. fLEX IILE INPUT
COUPLING,

START OF LOAD ELOOIIING HILL fLUID•FILN IEARINGS. ELIMINATE

CYCLE PuNCH PilUS IACKLASH. fLEXIILE COUPLING
DE·TUNE.

(OUPLIIIG, GEAR ROLLING HILL ELIMINATE IACKLASH. fLEXIILE

I

IACICLASH COUPLING DE•TUNE.

INDEXII:S NOTION AuTOMTIC COINING PRESS TUNED VI SCOUS DAMPER ,

II.MINATES'
Torsional Dynamics of Power Transmission Systems 393

Table 2 Analogy Between Rectil inear and Torsio nal Vibrat ions

Rectilinear vibration Tonio~~al vibratiOil

-
Symbol Unit SJD1bol Unit

Time t see t
rae!
;'
in
Displacement "' rad/sec
Velocity
"'· in/sec
in/see' ; rad/seel
Acceleration "' K in-lb/rad
Spring constant k lb/in
in-lb-see/rad
Damping coefticient e
r
lb-see/iD
dimensionleas
'r dimensionless
Damping factor
Maas ... lb-see'/in I lb-in-sec2/rad

Force or torque F = ms lb T = Ji' in-lb

Momentum mi lb-aee Ji in-lb-see

Ft lb-sec Tt in-lb-aee
Impulse
Kinetic energy !mi' Ib-in Fiz lb-in
}k:z:l Ib-in iKt2 !b-in
Potential energy
fFd.:z: !b-in f Td1 !b-in
Work

Natural frequency ... = .,fiJm rad/aee ... = ..fKi1 rad/aee

Equation of motion mz+ci+ k:t: = F 0 sinwt Ji + ,; + K1 = To sinwt

Initial conditions :t:(O) = "'•· z(O) = io 1(0) = '•· i(O) =io
= Ao-C...• sin (wdt +II>) = Ao-c...• ain (wdt + f>)
Transient responses "'• '•
... = yl- r2 w. ... = -/1-rs ...
'• = +ain(wt-11-)
"• = Xain(wl-11
Steady atate -)
responses To
X=
Fo
y(k- ,...Z)Z + (cw)l • = y(K- lwl)l + (..,)•
394 N.F. Rieger

Table 3 Torsional Natural Frequency

Equations for Simple Systems
CASE FREQUENCY EQUATION

1
~ Wn• ff
~ ll,)~ + r t.)~ :!:1 (w,.._I 1.)~)1__ +~u~ J
~
c..> D [ I
2

1 2
w~ :a k.fr ; w:- • KLJr,_ i r· lr.•I,Vr,

3
H 11 12
w,•O; Wz =u., = k(J, •Iz) 1 I,Iz.
1

4
~r:f:J 2
w,•O i (..)~ • 1\l k,k,~ (I,- n" I1)
I r.rLCK,- n•J<.&)

5
~~ I, w4 [ I,I 4 (Iz +n..I3 ~- ua.~f,Cla. +Iln.. •I.n1.)
11
!!~f.! • k',f4 (!,•I1 • n'f.,l + t,k'.. (I,•I.~ •
n•IJ •I.. n•) =-0
iii 13 1!5
11 12 13

6
~ w"(T,T2!3 )- tu'l[K,rl (rz •I 5)+K,!J(I,•I~,U
+ k,l<z. (I, +I1 •I~) =0

k,-I,w'l.

~
- 1(, 0 0
-k, l(, .. k,-r, ...... -I< ... 0
7
0 -k.. ka +k:s- rlldz =0
11 I,
-1<?»
12 13
0 0 -k3 K3 -I1 w"
 '(}
Table 3 Torsional Natural Frequency (l
(cont') Equations for Simple Systems
FREQUENCY EQUATION
[
CASE
~
~
K,-X.w" -k, 0 0 s.
~r n
en
-'<. IC,.,.:~•n:t, ~ea.n: ~n,·
8 0 n: lea. 0 =0 0
"'n:-r....• .....
0 n;l<. 0
04'
~
J,
Pr.--0 r. "·~..-lj ...~

I~ r, i
T, ~ t.·I,w' -Jr. 0 0 0 ~
9 -~ lr,•~li:a•Alr. ~k& 0 "1kt 8
0 <ke. ~(ka+(s•lJ) •ltl\~ 0 :0
0 0 - t, ~ n:cr,-r,..•) 0 m:
0
[ft~ ::s
0 n4tf 0 0 ~(11-I.c.') en
r .. '<
en
g
k,-r,w• "*• 0 0 0 ~ ~
10 [}1}-IF{]r, •ta IC,+k&•Jiw' •b 0 0
0 ·ta n;t, ~te. +ta Kanl t.tlt 0
n.~r., 0 0 t<,nJ n;kJ-q;I3 w1 0
r, I.z . 0 0 "'~· 0 "'~.Z.,.,'c.•
J;~ r.
r~ ~ l<z
lei·liw" -lc· 0 0 0
11 I.e~ ilu~O -k· lee•ta.• ts·l'all - ka 0 -IC's
I<~ Ks 0 -t. ka•tcs·.r,..• -t~ 0 =0
0 0 -k• ta•a:..-:r....• -1:"4
1<4 IS~ 0 -ks 0 •k1 t'4 •t',-JS..:.~
Ei' Is
t!, ~
\0
1.11
 CHAPTER 11.1

FREE AND FORCED VIBRATIONS OF TURBINE BLADES

H.lrretier
University of Kusel, Kusel, FRG

ABSTRACT

The free and forced vibrations of turbine blades are

considered. First, various models for the descri~tion
and the calculation of their dynamical behaviour are
presented and methods of solution are disc~ssed. Second~
results and parametric studies are considered which show
the influence of the most important blade parameters on
the vibrations.

1. INTRODUCTION
One of the important components of steam and qas turbines are the
blades. They re attached to the rotor surface to conv~rL
the fluid energy of the gas or the steam into rotational energy
of the rotor. A typical bladed rotor is shown in Fig. 1
In the following paper the vibrational behaviour of turbine
blades is considered whereas it is assumed that the blade
vibration is uncoupled from the disc and the shaft. This assump-
tion is valid in the most real cases. Otherwise, the coupled
vibration of the blade and the disc or the complete system of the
turbine rotor must be considered.
398 H. Irretier

fig. 1: Rotor of the 1200-MW-Turbine (KKU Biblis, FRG)

2. SOURCE OF BLADE VIBRATION PROBLEMS
The main source of turbine blade vibration problems ar1s1ng in
turbomachines is the fact that the distribution of the gas or
steam load on the blade is not constant with respect to the rota-
tional angle but varies during one rotation of the blade. This
angular dependence of the blade load is caused by various circum-
stances. The main reason for that is the fact that the rotor
blades run behind a row of still standing stator blades which are
fixed in the casing of the machine (Fig. 2). Thus, the rotor
blade is loaded in tangen-
tial and circumferential
o• 360" direction by fluid forces
L _ .0. .0. .O__j
~ ~--------~~-- ..i._ stator blades which are lower than the
1!:s - rotor blade
average value in the wakes
P rotation of the stator blades and
higher than the average
Fig. 2: Nozzle excitation of a value in the nozzles bet-
rotor blade ween the stator blades.
Another reason for a non-
uniform blade load during one revolution can occur by partial
admission e.g. the loading of the rotor blades by fluid forces in
one or more sectors around the machine while the other sectors
remain unloaded. In both cases, the load on the rotating blade is
periodic and can be described by the Fourier-series
Free and Forced Vibrations 399

p = 'f P cos (n t - 1/J)

p(t) = (2 • 1>
v=O v v
where t is the time, P are the Fourier<oefficients, which can
vary along the blade a~is, 1/J is a phase angle and n are the
excitation frequencies of th~ periodic blade force. Yhese exci-
tation frequencies are given by
n = vzn
\)
v = 1 , 2, 3 • • • ( 2. 2)

where z is the number of periodics of the blade load during one

revolution (for the nozzle excitation for instance z equals the
number of the stator blades) and n is the angular speed of the
machine. The lines of excitation frequencies corresponding to
equation (2.2) are plotted in the Campbell diagram which is shown
in Fig. 3.
On the other hand, each rotor blade
possesses a spectrum of eigenfrequen-
cies which generally depends on the
angular speed because of a stiffening
effect due to the centrifugal force
field. Thus, the eigenfrequency-angu-
lar speed curves are parabolic func-
tions as they are shown in Fig. 3, too.
At each point of intersection of one
eigenfrequency curve with one of the
excitation lines in the Campbell-dia-
'Fig. 3: Campbell-diagram gram resonance vibrations occur. These
resonances can only be avoided if the
for a rotating blade· corresponding Fourier-coefficient P
is very small and/or the damping is
high enough. Both requirements are mostly not given in real turbo-
machines. Thus, the only way to avoid strong resonance vibration
is the possibility to keep the eigenfrequencies of the blade
sufficient far away from the excitation frequencies. However, in
many turbomachines it is not possible to avoid resonance because
the blades run with varying operational speeds and various reso-
nance vibrations occur during the run-up or the shut-down of the
machine. The described circumstances show that the design of
turbomachine blading from the vibration point of view must in-
clude three different considerations:
(1) The determination of the amplitudes and the frequencies
of the aerodynamic forces acting on rotor blades,
(2) the determination of damping values and
(3) the investigation of the eigenfrequencies and the mode
shapes of the rotor blades.
While the first two points are discussed in detail in other chap-
ters this paper deals predominantely with the third point and
additionally with the calculation of the forced response of tur-
bine blades due to excitations as described above.
400 H. Irretier

3. MECHANICAL MODELLING AND MATHEMATICAL DESCRIPTION OF TURBINE

BLADES
The first problem to solve in the determination of the eigenfre-
quencies and the mode shapes of turbine blades is to find a suf-
ficiently accurate mechanical model. This model depends on the
shape and the size of the considered blade. In Fig. 4 blades are
shown which are marked by a typical beam-like shape. The lower
natural vibration of such
blades may be described by
more or less extended beam
theories. In contrast to the
blades shown in Fig. 4, other
turbine blades may have typi-
cal shell-like shapes or can
only be described by three-
dimensional models. Such blade
types often occur in modern
gas turbines with very high
inlet temperature so that the
blades must be cooled. For
this purpose the blades are
designed as composed shell
structures or, as shown in
Fig. 5, in a monoblock form
which is penetrated by cooling
drills.
The examples of turbine blades
shown in the Fig. 4 and 5 indi-
Fig. 4: Beam-like turbine blades/1/ cate that the right mechanical
model to describe their vibra-
tional behaviour may be quite
different and depends substan-
tially on the shape of the con-
sidered blade. Consequently,
the mathematical description
is also dependent on the con-
sidered type of blade.
In the following chapters
various kinds of mechanical
models of blades are discussed
and the corresponding mathe-
matical description is given.
While the beam-like blades are
considered in detail, the
shell-like and three-dimen-
sional blades can only be dis-
cussed shortly. This is also
valid for extended blade row
Fig. 5: Cooled, three-dimensional constructions e.g. shrouded
blade of a gas turbine blade stages or blade packets
etc.
Free and Forced Vibrations 401

Fig. 6: Beam-like blade; notations

The blade can perform bending vibrations in the plane y, z and
torsional vibrations around the x-axis. In the case, that the
x-axis of the centre of gravity of cross-section is equal to the
centre of shear or if the distance between both is neglectable
- which in the most practical cases holds true - the bending
vibration and the torsional vibration are uncoupled from each
other. Otherwise, the bendina and the torsional vibration are
coupled. -
Depending on the fact which constructive properties of the blade
(tapering, twisting, etc.) are included in the mathematical de-
scription of the mechanical model shown in Fig. 6, various ty~es
of equations have been developed in the past to calculate the
natural frequencies, the mode shapes and the forced response of
beam-like turbine blades considered in this chapter. A complete
survey on that is given in /2/ and /3/.
3.1.1 SENDING VIBRATIONS
Depending on the blade pro~erties included in the model, various
mathematical descriptions of the bending vibrations have been
developed and applied in the 9ast. The simplest model which neg-
lects the rotation, the twisting, the shear deflection and the
rotary inertia are of course the differential equations
2 2 2
_ a (EI a v) = A a v
a"1 za7 P atZ (3.1a)

a2 a2w a2w
- ~ (Ely ~) pA (3.1b)
dX~ axL ~
of the classical Bernoulli-Euler beam theory. However, these
402 H. Irretier

equations do not take into account the important effects of rota-

tion, twisting, rotary inertia and shear deformation. If in addi-
tion these influences are also included in the model, the com-
plete set of equations of motion is /4/

a a a * a 2 a2
ax [ KGA(a~ + Yz)] + ax(nx a~)+pAA (vcosa-wsina)cosa= ~
at
2 2
a a ayz ay a Yz ay
KGA(~xv + Yz) a~x(Eiz ~X + EI yz _l)=-p(I z~- Iyz~)
a a ax at~ at~

a! [ KGA(~~ + yy)] + a!(n: ~~)-pAA2 (vcosa-wsina)sina=pAaat2 2

aw a aYz ~a a2Yz a2Yy
KGA(ax + 'Yy) ax(EiyzTx + Ely ax) =-p( -Iyz7 + IY7)'
(3.2a) - (3.2d)
where
*
nx = pn
2 L
f A(R+~)d~ (3.3)
X

is the static normal force in the blade caused by rotation, v and

w are the blade displacements, Yv and Yz are the angles of cross-
section in the sense of Timoshen~o•s beam theory and the other
parameters are as shown in Fig. 6.
The set of equations (3.2a) to (3.2dl for the twisted, staggered
and rotating blade including the shear deformation and rotary
inertia effect is the mostly developed mathematical model to de-
scribe the bending vibrations of turbine blades when there is no
coupling to the torsional vibrations. This set of equations was
first given in /4/ and proved to be very reliable in comparison
to test results.
Introducing the displacement vector
T
~=[vyzWYy] (3.4)
the equations of motion (3.2) for the vibrating blade can be
written as
(3.5)
where Mis the mass matrix including the transverse and rotary
inertia terms and K is a matrix of differential operators with
respect to x. This-stiffness matrix K can be divided into two
terms
Free and Forced Vibrations 403

2
! = ~I + n .!$a (3.6)

where K describes the stiffness properties of the bent, non-

rotati~Iblade whereas Kn includes the stiffening effect of the
centrifugal force fiel~ue to rotation.
Equation (3.5), which is the differential equation of the contin-
uous, non-discretized turbine blade, describes the natural vibra-
tions of the blade. Its l-th natural vibration is given by
(3.7_)
where ~ =~(x) denotes the mode shape and wt the corresponding
circular frequency. Both are solutions of the eigenvalue problem
-w/ !i ~ + .!$_ [~] = .Q. (3.8)
which follows from equation (3.5) where it must be payed atten-
tion to the fact that the stiffness matrix K is a function of the
angular valocity n of the blade. Thus, the eigenfrequencies w
and, on principle, even the mode shapes ~ are dependent on t~e
rotational speed of the machine.
3.1.2 FREE TORSIONAL VIBRATIONS
If the assumption that the centroid and the centre of shear of
the blade cross-section are equal to each other is still valid,
the torsional vibrations are uncoupled from the bending vibra-
tion considered above. The simplest model to describe the tor-
sional vibration of the blade is the well-known equation
a <Kp a 2
ax (GIT ax) = piP~
at
(3.9)
which is a wave equation of a drilled shaft. IT is the torsional
rigidity of the cross-section while I denotes its polar moment
of inertia. Of course, this equation Reglects all refined effects
on the problem like the increase of the rigidity of the blade due
to the centrifugal force field, additional inertia forces because
of the formulation of Newton's law in the rotating reference sys-
tem and the twisting effect. If these influences are taken into
consideration, too, the equation of motion of the torsional blade
vibrations becomes :

a! (GIT+a: I p+E (*)\)~)-psi (I z- IV )cos2 (<><-y) • =pip::~

* * (3.10)
where ax = nx/A is the static stress in the blade due to rotation

+
2
2 22 I
ly = ~ (n + r,: ) dA - (3 .11)

is an area moment which describes the increase of the torsional

404 H. Irretier

rigidity of the blade due to twisting /5/.

Introducing a displacement vector w = [~] the equation of motion
for the torsional vibration of the blade also can be written in
the form of equation (3.5) and for harmonic natural vibrations
the angle ~ of torsion is of the form which is given in equation
(3.7) which again yields an eigenvalue problem of the for~ shown
in equation (3.8). It is obvious that the eigenfrequencies wl and
the mode shapes Wf of the free torsional blade vibration are also
dependent on the~rotational speed n of the system.
3.1.3 FREE COUPLED BENDING AND TORSIONAL VIBRATIONS
In the two preceding chapters it was assumed that the centroid
and the centre of shear of the cross-section are equal to each
other or that the distance between both is small in comparison to
the other sizes of the cross-section.
In this case the both bending vibrations and the torsional vibra-
tions are decoupled and the eigenfrequencies and the mode shapes
of one of the natural vibrations can be calculated independent
from the other. However, in some types of turbomachines blades
are used which have a large curvature so that this assumption is
no more valid. Now, the bending and the torsional vibration are
coupled and influence each other. The equations of motion must be
extended to take this coupling effect into consideration.
There are a lot of models developed in the past to include this
coupling effect. Probably the first set of equations including
this effect was applied in /6/ and reads
- a2 2 a2
- -2 ( EI a v) = pA ~
z~
(v + cz~) (3.12)
ax at
a2 2 a2
- -::---2" ( EI a w) = pA ~ (w - c ~) (3.13)
ax Y a7 at Y
a2 czv - cyw
= Ip ~ (~ + .2 (3.14)
0 lp

where cv and cz are the coordinates of the centre of shear of the

cross-section and ip is its radius of gyration (Ip = i~A). Of
course, these equations neglect the effects of shear deformatio~
rotary inertia~· rotation of the blade and twisting. Refined
models to include also these effects where given in several
references like /2/, /3/, /5/, /7/, /8/ and /9/.
3.1.4 FORCED VIBRATIONS
Following the considerations in the preceding chapters we recog-
nize that the equations of motion of the free decoupled and even
coupled bending and torsional vibrations of beamlike blades can
be described by the standardized equation (3.5). If now, in
addition. also the excitation of the turbine biades as described
Free and Forced Vibrations 405

in chapter 2 should be taken into account to calculate the stress

response of blades, external transverse loads acting in the y-
and z-direction must be added up in the equations for the bending
vibrations and an external torsional moment along the x-axis for
the torsional vibrations. This leads to the general equation of
motion
(3.15)
in which in addition a damping term is included which is assumed
to be proportional to the velocity.
The general method to solve this linear differential equation
is the modal transformation technique /10/, /11/. This technique
transforms the displacement vector w in the space of mode shapes
by the relation -
m
w = L ~1 1 , (x) qk(t) (3.16)
- k=1 --1\

where qk are the time-dependent modal coordinates of the problem.

Introducing this modal series and paying attention to the orthog-
onality relations of the mode shapes, the general equation of
motion (3.15) is transformed into
(3.17)

c.tk = lL !'!.e.T f [ ~] dx (3.18)

and

r.e. = lL !!.e.T E. dx (3.19)

describe the damping coefficient and the transformed blade load,

respectively. From equation (3.18) it is obvious that generally
the damping couples the differential equations (3.17) for the
modal coordinates q1 if no special assumptions are introduced at
this point of consi~eration. The basic idea of these assumptions
is that in the most practical cases the damping effect is so
small that there is no coupling between the various types of
natural vibrations of the system. Under these circumstances it is
possible to replace equation (3.17) by the uncoupled equation
(3.20)
where the introduced factor ~ is the viscous damping Factor. If
structural damping is used, where the damping matrix 1s given by

-·C = n-
.!l K, (3 .21)
whPrP
406 H. Irretier

n is the loss factor of material and n the frequency domain pa-

rameter, the transformed equations of motion are
- + w.e.2 ( 1 + n a )
q.e. n at q.e. = r.e. • (3.22)

To calculate the vibration response of the blade the next step is

to consider the solution of the non-homogeneous, linear differen-
tial equation (3.20) or (3.22). This solution depends on the form
of the right hand side r.e.· In a lot of practical cases - of
course there are also important exceptions of this - the external
blade load can be separated in a product form
£ = f(X) .f.{t) (3.23)
where P(x) is a diagonal matrix which includes the distributed
load on the blade and f(t) is a vector which contains the time-
dependent functions of excitations which come into the problem
as considered in chapter 2. Thus, the transformed blade loads as
pointed out by equation (3.19) are
(3.24)
where
l T
Bt = 6~ f dx (3.25)

We notice that the character of time-dependence of the blade load

is completely described by the vector f(t).
The most convenient way to solve the equations (3.20) and (3.22)
for various types of excitation (3.24) is the frequency domain
method. This method starts from the Fourier-spectrum
~.e.(n) = ~ j r.e.(t)
-oo
e-jnt dt (3.25)

of the time-dependent load function r • In a lot of practical

cases these spectra are given directly. In other cases, they
must be calculated from the time-dependent blade load using a
Fast-Fourier-algorithm. Transforming the equation of motion (3.2m
or (3.22) also into the frequency domain yields
[- n2 + j • 2c;,.e.w.e. n + w~] ij".e.(n) = ~.e.(n) (3.26)
and
[ -n2 + w~ ( 1 + jn) J 'O'.e.(n) = ~.e.(n) (3.27)
from which the vibratory response
(3.28)
for the .e.-th modal coordinate of the blade can be calculated
Free and Forced Vibrations 407

using the transfer functions

1
ll.e. = """w~..--_-n....2...-+-j-·-2-z;i-wi-n (3.29}
or

1li = -w~.....--(-1-+-jn-}---n.,..2 ' (3.30)

for viscous or structural damping, respectively. If necessary,

the modal coordinates q in the time domain follow from the back-
ward Fourier-transformation
00 •

q.e.(t} = f lf.e.(n} eJnt dn (3.31}

-oo

Hith Qr(n} or qi(t} the displacements of the blade follow from

equation (3.16} in the frequency or time domain. Using the con-
stitutive equation relating stresses to displacements yield the
blade stresses due to the forced vibrations, from which a life
fatigue analysis can be performed.
It is important to discuss one point at this stage of considera-
tion. It was pointed out in the chapters above that generally the
natural frequencies of turbine blades depend on the rotational
speed. Thus, if the rotational speed of the machine changes for
instance during a run-up or a run-down of the machine the natural
frequencies of the blade change, too. If forced vibrations are
considered during such periods of blade operation this changing
of the natural frequencies must be taken into account and the
application of equation (3.28} should be restricted only on
sufficiently short time periods whereas in each period the actual
eigenfrequency w must be considered to calculate the transfer
function R1 • An Ilternative possibility to overcome this difficul~
ty is to solve the equations of motion (3.20} or (3.22}, respec-
tively, not in the frequency domain as described above but in the
time domain by a direct numerical integration using a set of
Runge-Kutta-integration formulars. During this time integration
it is possible to take into account the actual eigenfrequency wr
for each time t because the dependence n(t} is a given parameter
of the problem.
3.2 SHELL-LIKE BLADES
A beam-like model for a turbine blade is only sufficient for the
consideration of the lower modes of blades which have cross-
sections much smaller than the blade length. Otherwise, extended
theories must be applied. For such types of blades, for which
still one dimension, the thickness h, is small in camparison to
the two other dimensions, shell theories have been successfully
applied to compute the free vibration behaviour. It would exceed
this chapter to discuss in detail all the approximations which
have been applied in the past. The models start from simple
cantilever cylindrical shells with constant thickness and end
408 H. Irretier

with complicated finite element models for rotating blades with

non-constant thickness and a twisted shape. The influence of pre-
stressing due to the rotation is more or less accurately included
in the applied models.
A complete survey on the subject is given in /13/. Details con-
sidering various shell theories are discussed in /14/ and in /15/
a comparison of bea~ and shell theories for the vibrations of
blades is given.
In comparison to the free vibration problem of shell-like blades
the response of such blades due to forced vibrations was consid-
ered much less in the literature. The convenient method to con-
sider this problem is also the technique of modal transformation
which was described in detail in the chapter 3.1.4 by using the
eigenfrequencies and the mode shapes from a free vibration analy-
sis.
3.3 THREE-DJt.1ENSIONAL BLADES
The only suitable method to deal with free and forced vibrations
of three-dimensional blades is the method of finite elements.
After intensive investigations of this method in the sixties and
seventies today the finite element approximation is a well-estab-
lished technique in the dynamic analysis in mechanical engineer-
ing sciences. Many efficient programs have been developed since
then and are available for the application of a vibration analy-
sis of three-dimensional blades. In the most cases shell or tetra-
hedron elements are used to calculate the eigenfrequencies and
mode shapes of such complicated blades. Some programs also allow
to take into account the important effect of the centrifugal
force field. By the steps of a finite element procedure, the
problem is formulated in a form given in equation (3.5) and the
eigenfrequencies are calculated from equation (3.8). To determine
the eigenfrequencies and mode shapes from this equation a lot of
efficient methods of eigenvalue calculation have been developed
parallel to the finite element method. Details on both are given
in /16/, /17/ and /18/.
In addition to the free vibration analysis, programs also have
been used to perform a forced vibration analysis. The most ade-
quate method is again the modal analysis technique which is de-
scribed in chapter 3.1.4 for the beam-like blade. Applications of
this method for the analysis of three-dimensional blades are
given in /19/ and /20/.
4. ~1ETHODS OF ANALYTICAL AND NUMERICAL SOLUTIONS
The various types of mechanical and mathematical models require
different techniques of solution. Closed form solutions are only
known f~r the very simple case of the free bending and free tor-
sional vibrations of beam-like blades without the consideration
of twisting, tapering and rotation i.e. for the equations (3.1a)
and (3.1b) for the bending and (3.9) for the torsional vibrations,
respectively.
Free and Forced Vibrations 409

If the refined effects for the beam-like blade models are taken
into account, only approximate solutions are available. The
classical procedures are the Rayleigh-Ritz and Galerkin-method.
They are described in /4/ and /15/. In the last twenty years, how-
ever, computer-aided numerical methods are preferred to calculate
the natural frequency and mode shapes of beam-like blades. One
classical procedure is the method of transfer matrices while more
modern techniques are the method of finite elements and a direct
numerical integration of the differential equations by a Runge-
Kutta-method and a shooting technique. A large amount of refer-
ences to the first method is given in /3/ while the second
technique was used in /5/ and /21/.
The suitable method for the calculation of the free vibration of
shell-like and three-dimensional blades is today the method of
finite elements which was described in the chapters 3.2 and 3.3.
For all three types of blades the technique of modal transforma-
tion is the most applied method to calculate forced response of
blades due to the various types of excitation. Details on this
have been described jn the chapters 3.1.4, 3.2 and 3.3. For
details of the application and numerical problems is referred to
the numerous literature in this field.
5. RESULTS AND PARAt·1ETRIC STUDIES
General statements concerning the dependence of the vibrational
behaviour of turbine blades on the constructive parameters are
only possible for the eigenfrequencies of beam-like blades. For
their forced vibrational behaviour and in particular for shell-
like and three-dimensional blades the connections are so compli-
cated that only some typical results can be discussed here.
5.1 BEAM-LIKE BLADES
Besides the mode shapes of the bending and torsional vibrations
of blades, the influence of the constructive parameters on their
eigenfrequencies is considered in particular. Moreover, two re-
cent results concerning the forced vibration response of the beam-
like blades subjected to nozzle excitation and partial admission,
respectively, are presented.
5.1.1 FREE BENDING VIBRATIONS
5.1.1.1 MODE SHAPES
As pointed out in the chapter 3.1.1 the most simple model for the
free bending vibrations of a turbine blade with constant cross-
section, without twisting and rotation and neglecting the shear
influence and the rotary inertia effect are the differential
equations (3.1a) and (3.1b) for an Euler-Bernoulli-beam. The
corresponding mode shapes are those of a cantilever and consist of
harmonic and hyperbolic functions /10/.
If the refined effects are included in the consideration, the
blade mode shapes change more or less considerably. The influence
410 H. Irretier

of the rotational speed and therefore of the centrifugal force

field on the mode shapes is small in a wide range of operational
speeds. Even the influence of shear deformation and rotary inertia
affect the mode shapes only a little except the higher vibration
modes where these influences dominate. In contrast to that, the
tapering of the bTade and in particular the twisting have a large
influence on the blade mode shapes.

x, ,' x2 2.

Fig. 7: Mode shapes of the

bending vibrations of a
tapered and twisted steam
tur~ine blade /22/ IXz . 3.
From the mode shapes of a tapered and twisted turbine blade shown
in Fig. 7 these effects are visible. Yhile the first mode is
similar to that one of a cantilevered Euler-Bernoulli-beam the
mode shapes two and three are characterized by complicated spatial
_curved lines. This result has ~articular consequences when the
forced vibrations are considered because an excitation in the
direction x2 can cause strong vibrational response in the perpen-
dicular direction x3 and vice versa. This phenomenon must be taken
into account for refined constructions of turbine blades with
damping wires or shrouding.
5.1.1.2 EIGENFREQUENCIES
In contrast to the mode shapes of beam-like blades, the eigen-
frequencies are considerably influenced by a stiffening effect of
the blade rotation. This is shown in Fig. 8 where the first three
eigenfrequencies of a slender rotating beam are plotted as a
function of the angular velocity n. Parameters on the curves are
the stagger angle a ~nd the ratio £ of disc radius to blade
length. We notice from the curves that there is a parabolic in-
crease of the eigenfrequencies versus the angular velocity. This
effect is well-known from Southwell's theorem which yields for
the rotating beam
2 02 2
wl. = wl. + Bi2 (5.1)
where gris the l.-th eigenfrequency for the non-rotating case and
B£is a centrifugal force factor which considers the stiffening
eTfect of rotation. As investigated in /4/ this factor can be
approximated for l. = 1 by
(5.2)
Free and Forced Vibrations 411

75.0

).I
l.,
• -=~:?
50.0 tn E•.B,
L l.•4;:-Vf
l?n Vf'
~
.!:-.so n•• "T,, T

-
1,0

r1' .J:•1
25.11- l.z I
(~-·-

l.,
~·o"
•lief
,:o2
,., -
.o
.o 1,0 2.0 3.0 4.0 5.0

Fig. 8: Eigenfrequencies of the bending vibrations of a slender

rotating beam as function of angular velocity
for the first eigenfrequency of bending vibrations of slender un-
twisted and even twisted beams. For the higher eigenfrequencies
the relationship (5.1T can only be used in a qualified sence. For
this case, useful approximated values for Blare given in /23/.
Transversed to real turbine blades with beam-like shape the re-
sults of Fig. 8 indicate that there is a large influence of the
rotation in particular for long and slender blades which are
fixed on a disc with large radius. For short blades on a small
disc this influence decreases.
The influence of taper on the eigenfrequencies of an Euler-
Bernoulli-beam-like blade on the eigenfrequencies is considered
in Fig. 9.
). 1.3-r-------------------.
If.; L~
1.2 rL.:JlPhg
----·~
---------
1.1
l., .---
1.0 .......-

.9

.8

Fig. 9: Eigenfrequencies of tapered beam-like blades as a

function of tapering parameter
412 H. Irretier

The curves show that the first eigenfrequency increases with

blade taper while the second and third eigenfrequency decrease.
The magnitude of the increase or decrease of the eigenfrequencies
can be considerable in practice because blades with a decrease of
cross-section about 50 or more percent are used in turbomachinery.
If in addition to the effect of taper the rotational influence on
the eigenfrequencies is considered it can be shown that the eigen-
frequency of the tapered beam increases less than for the un-
tapered one which has the same uniform cross-section as the
tapered beam at its root.
The influence of shear deformation and rotary inertia on the
eigenfrequencies of the bending vibration of beams had been fre-
quently studied in the past initiated by the classical work in/24~
Introducing this effect in the consideration of beam-like blades

----0--
fixed at the root and free at the end, numerical results show
that this effect can be often neglected for the first eigenfre-
quency but is of importance for the higher ones.
>.,
1.0 .----------====::;·====-=---,
>.~-
>.,~-.~ // ~ ·---
.~

/
0

.8

.7

.6

•5 .0 10.0 20.0 30.0 40.0 6y 50.0

-Fig. 10: Influence of shear deformation and rotary inertia on the
eigenfrequencies of bending vibration of beam-like blades
In Fig. 10 the eigenfrequencies of a cantilever blade with these
effects in account related to the eigenfrequencies neglecting
both effects are plotted as function of the ratio of the blade
length L to the radius of gyration i =~. We notice from the
curves that the influences of shear ~efo~ation and rotary inertia
are relatively small for L/iv-ratios greater than 30 but become
more and more important for shorter blades.
:A further effect which is very important for the eigenfrequencies
~~tra of beam-like turbine blades is the influence of twisting.
The curves in Fig. 11 show the change of the first three eigen-
frequencies of non-rotating blades without shear deformation,
rotary inertia and tapering as a function of the angle of twist
between the blade root and the blade· tip while a linear twisting
along the blade is assumed. The plot shows that the influence of
Free and Forced Vibrations 413

---X
twisting is completely neg-
lectable for the first
~X eigenfrequency while the
iyo =0.2 second one decreases and
'zo the third one increases
20r-------r-------+-----~ considerably. This result
is a consequence of the in-
creasing coupling between
the two bending vibrations
in y- and z-direction with
10 increasing twisting angle.

l
~-------~-------~-------~
01 I I I
Fig. 11: Influence of twisting on
the eigenfrequencies of the bending
vibrations of beam-like blades
5.1.2 FREE TORSIONAL VIBRATIONS
5.1.2.1 MODE SHAPES
The mode shapes of a cantilever with constant cross-section per-
forming free torsional vibrations are described by harmonic func-
tions which are the general solutions of equation (3.9). For
turbomachinery blades the torsional mode shapes are affected by
tapering, twisting and the other refined influences in a compli-
cated manner. A typical mode is shown by the holo~ram in Fig. 12
which 1s taken from
/25/. We notice that
the lines of equal
displacements are no
more parallel to the
edge of the blade.
Fig. 12: Mode shape of the first torsiona
vibration of a twisted turbine blade /25/ If, in addition, the
torsional vibrations
are coupled to the bending vibrations as described in chapter
3.1.3, the mode shapes are even more complicated and no general
description of the influence of the blade parameters on its mode
shapes is possible.
5.1.2.2 EIGENFREQUENCIES
As for the bending vibrations also the eigenfrequencies of the
torsional vibrations of turbine blades are considerably influ-
enced by the effects of twisting, tapering and rotation. As an
example, Fi~. 13 shows the influence of the rotation on the
fundamental eigenfrequency of the torsional vibration of a beam-
like blade. We notice again a parabolic increase of the eigen-
frequency. However, in comparison to the bending vibrations the
increase of the torsional vibration eigenfrequencies is much less
414 H. Irretier

2.0
A.,
tQ !yo =0,25

ltJ
1.9 lzo
E:ft
L
.!!2
lpa
=0.243
R L
1. 8

1. 7

1.&

1,5
.o .1 .2 .3 .4

·Fig. 13: First eigenfrequency of the torsional vibration of a

slender rotating beam as function of angular velocity
for real turbomachine blades. Therefore, this influence is often
neglected in practical calculations.
As to the influence of other properties of the blade on its
torsional frequencies spectrum is referred to the numerous
literature in particular to /2/, /3/, /4/ and /21/.
5.1.3 FORCED VIBRATIONS
In contrast to the free vibration analysis of turbine blades, the
forced vibration problem was considered much less in the litera-
ture. Some typical results of the calculations of the blade re-
sponse due to nozzle and partial admission excitation are con-
sidered in the following chapters.
5.1.3.1 PERIODIC EXCITATION AT CONSTANT ROTATIONAL SPEED
The usually considered problem of forced blade vibration 1s the
dynamical behaviour of the blade running with constant rotational
speed. The blade response is calculated as described in chapter
3.1.4 where. various models for the description of the bending and
torsional vibrations of the blade are used. In /27/ a numerical
model for a beam-like blade is applied based on a Timoshenko-
beam-theory where the coupling between the bending and torsional
vibrations is taken into account. For a typical steam turbine
blade the eigenfrequencies are calculated and plotted in a
Campbell-diagram which is shown in Fig. 14. For the first three
coupled modes various resonance points in the rotor speed range
between 400 and 2000 rpm are visible. The cross-sections at
,different distances from the blade root are also plotted in
Fig. 14.
For an excitation by z = 24 nozzles - the detailed data of the
forces are given in /27/ -the blade response is shown in Fig. 15.
It can be seen that the amp~itudes reach their peak values at the
points when the nozzle passing frequency lines intersects the
Free and Forced Vibrations 415

6000

sooo

400 100 12 00 1600 l 000

Rotor Speed, A . P.M.

Fig. 14: Turbine blade and Campbell-diaqram /27/

I O· Z

curves of natural fre-

10-l quencies. It can also be
observed from the curves
,
-· that the intensity of
,' \\-_ \\- v-W [mm]
< 10
resonance depends on the
JJ\~/
"'v
...
v
order of the nozzle
;; 10- s
- I
I ....
....... t'
II
passing frequency. He
--==~1!!l~L1------ .) ' ...
v
~,... ~J
~ observe that the re-
'

E' ,o-' nance with the first

•
J1\
I

coupled mode at 1400 rpm

l0- 7
yields a strong torsional
f.--_.......-/ -~~ad] - response while at the
10- ·
1100 1400 1500 1100 ~oo other resonances in the
Rotor Speed • A , P. M. considered speed range
Fig. 15: Stationary blade response due only bending responses
to nozzle excitation /27/ occur. This result is
caused by the special
geometry of the blade shown in Fig. 14.
The most important problem in the calculation of blade responses
due to forced vibrations is the determination of the exciting
forces and of the damping values. However, in recent times suffi-
ciently accurate data for these parameters are given in numerous
papers for a lot of practical cases.
5.1.3.2 TRANSIENT EXCITATION DURING RUN-UP AND RUN-DO~lN
As an example for a turbine blade, which is subjected to transient
excitation, a rotating beam is considered which runs up from one
rotational speed to another while it is loaded by partial ad-
mission. The considered system is shown in Fig. 16 from which two
90°-sectors of partial admission and the blade load in tangential
direction are visible. The blade runs up from a rotational fre-
416 H. Irretier

t:O x,
\ Po
\
0.9L-

blade force

Fig. 16: Rotating blade and load due to two arcs of partial
admission /26/
quency of f = 10 Hz up to a rotational frequency of 60 Hz as
shown in Ca~pbell-diagram in Fig. 17 where the resonance points
during this run-up are indicated for the first two eigenfrequen-
cies of the blade. The stress response of the blade at its root
f0 [Hz]
60 -----------=""r------

20 -------+ f [Hz] =2 t [s]

I

I
0 + 10

-10.0 -5.0
I I

~,2.82 9.41 1 -resonance points

I I i . z-118 1~
600.0 -:r-~.J..+t.....:........:~~-7----~iF-------:::::=o"'
6

f 1[Hz)

400.0
300.0

2oo.o l-A~~~--
2

1oo.o~~~;;;;~~:::;~:i::::~::::::~
.0
.0 10. 25.0 50.0 60. 75.0 fo[Hz) 100.0

Fig. 17.: Campbell-diagram of a simulated run-up of a turbine

blade loaded by partial admission /26/
Free and Forced Vibrations 417

~1 = .01 ~2 =.01
t, 2 t0 f2

6 f0
10 f 0
14 t 0

20. 150. 200. f.[Hz] 250.

Fig. 18: Stress response of a blade at its root during a run-up

loaded by partial admission /22/, /28/

plotted in Fig. 18 which shows for various time windows of the

run-up the blade response spectrum up to 250 Hz. \·le notice the
various harmonics of excitation which increase from the values
before the beginning of the run-un to the values of its end.
These harmonics cut at different times theparabolically-growing
eigenfrequency lines of the blade. Thus, at these points reso-
nance vibrations occur which are indicated by the significant
responses.
The results plotted in Fig. 18 are taken from the Turbine Blades
Simulation Program TUBSHI which is under development at the
University of Kassel /22/. For more detailed information, also
for the considered example of this chapter, the reader is
referred to /26/ and /28/.
5.2 SHELL-LIKE BLADES
As to the vibrations of shell-like blades only a few results can
be given here. As a typical example Fig. 19 shows the Campbell-
diagram and the mode shapes of a blade of the final stage of a
steam turbine. We notice the more or less complicated shape of
the mode of the blade especially the higher bending modes.
Strong distinction between bending and torsional modes is no more
possible because of the complicated shape of the blade. The eigen-
frequency curves show that those frequencies corresponding to
more or less typical bending modes increase much more with the
rotational speed than the torsional frequency does. This is also
418 H. Irretier

a typical result which can be

generalized for all beam- or
f (Hz) shell-like blades.
5.3 THREE-DIMENSIONAL BLADES
As for the shell-like blades in
the previous chapter also for
three-dimensional blades only a
100 ---- few typical results can be de-
scribed here. Fig. 20 shows a
Sn cylindrical hollow blade with
aerodynamic profile which ~1as
investigated numerically as well
as experimentally /30/. For the
calculations the method of
finite element was used. The
applied model is shown in
Fig. 20 and has about 600 de-
n grees of freedom. Some calcu-
lated and measured eigenfre-
0 1fXXl n [rpm) :mJ
quencies and mode shapes are
shown in Fig. 20, too. We notice
Fig. 19: Campbell-diagram and for them a good agreement bet-
mode shapes of a shell-like ween the numerical and experi-
blade of the final stage of mental results.
a steam turbine /29/ Finally, a complicated finite
element model using the NASTRAN

\
l~ ~
~ f' I:FEIOI4Hz EX 1054Hz

~
2; FE 1713 Hz EX 1717 Hz
~
\ ~

3:~E 1!115 Hz EX 1175 Hz

·Fig. 20: FE-model, eigenfrequencies and mode shapes of a

cylindrical hollow blade /30/
Free and Forced Vibrations 419

computer program of a twisted hollow blade is shown in Fig. 21.

The calculated eigenfrequencies as function of the rotational
speed are shown in the Campbell-diagram which is plotted in
Fig. 21, too. Again, the typical effect of the rotation on the

lOO

~so
fiHzl
200

--
240 3n
:!20

:oo
180

luO 2n

140

1~0

100

80 n

bO

20
Q
0 2000 lOOO n I rpml sooo

Fig. 21: FE-madel (NASTRAN) and Campbell-diagram of a twisted

hollow blade /31/
eigenfrequencies is obvious. From the results described in this
chapter we can conclude that today it is possible to give numer-
ical results for the natural vibrations of turbine blades even
for very complicated shapes and for refined influences for
instance like the effect of rotation. The most adequate method
to consider these problems is the finite element technique as
shown with the results in this chapter.
6. CONCLUSIONS
Free and forced vibrations of turbine blades had been considered.
Various types of models were introduced for beam-like, shell-like
and three-dimensional blades. It it obvious from these. models
that even complicated shapes of turbine blades can be considered
for calculations of the eigenfrequencies and the corresponding
mode shapes. As to the forced vibration problem of turbine blades
much less results are known in the literature. However, the
development of reliable numerical models is part of several
investigations today. Some of the early results had been de-
scribed in this paper.
420 H. Irretier

7. REFERENCES
I 1I BLOHf·1 & VOSS-Prospekt 11 Schaufeln fUr Dampfturbinen .. ,
Blohm &Vo6, Hamburg, 1976
I 21 Rao, J.S.: Turbine Blade Excitation and Vibration.
Shock and Vibration Digest 9 (1979) 15 - 23
I 31 Rao, J.S.: Turbomachine Blade Vibration.
Shock and Vibration Digest 12 (1980) 19 - 26
I 4/ Bohm, J.: Theoretische und experimentelle Parameterstudien
an schwingenden Turbinenschaufeln im Fliehkraftfeld.
Fortschr.-Ber. VDI-Zeitschrift 11, 29 (1979) 1 - 136
I 51 Montoya, J.G.: Gekoppelte Biege- und Torsionsschwingungen
einer stark verwundenen rotierenden Schaufel. BBC Brown
Boveri Mitteilungen 53 (1966) 216 - 230
I 6/ Rao, J.S.; Carnegie, W.: Solution of the Equations of
Coupled Bending-Bending-Torsion Vibrations of Turbine Blades
by the Method of Ritz-Galerkin. Int. J. of Mech. Sci. 12
(1970) 875 - 882
I 71 Carnegie, W.: Vibrations of Pre-twisted Cantilever Blading.
Proc. of the Instn. Mech. Engrs. 173, 12 (1959) 343 - 374
I 81 Carnegie, W.: Vibrations of Pre-twisted Cantilever Blading
Allowing for Rotary Inertia and Shear Deflection. J. of
Mech. Engng. Sciences 6, 2 (1964) 105 - 109
I 9/ Fu, C.C.: Computer Analysis of a Rotating Axial Turbo-
machine Blade in Coupled Bending-Bending-Torsion Vibrations.
Int. J. Num. Meth. Engng. 8 (1974)
/10/ r~eirovitch, L.: Analytical Methods in Vibrations. Collier-
Macmillan Ltd.; London, 1967
/11/ Bishop, R.E.C.; Gladwell, G.M.L.; Michaelson, S.: The Matrix
Analysis of Vibration. Cambridge University Press, Cambridge/
London/New York/Melbourne, 1965
/121 Hurty, W.C.; Rubinstein, M.F.: Dynamics of Structures.
Prentice-Hall Inc., Englewood Cliffs N.J. (USA) 1964
/1l' Lehsa, A.W.: Vibrations of Turbine Engine Blades by Shell
Analysis. Shock and Vibration Digest 12, 11 (1980) 3 - 10
/14/ Vogt, H.J.: Zur Berechnung der Eigenfrequenzen und Eigen-
formen von Schaufeln thermischer Turbomaschinen mit Hilfe
eines gekrUmmten finiten Schalenelementes. Dissertation,
TU Hannover (1974) 1 - 87
Free and Forced Vibrations 421

/15/ Leissa, A.W.; Ewing, M.S.: Comparison of Beam and Shell

Theories for the Vibrations of Thin Turbomachinery Blades.
ASME Int. Gas Turbine Conf.,London (England), April 1982;
ASME Publication 82-GT-223 (1982) 1 - 12
/16/ Zienkiewicz, O.C.: The Finite Element f.1ethod. McGraw-Hill
Book Company Ltd., London (England), 1977
/17/ Meirovitch, L.: Computational Methods in Structural Dynamics.
Sijthoff and Moordhoff, Alphen aan den Rijn (The Netherland~
Rock vi 11 e MD( USA), 1980
/18/ Bathe, K.J.; Wilson, E.L.: Numerical Methods in Finite·
Element Analysis. Prentice-Hall Inc., Englewood Cliffs
N.J. (USA), 1976
/19/ Steele, J.l~.; Lam, T.C.T.: Stress and Fatigue Analysis of
Steam Turbine Blades with ANSVS. Proc. ANSYS Conf.,
Pittsburgh PA (USA), April 1983, 4.48 - 4.65
/20/ Lam, T.C.T.: Computer Simulation of Fatigue Damage. ASME
Conf. on Computers in Mech. Eng., Chicago (USA), August 1983
/21/ Irretier, H.: Die Berechnung der Schwingungen rotierender,
beschaufelter Scheiben mittels eines Anfangswertverfahrens.
Dissertation, TU Hannover (1978), 1 - 129
/22/ Irretier, H.: Turbine Blade Simulation Program TUBSIM.
Institut fUr Mechanik, Universitat- Gh Kassel, 1985
(unpublished)
/23/ Traupel, W.: Thermische Turbomaschinen, Bd. 2. Springer-
Verlag Berlin/Heidelberg/New York, 1977
/24/ Timoshenko, S.; Young, D.H.; Weaver, W.: Vibration Problems
in Engineering. John Wiley &Sons Inc., New York, 1974
/25/ Vogt, E.: Hybride Schwingungsformanalyse an Turbinenschau-
feln. Fortschr.-Ber. VDI-Zeitschrift 11, 59 (1984) 1 - 118
/26/ Irretier, H.: Computer Simulation of the Run-up of a
Turbine Blade Subjected to Partial Admission. ASME Conf.
Mech. Vibr. and Noise, Cincinnati OH (USA), Sept. 1985;
ASME Publication 85-DET-128 (1985) 1 - 12
/27/ Rao,J.S.; Jadvani, H.M.: Free and Forced Vibrations of
Turbine Blades. Proc. ASME Conf. Mech. Vibr. and Noise,
Dearborn MI (USA), Sept. 1983, 11 - 24
/28/ Irretier, H.: Transient Vibrations of Turbine Blades due to
Passage through Partial Admission and Nozzle Excitation
Resonances. IFToMM Int. Conf. on Rotordynamics, Tokyo (Japan)
Sept. 1987
422 tl. Irretier

/29/ Pan-Report "Institute of Fluid-Flow Machinery", Polish

Academy of Sciences, Gdansk {Polen), 1978
1301 Gill, P.A.T.; Ucmaklioglu, M.: Isoparametric Finite Elements
for Free Vibration Analysis of Shell Segments and Non-
Axisymmetric Shells. Journal of Sound and Vibration 65,
2 {1979) 259 - 273
/31/ Aiello, R.A.; Hirschbein, M.S.; Chamis, C.C.: Structural
Dynamics of Shroudless, Hollow, Fan Blades with Composite
In-Lays. ASME Publication 82-GT-284, {1982) 1 - 7
 CHAPTER ll.l

FLOW PATH EXCITATION MECHANISMS FOR TURBOMACHINE BLADES

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACI'

The sources of non-steady forces in a turbine stase

are reviewed. Procedures for line-vortex stase
flow analysis and for actuator disk analyses are
described. with details of analytical contributions
to these areas. Experimental procedures for
measurements of blade non-steady forces in the
laboratory and in the field are described.
Selected test data are included. Recent develop-
ments usins the rotatins water table approach are
reviewed.

11.2.1 Introduction

Non-steady forces arise from local chanses (distortions) of

the fluid flow. from trailins wakes. and from flow
instabilities such as Karman vortices and rotatina stall. In
a turbomachine stase such disturbances are co-only
associated with flow obstructions such as nozzles. auide
vanes. diaphrasm joints. and certain downstream pressure
conditions. During rotation. such stationary flow distur-
bances may cause harmonic forces to act on the aovina blades.
Fisure 1 illustrates the interaction between stationary auide
vanes and a moving blade row. showing the manner and sources
from which flow excitations may arise. Table 1 lists several
important flow excitation sources. One major source of
excitation is passase of the moving blades throush the nozzle
wakes. This sives rise to blade excitations at nozzle
passins frequency (NPF) and at hisher multiples (2x. 3x.
424 N.F. Rieger

etc.) of NPF. The strength of these excitation harmonics

depends on factors such as the pressure drop across the flow
guides. ratio of number of guides to number of blades. axial
spacing between the flow guides and blades. guide vane
trailing edge thickness. local flow Mach number. and so on.
Such excitations have been observed on the moving blades in
the tangential. axial. and torsional directions. Similar
excitations also act on the non-rotating guide vanes.

Low pressure stage excitations arise mostly from low

frequency harmonic distortions of the flow field. These
excitations are shown in Figure 1. Harmonic flow distortions
may arise from the inlet flow guide passage spacing. e.g. •
from guide vane pitching and/or gaging errors during manu-
facture or assembly of the guide rows. These excitations
also occur from construction features such as diaphragm
joints. and from radial or angular eccentricity of the moving
row. Low pressure moving blades are airfoil-shaped. i.e ••
relatively thin with small camber (and hence small turning
angle). as in Figure 1. Low frequency per-rev excitations
may be of the same magnitude as the high frequency NPF
excitations for these stages. LP blade excitation spectrum
data may be obtained using a rotating pressure transducer on
a moving blade. or by analog sumulation of the stage flow
using a rotating water table. as described in Section 11.2.4.

At present • pratical computer methods for predicting

magnitude of the excitation on turbine blades are in a
developmental stage. The fundamental problem is highly
complex: the stage geometry creates a three-dimensional flow
field in which interactions occur between the nozzle flows
and the flow in the moving blade passages. Flow-field
analyses of this problem involve severe analytical diffi-
culties. Up to the present such analyses have mostly been
restricted to two-dimensional potential flow studies. The
wake/boundary layer interaction is thereby excluded from such
formulations of this problem. though such effects may be
significant. Further difficulties exist with available
methods for handling effects from the high turning angles of
impulse blading. Localized transonic flow conditions may
occur in HP blade passages. and such effects are beyond the
scope of all but research analysis. Details of some
approaches which have been developed for non steady flow
calculations are given in Section 11.2.3.

The calculation of LP stage exciting forces is simpler than

for an HP stage because smaller blade turning angles are
involved. and the stage pressure ratios may be lower. Non-
steady line vortex theory may then be used as an approximate
method for evaluating the high frequency exciting forces on
Flow Path Excitation Mechanisms 425

low camber airfoils. while actuator disk techniques (and

others) have been applied to predict the low frequency per-
rev harmonic forces.

The inherent danger from non-steady forces acting on the

moving blades is evident from the Campbell diagram shown in
Figure 2. This diagram is used to identify potential
resonant operating conditions. Blade frequency is plotted as
ordinate against machine rpm as abscissa. Blade natural
frequenc i.e s are plotted as characteristic lines. Speed-
related excitation harmonics are shown as the 1x. 2x.- •• •
per-rev excitation lines. The intersection of any per-rev
harmonic with any natural frequency is potentially dange~ous.
because it represents a resonant condition. Sustained
resonant operation under such conditions constitutes a
possible failure hazard. Whether danger actually exists
depends on three additional factors: (a) excitation mag-
nitude. (b) damping associated with the excited mode. and
(c) the phasing of the exciting forces relative to the blade
group. Non-steady flow studies are directed toward defining
the magnitude of the exciting forces in the spectrum of a
given stage. for nozzle wake harmonics. and for the lower
per-rev harmonics. together with details of their phasing.

A further source of excitation in BP turbine stages arises

from partial admission operation. Here. the inlet flow
enters via one or more n~zzle sectors around the inlet
circumference: see Figure 3. As the blades pass throuah the
flow admis.sion arcs. they are subjected to transient loading
from the inlet gas jet. followed by full unloading in the
inactive (non-flow) sectors. one or more times per rotor
revolution. The steady-state blade load is readily
determined. but details of the associated time-dependent
transient loadings are more difficult to obtain. because
impulse harmonics are involved. An experimentally-deterained
load profile for a partial admission stage is shown in Fiaure
4'. Until recently. there has been little research infor-
mation available on partial admission blade loadinas. due to
the obvious difficulties of making force measurements in
high-pressure. high-temperature turbine stages. Blade strain
gage data is more readily obtained. and this has been used to
calibrate the force-time profile. Some comments and infor-
mation on partial admission loads on turbine blades .are given
in Section 11.2.4.

11.2.2 Notation

tan 8
sonic velocity in gas ("yg RT)
constant in reference [19] see below
constant in Whitehead's analysis [13]
axial distance between blade rows
426 N.F. Rieger

b constant in Whitehead's analysis [13]

s
c blade chord
C(a) Theodorsen function (defined in te:~:t)
CL lift coefficient normal to chord

2
DM
dimensionless force (defined in te:~:t)
dimensionless force (defined in te:~:t)
constant in reference [19] see below
F constant in reference [19] see below
Fr Froude Number
G Glauert e:~:pansion (defined in te:~:t)
hm constant in reference [17]. Also water depth
i
J(a)
fl
Bessel function of first kind
K(a) modified Bessel function of second kind
k constant in reference [8]
1 disturbance in wavelength 2nVs /1
L non-steady a:~:ial force
Llt non-steady tangential force
Ly steady lift
Lo unsteady lift
M Mach number
N rpm
p pressure
p constant in reference [19] see below
q blade velocity due to vibration
r blade mean radius
a• gas constant
8(0') Sears' function (defined in te:~:t)
s blade pitch
t time
T temperature
T' (a) Holmes function (defined in te:~:t)
T(a) Horlock function (defined in te:~:t)
u non-steady a:~:ial velocity of gas
u mean a:~:ial velocity of gas
v non-steady tangential velocity of gas
v mean tangential velocity of gas
v. blade tangential velocity
general non-steady gas velocity
w
w mean gas velocity relative to blade
:X(a) function in reference [11]
lt a:~:ial direction
y tangential direction
Y(a) function in reference [11]
a angle of attack
IS mean flow angle relative to axial direction
r specific heat ratio (Cp/Cv)
8 constant in Whitehead's analysis [13]
e stagger angle
eLE lead edge blade surface angle relative to a:~:ial
direction
Flow Path Excitation Mechanisms 427

trail edge blade surface angle relative to axial

direction
constant in reference [19] see below
Whitehead's reduced frequency parameter,1)c/U sec ~
disturbance frequency 2nV /1
s
location of center of pressure of blade
location of center of twist of blade
3.14159
mass density
reduced frequency parameter relative to semi-chord
"'Jc/2W
I potential function
~ constant in reference [19] see below
Cal reduced frequency parameter relative to pitch 2ns/l

Constants:

Horlock, Greitzer, and Henderson [19]

B =1< 2Vs sec8{[2(A4-2A2-1) + D - 2'A(1+A2 )]

+ i[DA+(1+A2 ><f-4A3 )]}

D = 4A2 -"fi 2
F = [ (2A2+ )bA-D) + i (2A3-'¥1]

P = [(2A2+Yl> + i(2A 3 -~l

~ = (1+A) [1+2a(1-A)]
:K = 2rr/1
Whitehead [13]

Bs = (l+b s ) I (1-b s )
a = arc tan U/Vs

Henderson - Horlock [17]

Let tan eLE = g1

h1 = '2 - 81
428 N.F. Rieger

2 2 2 h3
g1 h1 g1 2g1 h1 g1h1
h5- -+ -+ -+ + - - + ..!
2 3 2 3 3 15

2 h3
g1h1 g1h1 1
h6- - - + - - +
2 2 8

h3
2 1
h7 - '1 + g1h1 +
3

2 2
h8 - 12 - 11

11.2.3 Non-Steady Force Theories

Information Sources;

Several recent reviews of the excitation literature have been

published. Sisto [1] has summarized the status of non steady
flow analyses, with emphasis on vortex theories. This is a
brief but useful introduction to the excitation literature of
turbine LP stage flows. Jlao [2] [3] has catalogued vortex
analyses with reference to turbine blade vibration, in two
general survey papers. Samoylovich [4] has published (in
Russian) a monograph concerned with turbine blade excitation
and vibration, which reviews excitation sources and describes
blade measurement and test procedures for dynamic pressure
distributions and for blade response. Osborne [5] has aiven
an excellent-comparative review ot developments in non-steady
coapre ss ible flow theories. Exist ina procedures for
calculating non-steady interactions between blade rows are
reviewed and compared with results from the matched asymp-
totic expansion approach for hish subsonic flows (1(}0.9).
Gostelow [6] has reviewed steady state compressible theories
for potential flow throush cascades with reference to the
Flow Path Excitation Mechanisms 429

transonic flow problea. and haa indicated soae proahina

numerical techniques for ita solution. Crofoot [7) has
presented a coaprehensive state-of-the-art review for non-
steady forces on turbomachine blades.

Vortex Theories:

l:eap and Sears [8] represented liftina surfaces by line

vortices to predict the non-steady forces actina on both the
aovina blades and the stationary blades of a turboaachine
staae such as that shown in Fiaure 5. The flow throuah the
cascade h aaauaed to be invhcid and incoapresaible. The
airfoils are assumed to be isolated i.e.. the non-stetdy
circulation of neiahborina blades ia nealected when calcu-
latina the non-steady effects on a particular airfoil. For
rotor blades however. the effects of the vortex wakes shed by
the stators are included. The rotor and stator blades are
assumed to be flat plates thouah they aay be thin. sliahtly
caabered. liahtly loaded airfoils. The implication of theae
assumptions is that the flow can tolerate only aaall turnina
anales as it passes throuah the blade rowa. Thia analyaia ia
therefore more suited to low solidity LP turbine ataaea than
BP turbine stases.

The stator wakes are aodeled as auats which decay in ampli-

tude as they propaaate downatreaa. The velocity of the auat
perpendicular to the chord is assuaed to be of the fora.
w = w e-i~t e-iky/Y (1)
0

where~ ia the stator vane paaaina frequency. Y ia

the free atreaa relative velocity. and t ia an arbitrary
constant. The lift fluctuation ia then aiven by:

L • x p c Y w0 S(a.1)e i'llt (2)

where p is the fluid aass denaity. c is the chord leaath. and

S(a.1) is the aodified Sears fuactioa. aivea by:

S(a.1) • 1(1) C(a) + i(a/1) 1 1 (1) (3)

where the 1 (1) are Beaael fuactioaa of the firat kiad. aa4
C(a) is thentbeodoraea fuactioa. The Theodora•• fuaotloa la
aiven by:

C(a) • 1:1 (ia) I [I: (ia) + 1:1 (ia)) (4)

where 1: (ia) an aoUUed Beaael fuaotioaa of tle aeooad

kind. Tle arauaeata of the aodifled Seara fuaotloa are t••
reduced frequency (a • ~ c/2Y) aad the h•CIU•••Y aaaoolated
·with the decay term (1 • l:o/Y).
430 N.F. Rieger

The results shown in Fiaure 6 are for the first two harmonics
(a = 1.2) of the lift ratio (unsteady lift/steady lift)
plotted aaainst the axial spacing ratio (b/c) and the pitch
ratio (Sr/Ss>• The dashed curves correspond to an elliptical
steady load distribution on both the rotor and stator blades.
while the solid curves are for an elliptical distribution on
the stator blades. and a flat-plate distribution on the rotor
blades. The authors concluded that 'the non-steady part of
the lift may be as large as 18.. of the steady lift and
therefore may be of practical importance.' This low value is
a feature of the simple model chosen. It is by no means an
upper liait for practical blade conditions.

In a subsequent study [9] l:emp and Sears considered the

effect of the viscous wakes of an upstream .stator on a
downstream rotor blade as the blade passes through the stator
wakes. The ass'Gilptions of thin airfoil theory were again
imposed. The wake behind each stator blade is the same as
the wake behind an isolated airfoil. The wake consists of an
inviscid. symmetrical. shear perturbation of the undisturbed
stream. If the wake velocity in the streamwise direction is
assumed to be constant. the non-steady lift on the rotor due
to viscous wakes is aiven by:
L • no Ww S(a)ei~t (5)
0

where p is the aass density. W is the free stream velocity

relative to the rotor blade. and c is the chord lenath. The
non-steady lift coefficient CL is aiven as:

CD

where G is the Glauert expansion for the non-steady velocity

normal !o the chord. S(aa) is the Sears function for reduced
frequency (a • V c/2w). and )} is the disturbance (stator
passina) frequency. Comparing the results of this analysis
with those obtained previously suagests that the viscous-wake
effects on a rotor blade may approach the maanitude of the
circulation-induced non-steady lift.

Borlock [10] [11] noted that the l:emp-Sears analyses

considered only velocity perturbations normal to tho chord.
ae then derived an expression for fluctuatina lift due to a
streamwise velocity perturbation. The l:emp-Sears assumptions
for thin airfoil theory were aaain used. Since the flow
inlet velocity is not necessarily parallel to the chord.
Flow Path Excitation Mechanisms 431

small ansles of attack (a.) were thereby included in this

analysis. The non-steady velocity parallel to the chord is
assumed to be of the form:

X
i l1(t - - )
u = u0 e U (7)

where Y is the stator passins frequency, and the coordinate x

is in the direction of the chord. The fluctuatina lift is
then siven by the expression:

L = 2n U u 0 a. p e i~t T(a) (8)

where p is the mass density, w is the maxiaum value of the

non-steady chordwise velocity, ca. is the ansle of attack, and
T(a) is the Horlock function:

T(a) = :X(a) + iY(a) (9)

where, :X(a) = (2-a) J(a) - b J 1 (a)

Y(a) = (a+l) J 1 (a) - b J(a)

and K /(K +K1 ) = a + ib

0 0

where J (a) are Bessel functions of the first kind and K are
modifiel Bessel functions of the second kind. The resulfs of
this analysis may be combined with the results of the first
Kemp-Sears analysis to obtain the total circulation-induced
lift fluctuation on a blade due to a aeneral periodic
disturbance of the free stream flow. This disturbance w is
resolved into components perpendicular to the chord wt and
. wc as shown in Fiaure 7. In aeneral,
parallel to the chord
the velocity of the disturbance can be expressed as a Fourier
series:

w='\w sin (10)

L on
1
n

where 1 is the wavelensth of the disturbance. For the n~

component of this series, the fluctuatina lift is:

L=2x p Usee' w
on
ei~t[S(a) sin,-aT(a) cos' ]
a (11)

where U is the axial velocity, ' is the aean flow anal•• " is
excitins frequency, and a is reduced frequency. 'Dle abo••
expression for fluctuatina lift applies to flat plate
airfoils. To account for camber Horlock incla6ed the Bolaes
function, T' (a), which was derived for an airfoil of para-
432 N.F. Rieger

bolic camber. The total lift fluctuation for the nth

component of the Fourier series includins all the above
effects is:

~- 0 0
2
p U 1 ei1 tv S(a)+u [aT(a)+( Ymax)T'(a)]
1
(12)

where y is the m:axiaum airfoil thickness and c is the

airfoil ~ord. Naumann and Yeh [12] independently derived an
expression relatins the unsteady lift of a cambered airfoil
to a seneral flow perturbation.

Actuator Disk Method:

Whitehead [13] introduced a procedure for calculatins the non

steady blade force reaultins from a siven harmonic distur-
bance profile in the incomins flow. Fisure 8 shows·the blade
row considered as a thin 'actuator disk,' with the associated
velocity chanses resultins from the momentum transfer throush
the disk. It is also assumed that the blades vibrate nearly
in-phase, and that the blades are flat plates at stasger
anale e to the axi~l flow. The blade row is considered to be
narrow, such that the time for the fluid to pass through is
saall compared with the disturbance frequency acting on the
blades. This assumption is expressed in terms of the reduced
frequency parameter 1 • V c/Uaec,, where "Y is the frequency
of the non-steady excitation, c is the chord length, and U is
the aean velocity of flow in the axial direction. The
reduced frequency is assumed to be close to zero in this
analysis. Using these assumptions, Whitehead obtained a
siaple expression for the dimensionless non-steady force
acting on any blade.

· 'lhi tehead' s procedure depends only on a knowledge of the

blade/cascade proportions and seometry, and the steady and
non-steady flow velocities. The above theory is restricted
to oases where the reduced frequency 1 is small, i.e., 1«1.
It is therefore well-suited to the analysis of low frequency
harmonics of high inlet velocity machines with narrow blades,
but it is not suited to analysis of nozzle wake excitations
from low inlet velocities. To relax this restriction
Whitehead [14], Borlock and Danyshar [15], and Smith [16]
have developed more advanced procedures which are suitable
for 0<1<1.0.
Henderson and Borlock [17] have approximated the unsteady
lift on airfoils for cascades having low pitch-to-chord
ratios. The pitch is also assumed to be small. with respect
to the wavelensth of the inlet disturbance, Fisure 9. The
rotor blades are thin and may have considerable camber but
the lift coefficient is assumed to be not large due to the
Flow Path Excitation Mechanisms 433

low pitch to chord ratio. The flow is assuaed to be two-

dimensional, inviscid, and incompressible. Sinusoidal azial
disturbances are considered, of finite frequency paraaeter a,
baaed on the blade chord, but of low frequency paraaeter M,
based on blade pitch.

The basis of this theory is the so-called pitch-averaaina

technique due to Borloct and Marsh [18] applied to the
unsteady equations of motion for the fow throuah the aoviaa
blade passage. The total force in the tanaential and azial
directions was obtained by intearating the pressure
difference between the pressure and suction faces of the
blade alona the chord. For an inlet velocity of:

u =U+ u0 sin V ( t - y/V) (14)

the unsteady tanaential component of lift is:

L
_L_
pcv2
... t[2 c-
• v
2uoUs
- -- h
1
sin w] sin1t
,v2c 1
(15)

4u U u h4 }
+{[ y2° h 3 (1-cos M) + - 0- - sin w] cos~t
.., sec' Vsec'
A a imilar ezpreaaion "141' obtained for the unsteady azial
component of lift, Ly /pc~.

The symbols used are defined in the Notation section. The

total unsteady lift for a turbine cascade is the vec.tor sua
of the tangential lift and the azial lift components:

L
(16)

A comparison of results obtained usina the above theories is

shown in Figure 10.

The above calculation procedures have been further developed

by Borloct, Greitzer, and .aenderson [19], who presented an
alternative 'semi-actuator disk' aethod. This procedure
assumes that the blade row is made up of a closely-spaced set
of flat plates set at an anale 8 to the azial direction, and
that:
434 N.F. Rieger

a) Far upstream there is a harmonic disturbance in the

axial velocity.

b) The stream function is continuous at the leadins

edse of the blades with a discontinuity in slope.
c) Continuous stasnation pressure exists at the
leadins and trailins edses.
d) Continuity of velocities exists at the blade
trailins edse.

These authors obtained the fol1owins expression for non-

steady lift coefficient:

.. neD
s

(17)
B2acose
+ i( 2 -F)}
v~ tane
This result applies for the case of a seneral stasser ansle
e. where:

L1 =- pVssl(. sece (F-iP) up i( "'t- 'l(y)

(18)
1
L2 • - ps cose {2Bc-i(3<. 2 o2 ')' seo 2 8)P}exp iV t
2

and where P. F.3<. and Bare defined in the Notation section.

Results obtained by the semi-actuator disk method have been

oo11pared with data from Smith [16) in Fiaures 11 and 12. from
[19). It is evident that the semi-actuator disk theory sives
aood results when the apace/chord ratio is low. and when the
reduced frequency a is fairly small.

11.2.4 8zperiaental Studies

TJpea of Blade-Force Tests and Experiments:

Meaaureaents for non-steady blade forces and stresses have

been conducted as follows:
a) In-situ teats on turbine and coapressor blades.
rotatin& and non-rotatin&• uaina strain aaaes.

b) Bxperiaental turbine ataae studies in wheelboxes

us ina strain aaaes and developmental holoaraphic
techniques.
Flow Path Excitation Mechanisms 435

c) Wind tunnel tests on blade cascades.

d) Water table measurements using strain gages.

e) Conducting-sheet analogy and electrolytic tank

analogy.

These studies may be divided into several categories:

a) · Direct measurement of blade response in a rotating

stage flow test.

b) Laboratory measurement of blade response under

controlled stage flow conditions.

c) Quasi-static studios of staae flow for specified

nozzle-blade static confiaurations.

d) Nozzle cascade wind tunnel measurements.

Direct measurement of blade response to excitina forces is a

meaningful procedure for non-steady force estimation only
whore the blade dynamic response properties do not substan-
tially interfere with the blade excitation frequencies. Any
tendencies toward blade resonant response will aive incorrect
excitation data. The strain aaaes used in such testa aaat be
calibrated to aive load readout. Direct readout of forces
from strain aaaea is difficult to achieve because the blades
themselves are very stiff and the strain aaaea have limited
sensitivity and signal/noise properties. Good sianala may be
achieved under blade resonance conditions. but non-resonant
strain aaae data from blades are rarely taken in practice.
For this reason in-situ blade testina is most coaaonly used
to test/define the Campbell diagram over the speed ranae
under operatina load conditions.

Laboratory measurement of blade responte. Such measurements

may be undertaken in several different ways. Beater box
tests are colllllon blade verification procedures. They are
utually conducted in vacuum (to minimize power and heatina).
without gas flow. Blade natural frequenoiet can be excited
under such circumstances. utually with electromaanets. Blade
tests in experimental turbines have alto been conducted e.g ••
by Wagner (20] with water jet excitation to measure blade
a roup damp ina. and by Part ina ton (21J to meuure blade
excitation and ttretaes: see later. Other laboratory tetts
have been conducted on a rotatina water table model. utina
the hydraulic analoay to model the flow through one or more
turbine staae aeometriet. Thete tettt ha~e provided data on
non-steady force (and torque) spectra. for a wide ranae of
flow parameters. and are ditcutsed later in this tection.
436 N.F. Rieger

Stationary blade-nozzle stau tests. Several stationary

stage tests have been reported in the literature e.g.,
l:earton [22], llalavard [23]. Some indication of the flow
through a stage is usually sought by measuring blade forces
at series of increments across the nozzle exit. QBasi-static
testing of this type does not match the velocity conditions
in an actual stage, and the results are therefore considered
to be unrepresentative of stage flow conditions, and of wake
effects.

Nozzle cascade wind tpm1el tests. Some indication of the

excitation likely to arise from a given nozzle row or guide
row is sometimes sought using a model row of guide vanes in a
wind tUilllel (frequently installed as corner turning guides).
Tests of this type do not give blade forces, but they can
give guidance of BP stages where the pressure drop occurs
primarily in the nozzle and not in the blades. Velocity
traverses from such models have been reported by several
authors. Samoylovich (4] hu discussed wake properties in
detail.

Non-steady Blade Stress Measurements:

Jleasurement of blade dynamic stresses is now a relatively

straightforward procedure. In the field such measurements
are conducted on a 'need' basis because of the high costs
involved in monitoring the number of challllels required to
obtain representative information. Relatively few blade
stress reports of general usefulness have been published
despite years of such testing. Partington [21] used an
experillental turbine test to determine tho sources of low
pressure blade excitation and stress. Excitations arising
from tho following factors and others were investiaated:

a) Obstructions in inlet and exhaust flow fields.

b) Nozzle and blade exit profile changes.

c) Axial distance between guides and blades.

d) Gaging modifications in nozzles and blades.

The blades and nozzles of the test turbine were instrumented

with strain gages, and a slip ring assembly at the end of the
rotor was used to obtain the b.lade strain gage signals.
Partington found that:

a) Larae differences in stress at similar blade

locations were found between corresponding blade
aroups.
Flow Path Excitation Mechanisms 437

b) Small differences in stress were fo1Ul4 betweea

blades of the same aroup (presuaably for ia-phase
modes).

c) Stress aenerally iacreases with output torque. but

decreases with increasiaa eXhaust pressure.

d) Plots of stress amplitude vs. rpa aear resoaaace

showed the first mode respoase to be co-,licatod by
multiple stress peaks. This was apparoatly caused
by coupliaa between differeat blade aroups aroUDd
the disk. Bach blade aroup is evidoatly aot aa
independeat vibration systea.
This latter point has beea observed by maay other
investiaators. Partiaaton also made blade aroup da.piaa tests
aad found aroup loaarithaic decremeats ia the order of 0.011
and 0.025. Heyman [24] conducted siailar tests to doter.aiae
the iaflueace of adal aap between blades aad aonles at
various wheel speeds oa blade dyaamic stresses. Strain aaaea
were aaain used to measure the stress reapoase.

Teat Tarbiae Studies of Blade Forces:

Test turbiaes are used frequently ia blade developaeat work.

Pressure profiles upstream aad doYDstre'am of blades aad
aozzles have beea measured by pitot nozzles aad by pressure
transducers. Descriptioas of teat turbiaea aad flow
measuriaa equipmeat have beea aivea by Reaaudia aad so.. [25]
aad others. Samoylovich [4] shows drawiaas of two teat
turbines and aives details of pressure iaatruaeatatioa aad
strain aaae results obtained. with details of local pressure
vs. tiae at various locatioas across the blade wake. The
Westiaahouse test turbiae facility in Lester. Poaaaylvaaia is
desiaaed to operate with either steam or air as tho workiaa
fluid.
Studies of Flow ia Tarboaachiaes by the Hydraulic Aaaloay:

The earliest iavestiaatioa of turbine ataae flow ualaa the

hydraulic analoay appears to have beea that of Prelawerk [26)
la 1942. This paper contalaod photoarapha of aubaoalo. low
aupersoaic. and hiah auporsoaic flows throuah a ataUoaar7
cascade. For tweaty years. the atatioaar,. caacade repre-
sented the state-of-the-art for h7draulic aaalOJJ' at.. lea of
flow ln turbiaoa aad coaproaaors. haulta froa aeverd
iaveatiaatloaa of this type have beea co-,are4 favorabl7 wltl_
cascade wia4 tuaael teat reaulta: see Bo,.t [27). •re
recent studies. have used both statioD.&rJ' bla4ea aa4 aovf.aa
blades to aore accurately 80del the flui4-atnotue later-
actioa ia a turbiae ataae. Heea aad Jaaa [28) at. .le4 flow
in a partial adaiuioa turblae ataae ulaa a track u.4
438 N.F. Rieger

carriase apparatus. Fisure 13. The water depth was measured

alons a blade passage. A theoretical depth at each position
was obtained from one-dimensional water flow theory. Typical
results plotted against theoretical values are shown in
Fisure 14. 1ohnson [29] has described a two stage axial flow
compressor study made at the General Electric Research
Laboratory.

1ohnson's model employed two rows of rotor blades which moved

relative to adjacent stator rows. The apparatus was used to
study off-desisn conditions as well as the effects of varying
rotor and stator ansles and spacings. Slow-motion moving
pictures were taken during testins. Similar tests have been
performed on turbine stages using a similar apparatus.
Rhombers [30] made a qualitative comparison of results
obtained from a water table model with results obtained from
a rotatins transonic air cascade. Shadowgraphs of the flow
downstream of a transonic air cascade are given. Good
correlation was found for the relative locations of the shock
waves obtained by each method. Owczarek [31] investigated a
periodic wave phenomenon occurring from stage interactions
between turbine blades and stators. This phenomenon occurs
when a pressure wave is generated on the leading edge of the
rotor blades. Owczarek constructed a rotating radial inflow
water table model of a suitable configuration for this
phenomenon. according to theory. Photographs were taken of
the flow between the stator and rotor which showed waves
(analosous to gas shocks) which propagated as predicted by
theory.

Rotating Water Table Tests of Blade Forces:

Nop-Steady Forces from Nozzle Wakes. A turbine stage was

constructed [32] and tested [33] to determine the magnitude
of the exciting forces associated with a certain nozzle
design which was associated with in-service failures. The
stage geometry is shown in Figure 15. A typical force
response amplitude vs. frequency spectrum from the moving
blade strain gage output is shown in Figure 16. A large
amplitude spike is evident at nozzle passing frequency (NPF).
Such spikes represent the harmonic force which is applied to
the moving blades at NPF. Harmonics of the NPF are also
present. The test results are shown as a plot of dimension-
less non-steady force amplitude vs. velocity ratio for
several pressure ratio tests in Fisure 17.

Transient Forces From a Partial A4mission Stage. This test

was conducted to measure the transient load variation on a
moving blade as it passed through the flow from a nozzle arc
in a partial admission stage. The general form of the load
trans'ient on a moving blade in an actual partial admission
staae is well known from prototype strain gage tests. This
Flow Path Excitation Mechanisms 439

water table test showed the desree to which tho water table
load transient resembled the load transient obtained from
actual bladins tests. Tho staso parameters and typical
results are shown in Fisure 4. The followins similarities
can be observed:

a) Relative slope of the inlet response

b) Inlet response spike aasnitude

c) Outlet response spike masnitude

d) Relative masnitudo of main curve

Subsequent partial admission tests have demonstrated that

this correlation is typical, and that quantitative co.,arison
(e •I•, peak/averaso) is representative of that observed in
practice.

11.2.5 References

1) Sisto, F., 'Nonsteady Flow &citation in Steam Tarbine

Stases,' Proceedinss, Workshop on Improved Tarbine
Availability, Electric Power Research Institute, Palo
Alto, California, 1anuary 1977~

2) Rao, 1. S., 'Turbine Blade &citation and Vibration,'

Shock and Vibration Disest, Vol. 9, No. 3, pp. 15-22,
March 1977 •

3) Rao, 1. s., 'Turbomachine Blade Vibration,' Shock and

Vibration Diaest, Vol. 19, No. 5, pp. 3-10, May 1987.

4) Samoylovich, G. s., ''Vibration Problems in llotatin&

Turboaachines, 'llachinostroya Publishin& Bouse, Koscow,
1975 (in Russian).
5) Osborne, C., ''Compressible Unsteady Interactions
Between Blade Rows', AlAA 1ournal, Vol. 11, No. 3, pp.
34G-346, March 1973.

6) Gostelow, 1. P., 'Review of Co.,ressiblo Flow Theories

for Airfoil Cascades,' Trans. ASKE, 1ournal of EDaineer-
ins for Power, Series A, Vol. 95, No. 4, Pit• 281-292,
October 1973.

7) Crofoot, 1. F., 'Theories and Bzperiaents for Deter-

mination of Non-steady Loads on Tarbouchine Bladea,'
M.S. Thea is, Rochester Institute of TechnolO&J'•
Rochester, New York, February 1979.
440 N.F. Rieger

8) l:emp, N. H., Sears, 1J. Jt., 'Aerodynamic Interference

Between Kovins Blade Rows',' 1ournal of the Aeronautical
Sciences, Vol. 20, No. 9, pp. 585-597, 1953.

9) l:e~, N. H., Sears, lf. Jt., 'The Unsteady Forces Due to

Viscous Wake a in Turbomachines,' 1ournal of the Aero-
nautical Sciences, Vol. 22, No. 7, pp. 478-483, 1955.

10) Horlock, 1. H., 'Unsteady Flow in Turbomachines,'

Proceed ins•, Conference on Hydraulics and Fluid
Kechanics, Institution Of Ensineers, Australia, pp. 221-
227, November 1968.

11) Horlock, 1. H., 'Fluctuatins Lift Forces on Aero foils

Kovins Throush Transverse and Chordwise Gusts,' Trans~
ASIE, 1ournal of Basic Ensineerins, Series D, pp. 494-
500, December 1968.

12) Naumann, H., Yeh, H., 'Lift and Pressure Fluctuations of

a Cambered Airfoil UDder Periodic Gusts and Applications
to Turbomachinery,' Trans. ASME, 1ournal of Engineering
for Power, Series A, pp. 1-10, 1anuary 1973.

13) Whitehead, D. S., 'Vibration of Cascade Blades Treated

by Actuator Disk Methods,' Proceedinss, IllechE, Vol.
173, No. 21, pp. 555-574, 1959.

14) Whitehead, D. S., 'Force and lloment Coefficients for

Vibratina Aerofoila in Cascade,' Aeronautical Research
Council, It and J( 3254, 1960.

15) Horlock, 1.H., Daneshyar, H., 'Stagnation Pressure

Chansea in Unsteady Flow,' Aeronautical Qnarterly, Vol.
22, Part 3, August 1971.

16) Smith, S., 'Discrete Frequency Sound Generation in Axial

Flow Turbomachinea,' Aeronautical Research Council, R
and II 3684, 1972.

17) Henderson, R. B., Horlock, 1. H., 'An Approximate

Analysis of the Unsteady Lift on Airfoils in Cascade,'
Trans. ASKB, 1ournal of Engineering for Power, Series A,
pp. 233-240, October 1972.

18) Horlock, 1. H., Karsh, H., 'Flow Models for Turbo-

machines,' 1ournal of Mechanical Engineering Science,
Vol. 13, No. 5, pp. 358-368, 1971.

19) Horlock, 1. H., Greitzer, B. II., and Henderson, Jt. B.,

'The Response of Turbomachine Blades to Low Frequency
I~let Distortions,' Trans. ASIIB, 1ournal of Ensineering
·for Power, Series A, 1976.
Flow Path Excitation Mechanisms 441

20) Wagner. 1. T.. 'Blade Damping Tests.' Westinghouse

Engineering Report EC401, NOBS N00024-67-C-5494, llay
1969.
21) Partington, A. 1., 'Experimental Study of the Sources of
Blade Vibration Excitation in a Low Pressure Turbine·;•
Westinghouse Electric Corporation Development Engineer-
ing Dept., Lester, PA, Contract NOBS-94380, Report No.
EC 384, April 1971.

22) Kearton. W. 1., Steam Tprbine Theory and Practice, Sir

Isaac Pitman and Sons Ltd., London, England, 1958.

23) Malavard, L., 'The Use of Rheo-Electrical Analogies in

Certain Aerodynamical Problems,' 1ournal of Royal
Aeronautical Sciences, Vol. 51, pp. 739, 1947.

24) Heyman, F. 1., 'Turbine Blade Vibrations Due to Nozzle

Wakes,' ASME Public at ion, Paper No. 68-WA/PWR-1,
Presented December 1-S, 1968.

25) Renaudin, A., Somm, E., 'Quasi-Three-Dimensional Flow in

a Multistage Turbine Calculation and Experimental
Verification,' Flow Research on Blading, Elsevier
Publishing Company, London, England, pp. 51-88, 1970.

26) Preiswerk. E., 'Some Applications of the Method of

Hydraulic Analogy.' Publ. Scientifiques et Tech. du
Ministere de l'Air, Paris, France. 1942.

27) Hoyt, 1. w•• 'The Hydraulic Analogy for Compressible Gas

Flow,' Applied Mechanics Review, Iune 1962.

28) Been. H. K•• Mann, R. W., 'The Hydraulic Analogy Applied

to NonSteady Two-Dimensional Flow in the Partial-
Admission Turbine' , Trans. ASME, 1ournal of Basic
Engineering. Series D. pp. 408-421, September 1961.

29) 1ohnson. R. H., 'The Hydraulic Analogy and its Use with
Time Varying Flows,' GE Research Laboratory Report, 64-
RL-(3755 C), Schenectady, NY, August 1964.

30) Rhomberg, E., 'Investigations into Rotating Blade

Cascades for Transonic Flow, ' The Brown Boveri Review,
Vol. 51, No. 12, pp. 762-773, December 1964.

31) Owczarek, 1. A., 'On a Wave Phenomenon in Turbines,'

Trans. ASME, 1ournal of Engineering for Power, Series A,
pp. 262, 1uly 1966.
442 N.F. Rieger

32) Rieger, N. F., Wicks, A. L., 'Design and Development of

Water Table Analog for Dynamic Forces in Turbine
Stages,' ASME Paper 78WA-DE16.

33) Rieger, N. F., Wicks, A. L., 'Non-Steady Force Measure-

ments and Observations of Flow in Three Turbine Stage
Geometries,' Trans. ASME Vol. 100, pp. S2S-S32, October
1978.
Flow Path Excitation Mechanisms 443

Figure 1. Interaction Between Guide Vanes and Moving Blade

Row Showing Sources of Flow Induced Excitation

FReouENCY <Hz>

RotATIONAl. SPEED (RPid.

Figure 2. Campbell Diagram for LP Stage Blade Group

444 N.F. Rieger

Inlet Nozzles

~p;pppj~
-- (( (( (( (( (( (( (( (( --- Moving
Blades

Figure 3. Typical Admission Arc in Partial Admission Stage

Tftt Conditions
Type of staae Impulse
Number of IBOVIng b'-des 1Zl
Number of .,-rtlal admission arcs 3
Number of nozzlft In arcs 10,10,12 {corresponds to 7& nozzle~
In complete drcle}
Stage pressure ratio 1.58
Stage velocity ratio o.u
Number of Instrumented blades 1
lnst. Force Direction of force measurement Tangential
Avg. Force
lnst. Force
20 Enter Leave Avg. FOrce NPF
Ripple

Blade Pitch
I I I I I I I I I I I

0 2 • & I 10 12 1• 0 2 • & I 10 12 1•

Figure 4. Comparison of Water Table Partial Arc

Result with Typical Steam Turbine Result
Flow Path Excitation Mechanisms 445

Figure 5 • Kemp-Sears Cascade Configuration

. 20r---,----T----~---,

m=1

o L...!.1:::::::5:~:Lmt:Z.J
0 0.2 0.11 0.6 0.8

b'/c r
Ratio of Unsteady Lift to Steady Lift vs. Spacing Ratio (Stator
Pitch Equals Rotor Pitch).
Figure 6. Results From Kemp-Sears Analysis
446 N.F. Rieger

~"'LA(_ ''-LJ_;/'~ x.
VELOCITY PROFILE
w • w e1v(t - &>
0

Figure 7. General Unsteady Gust Velocity Profile

Upuru• Oowni\reaa

~ y • weiwt

ue 1 "''
~
U•
y • veiwt
U• ue iwt

At\UI\Or Dht

Figure 8. Actuator Disk and Velocity

Flow Path Excitation Mechanisms 447

f1oVJNI 11.\DII

Figure 9. Cascade Notation of Henderson-Horlock

JIG

--
cr•J.OC'"'ltehea412ll 0.:1

0.5 O.J
..,...- --

o.a
o.y /"""' cr•O

/
/
O.l
O . J , _ Neuar-·llodocki:IJI

I
cr•O /
-o.4 -0.) -o.J -0.1 0.1

Flat plate airfoil•

Stat,.~ aDtlaa 45" -o.l
P&tch•cho~ ~atloa 1.0

Figure 10. Real and Imaginary Parts of Unsteady

Lift, Henderson-Horlock
448 N.F. Rieger

-. . . _ >
-.
...
'

. . _....L>- .
....... 0=0.11

-.. _.._
c:
u
...
..
c:
1. 0
-- O•O.l
.... ....

-- .. --
--- ..... -------
---

--,..
c3
::J
0.1
-----P- ---------. -
0•0.1

-
i
••
c:
0.6
Oc0.01
-
...
~
--
---
Semiactuator Disk Analysis
Ref. II
-
0

-•
•
'9
cDl
0.111 f- Stagger Angle II!'
Flat-Plate Airfoils at Zero
Incidence
::E 0.2 -
.
0.6
I

o•• 1.0
.
0 0.2
Space-Chord Ratio S /C

Figure 11. Comparison of Predicted Magnitude

of Unsteady Lift Coefficient

L a "'ICOW c elvt
0 Ill · - -

_ Set11lactuatot Disc o.s

_ Isolated Airfoil
Stagger Angle 10•
0.11
Flat-Plate Airfoils at Zero '\ r-

SC•t.•W
Incidence \ ;::
'' 0

o. 3 ~
IC

SC=- \ ~ It
1 I I I o. 2
' I I I
SC=O·~I'
\ I I
I

''· '~ :: o. 1
...
', ~·
\'I
~------L-------~------~---~------~------_.------..JO
-o.6 -o. s -o. 11 -o. 3 -o. 2 -0.1 0
IL/Lokeal

.Figure 12. Comparison of Semi-Actuator Disk

and Isolated Airfoil Theories
Flow Path Excitation Mechanisms 449

Figure 13. Partial Admission Test

Rig. Heen and Mann

~ l
....
:.
1: •
.
0
c.
5

Water Height (in)

.75~ -
,.so~~ r.
• :zs~ -~
00 I ~ I ~ I ~ I ~ o (o 1 ~~ I
1,4
Dist•nc:e In Bl•de Spac:ing5
Experimental
Theoretical -----

Figure 14. Water Depth vs. Distance for Several Locations

in Moving Blade Passage from Heen and Mann.
NOTE: Position 1 at Lead Edge, Position 7 at
Trail Edge
450 N.F. Rieger

Figure 15. STI Water Table Apparatus

Per-rev
Harmonics
Non-Steaay N.P.F.
Blade Force

Frequency, Hz.

Figure 16. Tangential Load Magnitude vs. Frequency

Flow Path Excitation Mechanisms 451

,--·--·--···--···-··
i 1$1 "'""·
···--...................
,CRt( IIAlJII l
I:DW·!.ll~1
:

I vs. !
! Yti.C:J n RATJII j
i
,! nu ~
j D11lt1JOII: h"C••u•l
j rcusuat tiiTJOS: ::::~!
! NOtl\. IDIItlh.SIU, AI.OJ ll.UI a:ui
i HOIIII. Y(l.RATJD: .)000 I
: !
~ ;_""""-"'"-"""'"--'-""""""""'-l

Figure 17. Normalized Tangential Force

Ratio vs. Velocity Ratio
452 N.F. Rieger

Table 1. Flow Path Excitation Sources and Harmonic Ranges

IW.t Banopio Trpiaal SOp[R!!

liP Blah per-rev (40 x) Nozzle tolereuoe haraoulca
Upatreaa wake deaeueratioD
Structural turbuleuce

Nozzle paaalua frequeucy Nozzle waku

2 X NPP Baraouloa froa Dozzle wakea
3 X NPP Baraouica froa uozzle wakea
IP Medlu. per-rev (20 x) Nozzle tolerauce haraoulca
Up1treaa wake deaeueratlou
Structural turbuleuce
NPP Nozzle wak01
LP ODe per-uv Relative diaplaceaeut uozzle blades
Twoper-nv Dlaphraaa jclut1
Multiple per-rev Structural 1upporta iD flow path
Jradlu per-rn Nozzle tolerauce haraouica
Blah per-rn Nozzle tolerauce haraoulca
Upatreaa wake deaeueratioD
Structural turbuleuoe

Table 2. Analogous Quantities for Gas and Water Flows

Deualty ratio (p/p 0 ) • Beiaht ratio (h/h0 )

Teaperature ratio (T/T0 ) • Beiaht ratio (h/h0 )

2 2
Preaaure ratio (p/p0 ) • (Beiaht ratio) (h/h0 )

Velocity of aouud (a• (Yili> • Gravity wave propaaatioD

velocity (c• {it>

Jraoh DUbtr (M)

 CHAPTER 11.3

THE DIAGNOSIS AND CORRECTION OF STEAM TURBINE BLADE PROBLEMS

N.F. Rieaer
Stress Technoloay Incorporated, Rochester, New York, USA

ABSTRACT

Several important types of turbine blading failure

are discussed. A review of the basic causes of
blading failures, e.g., fatigue, corrosion, stress
corrosion cracking, erosion, etc., is given.
Procedures for calculating stresses associated with
such failures and the number of cycles to failure
are reviewed. The familiar Goodman diagram
procedure for cycles to crack· initiation is
included, plus a fracture mechanics approach to
give the number of propagation cycles. Four case
histories of turbine blade failure are presented in
detail, with operating conditions, diagnostic
procedures used to determine failure cause, and the
remedies chosen to avoid further blading failures.
Thirteen references to the subject literature are
included.

11.3.1 Introdpction

Service failures of turbine blades are infrequent but costly

events. Electrical utility records show nearly 30 percent of
all steam turbine forced outages are attributable to blade
problems such as cracking, erosion, blade fracture, etc. The
duration of blade related turbine outages may range from
several days for a simple blade replacement in a small unit,
to several months for a significant blading failure involving
consequential damage in a large unit. Outage duration is
obviously influenced by the availability of replacement
parts. as well as repair time. Similar circumstances apply
to industrial drive turbines and to marine propulsion
turbines.
454 N.F. Rieger

This chapter discusses several causes of steam turbine

blading failures, identifying a number of important factors
relating to these failure causes. Certain corrective
measures which have been uses successfully in the past to
overcome such problems are indicated. Knowledge of potential
problem areas and of corrective measures is of value to
designers and turbine operators attempting to avoid similar
problems. Procedures for the analysis of stress related
blading failures involving both high cycle and low cycle
fatigue are described. These procedures permit quantitative
assessments to be made of cases involving major stress
related failure mechanisms such as fatigue, corrosion
fatigue, and stress corrosion. Several case histories of
blade failures are described, together with practical
remedies which were used to overcome these failures.

Most blading problems present an unclear variety of evidence

when the turbine is first opened. The first task when
looking for the failure cause is to carefully record and
evaluate the failure data and operating conditions which the
blading has experienced. Identification of the failure cause
is the first major step toward prescribing an effective
solution. However, problem diagnosis may be a secondary
objective in the period immediate following the failure. The
turbine operator usually wants to 'get running' again,
typically using some interim arrangement such as re-blading
with available replacements or removal of the damaged row.
Properly utilized, this situation offers a valuable oppor-
tunity to conduct a more thorough investigation and diagnosis
of the problem as a basis for a more permanent fix to be
installed at a future outage.

11.3.2 Blade Loading Conditions

The ability of a blade to support its applied loads depends

on:

a) Strength of blade material in its environment.

b) Magnitude and distribution of steady mean stresses.

c) Magnitude and distribution of alternating stresses.

d) Loading h~story, including power and over speed

conditions.

Strength of the blade material is influenced by possible

corrosive effects in the environment, by mean and alternating
stress levels, by transient peak overload conditions, and by
the number of applied load cycles which the most highly-
stressed blade regions are capable of withstanding, including
any residual effects of the blade forming process. Turbine
Steam Turbine Blade Problems 455

blades may be subject to complex three-dimensional stress

conditions at their attachments to the rim of the disk. Both
the mean stresses and the alternating stresses have localized
maximum values at such locations. i.e.. at the cover-tenon
junction. at the tiewire attachments. at certain locations in
the airfoil section. at the airfoil-platform junction. and in
the blade root/disk attachment region. At each location the
mean stress value depends on centrifugal and steam loading
conditions. Alternating stresses result from steam (and
other) dynamic stimuli. from the modal response properties of
the blade group. from the degree of modal damping involved.
and from the extent of the dynamic coupling which occurs
between the stimulus and the blade group modes. Loading
history involves such factors as overspeed events. base load
operation vs. peaking operation. and machine MWe load
profile. It may also involve more subtle dynamic factors
such as circumferential pressure distribution resulting from
exhaust hood geometry. from reheat/extraction port locations.
size. arrangements. from manifold strut arrangements. water
ingestion. condenser vacuum rupture. and electrical line
switching transient conditions.

11.3 .3 Diagnosis of the Failure Mechanism

A variety of vibration sources may exist within a turbine

stage. but in most instances the resulting blade vibration
amplitudes and associated stress levels are small and
insignificant. Occasional cases are observed where the
vibration amplitudes of the blades have been shown to be
large enough to cause blade failure by fatigue. More
frequently. the blade vibration plus some other significant
factor. e.g •• corrosion. or residual stress. was also
involved. Correct diagnosis of the failure mechanism is
necessary before modifications can be proposed with
confidence. The following questions may provide useful
information when seeking the cause (s) of a given blade
failure:

a) Does the evidence suggest that failure was due to

(a) high cycle fatigue. (b) low cycle fatigue.
(c) stress corrosion. (d) corrosion assisted
fatigue. (e) erosion. (f) creep. (g) other
sources. e.g •• water ingestion?

b) What were the tangential. axial. and group natural

frequencies of the blades under operating
conditions? What were the associated mode shapes?
Is the Campbell diagram available?
456 N.F. Rieger

c) Does the Campbell diasram indicate the possibility

of resonance between any blade sroup mode and any
per-rev excitation harmonic (1x, 2x, etc.)? Which
mode is most likely to become resonant considerins
statistical scatter of natural frequency values?

d) Was resonance possible between any blade sroup mode

and any harmonic of nozzle-passins frequency (1 x
NPF, 2 x NPF, etc.)?

e) Where were the failure initiation sites located?

Is there evidence of local damage from corrosion,
erosion, impact, or other initiatins cause in that
res ion?

f) What are the operatins load and speed cycle history

details for the machine since the orisinal spin pit
provins tests? How many overspeed sovernor trips
have occurred? What speeds were reached in such
cases? How Ions were the blades at overspeed, and
could they have been resonant in this condition?

g) What was the chemical history of the steam operat-

ing conditions? Where was this sampled?

Other circumstances such as location of Wilson line in the

machine, electrical network load variation details, and unit
thermal cycling profile details may also be important. Some
known features of several important blade failure mechanisms
are discussed in the followins section. It is evident that
monitorins of speed and load to identify any transient
conditions ca» provide important diasnostic information in
such instances.

11.3.4 Types of Blading Problems

Fatisue:

Fatisue in turbine blades is broadly classified as either

high cycle fatisue or low cycle fatisue. High cycle fatisue
is often associated with a locally hish mean stress level and
moderate dynamic stresses. With hish cycle fatisue, a larse
port ion of the time to fa i1 ure is taken up with the
initiation of the fatigue crack. When a crack develops, the
stresses at the crack front are much increased, and crack
propasation usually takes place quite rapidly under the same
alternati~g blade load conditions which caused the crack to
initiate. Several known causes of steam turbine blade
failure are listed in Table 1.
Steam Turbine Blade Problems 457

Low cycle fatigue is commonly associated with fewer load

cycles applied through a much larger strai~ range than that
which causes high cycle fatigue failure. A typical LCF
failure cycle would involve strains from zero to a maximum
such as the start-stop cycling of a blade or disk due to
centrifugal force. For any turbine blade in which the local
maximum stress exceeds the material yield point during such a
load cycle. fewer of these load cycles will be needed to
cause a crack to initiate (and subsequently to propagate).
compared- with the .number of cycles and propagation rates
observed in high cycle fatigue.

Many sources of harmonic excitation exist within a steam

turbine stage. Steady harmonic excitations are continuously
applied to the blades from sources such as nozzle wake
excitations. Under resonant conditions. these excitations
may cause large dynamic stresses to occur due to the low
damping which is known to exist in most turbine blades.
Transient blade excitations of large magnitude may also
result from network electrical fault conditions at the
generator. or from partial steam admission on startup. and so
on. In many observed failure instances the blade vibration
is observed to be intermittent (from the beach marks on the
fracture surface). This tends to cause slower overall crack
growth rates. but the growth increment is often larger where
large dynamic stresses are involved.

High cycle fatigue failures can usually be recognized by

visual inspection from the characteristic pattern of lines
(called beach marks) which radiate from the crack initiation
site. The surface may be uniformly polished from the rubbing
of the crack surfaces against each other during vibration.
The surface may also be tin ted. depending on whether
corrosion has resulted from the gas/ steam environment.
Frequen_tly there is intermittent growth of the fatigue
surface. This indicates intermittent growth of the fatigue
crack. showing that crack-driving excitation was not
constantly applied throughout the blade fatigue life. See
Figure 1.

Many high cycle blade failures originate at some structural

discontinuity or stress raiser. Such failures are frequently
related to high local steady stresses. e.g •• from centrifugal
blade loading. as well as high dynamic stresses. With high
steady stresses. more moderate vibratory stresses may cause a
crack to initiate and propagate from the stress raiser. The
same conditions can also cause an existing crack to propagate
and grow until the component fails.

Low cycle turbine blade fatigue failures are frequently

associated with corrosion or high temperature. The influence
of these effects on fatigue is discussed later. Where cyclic
458 N.F. Rieger

stresses alone have led to low cycle fatigue failure, the

progressive development of the crack can often be identified
in electron microscope photographs, Figure 2.

Corrosion:

Corrosion assisted failures have occurred in the blade

attachment region and in disk attachment, as well as in the
vane, tiewire, tenon, and cover sections of the blade. Such
failures typically occur at points of high operating
stresses: the presence of dynamic stress is !Q1 required for
stress-corrosion failure to occur. However, corrosion
fatigue may occur where large dynamic stresses are applied
with high steady stress in a suitable environment. This
aspect is discussed in the next section.

Corrosive attack on blade and disk materials may arise from

chemical impurities in the steam, such as sodium and
potassium chlorides, sulfides, and carbonates. These
substances usually exist in the steam in small quantities.
Efforts a~e made to reduce chemical impurities by feedwater
treatment. The effect of even very small quantities (parts
per billion) may be concentrated by entrapment within grooves
and cracks. Where such entrapment occurs at or near high
stress regions, stress-corrosion may result. Evidence of
concentrated corrosive attack may range from general degra-
dation of the surface quality to corrosion failure of a
component. See Figure 3.

Corrosion coupled with component stress and steam or water

erosion may result in sufficient deterioration of the blade
airfoil surface to affect the operating performance over a
period of time. Corrosive buildup of deposits on blade
surfaces can adversely affect stage operating efficiency by
several percent. This has occurred in large utility steam
turbines, process turbines, and geothermal steam turbines.
Stress accelerated breakdown of surface quality is another
important practical source of turbine blade degradation.

The possibility of stress-corrosion cracking and failure of a

given component may be assessed by fracture mechanics
procedures. Visual data which suggests this type of failure
are the presence of white or gray chloride, sulphide, and/or
carbonate deposit (nodules) forming a local coating over the
surfaces of the moving and/or stationary blades. Signs of
corrosive pits (small or large) ma_y be evident especially
near known regions of high stress, e.g., notches. Additional
data can be obtaine4 from microscopic examination of the same
surfaces, Figure 3. This may reveal additional corrosive
degradation of the surface, and a variety of small and large
pits. Medium power microscope studies of sections through
the surface may reveal that the progress of the crack has
Steam Turbine Blade Problems 459

been aided by corrosive attack along the grain boundaries

( intergranular cracking). The most informative source of
such data is photographs from the scanning electron micro-
scope which reveals the presence of corrosive attack as large
nodules of corrosion products. See comments in case
histories.

A recent paper by Jonas [1] discusses the general problem of

corrosive attack on components from steam impurities. Table
2 herewith is taken from this paper. It identifies the
locations, component materials, and associated chemical
deposits observed at the sites of a variety of turbine plant
problems. Three regions are specified as most susceptible to
corrosion: (a) regions where metal or steam temperatures
are around the melting points of corrodents, e.g., NaOH T =
604°F, (b) regj.op.s immediately ahead of, or at fi,st
condensation, e.g., LP turbine stage at Wilson point
(pitting, stress-corrosion, and corrosion fatigue of blades
and disks occurs most often in this region), and (c)
superheated metal surfaces where impurities can concentrate
by evaporation and drying. Long term (24,000 hours) tests on
certain turbine steels at 150°F in a 28 percent NaOH solution
have shown that stress corrosion cracking may occur a~
stresses as low as 30 percent of the material yield strength.

Recent utility turbine research programs have begun to

develop comprehensive methods for chemical monitoring of
steam turbine plants. A useful description of appropriate
tests and water/steam monitoring requirements is given in the
above paper by Jonas.

Corrosion Fatigue:

Corrosion assisted fatigue is probllbly the major source of

steam turbine blade fatigue failures. Corrosion fatigue most
frequently occurs in a corrosive environment where high
steady s_tresses are applied together with high alternating
stresses. Turbine blade vibration tests by many investi-
gators have shown that even under conditions of nozzle
resonance, blade types which have been known to fail during
operation often do not develop dynamic stresses of sufficient
magnitude to cause fatigue without the presence of corrosion.
Such conditions suggest that some other factor must be
involved, and these failures can frequently be explained
where it can also be shown that significant corrosive attack
must also occur along with fatigue conditions in the same
operating environment. Laboratory test evidence has demon-
strated that the material fatigue strength (endurance limit)
may be immediately reduced by as much as seventy percent by a
sufficiently aggressive chemical environment. On occasions,
such chemical environments appear to have existed in, (a)
certain marine turbine steam conditions, e.g., from sodium
460 N.F. Rieger

hydroxide in the make-up water. (b) from inadequate de-

mineralizers in main utility stations. and (c) process steam
turbines. Otherwise unexplainable blade fatigue failures can
be accounted for quite readily where such circumstances can
be shown to exist. Corrosive environments may accelerate
both high cycle fatigue and low cycle fatigue if the operat-
ing conditions and chemical concentrating mechanism are
right.

Corrosion fatigue in turbine blade steels has been studied in

depth in recent years. and certain fracture mechanics data
have been given, e.g •• by Clark [2]. Figure 4 shows the rate
or crack growth da/dN vs. stress intensity factor AK for a
304 stainless steel in a three percent caustic environment.
Crack growth for the same steel in air is also shown for
comparison. It is seen that the difference in crack growth
rates for the same AX: value. i.e •• stress level, is
approximately 3:1 for the caustic environment. this is
another way of stating that the fatigue strength of the test
components under the corrosive attack shown in this instance
was only about one-third of the fatigue strength of the same
components in air, i.e •• seventy percent reduction as noted
above.

An important initial source of corrosion fatigue may develop

on site when rotors are left exposed to an aggressive
environment without other protection for a considerable
period prior to erection. Particularly damaging is the case
where the protective coating has been removed from a rotor
and blades. which are then left exposed to the moist, outdoor
environment near a river or even the sea. The initial
chloride pitting which may occur in such circumstances may
later provide nesting sites for steam, etc •• impurities in
high stress regions which can accelerate the tendency toward
blade fatigue.

Erosion:

Surface erosion can be a significant problem in all stages of

a turbine. Surface erosion from hard particles (usually
boiler exfoliation) can damage the HP and IP stage blading,
and wet steam erosion can damage the leading edge of the long
LP blades, usually from mid-height out to the cover. Cover
damage from wet steam erosion can also be significant. Such
erosion is caused by high velocity particles of steam
condensate striking the blade lead edge over a period of time
and eroding the material away. Certain wet region blades
have been designed for many years with an erosion shield
(stellite strip) which is bronze welded on to the blade lead
edge to protect against wet steam erosion.
Steam Turbine Blade Problems 461

Signs of early erosion may often be observed on those blade

leading edges which project noticeably out into the incoming
steam, beyond the other blades in the same row. This
condition may occur from minor misalignment on assembly, and
this erosion is of no special significance unless i~
continues and presents an evident major damage problem.
Replacement of damaged erosion shields is a straightforward
procedure which can now be undertaken in the field, as well
as in the manufacturer's shop. Suitably located LP moisture
separators also help to decrease the rate of blading erosion.

Exfoliation of tube scale is another form of erosion which

occurs in boiler tubes, superheater tubes, inlet steam pipes,
and from condenser pipes. The scale develops from oxidation
and corrosive attack from the feedwater condensate and from
steam impurities. The scale is eroded away, and on passing
the turbine, may damage the blading and may accumulate in the
drains. Geothermal turbines are especially prone to scaling
and exfoliation damage because of the high corrosive and
impurity content of the inlet steam. Erosion products should
be monitored as part of the turbine system chemical monitor-
ing program.

11.3.5 Stress-Related Blade Group Cracking Theory

The following theoretical approach is general and may be used

to develop an understanding of possible causes for specific
cases of blade cracking which appear to involve fatigue,
corrosion fatigue, or stress corrosion cracking. This method
follows from the approach which was pioneered by Prohl in
reference [3].

Consider a group of blades rotating in an axial flow turbine,

Figure 6. The total stress at any location is due to two
sources, the steady mean stress a and the dynamic or
alternating stress a • At any instantmduring operating where
a single harmonic rfsponse component predominates, the total
stress is given by:

at = am + a a cos wt

where w is the circular frequency of the alternating stress.

The mean stress results from the combined action of the
centrifugal load due to turbine rotation, and from the steady
blade bending load from the steam forces which drive the
462 N.F. Rieger

turbine. 1 These mean stress components combine to give the

nominal steady extreme fiber stress, a at the blade (or
mo
root section):

a
mo = aco +abo
= P/A + .MC/1

where P is the centrifugal load from blade rotation acting at

the section, A is the section area, .M is the local bending
moment due to the steam load on the blade, c is the extreme
fiber distance from the neutral axis, I is the appropriate
second moment of area of the blade section, and a and a
are the nominal centrifugal and bending stress'&~ at tkg
section (no stress concentration effects). The mean stress
remains constant for a given _blade arrangement at the
specified speed and power output. Corresponding expressions
may be wri~ten for steady stresses in the blade group cover
and tenons.

Alternating stresses in the blade may arise from several

causes, of which harmonic excitation from the nozzle wakes is
widely recognized as one potentially significant contributor.
The frequency of excitation f from the nozzle wakes is given
by:

f = Nk 2nw cycles/sec. (Hz)

where -N is the rotor speed rev/sec and k is the number of

uniformly spaced nozzles2 around the 3600 circumference~
Alternating stresses from nozzle excitation depend on several
factors:

a) The magnitude of the steam exciting force or

hl!ormonic stimulus, expressed as a stimulus factor
s.
b) The damping of the blade and its attachment
expressed as a logarithmic decrement &.

!Additional forces from torsion, centrifugal untwist,

etc., may of course also apply.

2k may also be considered as the number of per-rev

harmonic waves around the diaphragm circumference. This
allows per-rev nozzle excitations to be considered.
Steam Turbine Blade Problems 463

c) The response factor K which is a measure of the

ability of the blade group to accept energy input
from the nozzle stimulus.

Blade harmonic stimulus is usually expressed as a proportion

of the steady steam bending force F acting on the blade.
i.e •• S = AFIF. where AF is the amplitude of the time-varying
steam force. In practice. values of S may range from below
0.02 in smooth-running stages to above 0.20 in rough stages
with off-optimum conditions. See references [4] and [5].

Damping in blade groups can arise from several sources such

as rubbing friction in the attachment areas (root. cover).
from material_ hysteresis and from gasdynamic effects on
longer blades. The magnitude of blade group logarithmic
damping value & may vary substantially from one appl_ication
to another. but the general range is from about 0.002 to
0. 030 for conventional AISI 403. 12 chrome steam turbine
blades. depending to some extent on blade geometry and the
mode of vibration involved. See references [6] and [7].

The resonant response factor K depends on the excitation

parameter E = (nq/m). where n is the harmonic number (n = 1.
first order: n = 2 second order. etc.). q js the number of
nozzle inlets and m is the number of blaftes. The variation
of the response factor K ranges from 1.0 under conditions
where the blade group can receive strong energy input from
the nozzle stimulus down to zero where the phasing of the
stimulus is such that the blade vibrations are not readily
excited.

As each turbine stage may develop many excitation harmonics.

and each harmonic may act on several blade group modes. the
influence of the excitation parameter on each blade group
response in the frequency ranges of interest should be
examined. A convenient procedure for determining the
excitation harmonics and blade modes of interest was first
given by Campbell [8] in which the natural frequencies of the
blade modes are plotted as ordinate and the rotor speed is
plotted as abscissa, see Figure 9. Radial lines from the
origin corresponding to once-per-rev (1x), twice-per-rev
(2x). etc.. nozzle passing frequency (Nk). twice NPF (2nK).
etc •• are also plotted. Speed regions of intersection
between blade frequencies and excitation harmonics are then
noted. with particular reference to regions of sustained
operation, e.g •• operating speed range. The Campbell diagraa
shown in Figure 9 indicates the possibility of blade
resonance in the axial torsional mode with the 2x NPF
excitation. and also excitation of the secon4 type tansential
mode by NPF. See Case History 1 for details.
464 N.F. Rieger

Resonant stresses are related to stimulus S, damping &, and

blade group dynamic response factor K, by the expression:

where a is the nominal resonant alternating extreme fiber

stress ~~ the blade (or root) section, and the ab is the
nominal bending stress at the section, defined prfviously.
It is evident that the practical combinations of blade
damping, nozzle stimulus, and dynamic response factor may
lead to resonant stresses a at certain blade cross sections
which could approach or a~reatly _exceed the nominal mean
bending stress ab at that section. It should further be
noted that in _p~actice, the resonant condition is often
sharply defined. Sustained operation at the resonant peak
condition is therefore unlikely to occur for long periods,
though some lesser stress magnification should always be
expected for operation in this region.

To determine whether the stress conditions at a given

location could be re spons ib le for blade cracking during
operation, it is necessary to compare the local stresses with
the appropri~te strength criterion for the blade material at
that section. To obtain an indication whether the nominal
stresses a and ab are likely to initiate high cycle
fatigue crac~ing, a pr8cedure due to Heywood [9], and. adapted
by Rieger and Nowak [10] for steam turbine blades using the
Goodman diagram may be used as follows. The appropriate
fatigue life envelope for unnotched specimens in the steam
environment is modified in a specified manner to account for
mean stress.. local stress raisers, . cycles to failure, and
size effect. The new (notched, etc.) fatigue envelope then
becomes the crack initiation criterion against which the
nominal stresses a and a are compared. A point falling
outside this regio~0 can bea 0 expected to initiate a crack in
the number of cy?les assumei in the c!lculation 7of the
notched envelope, 1.e., N = 10 • • • 10 • • • 10 , etc.
For high cycle fatigue, crack initiation commonly represents
the larger portion of the fracture life, and the time to
propagate the crack (which is not cons ide red in _this
approach) represents the remainder of the component life.

Where the initial defect size is known from inspection or can

be assumed, an alternate approach using fracture mechanics
procedures may be used to estimate the number of load cycles
required to propagate a crack, and cause failure of the
defective component. Suitable materials data obtained from
tests· on fracture mechanics specimens within a similar
environment and loading, in accordance with standard ASTM
Steam Turbine Blade Problems 465

testins procedures, is required. The rate of crack propa-

gation da/dN is related to material properties A, n., stress
intensity range AX:, and applied stress ratio by the
expression:

da A [AK]n
-=
dN U-R> 0 • 5

is the range of stress intensity factor = cAa na

A, n are material properties

c is a geometric factor for the crack model involved

Aa is the stress range, amax - amin

a is the crack length

N is the number of stress cycles

For a crack in a notch-free region:

a ao

Aa = 2aao

For a crack in a sharply notched region:

amax = ~tamo + Kata ao

466 N.F. Rieger

a
A.a = 2Ktaao

a and a i are the maximum and m1n1mum values of the total

s'f~\ss atm he location in question. A typical relation
between da/dN and AI for a 4340 steel in a three percent
caustic solution is shown in Figure 4. This chart does not
include the effect of stress ratio R. In most cases the.
stress field in the body changes with distance into the body.
This causes the stress at the crack tip to change and so
influences the rate of crack propagation. Calculations which
must include the effect of the above factors are most
conveniently performed with a suitable fracture mechanics
computer program, such as the CRACKS [11] or BIGIF [12]. The
end result of such a calculation is a value for a specific
number of cycles to propagate the crack to failure when Kmax
= K1 c (fatigue) or KI.SCC (corrosion fatigue), or until a
length is reached at wli1cn the crack stops propagating.

This fracture mechanics procedure is suitable for a time-to-

failure analysis at any location in the blade group with a
crack of known (or assumed) proportions, and for any material
for which suitable fracture mec4anics data is available for
the blade operating environment. Where such input data is
difficult to obtain or specify, the above procedure may be
used to provide a bounding analysis suitable for determining
the performance of a hypothetical crack under assumed minimum
or maximum stress and material conditions based on
experience.

11.3.6 Case Histories of Blade Failures

Case 1 Nozzle Resonance of HP Marine Turbine Blades [13]:

Both rotors were 55,000 SHP turbines operatin~ at 3500 rpm.

The blading of both HP rotors sustained damage. The complete
blading of the ninth stage of the starboard turbine was
missing and eight blades were missing from stage 10. On the
port rotor, seven blades were missing from the ninth stage
and there was some cracking in stages 8 and 10. The cracks
occurred near the vane-platform junction where the vane
overhung the platform. In most instances the cracks appeared
to have propagated from beneath the overhung trailing edge,
horizontally into the blade airfoil section, Figure 7 (a).
One blade only was broken at mid-height, in the eighth stage
of the port turbine.
Steam Turbine Blade Problems 467

A comprehensive investigation was made of the failure.

Frequency calculations and vibration tests were performed on
the original blades, and on the modified blades. Much
evidence was found to show that the failures were due to high
cycle fatigue from vibrations in the second type tangential
(out-of-phase) mode, Figure 8 (a). The cracking pattern
corresponded to the calculated distribution of modal ampli-
tudes shown, both in magnitude and location. The Campbell
diagram Figure 9 showed that the eighth, ninth, and tenth
stages could resonate at propeller shaft speeds between 90
rpm and 174 rpm. The ship operating log, Figure 10, showed
2610 minutes of operation at 148 rpm. In this condition, the
ninth stage could resonate in its second tangential mode.

Several remedial changes were included in the blading

redesign. The overhung stress raiser was eliminated by
smoothing the vane into the platform of the replacement
blades, Figure 7 (b). Tiewires were added by brazing to
eliminate the second tangential mode from the range of nozzle
resonance, Figure 8 (b). Diaphragm changes (decreased number
of nozzles) were also considered, but were not used because
such changes were shown to be eff~ctive in removing the
eighth stage alone from resonance. Also, larger nozzle
passages frequently lead to higher exciting forces.
Alternatively, an increased number of nozzles might have been
considered.

Case 2 Nozzle Resonance in Ninth Stage of Process Steam

Turbine:

A rash of shroud and vane cracking incidents had occurred in

several process turbine drive units. Generally the shroud of
a six-blade group had cracked, and also several vane sections
near the platform in the eighth stage. The blade surface
adjacent to the failure site was pitted, and white, solid
deposits of unknown type (of contaminant) were attached to
the blade surface. A diaphragm change from thirty-four inlet
nozzles to forty-six inlet nozzles had increased the blade
life somewhat (six weeks to forty weeks), but had not
eliminated the problem.

Analysis showed that the second in-phase tangential mode of

the blade group coincided with NPF at full operating speed
using the thirty-four nozzle diaphragm. Further, the
Campbell diagram, Figure 11, showed that although a change
from thirty-four nozzle openings would cease to excite the
second in-phase tangential group mode with the second NPF
harmonic (2 x 234), the (1 x 46) NPF harmonic would then
excite several second type tangential (out-of-phase) modes.

The fix proposed in this instance was a blade profile re-

design which removed the blade groups from the troublesome
468 N.F. Rieger

resonance harmonics within the operating speed range

identified above, without introducing other resonance
problems. The blades were also de tuned as sho~ in Figure
12, so that the excitation factor E would be 0.183 an(! the
corresponding response factor K would then become zero. In
the original design, E was 0.352 and the resonant response
factor K had been 0.275. The new blades were therefore less
responsive to nozzle stimulus.

Other factors requiring attention were the chemical content

of the steam and the location of the Wilson line (dry/wet) in
the turbine. Suitable de-mineralizers should have been pre-
sqr:i.bed and thoroughly maintained in view of the extreme,
i.e., coated corrosion fatigue situation which existed. If
the Wilson line corresponds to the ninth stage, care should
be ta~en to shield the blade attachment regions in some
manner. The chemical functioning of the turbine steam system
should have been monitored following the re-installation to
ensure that solids and impurities were within acceptable
limits, see Jonas [1].

An apparent alternate fix would have been to de tune the

blades using a bronze welded tiewire, as was done in Case 1.
This would suppress the out-of-phase tangential modes, and
the forty-six nozzle diaphragm would not excite the second
in-phase tangential mode, as noted. This was not do~e as the
blade required re-design to remove the vane overhang.

Case 3 Stress Corrosion in a Fifth Stage 200 lfw Utility

Turbine:

Catastrophic rupture occurred in the blade root section of

thirteen fifth stage axial entry blades after eleven months
of on-line operation, with considerable consequential damage
to flow guides and to blades in adjacent stages. The failed
row contained 324 moving blades, each about six inche$
average vane height, arranged in groups of five and six.
Blade material was 403 stainless steel with nominal UTS 105
ksi, and yield stress 85 ksi in air. The adjacent inlet
nozzle row had 240 nozzles. Extensive white-colored chemical
deposits with average ph value of eleven, coated the general
resion (nozzles, moving rows), near the failed stage, Figure
13. Pitting in t~e remaining blades of the row ranged from
sligh,t to severe. 325 additional cracks were found (173
blade, 152 disk steeples) of varying sizes,_ mainly in the
contact hook regions of the blade attachments. The location
of the row coincided wi~h the location of the Wilson line of
the rotor at full power.

Metallographic examination showed branched intergranular

cracking followed by transgranular cracking inward from th~
highly stressed blade root notch surface, Figure 14.
Steam Turbine Blade Problems 469

Scanning electron microscope studies showed pitting in the

vicinity of the primary fractures and_ secondary intergranular
fractures linking the corrosion pits. The fracture surfaces
were relatively clean, indicating that little rubbing or
polishing had taken place since cleavage occurred. This
suggests that no predominant HCF or dynamic stress mechanism
was involved. The high ph white coating was composed of
NaOH, and Na 2 co 3 • This indicates that the initial cracking
had been asusted by corrosion in the highly stressed hook
region of the blade root. Furthermore, the original material
away from the corrosion sites still had the strength and
impact properties required in the original material speci-
fications. A typical SEM photograph of the fracture surface
is shown in Figure 15.

Further investigation revealed that the boiler feedwater

chemistry during operation had contained dissolved solids,
iron, and sodium (77 ppb compared with 30 ppb specified) in
excess of prescribed limits, despite the use of feedwater de-
mineralizers. Stress corrosion of the blade root material
under high stress conditions was the primary cause of this
blade failure, based on (a) high sodium and other salt
deposits, (b) wet/dry Wilson line at failure location, (c)
widespread cracking in the vicinity of the high stress
locations, (d) general pitting of adjacent surfaces, (e)
absence of plastic flow and beach marks on the failure
surface, (f) corrosion products seen in many SEM ~cans, and
(g) corrosion fractography observed in sectioned failure
regions. The principle remedy was improved steam quality and
boiler feedwater chemistry by improved de-mineralizer control
and monitoring.

Case 4 Fatigue Failure of Fourth Stage Marine LP Turbine

Impulse Blades:

A single side entry fourth stage blade failed catastroph-

ically in the root section at the first hook after thirteen
months of service at sea. On inspection it was found that
eleven additional fourth stage blades were also cracked in
the same region, and that all cracked blades were end blades
of seven-blade groups. In addition, a total of eight disk
root sections were found to be cracked in the same stage.
These failures were found by magnetic particle inspection.
Hardness and chemistry checks of the blade and disk root
sections were also made. Hardness was found to be in the RC
20-21 average, as required. Macro-etched sections of the
cracked root were carefully inspected at 20x magnification.

The failure mechanism was found to be high cycle fatigue, as

indicated by the beach mark progression of the crack front
along several regions of the failed surface. The multiple
crack origins indicate a wide distribution of the initial
470 N.F. Rieger

crack driving stress mechanism along the root hook notch. No

oxides 9r corrosive deposits were observed on the fracture
surface. Microexamination showed that the root cracks wer~
relatively straight and transgranular without branching.
This further snggests that pure fatigue was the cause of
failure, though no electron microscope studies were made on
this occasion.

Several possible causes of these fatigue failures were

identified. First the tangential out-of-phase group modes
were found to lie close to nozzle resonance, due to an
inaccurate design estimate of blade root stiffness. Second,
the root fillet radius of 0.031 inches was quite small, and
this gave rise to magnified fillet steady and dynamic
stresses. Third, variations in nozzle geometry from stage
arrangement around the nozzle row were foun4 to give 4:1
variations in the magnitude of nozzle stimulus. This raises
the additional possibility of per-rev stimulus from this
nozzle geometry variation, but this possibly was not
considered further. Fourth, a small amount of stress
corrosion may have occurred judging from minor discoloration
observed on the crack surface • Fifth, a further source of
significant excitation was thought to have occurred from four
condensate extraction ports located around the circumference
adjacent to the fourth stage moving blade row.

Design modifications were made as follows: (a) long-arc

shrouding was introduced to suppress the troublesome resonant
group modes, (b) the notch fillet radii were increased to
0.060 inches to reduce the fillet stresses, (c) the inlet
nozzle geometry was made uniform around the diaphragm and the
number of inlet nozzles was increased from 92 to 120, and (d)
a flow-smoothing baffle was inserted to remove the_ flow
disturbances created by the four extraction openings. No
further failures have occurred since the introduction of
these modifications.

11.3.7 Conclusions

o Turbine blade problems may result from design, manu-

facture, materials properties, steam/gas quality
(erosion), steam/gas chemistry (corrosion, exfoliation),
and abusive operation (water extraction, condenser
flooding). Design practices, operating practices, and
turbine plant spec ificatio~s should address each of
these potential problem areas.

o The primary diagnostic tools for analysis of turbine

blade failure causes are:
Steam Turbine Blade Problems 471

a) Electron microscope to investigate failure

mechanism and identify the role of corrosion in the
failure.

b) Surface microscopy and section microscopy for

defining the cracking mechanism, and for basic
material quality assessment.

c) Water/steam chemistry records to determine role of

corrosion.

d) Blade group and disk natural frequencies, modes and

static/dynamic stress calculations, to determine
role of operating stresses.

e) Fracture mechanics testing of failed component

material to determine quality of material supplied,
corrosion res is tanc e, and crack propagation
characteristics in operating environment.

o The failure surface and fracture sections should always

be examined microscopically when the causes of a failure
are being sought. Examination by Scanning Electron
Microscope is a valuable additional aid for determining
whether the failure is associated with corrosion,
corrosion fatigue, or fatigue alone.

o Each stage of blading should be checked as to whether

nozzle resonance and per-rev resonance may occur during.
operation. A Campbell diagram should be developed
containing the first six modes of the blade group for
each stage. The calculated modes and frequencies should
be checked by vibration tests performed in the manu-
facturer's shop, on several blade groups around the
circumference of each stage.

o High cycle fatigue is usually related to some resonant

operating condition. HCF may be identified by the
presence of polishing, beach marks, crack propagation
increments undet SEM, and final static rupture on the
failure surface. Multiple staining lines indicate the
occurrence of intermittent crack propagation from short
periods of high dynamic stresses.

o Corrosion fatigue may be induced by concentration of

steam impurities acting on high stress regions in the
presence of dynamic stresses. The source of such
corrodents may be in the steam itself, in the steam
chemistry control apparatus, in the steam chemistry
specifications, or in the original turbine erection
environment.
4 72 N.F. Rieger

o Erosion of turbine blades and stage inlet guides may

occur from boiler and tube exfoliation, and from wet
steam impact. Such erosion may lead to performance
degradation, and to degradation and failure of the
working components.

11.3.8 References

1) Jonas, I., 'Turbine Steam Purity,' Combustion Magazine,

p. 11, December 1978.

2) Clark, Jr., W. G., 'Evaluation of the Fatigue Crack

Initiation Properties of Type 403 Stainless Steel in Air
and Steam Environments,' ASTM STP 559, ASTM, pp. 205-
224, 1974.

3) Prohl, M. A., 'A Method for Calculating Vibration

Frequency and Stress of a Banded Group of Turbine
Buckets,' Trans. ASME, pp. 169-180, January 1958.

4) Heyman, F. J., 'Turbine Vibrations Due to Nozzle Wakes,'

ASJIE Publication Paper Number 68-WA/PWR-1, Presented
December 1-5, 1968.

5) Rieger, N. F., Wicks, A. L., 'Non-Steady Force Measure-

ments and Observations of Flow in Three Turbine Stage
Geometries,' Journal of Engineering Power, Vol. 100, p.
525, October 1978.

6) La zan, B. J., Damping of Materials and Members in

Structural Mechanics, Pergammon Press, Incorporated, New
York, 1968.

7) Wagner, J. T., 'Blade Damping Tests,' Westinghouse

Engineering Report EC-401, NOBS N00024-67-C-5494, May
1969.

8) Campbell, W., 'Tangential Vibration of Steam Turbine

Buckets,' Trans. ASME, pp. 643-671, 1925.

9) Heywood, Desianin& Against Fatigue of Materials,

Reinhold Publishers, London, England, 1962.

10) Rieger, N. F., Nowak, W. J., 'Analysis of Fatigue

Stresses in Turbine Blade Groups,' Workshop on Improved
Turbine Availability, Electric Power Research Institute,
Palo Alto, CA, January 1977.

11) CRACKS: 'A FORTRAN IV Digital Computer Program for

Crack Propagation Analysis,' USAF Flight Dynamics
Laboratory, Wright-Patterson AFB, Dayton, OH, March
1970.
Steam Turbine Blade Problems 473

12) BIGIF: 'Fracture Mechanics Code for Structures,'

Program Users Manual, Failure Analysis Associates, Palo
Alto, CA, December 1978.

13) Fleeting, R., Coats, R., 'Blade Failures in the HP

Turbines of R.M.S. Queen Elizabeth 2 and Their
Rectification,' Trans. Marine Engineers, London,
England, October 28, 1969.
474 N.F. Rieger

Figure 1 Fatigue Surface with Staining Pattern

Figure 2 Electron Micrograph Showing Fatigue

Striations in 4340 Steel

Figure 3 Corrosion Pits on Root Hook

Near Fracture Initiation Site
Steam Turbine Blade Problems 475

• DATA IN AIR a DATA IN 3%

NaCl SOLUTION
AT 6 CPM

da
dN
INCH PER
CYCLE

° 10
1
-6~~+-~~~~~~
2 4 6 c 10?
.6.K KSiffN'

Figure 4 Crack Growth Data for 304 Steel in Air

and Three Per Cent Solution

Figure 5 Medium Pressure Blade Profiles

in Axial Flow Turbine Stage

Second Tanqential Second Axial-torsional

Tang. Group (typ.) First Axial-torsional

llil1
First Tangential
\CCC&~
UUt
First Axial

Figure 6 Mode Shapes of Turbine Blades in Groups

476 N.F. Rieger

Redesign

Vane
section

Original
Design Crack Overhang
origin removed
in redesign

FIGURE 7. DETAIL OF FAILED BLADE AND MODIFIED

REPLACEMENT BLADE, VANE OVERHANG REGION.

(a)

(b)
Figure 8 Second Type Tangential Group Mode (a)
Second Type Tangential with Tiewire (b)

Figure 9 Campbell Diagram for Fatigued

Marine Turbine Blade
Steam Turbine Blade Problems 477

Figure 10 Turbine Log. Times at Various Speeds

6
znd AXIAL

2
34 NOZZLES

:J

SPEED (1000 RPM)

Figure 11 (a) Campbell Diagram for Failed Blades

478 N.F. Rieger

5
N
Q
-4

~-+--~--+---+ 1>1 AYI AL

l---1-'--~:...._--+---+ IS l ~ 'I'; ,

11, ' R'J •

.. of r r.
0
0
'
)Pe ED ' I' · , ,,;'I'

Figure 11 (b) Campbell Diagram for Replacement Blades

K
.2
~
rlt"W (
SL A ~Jl
I

.r .·,

Figure 12 Resonant Response Factor Re-Design

Figure 13 Blade Vane Sections Showing

White Chemical Coatings
Steam Turbine Blade Problems 479

Figure 14 Crack Origin and Propagation From

Stress Raiser into Component

Figure 15 Electron Micrograph Showing NaOH

Nodules on Surface of Crack

Figure 16 Multiple Origin Cracking

480 N.F. Rieger

Figure 17 Crack Path - Straight, Intergranular

(a) Eight-Per-Rev Harmonic Excitation Around

Nozzle Circumference

(b) Long-Arc Shrouds Each Spanning One

Excitation Wave at all Positions
Figure 18
Steam Turbine Blade Problems 481

...
I _L_:
rLLL.£ '////, rr//L rLLLL r..tlliLI.J
I I

/:'
.....

I
Goo()
'

-
, I
i
I '
..; 44100 I 1
~-
:r: --:
>, li'. 'fJ. r ~
u
A'u'
I _I
c:
Q)
::> I !
aQ)
'LL.L
.... 2ooo
--1
iL I JAX'
LL

1/ A!
0 T1.
0 1000 ZJXIO -
Speed RPM. "* &1100

Figure 19 Campbell Diagram, 4th Stage, 96

Nozzle Diaphragm - Failed Row

Table 1 Possible Causes of Turbine Blade Failures

FAIUJIE
IIECRANISJI SOURCE LOCATION

Fatiaue Unsymmetrical staae General

flow
Nozzle resonance
Partial admission
Torsional transients
Neaative sequence
currents
Excessive condenser
pressure
Corrosion Excessive corrosion General
aaents in steaa/
feedwater
Concentration aechanisa Wilson Line
Corrosion Corrosion plus General
Fatiaue vibration source
Erosion 'fet Steam LP stases
Erosion shield.
cover
Wilson Line
Ash Deposit Coabustion residue in Blade lead edae
aas streaa
Boiler. pipewall BP staaes
ezfol iat ion
later Moisture separators Adjacent LP
Inaestion Condenser staae•
482 N.F. Rieger

Table 2 Industry Experience Stress

Corrosion and Corrosion Fatigue
PART MATERIAL Y.S.,KSI ENVIRONMENT

Pipe expansion Inconel 600 35 Melted caustic

joints 304SS 30 611 to no•c
HP turbine bolts Re 26 High sodium
Pyromet 860 in deposits
Steel
Incoloy NaOH-NaCI in deposits
LP stationary Welded 304 Na 1 K~Clidentified in
blades craci:s
LP stationary 304 30 Caustic and chlorides
blades
IP rotating blades Re 26 Caustic in steam and
deposits
LP rotor lCrlJolo 1/4 V 87 Caustic carry-over,
140°F 3 psi at satur-
ation
Wheel dovetails NiCrMoV 97 Caustic in deposits
in boiler feed
pump turbines
Shrunk on wheels lCrl/2 Mo 123 Caustic wash
Shrunk on discs 3Cr 1/2 Mo 117 Up to 215 ppm of sodium
hydroxide in boilers
Small turbine rotor CrMoV 126,133 Caustic and chloride in
discs deposits
HP inner cylinder
horizontal joints
HP rotor dovetails CrMoV Hydrogen sulfide (229
ppb in condensate) from
sodium sulfite treat-
ment
LP rotor NiMoV
HP discs NiCrMo 117 Caustic from demineral-
izer in deposit
HP blade pins Re 26 High caustic in
feedwater, steam and
deposits source
demineralizers
HP seal springs Inconel 718
HP bolts AISI 4130
LP discs NiCrMoV
HP inner cylinder, CrMoV NaOB, NaCl in deposits
horizontal joints Source: demineralizers
Two LP discs (ESCOM) 2.. NiCrMoV 114
Four LP discs (SEVC) 2.. NiCrMoV 114
Two LP discs NiCrMoV 130 Caustic suspected
Steam Turbine Blade Problems 483

Table 2 Industry Experience Stress

(cont') Corrosion and Corrosion Fatigue

PART MATERIAL Y.S.,KSI ENVIRONMENT

Last minus one 12Cr, 17-4 PH Chlorides and sulfates

rotating blades in deposits
(fossil turbine)
Single cylinder of Cast iron High oxygen and a small
turbine sulphuric acid from
hydrolyzed Na 2 so 4 plus
oxygen
LP blades 12Cr Mixture of organic
acids and acetic acid
First row LP 12Cr
LP discs and blade Low alloy steel
fastenings
LP blades and Stainless Steel Organic acids (acetic,
shrouds (magnox propionic, butyric)
reactor unit) from surface cooling
water organic
compounds (humic
acids)
Last row LP 12Cr hardened Inorganic acids in
rotating blades steam, ferrous
chloride and sulfate
in deposits
Stationary blades Hydrochloric acid in
steam (seawate;:
cooling + powder)
 CHAPTER 11.4

AN IMPROVED PROCEDURE FOR COMPONENT LIFE ESTIMATION

WITH APPLICATIONS

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

A general procedure is described for the calcu-

lation of fatigue initiation life. This procedure
was developed for the investigation of component
fatigue failure, but it is also suitable for the
design of new components. The procedure incor-
porates the effects of steady stresses, dynamic
stresses from excitation spectra, and material
cyclic properties. The Rainflow Cycle Counting
procedure is used to incorporate complex response
waveforms, and the Local Strain Approach is used to
determine the amount of fatigue damage from each
harmonic component. The resulting cumulative
damage is evaluated using Miner's law, to provide
an estimate of component life under given stress
and thermal load conditions. Several examples
which describe the application of this procedure
are given.

11.4.1 Introduction

This paper describes a comprehensive procedure for estimation

of component fatigue initiation life. By initiation life is
meant the period between the first loading of the new
component through the initiation of a fatigue crack at the
location under investigation. This period is here referred
to as crack 'initiation life,' and it is distinguished from
crack 'propagation life,' in which the crack propagates until
component failure occurs by rupture. The purpose of the
paper is to explain this initiation life procedure in detail,
and to show that all major factors which influence fatigue
crack initiation are included in the calculation sequence.

This procedure was originally designed to meet the needs of

component failure investigations, through a comprehensive,
computerized procedure which gave improved estimates of
486 N.F. Rieger

component initiation life under practical operating

circumstances. During such studies, information on material
properties can often be obtained only approximately from
records, and possibly by the testing of coupons from failed
components. During failure investigations details of
relevant operating circumstances are usually obtained from
operating records and from discussions with plant engineers.
This life analysis approach therefore does not contain any
in-built factor of safety, because the requirement is for an
accurate estimate of the life of the failed component. The
same procedure can be used under design conditions to provide
'true' life values, within current limits for such estima-
tions. This procedure can then be used to design a life
factor-of-safety into the component, through design
adjustments which give lifetimes suitably beyond the required
service life. Any difference between required life and
anticipated service·life then becomes the factor-of-safety of
the design.

It may be argued that the well-known Goodman diagram

procedure leads to similar results, and is therefore adequate
for design purposes. The conventional Goodman diagram is
however, limited to constant life characteristics, and to
single component harmonic stresses. The method proposed in
this paper is able to address the problems of life estimation
in much greater detail, because it includes the following
factors into a single life result:

a) Multi-component vibratory stresses are included

directly using :Rainflow cycle counting to prepare
the cyclic data for Miner's law, or other multi-
component cumulative damage criterion.

b) The proposed procedure is based on the Local Strain

method. Strain life data can be obtained for
actual material samples under precise strain
loading.~ and temperature and chemical environment
control. Life tests can then be run to many times
the anticipated lifetime.

c) Strain-life curves are developed using test

specimens and loads which simulate actual stress
levels, :R-ratios, and temperatures experienced by
the component during service life. These strain
life results replace the conventional S-N curves,
and they can include creep effects if needed.

d) Goodman envelopes are often fot available for

fatigue lives of less than 10 cycles. Further-
more, such data is invariably stress-based rather
than strain-based, and may involve wide scatter
without strain-controlled testing. Stress-based
Procedure for Component Life Estimation 487

procedures also lead to complications when the

calculated elastic stresses exceed the material
yield point. Goodman diagram data for high
temperatures is only available in rare instan~es.

The fatigue life procedure described in this chapter is

presented with definitions of the major terms involved. For
convenience i t is assumed that results for both steady
stresses and dynamic stresses are available in sufficient
detail from finite element calculations. in order to commence
the initiation 1 ife procedure. Such stress calculations
should address all major loading aspects. including component
validation tests. anticipated service load (and overload).
and also temperature influences where these affect material
properties and stresses. Elastic s1;resses are adequate for
this purpose, as will be described. Several examples from
practice in which this procedure was used are described.

11.4.2 Notation

A Cross-section area under uniaxial load

A0 Cross-section area at zero load

b Plastic strain life exponent

c Elastic strain life exponent

D Damage

D. Damage under given loads

J
E Modulus of elasticity
,
K' Cyclic strength coefficient af'/af'D

Kt Theoretical stress concentration factor

Ka Stress concentration factor

K Strain concentration factor

8

Nf Cycles to crack initiation

n.1 Number of cycles at strain level

Ni Cycles to cracking at strain level

2Nf Stress reversals to crack initiation

n' Cyclic strain hardening coefficient b/c

488 N.F. Rieger

P Applied steady load

q Notch sensitivity factor

R Ratio ami n /a max

Aa Stress range

Aa Strain range

af True fracture ductility

at' Fatigue ductility coefficient

ae Elastic stress: from linear F.E. calculations

af True fracture strength

am Mean stress

aT Total stress

af' Fatigue strength coefficient

am Mean strain

aT Total strain

11.4.3 Definitions

Engineering Stress

True Stress

a = P/A = a0 (1 + a0 )

Engineering Strain

a0 = a0 IE = A1/1

True Strain
Procedure for Component Life Estimation 489

True Fracture Strength

af = Pf/A (plus correction for triaxiality)

True Fracture Ductility

Half Cyclic Strain Range

1/2Ae = 1/2Ae P + 1/2Ae

e

Half Plastic Strain

Ae (Aa/2K') 1 /n'
p

Half Elastic Strain

Ae e = (Aa/2E)

Half Elastic Cyclic Stress

1/2Aae = af' (2Nf)b

Half Elastic Strain

Half plastic Strain

R-ratio

R = amin I amax

Notch Sensitivity

q = (K f - 1)/(K t - 1)
490 N.F.Rieger

Fatigue Factor

Kf = {1 + (Kt 1)/(1 + a/r)}

a = constant, r = radius

Neuber's Rule
2
(ae = aEa)

11.4.4 Previous Work

The procedure described herein is original, though portions

of it have been described previously by several authors [1]
[2]. Development of the Local Strain method is primarily due
to Morrow, Socie, and Dowling [3] (4] [S), though many other
studies and developments have been made concerning this
procedure in recent years. The Rainflow cycle counting
procedure has been described and discussed by Dowling [6]. A
recent assess~ent of the current accuracy of the procedure is
given in [7]. Valuable studies of cumulative damage have
recently been repo~ted by Conway, Stentz and Berling [8], and
by Battacharya [9].

11.4.5 Factors Affecting Component Life

Steady stress increases the local strain level at the

location under study, and load cycling (dynamic stress)
occurs above and beyond this strain level. Stress cycling
about the steady stress value within the elastic region dis-
sipates relatively little ene-rgy, and leads to relatively
little cyclic damage. Stress cycling about the steady stress
va 1 ue in the plastic reg ion leads to greater energy
dissipation, and to greater cyclic damage and loss of fatigue
life.

Dynamic loading may be classifjed as (a) simple harmonic,

(b) general periodic, e.g., saw-tooth waveform, (c)
transient or impactive, (d) spectral or random. General
dynamic load spectra may contain any or all of the above
types of loading. In practice, component loadings obtained
from testing are commonly presented as frequency spectra
containing multiple harmonic components.

D;ynamic strain causes the cyclic damage which can lead to

crack initiation. Cyclic strains may result from low cycle
loading, e.g., machine start-stop sequence, and from high
cycle loading, e.g., forced vibration. Cyclic strain damage
can often be observed from electron micrographs as a pattern
Procedure for Component Life Estimation 491

of ridges in the material fracture surface. Dynamic strains

are caused by the effect of the loading spectrum on the
dynamic properties of the structure. Resonance may then
cause certain harmonic components to dominate the dynamic
strain spectrum.

Material strength determines the number of strain cycles a

component can withstand at a given location at a given
frequency. under the applied preload steady stress. Charts
of material strength data are prepared from fatigue tests on
standard samples of the material having similar chemical com-
position. material properties. and heat treatment as those of
the prototype. A recent development is the strain life
curve. as shown in Figure 1. This approach is used in the
present initiation life procedure.

Component life history. The loading to which a component is

exposed will strongly influence both the steady preload
strain level and the dynamic strains which occur at a given
location. The time to which a component is exposed to such
loadings influences the fatigue initiation life of the
component. Many comprehensive component load histories
involve repetitions of parts of the overall loading sequence.
as shown in Figure 2. Important aspects of any load history
are:

a) Initial startup cycle (to 100% speed).

b) Transient excitation of structural modes during

runup and shutdown.

c) Initial overspeed test cycle (to 12()1fo speed or

more).

d) Several balancing speed cycles in spin pit (to 100%

speed).

e) Thermal tests at speed. Various start-stop cycles

(100% speed).

f) Initial startup cycles in plant (to 100% speed).

g) Overspeed trip settings in plant (typically 110%

and 112% speed).

h) On-stream or on-line startup (to 100IWI speed.

several times).

i) On-load (to 100% speed).

j) Process establishment. (Several 0-100% cycles.

thermal transients).
492 N.F. Rieger

k) Process dynamic loading (gas pulsations, torsional

transients. etc.).

1) Process load variations (load, _speed, temperature,

f1 ow. etc • ) •

m) Planned outages (lOOC!b-zero-100% speed; thermal

transients).

n) Unscheduled outages. Process problems, failures,

etc.

Each of these factors must be incorporated into the load

history of the component, as applicable. Previous load
history details are often assembled by estimating probable
cycles from typical procedures, records, and operator
recollections. Anticipated loadings will include items from
the above list, and are arranged as typical load-blocks to be
applied to the component: see Figure 2. Note that process
thermal history and stresses may be an important aspect of a
given history.

Cumulative damage. The applied load history contains both

low cycle and high cycle load components. Low cycle loading
typically involves the largest strain ranges, and provides
the pre-strain about which the dynamic cycling occurs. High
cycle loading typically contains multiple frequencies, each
with different load magnitudes. This results in many more
load cycling harmonic components, but these components
usually have considerably smaller magnitudes than the low
cycle components. All loadings together result in cumulative
damage to the complrnen~. for which the effects can be
assessed using the Rainflow cycle counting method, as
described later under Cycle Counting Procedure. The
influence of load cycling on the stress-strain curve is illu-
strated in Figure 3. Each time a cycle is completed in the
manner shown, it counts a~ one load cycle having a specific
strain range and frequency. A computer program keeps account
of the number of load cycles which are completed at a given
strain level, in the procedure described herein.

11.4.6 Calculation Procedure for Initiation Life

Prerequisites, The following input data are needed for

conditions at the location being considered, to perform the
initiation life calculation.

a) Steady stress value corresponding to each load

level in the load history. Precise data from
finite element calculations is preferred.
Procedure for Component Life Estimation 493

b) Dynamic stress value corresponding to each load

condition in the load history. Finite element data
is preferred.

c) Material properties for the component material

obtained from strain-life tests on samples, in the
operating environment, at each operating
temperature in the load history. A cyclic
stress-strain curve for the material is also
required.

d) Load history details for number of cycles during

each of the life history conditions (a) through (n)
are listed previously.

Examples of such data are given in the case histories given

later in this chapter.

Cyclic stress-strain curve, Where a material is loaded

cyclicly, the stress-strain response does not follow the
well-known pull-test or monotonic load deflection curve. The
needed cyclic stress-strain data is obtained from the locus
of the hysteresis loop peaks, as shown in Figure 4. This
data is obtained from laboratory tests on material samples
under simulated operational conditions (preload, temperature,
environment). The cyclic stress vs. strain curve is then
curve-fitted in the computer for life calculations, using the
following relations:

1/n'
.As .Aa .Aa
= + (ii,) (1)
2 2E

where K' = af'/sf ,n'

and n' b/c

Figure 3 shows how preload and load cycles influence

component strain cycles, in the elastic and plastic regions
of the stress-strain curve.

Stress history. The required steady stress distribution is

calculated using some linear elastic procedure, such as the
finite element method. This provides high computational pre-
cision in the stress results, and it allows stress conditions
for other operating conditions (such as lower speeds or
higher temperatures) to be found by proportioning the results
obtained from a single calculation, in accordance with the
known structural or thermal laws.
494 N.F. Rieger

Dynamic stresses are found in a similar manner, taking into

account the relation between the exciting frequencies in the
load spectrum and the natural frequencies in the structural
response. Changes in vibration amplitude are again propor-
tional to the applied forcing, as mentioned above. Changes
in dynamic stress magnitude due to changes in exciting
frequencies. e.g.. due to speed changes. or in the process
gas excitations, can be addressed by use of equation 2. as
follows:

(2)
a
n

where a is the dynamic (modal) harmonic stress, a is the

resonanf stress for the mode. e
is the modal dampingnratio. w
is the excitation frequency and w is the natural frequency
of the mode. n

The use of this expression can be demonstrated as follows.

The excitation spectrum shown in Figure S shows typical
process conditions corresponding to the dynamic portion of
the load history in Figure 2. If the exciting frequencies
and harmonic forcing are then changed from Condition A to
Condition B, the dynamic response of the structure in its
various modes of vibration will vary in accordance with
equation 2, as shown in Figure 6.

Strain life behavior, The strain life of the material as

determined by test specimens under controlled conditions is
governed by equation 3. as follows:

As a ,
-= f (2Nf)b + , (2N )c (3)
2 E 8 f f

This forms the basis for the fatigue initiation life

calculations. The specimen tests provide data which allows
(a) the fatigue ductility coefficient sf'• (b) the fatigue
strength coefficient af'• (c) the fatigue strength exponent
b, and (d) the fatigue ductility c, to be determined. This
then allows the strain life curve for the material, equation
3 to be calibrated. The resulting analytical expression is
then used to calculate the true strain at stress levels
resulting from the various aspects of the finite element
calculations.
Procedure for Component Life Estimation 495

The influence of mean stress on the strain-life behavior may

be considered in several ways. First. it may be considered
directly during laboratory testing. by performing tests at
different R-ratios. so that the resulting curve includes the
influence of mean stress directly. Secondly. the mean stress
effect may be included directly in the strain-life
expression. using the form shown in equation 3a:

As
(3a)
2

A third method is to use the Smith-Watson-Topper hypothesis.

which is described by equation 4:

a
As af'2 (2N >2b + a ' s ' (2N )b+c (4)
max = E f f f f
2

The second of these procedures. i.e. modified strain-life

equation. is used in the case histories which follow.

The ability to include the effects of calculated stresses at

levels which exceed the dynamic yield stress of the material
is another important aspect of the procedure presented here.
This step is accomplished using Neuber's rule. as follows:

(S)

This expression allows calculated linear stresses to be

realistically converted to dynamic stresses and strains.
using the dynamic stress-strain curve which is based on data
obtained from the cyclic strain tests. This data is next
converted into an analytical form using equation 3.

Equations 3 and 3a give the relation between dynamic stress

and strain throughout the component strain range tested.
When used with Neuber's rule. this convenient procedure
allows the linear stresses obtained by the finite element or
other method to be converted by the computer into practical
elastic-plastic strains. These operations are included in
the computer program developed for life calculations.

Cycle counting procedure, Under spectral loading the dynamic

strain conditions at critical locations in a component may
have very complex waveforms. in keeping with the dynamic
496 N.F. Rieger

spectrum at that location. Several procedures exist to deal

with this condition, of which the Rainflow cycle counting
procedure is well known. Simply stated, this procedure
consists of dividing the complex waveform into a sequence of
simple cycles, and then counting the number of stress cycles
within a given strain range. The resulting number is then
compared with the tested fatigue life of the material at this
strain level.

The complex waveform shown in Figure 7 can be sub-divided

into constituent cycles (or strain reversals) by drawing from
each significant peak a line in the direction of increasing
time until this line again intersects the complex waveform,
as shown. Each time this occurs, one cycle of damage has
been completed at a corresponding strain level. Cycles
within cycles may also occur, and the corresponding cyclic
damage is shown on the dynamic stress-strain curve in Figure
3. Computer sub-routines are available to accomplish this
cycle counting process automatically. When the process is
completed, for a given load history segment (corresponding to
a particular operating condition) the number of cycles
occurring at a given strain level is identified. A table of
such cyclic damage constitutes the output from the load-cycle
counting sub-routine.

Cumulative damage, The best known cumulative damage

assessment procedure is Miner's law, which states:

(6)

This criterion is used in the procedure described in this

paper. The number of cycles ni occurring at a given strain
level is obtained from the Rainflow cycle counting procedure
described above. The number of cycles to failure N.1 based on
test sample data, adjusted for mean stress effects, is then
obtained from the strain-life equation. Particular values
are then divided into corresponding ni values to form the
damage portion ni/Ni. This procedure is repeated for each of
the other strain levels to form the remaining cycle ratios
for the forcing response. These cycle ratios are then
summed, and this sum is compared with the criterion value
1.0, in accordance with Miner's law.

For practical loading conditions, the load segment time is

often relatively short, and represents only a portion of the
anticipated life of the component. Each segment of the load
history may therefore be interpreted as contributing a
portion of the total damage which the component can sustain:
see Figure 2. If the loading period j causes damage D., then
J
Procedure for Component Life Estimation 497

the total damage D = ED. for the period of operation may be

obtained from the summlation of the individual damage Di
components. Furthermore, the damage which occurs in I
specified time period Dt• representing the total number of
dynamic periods completed within the time period t may be
used to calibrate the operating life in years. For example,
if D = 0.25 damage units acquired in one year, then D = D
= 1.0 damage will be acquired in a total operating life ol
four years, under the same circumstances, in keeping with
Miner's law. Component failure could then be anticipated in
4 years.

With a computer program it is apparent that the above process

can take into account any number of cycles within a given
time frame, at any stress level. This damage can then be
compared with Miner's criterion to determine whether
component failure is likely to result within the specified
period. Alternatively, once the amount of damage, however
complicated, acquired within a specified time is known, then
the reciprocal of this value will give the anticipated life
of the component under similar circumstances. In this
manner, component life can be assessed, in a manner which
includes wide variations in load conditions and complex
operating conditions.

11.4.7 Case History 1: Shaft Keyway Fatisue Life

Problem: Consider a steel shaft made from AISI 4340 steel in

a compressor drive system, with a possible fatigue failure
point at a keyway. The material properties for the shaft
material were obtained from fatigue tests, and are listed in
Table 1. The loading history on the keyway, Figure 8, shows
both high and low stress cycles. Each startup includes a 20.
overload peak stress (point d) resulting from motor over-
shoot. During norma 1 operation, the shaft vibrates
torsionally at 1,000 Hz due to tooth-mesh excitation from the
drive gears. The resulting alternating torque varies by 6.7~
about the mean torque. The maximum elastic mean stress due
to the mean torque is 150 ksi at the keyway stress calcu-
lation.

The strain history of the shaft is shown in Figure 3. The

compressor driver operates 8 hours per day for 260 days per
year and has only one startup/shutdown cycle per year (points
c through g). The shaft is subjected to 2.88 x 10' vibratory
stress cycles each day (points e through f). The loading
history also includes the manufacturer's overload test
(points a through c) to about 133~ of the required operating
load. The elastic stress due to this overload is 200 ksi.
The local strain method is used to predict the fatigue life
of the shaft.
498 N.F. Rieger

The local stresses and strains are first computed from the
loading history (keyway elastic stress o vs. time) using
equations 1 and 3. Typical results are sum:arized in Table 2
as local stress vs. strain data. For instance, a peak stress
o • of 200 ksi (point b) during the manufacturer's overload
t:st results in local plastic deformation at the keyway, and
a local stress o of 123.7 ksi. This value of o is computed
from Neuber's rule, equations 1 and 5, using Newton-Raphson
iteration, along with the local strain 8. These computations
may be summarized as follows:

200 2
08 = = 1.356
29,500

08 = o(-) +
0
0 (~,) 1/n'
E

o = 123.7 ksi

08 = 1.356

8 = L.lli = L.lli-
123 • 7 - 0.011 in./in.
0

Because the shaft initially consists of material with zero

damage, the initial overloading by the manufacturer occurs as
a monotonic stress vs. strain curve. However, the subsequent
duty cycles occur as cyclic hysteresis loops as discussed
previously. Note that the overload test results in a
residual compressive stress of 69 ksi at point c, Figure 3.

Both the loading history and the local stress vs. local-
strain diagram show both low and high cycle loads. Low cycle
loading is represented by the hysteresis loop, points c
through g of Figure 3, whereas the high cycle loading
consists of fluctuating stress levels (between points e and
f). Although not detailed here, the total strain amplitude
aT and the mean stress om are respectively computed for the
low cycle loading to be 0.004520 in./in. and 18.3 ksi, and
for the high cycle loading to be 0.00539 in./in. and 71.1
ksi.
Procedure for Component Life Estimation 499

The fatigue damage due to both high and low cycle loading is
determined by solving equation 3 numerically. The results of
this computation are listed in Table 3, for fatigue damage
data. Table 3 also lists the damage per year which then can
be summed to determine the total fatigue damage which occurs
each year due to both types of loading. Thus:

1 n1
-= ---+ ~
N.r Nf(1) Nf(2)

260 7.488 X 101

= +
4.153 X 10-4 1.400 X 10-!3

= 0.1090

N.r = 9.17 yr

Finally, the fatigue life is predicted as the inverse of the

total damage per year, or N.r
= 1/0.1090 = 9.17 yr. This
fatigue life analysis shows that low cycle loading due to
machine startup and shutdown was the primary factor leading
to fatigue damage at the keyway.

11.4.8 Case History 2: Six Blade Group

Blade group model, Figure 9 shows a 3D finite element model

of a blade group consisting of six blades, a cover band,
tenons, vanes, platforms, and blade-root/disk-root sections.
Suitable boundary conditions were applied to the cover
section, and to the disk root section, to ensure that the
model would simulate the blade group response under steady
and dynamic loading. The model was then subjected to 3600
rpm rotational loading, and to a 10 psi steam pressure drop
applied to the vane upstream surfaces, to the cover
underside, and to the platform upstream surface. The
platform nodal displacements were obtained from this
calculation for use in further detailed 2D stress calcu-
lations.

The 2D gridwork shown in Figure 10 was used to obtain

stresses in the attachment region in finer detail. Very fine
elements were used in notch radius regions where higher
stresses are normally encountered. Twelve gap elements were
used to transfer the hook loadings from the blade root to the
disk, to account for the indeterminate support conditions,
and to study tolerance effects on attachment stresses.
Details of maximum stress found for both loading cases are
500 N.F. Rieger

given in Table 4. Figure 11 shows the deformed shape and

stress distribution for the root section with zero tolerance
mismatch when subjected to steady load.

Steady stresses. The maximum elastic stresses found in the

root attachment occurred in the upper hook radius region,
adjacent to the contacting surfaces. Maximum elastic stress
values found in the blade attachment are listed in Table 4.
These values were used to define the plastic strain using
Neuber's rule as described previously.

The overall effect of the machining tolerances was to

increase the upper blade hook maximum stress by 1~. Under
the same conditions, the maximum stress in the middle blade
hook and lower blade hook was reduced by 20'fo. The maximum
stresses in all hook-notch regions were greater than the
material yield stress, indicating that local plastic flow
will occur in those locations. The work hardened regions are
quite localized and are not expected to influence the results
greatly. The hand-calculated centrifugal force for a single
blade was 68,555 lbs. at 3600 rpm. This force was checked
against the maximum total force in the gap elements, and the
validity of the calculation was confirmed.

Dynamic stresses, Resonant stresses resulting from harmonic

forcing were calculated for the first 20 modes. A constant
harmonic forcing of 10'fo of the steam-bending loads in the
axial and tangential directions was applied, with appropriate
load phasing between the blades. The 2D root section model
was used to calculate accurate stress details for five modes
which showed significant vibratory displacements. The other
modes were not included because, even at resonance they
showed very 1i ttle response. These results are plotted in
the dynamic stress spectrum in Figure 12. Table 5 lists the
natural frequencies, and Table 6 gives dynamic stress results
for these five modes.

Dynamic stress was calculated for the first 80 per-rev

excitation harmonics, and the rms total dynamic stress in
each mode was obtained as the square root of the sum of the
squart!S from these stresses, to account for phasing differ-
ences. Figure 6 illustrates the procedure for detuning the
stress from two resonant modal responses. The excitation
spectra were based on water table test results for similar
stages. Forcing frequencies above 5000 Hz gave negligible
response amplitudes. Damping was assumed to vary from log.
dec, 6 = 0.015 at 1000 Hz to 6 = 0.025 at 4000 Hz.

The calculated results showed that the R1 mode and T22 mode
were responsible for the highest dynamic stress in the upper
hook-notch region. The calculated rms total dynamic stress
Procedure for Component Life Estimation 501

was 3335 psi. The resonant stresses in mode R1 and mode T22
were 5652 psi and 5387 psi respectively.

Fatigue analysis. The blade group had operated cyclicly, and

had experienced frequent starts and stops. Operating records
of startup, overspeed, trip settings, load rejection outages
and service hours were available for load history input
preparation. An average number of 170 start-stop cycles and
one 101ft overspeed test per year was used for nominal
operatioh, i.e., one load block. The fatigue life evaluation
was based on the startup overspeed block and the nominal
operating years as succeeding load blocks. Figure 13 shows
the type of load-life history used for the fatigue calcu-
lations.

A material test program was conducted to obtain fatigue data

at the blade operational temperature _of 635°F. This program
has been described by Morrow [10]. Fatigue life due to
start-stop cycles was obtained by testing specimens cut from
blade material. The cyclic strain-life curve obtained for
403 stainless steel at 635•F is shown in Figure 14. The
material data is summarized in Table 7. These data provided
the necessary information for the fatigue damage calculation.

Low cycle fatigue. Details of unit service hours showed that

a typical operating year has one 101 overspeed run, and 170
normal start-stop cycles. The elastic stresses in the
leading blade upper hook notch reg ion during 3600 rpm
operation were:

Case 1: Perfect contact all hooks at zero speed-138,800 psi

Case 2: Contact upper hook only at zero speed-140,240 psi

Case 2 corresponds to a maximum stress of 168,288 psi during

the 101ft overspeed test, and a maximum stress of 201,946 psi
during the initial 201ft overspeed tests. Elasto-plastic
stress and strains were obtained from these elastic stresses
using Neuber's rule.

Figure 13 shows the nominal stress history including the 201ft

overspeed spin pit test by the manufacturer, the 10 1ft
overspeed which is done once a year to set the governor, the
start-stop cycles at normal speed, and the vibratory stress
cycles. The corresponding stress-strain path for the upper
hook notch material follows the load history in Figure 13.
The material was loaded from its initial stage at point 0 to
the maximum stress of 93 ksi at point 1 during spin pit
test. After the 201ft overspeed test, the residual stress in
this region was -57.5 ksi in compression. Point 2 is the
material in the stage of residual stress and strain. During
the 101ft overspeed test, the material was brought to point 3.
502 N.F. Rieger

The stress due to this overspeed was 66.1 ksi. Duriug steady
load operation the stress-strain condition was at point 4.
The steady stress due to normal speed operation was 55.9
ksi. When the vibratory stress reached the positive peak the
total stress we~t to point 5. When the stress reached a
minimum value, the stress dropped to point 6. The stress-
strain condition went back to point 2 when the unit was
brought to zero rpm.

Table 8 summarizes the strain amplitudes and fatigue damage

rate for the 10.. overspeed cycle and the normal speed
cycles, The total damage in one operating year is:

This means that the blade would have a nominal life of 54

years in low cycle fatigue. In a pra~tical sense this figure
is regarded as statistically marginal.

Hiah cycle fatiaue, Per-rev excitations stimulate the blade

sroups to vibrate in different modes. The exact forcins
function for the bla6e group was not known at the time these
calculations were made. High cycle fatigue analysis was
based on calculated vibratory stresses obtained by the
procedure described previously, including the mean stress
effect, equation 3a,

The magnitude of the mean stress cs for high cycle loading

calculation is obtained using Neufer's rule. The elasto-
plastic stress due to steady forces at 3600 rpm was 55.9
ksi, The mean stress for vibration in this condition becomes
53 ksi. Hish cycle vibratory strains were found to be
predominantly elastic in nature. A_ Rainflow cycle countins
technique is used for the analysis, Damase was calculated
and accumulated linearly throushout the entire loadins block
usins Miner's law.

Cumulative damaae. Fatiaue damase calculations were made

asain assumins that resonance occurred in five sisnificant
modes. Table 8 shows the results of resonant stresses and
fat iaue damase for each of the five modes in one year.
Resonance in the T22 mode was found to cause the most
damaae. The resonance of Rl sroup rotational mode was also
danserous for the leadins blade in the aroup i f the blade
sroup experienced the loading in Load Case 1. The total
fatisue damage was computed by Miner's law, as follows:
Procedure for Component Life Estimation 503

where DB is fraction of the fatigue life used per year due to

high cycle strains, DL is fraction of the fatigue life used
per year due to start-stop cycles, and D. is total fraction
J
of the fatigue life used per year.

The results indicated that if resonance occurred in the Rl

mode, the leading blade life would be 56.7 months. If
resonance occurred at T22 mode, the same blade would have a
life of about 19.6 months.

A fatigue life of 19.6 months in the T22 mode appears to be

inadequate for blades operating under the above circum-
stances. The R1 mode life of 56.7 months also appears to be
statistically marginal. These estimated life values are both
consistent with the actual blade life observed in practice.

11.4.9 Conclusions

o A comprehensive procedure has been described and

demonstrated for the calculation of fatigue initiation
life.

o The procedure requires accurate input data on steady

stresses, dynamic stresses, material cyclic properties,
and component load history.

o The resulting fatigue initiation life value has

state-of-the-art accuracy for the various technologies
involved.

o The procedure may be used for new component design

evaluation. Component life vs. required life then
represents the factor-of-safety of the component.

11.4.10 References

1) Steele, 1. M., Lam, T. C., 'Improving the Accuracy of

Fatigue Analysis,' Machine Design Masazine, Penton
Publishers, Cleveland, OH, August 1, 1983.

2) Rieger, N. F., 'Blade Fatigue,' Invited·Address,

Ro tordynamic s Session Proceedings, Sbth IFToJOI
Congress, Theory of Machines and Mechanisms, New Delhi,
India, December 1983.

3) Socio, D. F., Morrow, 1. D., 'Review of Contemporary

Approaches to Fatigue Damage Analysis,' Chapter 8- l!!t
504 N.F. Rieger

and Failure Analysis for Improved Performance and

Reliability, J. J. Burke and V. Weiss Eds., Plenum
Publishing Corporation, 1980.

4) Socie, D. F., 'Fatigue Life Prediction Using Local

Stress-Strain Concept,' Paper Presented at 1975 SESA
Spring Meetina. Chicago, IL, May 1975.

5) Dowling, N. E., Brose, W. R., and Wilson, W. K.,

'Notched Member Fatigue Life Predictions by the Local
Strain Approach,' Fatigue Under Complex Loading:
Analyses and Experiments, SAE, Vol. 6, 1977.

6) Dowling, N. E., 'Fatigue Failure Predictions for

Complicated Stress-Strain Histories,' Jnl. of Materials,
JMSLA, Vol. 7, No. 1, March 1972, pp. 71-87.

7) Rieger, N. F., 'Factors Affecting the Fatigue Life of

Turbine Blades and an Assessment of Their -Accuracy,'
Proceedings, 55th Shock and Vibration Symposium, Dayton,
OB, October 1984.

8) Conway, J. B., Stentz, R. B., Berling, J. T., 'Cumula-

tive Damage Concepts,' Chapter 5, Fatigue, Tensile, and
Relaxation Behavior of Stainless Steel, TID-26135
U.S. Atomic Energy Commission, Technical Information
CeD'ter, Oak Ridge, TN, 1974. (Also Mar-Test, Inc.,
Cincinnati, OB).

9) Battacharya, A., 'Cumulative Damage of Carbon Steel

Specimens in Tension-Impact Fatigue,' Ph.D. Thesis,
Banaras Hindu University, Varanasi, India, 1984.

10) Morrow, J. D., 'Laboratory Simulation of the Low Cycle

Fatigue Behavior of the Book Region of a Steam Turbine
Blade Subjected to Start Stop Cycles,' ASME Proceedings,
Fourth National Congress on Pressure Vessel and Piping
Technology, Portland, OR, June 1983.
Procedure for Component Life Estimation 505

,•..
li

..=' ···:r
j

§
.
<il ...
···l

Figure 1 Total Strain vs. Fatigue Life

for Annealed AISI 4340

Factory
a Overs peed Safety Safety Safety
Trip Trip Trip

Stress
Process
Load

Time

Startup load load Load

Cycle Cycle I Cycle II Cycle Ill

Figure 2 Typical Load History -

Start and Process Cycles
506 N.F . Rieger

100

'iii
~ 50 Hysteresis
b loop
"'"'
~
iii
<ii
0
0
...J

Residual

- 50

0 0.005 0.010 O.ot5

Local strain ~. (in./in.)

Figure 3 Stress-Strain Load ing History

~ 1- One Block

f :::~·····~
(o) Incremento! Spectrum
Tom•

rOksi
0 .004

(b) Stren-Stroin Response (Oecreosing Steps)

Figure 4 Dynamic Stress-Strain Curve

Developed from Hysteresis Loops
Procedure for Component Life Estimation 507

Typical Vane
F Gas Pressure Passing
Dynamic Components Frequency
Force

Frequency
Figure 5 Excitation Spectrum

a'
•
t - - - - - - - - r l"
,)'::
I I
a'R ~ I I
Ill
Ill ~-----~\, I I
C1l
s..
+"
I\ j \
V')
/ \ / \ - tot• I dyNIRic ' ' ' " " "

u I \ /
.,_
E / v· ClyN•'c ruponse fOf" -ode 1
rtl
c: \
>, \
0
\
''
' ....... .............. _
,, ·Frequency f

Figure 6 Stress Detuning Procedure

€ , Strain

-i 1 ;L 2
31
3
~ 5
]4
6
7
1 18
9f ~ 10
ltzcycle
1cycle
112 cycle
1 cycle
I cycle
112 cycle

Figure 7 Rainflow Cycle Counting Procedure

508 N.F. Rieger

·;;; 20% overshoo1

~

::200
.,.,
e 1so
iii
u
.
·; ; 100
'ii

c
Time

Figure 8 Loading History for Keyway

Figure 9 3D Blade Group Model

Figure 10 2D Root Attachment Model

Procedure for Component Life Estimation 509

Figure 11 Stress Distribution at Attachment

Area with Zero Tolerance Mismatch

10~-----------------------------------------,

Figure 12 Response Spectrum: Stress

Amplitude vs. Frequency

RESIDUAL STRESS
a AND STRAIN
1
NOMINAL
STRESS

Figure 13 Applied Load History for Turbine Blade

510 N.F. Rieger

1.0000 .,.--...,----,---,---.,----.-- --"7-----r----,

I I
o1,
i
II ( • • - E (lN')Ib. 'JlN ,c
1 1
I i
o. 1000 1:---1\.----'---t--.,....---'-- ---+--·

....

0. 0010 l:---.,..--5tic-$t...lo...__;~.,.-~::::::;,._~-cl----l
/TIItudel

Plutlc Str•ln
Amplit<.de

Figure 14 Strain Life vs. Number of Cycles

for ASM Stainless Steel

Table 1 Material Properties for AISI 4340 Steel

NOdulus of elasticity, E 29,500 kai

cyclic strength coefficient, K1 167 ksi

cyclic strain hardening exponent, n 1 0.100

Fatigue ductility coefficient, e: f 1 1.060

Fatigue strength coefficient, a f 1 168 kai

Fatigue strength exponent, b -0.075

Fatigue ductility exponent, c -0.75

Procedure for Component Life Estimation 511

Table 2 Computed Local Stresses and Strains

Point on Local Local

Load History Stress, Strain
Diagram a (ks i) e:(in,/in,)

a o.o o.o
b 123.74 0,01096
c,g -69,18 0.00393
d 105.6 0.01297
e 61.1 0.01229
£ 81.1 0.01161

Table 3 Computed Fatigue Damage Data

Total Strain Mun Fuiauc Nwmcr of

Type of lqllitudc ,e: Strcu, Damage/Cycle, Cyclu/Yr. Damage/Yr,
Load ina (in,/in,) a a0 (kai) 1/Nf(i) Di 0 ifNf(i)

Low Cycle 0,004520 U,J 4,153:110- 4 260 0,10798

Hiah Cycle O,OOOH9 71,1 1,400x10-1J 7,48h10 9 1,0484d0- 3

Table 4 Maximum Stresses in Blade/Disk Attachment Area

....
MAXIMUM STIIESSES IN ILADE AND DISK ATTACHMENTS

Loc•tion c ... Dlok

Upper '-k c.... I JIIOO pol I'M70 pol

c... , 1'02110 pol 15Uto pol

Mtddle c ••• 1 IU7JO pol UUIO pol

'-k c... 2 II 2250 pol 1J2UO pol

-
lower c.... nuu pol 1SIUO pol
'-k c ... 2 103770 pol unu pol
512 N.F. Rieger

Table 5 Blade Group Natural Frequencies

MODEL MODE STI HAST RAN

A1 863 759
R1 990 888
6 Tl 1052 919
BLADES u 161l4 ---
s 3119 ---
A1 850 763
R1 901 816
12 Tl 1002 922
BLADES u 1041l 931
s 1320 121l9

Table 6 Detuned Dynamic Stresses in the Leading

Blade Upper Hook Notch Region
DYIWIIC STUSSES PSI

Al Rl Tl A3 T22 TOTAL, rms

WEl 503 857 6lJ 1980 ~ 1lli

SPECTRIJI

CASE 2 195 m 2~2 17~ 3866 !IZZll

SPECTRII1

Table 7 Material Properties of

ASM 403 Stainless Steel
Ttst DIU
.... _ [10)
ns•r

Elestlc Modulus, E, ksl 21,000

F•t19ue Strength Coefficient Exponent,

oj__jul 131.0

F•tlgue Strength Exponent, b -.013

F•tigue Ductility Coefficient tf 0. 311

F•tlgue Ductility Exponent, c -.51

~ycllc StNngth Coeff'lcient, K', ksi 151. 0

lcycllc Strain H•rdening Exponent n' O.IU

 a
n
n>
p..
t=
....
11>

0'
....
(")
0
Table 8 Fatigue Damage Evaluation for Six-Blade Group 3
"'0
0
::s
11>
::s
....
Mode Al R1 Tl A3 T22 I
t""'
;:.
Natural Frequencies (Hz) 863 990 1052 3849 4481 tTl
"'....§'
Stress Amplitude 3.45 5.65 3.70 2.37 5.39
~
( ksi) ....
Load
o·
::s
Hfgh Cycle Fatigue 1.21 X 10•J 1. 93 X 10· 1 2. 97 X 10-J 1.26 X 10· 4 5,45 X 10-l
Case
I Damage 1/Y f

Years to Failure Yf 825 5.19 337 7.91 X 10 3 1.84

Stress Am~litude 1.29 2.12 1.44 2.03 5.44

( ks i
Load High Cycle Fatigue 6.47 X 10· 8 1.06 X 10-~ 2.37 X 10-l 2.69 X 10· 5 5,97 X 10· 1
Case Damage 1/Yf
II

Years to Failure Yf 1.55 X 107 9,38 X 10 4 4,22 X 10 6 · 3,72 X 10 3 1.67

U1
....
~
 CHAPTER 11.5

DAMPING PROPERTIES OF STEAM TURBINE BLADES

N.F. Rieger
Stress Technology Incorporated, Rochester, New York, USA

ABSTRACT

This section describes a program of tests to

examine the damping properties of several types of
steam turbine blades. These studies were conducted
in a non-rotating damping test rig which used real
blades mounted in corresponding real disk root
attachments. Simulated centrifugal loading was
applied to the blades as a lengthwise axial pull.
The rig and the test program are described. The
damping properties of the blades studied were found
to conform primarily to effects associated with
material hysteresis damping. The influence of
attachment friction appears to be negligible for
test loads and vibratory amplitudes encountered in
practice. A small level of fluid damping from the
gas stream appears to be possible where the blades
are subject to working pressure loads.

ll.S.l Iptroduction

This section describes a series of experiments which were

undertaken to determine the nature and magnitude of damping
in steam turbine blades. The test program included high
pressure blades, intermediate pressure blades, and low
pressure blades. The program was undertaken using a special
type of test rig, Figure 1, which simulated the blade loading
under rotating conditions, by the application of a known pre-
load. A statistically significant number of identical blades
were involved in each test. This gave some indication of the
amount of test scatter which could result in the damping data
under controlled test conditions from differences between
blades and their assembly conditions within their root
attachments. The tests examined the influence of blade type,
attachment root type, surrounding environment (air, steam,
vacuum), axial load, mode of vibration, and other factors.
516 N.F. Rieger

The objectives of this study were:

a) To study the damping properties of several types of

steam turbine blades in actual root attachments
under simulated operating conditions.

b) To examine the nature of the damping which is

evident in the test results.

c) To evaluate the magnitude of the damping from the

decay traces for a variety of blade types, root
attachment designs, blade lengths, and operating
conditions.

d) To determine the relative contributions of the

various damping mechanisms to the total damping
under these conditions,

e) To evaluate the significance of applied root load,

vibration amplitude, root type, blade length, and
gas environment as possible contributing parameters
in each case.

Details of the test rig used, and of the test conditions and
results achieved are given in the following sections.

11.5.2 Previous Blade Damping Studies

There is surprisingly little information in the open liter-

ature on the damp-ing properties of steam turbine blades,
Blade manufacturers have developed extensive test data
relating to materials performance, and to blades and blade
group performance from vibration tests.

Grady [1] tested several dummy intermediate pressure blades

in disk attachments using a pull-test machine, and used a
dynamic shaker to excite the blades to specified force
levels, Attachment pre-stress conditions were changed by
varying the root interference fits, It was shown that an
optimum root flexibility condition could be developed which
would result in dynamic stresses about 1/3 to 1/2 of those in
a rigid (welded) joint, Where root stiffness is high, less
damping is developed, Considerable variability existed in
this optimum root flexibility condition, which was found to
depend on blade assembly procedures and on blade vibration
history.

Grady [2] performed tuning fork vibration tests in air to

measure blade material damping under simulated centrifugal
load conditions in a tensile test machine, Data was obtained
at zero centrifugal load through a range of dynamic stress to
Damping Properties of Steam Turbine Blades 517

30,000 psi. The results showed that the material log. dec.
values were highest with high dynamic stresses. A maximum
test log. dec. value of 0.033 was obtained. Damping
dec rea sed with increase in centrifugal load, and increased
with frequency of excitation.

Wagner [3] conducted a program of damping tests on rotating

steam turbine blade groups in a test turbine. The magnitude
of blade damping, the influence of root details and of blade
sizes on damping, and the effectiveness of certain damping
devices were studied. The test blade groups were rapidly
rotated through axially directed water jets which impulsively
excited them into vibratory motion. The blade response was
obtained with weldable strain gages and a slip ring assembly.
Substantial scatter was observed in the log. dec. results.
The damping for impulse blade groups was found to be greater
than for long reaction type blades. Slightly higher damping
with higher stress values was .observed. In one test the
damping remained substantially constant over 84,000 cycles at
a maximum stress value of 30,000 psi. Undershroud welding
caused an insignificant reduction in damping.

Rieger and Beet [4] found similar results to these of Grady

and Wagner. Their tests were performed on blade pairs
inserted in disk segmeDts, with an axial pre-load applied
during testing. log. dec. values increased in a linear
manner with increase in vibratory amplitude, and decreased
with increase in root load. Values of log. dec. within the
same general range as those obtained by Grady and Wagner were
obtained (6 = 0.01 to 0.04) for similar blades, at similar
vibration amplitudes. Rieger and Beck observed higher log.
doc. vahos in axial vibration (6 = 0.02 to 0.08) than in
tangential modes.

Brown [5] tested a blade row in a rotating test apparatus to

determine the blade row dampins values. A steam jet was used
for excitation, and both impulse and harmonic excitations
were used to excite the blades. The damping results were
presented as a ratio between the log. dec. decay test results
and the resonance bandwidth test result~. No absolute data
are given. Brown reported difficulty extracting the damping
results due to beating, multiple resonances, etc. Impulse
excitation was observed to initiate several modes simul-
taneously.

Gotoda [6] studied the influence of damping on steam turbine

blade vibrations. Be identified damping contributions from
the blade material, from structural effects, and from
aerodynamic damping. Gotoda used the resonance decay method
to obtain data, and showed that the 1 og. dec. inc rea sed in
proportion to dyna~ic stress. Tests were next conducted on
tangential entry tee head steam turbine blades, tightly
518 N.F. Rieger

fitted as single blades into the disk root groove. Higher

damping values were obtained in this case than in the tuning
fork tests. Damping was again found to vary linearly with
dynamic stress, and the increase in such values over the
material test ~esults was attributed to root friction. The
scatter in results was attributed to individual differences
between blades. At high stress this scatter was quite large,
and the log. dec. vs. dynamic svess relationship was no
longer linear, above a = 12.0 Kg/mm • Calculations by Gotoda
demonstrated that the value of stress index was 3, for the
blade material used. Further data was given for the
variation of log. dec. with stress for simple blades, blade
groups, and tuning forks. The form of the damping vs. stress
relation was linear for the blades tested, except as noted
above.

An early study of enhanced damping in hollow blades was made

by DiTaranto [7] who tested blades containing a core of 0.005
inch diameter radial wires. The damping in a stationary
blade with wires pre-stressed to 7,000 psi was an order of
magnitude higher than that of the stationary hollow blade
without wires for equal excitations. The damping in a blade
rotating at 8,000 rpm was further found to be twice that of
the stationary blade with wires stressed to 7,000 psi.
DiTaranto discussed the following methods for enhanced
damping: (a) insertion of particles and/or wires of dif-
ferent sizes into hollow blades, (b) use of two flat
cantilevered plates under normal force, rubbing together as
they vibrated, (c) root damping of the vibrating blade, with
normal force being the critical factor. Too great a force
was found to prevent relative blade root motion: (d) tuned
dampers, which were effective in narrow frequency bands.

Srinivasan, Cutts, and Sridhar [8] studied compressor fan

blade damping. Root damping was found to be negligible at
operating speeds for the dovetail root attachment structure.
The material damping in the titanium alloy blades used was
also found to be very small. The shroud interface contacts
were found to be the most useful source of damping in such
fan blade structures, but the authors indicate that damping
from shroud or platform rubbing may be difficult to define
for consistent results. Material damping log. dec. values
for titanium blades were found to range from 0.0006 to
0.0015.

Jones and Muszynska [9] investigated the influence of slip

between components in the blade platform region. The purpose
of this study was to determine the extent to which damping in
a blade geometry could be optimized by allowing contact
between the blade platform and the disk to occur at high
speed. Approximate non-1 inear equations of motion were
derived and solved using a harmonic balance procedure. It
Damping Properties of Steam Turbine Blades 519

was found that high levels of slip damping could be achieved

if the relative component st.iffnesses were properly selected.
This paper does not contain any significant experimental
results, but testing is proposed together with a spin pit
investigation, to validate the influence of practical
platform dampers.

Muszynska, Jones, Lagnese, and Whitford [10] analyzed the

non-1 inear response of a set of compressor blades mounted on
a rigid disk and interconnected with a dry friction coupling
near each blade platform. Each blade consists of a discrete
two-mass system connected to adjacent blades by a flexible
1 ink, with a dry friction contact. Effects from blade-to-
blade dry friction contacts are also included. A single
blade was initially investigated experimentally to obtain
modal values for the blade models. Several system effects
were examined, including blade mistuning and influences from
the magnitude and distribution of exciting forces and phase
differences between the exciting forces on adjacent blades.
Conclusions concerning the influence of dry friction in
reducing blade response amplitudes and on the influence of
mistuning on the resulting blade amplitudes are given. The
fundamental modes of the blades were accurately modeled and
their parameters were precisely identified. An optimum value
of the friction parameter was found to exist for which
response amplitudes are at a minimum for all blades. The
effect of blade mistuning was to introduce blade-to-blade
coupling through the stiffness elements connecting the
blades. This led to typical mistuned behavior in which
several response peaks were observed, scattered on both sides
of the tuned system resonant frequency.

Attachment Slip Studies:

Hanson [11] described a rig for testing gas turbine and

compressor blade damping properties, which consisted of a
disk carrying a single blade which could be rotated at speeds
up to 15,000 rpm. The blade could be excited by an air jet
or by the impact of a small ball. Blade decay rates follow-
ing impact were observed. Data is given on root friction
damping coefficient vs. rpm, and on material damping vs. rpm,
for several types of blade roots. For these blades fir tree
roots were found to give the highest rates of amplitude
decay. Pin root and wedge root blades gave smaller damping
values. In all instances the damping increased as rpm
increased.

Hanson, Meyer, and Manson [12] investigated a proposed

compressor blade friction damping device. Using the same
rotating blade apparatus, they demonstrated a 3:1 decrease in
blade vibration amplitudes when this device was used on
compressor blades. Similar platform friction devices were
520 N.F. Rieger

recently developed and tested during studies of t•lade

vibration associated with the space shuttle t1:rbopump
engines.

Beards [13] evaluated the likely effectiveness of root slip

damping for suppressing vibrations of compressor blades based
on an examination of the results obtained by Hanson [11],
Hanson, Meyer, and Manson [12], and Good~an and Klumpp [14].

Jones and Muszynska [15] developed a simple two-mass analy-

tical model to represent the vibrational behavior of a jet
engine compressor blade in its fundamental mode, allowing for
slip at the blade attachment interface. This work is an
attempt to develop a theoretical basis for earlier work by
Hanson, Meye2·. and Ma11son [;12], and others. Vibration tests
were carried out to experi1r.entally verify the a11alysis usir:g
a partic11lar blade geometry. A blade with a simple root
geometry was inserted in a heavy fixture contai11ing a
matching axial entry tee head root. The blade was loaded
with radial forces to simulate the centrifugal load on t)le
root during operation. Tests were conducted to determine the
response amplitude at various freq11encies corresponding to
several centrifugal loads. The non-linear equations of
motion with slip at the root were developed and solved by the
method of matched parameters. Correlation between analysis
and experiment showed that it is possible to model the blade
response using a simple two-degree system wi.th slip at the
root level. The response curves obtai11ed follow tlte response
for a linear model until slip occurs at tbe loaded interface.
Beyond this poillt the blade vibration amplitude ceased to
grow, and remained constant over a substantial range of
frequency.

Muszynska and Jones [16] investigated the dyna~ic response of

blades on a rigid disk allowing for iDterface slip at the
attachment illterface and blac!e bysteretic dampfr.g. Two tn·es
of mass model of a single blade in a disk were used. The
first model allowed for root slip and blade hysteresis, and
the aecond model allowed a damper concept which incorporated
platform slip to be studied. llesults were compared with
experimental response data for a single blade in a test
fi:r.ture under low level harMonic excitation. It was demon-
strated that the contribution of root slippage to blade
damping falls off with increasing rotational speed, but His
was compensated wH.l1 a suitably sized damper device iD the
second model.

Material Damping:

Material damping has been studied by many irtvestigators.

This literature has been discussed comprehensively by La zan
[17]. Rowett [18] investigated tbe torsional c1alt1I'ir g
Damping Properties of Steam Turbine Blades 521

properties of certain grades of steel shafting, and appears

to have first snggested the stress damping law D • Ian.
Kimball [19] obtained log. dec. damping data for several
grades of carbon steels and alloy steels. Lazan stndied the
general nature of material damping, and presented results for
almost 2000 materials in his treatise on this subject.
Lazan's results are presented in terms of the above material
damping law in chart form and in a data glossary, in which
specific damping energy is related to dynamic stress. An
application of Lazan's procedure to first mode vibration of a
turbine blade is given.

Lazan's results again show that log. dec. values will

increase with dynamic stress, i.e. with vibration amplitude,
where material damping is the dominant mechanism. Similar
results were obtained by Robertson and Yorgiadis [20] in an
early study of energy dissipation in several metals under
torsional and axial loadings.

u.s .3 Apparatus

Blade test rigs of the type shown in Figure 1 were used in

this program. The principle of the rig is illustrated in
Figure 2. Several rigs of different sizes were built to
accommodate different blade lengths. Damping properties of
steam turbine blades in disk segments with actual root
attachments were measured. Vibration decay tests were
conducted using blades which had been welded together in
pairs at their tips: see Figure 3. The purpose of this
blade pair arrangement was to carry the lengthwise tensile
pre-load which was intentionally developed using relative
thermal shrinkage (initially) and hydraulic load cells
(later) between the blade pairs and the rig frame to simulate
the blade centrifugal loading which is applied to the root
attachments during operation. Damping data was obtained from
the pre-loaded blade pairs tested in their lowest mode of
vibration, in both the tangential and the axial directions.
Details of vane lengths, root attachment types, location in
rotor, etc., involved in the test program are given in Table
1.

The primary requirements for the blade test apparatus were:

(a) to provide representative support conditions for the
blade and root attachment structure during testing, (b) to
apply a lengthwise load to the blade pairs during testing in
a manner which simulates applied centrifugal loading on the
blade, and (c) to isolate the blade pair from unwanted rig
or apparatus natural frequencies. It was also required that
the natural frequencies and mode shapes of the pre-loaded
blade pairs should be reasonably close to those which occur
in the grouped blades during operation in the turbine. This
requirement was achieved by mating a natural frequency
522 N.F. Rieger

calculation of the oriainal blade. and by then adding a

suitably proportioned flexure link between the blades. Fiaure
3. The flexure lint transmits the end-applied blade tensile
load. This arranaement ty~ically allowed the tangential and
axial natural frequencies to be approximated to within 15
percent. Some inaccuracies remain between the relative mode
shapes of these two conditions in the blade tip flexure link
reaion. but calculations showed that the mode shape details
in the attachment reaion and in the lower portion of the vane
are in aood aareement.

A requirement of each test ria was that the natural

frequencies and mode shapes of the ria should not influence.
or interact sianificantly. with any mode of the blade pair.
Prior to rig construction. a detailed finite element calcu-
lation was made to identify the first ten or more natural
frequencies of the test rig. both with the blade pair
inserted. and without blade pair inserted. Each ria was also
tested by rapping. to locate actual natural frequencies. The
test apparatus was finally modified as needed so that the
required blade natural frequencies occurred in dynamically
inert frequency reaions ('dead zones') of the test apparatus.

The environmental chamber shown in Figure 4 was built to

investiaate the influence of the surrounding gas environment
on the blade damping properties. Gasdynamic damping effects
from surrounding air. from steam (at various pressures and
temperatures). and in vacuum were tested using this chamber.
The chamber comple.tely enclosed the blade pair and disk
attachment segments. The desian and construction of the
environmental chamber used for evaluation of the 25 inch
vane. low pressure. curved axial entry blades involved the
follow ina:
a) A simulated low pressure stage environment between
0.8 and 1.5 inches of Hg and lOOoF.

b) A steel tube which surrounded the blade pair and

attachment. using a pair of flexible bellows at
either end to accommodate ria thermal and other
expansions. from the steam tests and applied pre-
load.

The blade pairs were mounted in the rig in the usual manner
and durina testing were pre-loaded usina hydraulic load
cells. All blade and attachment components were contained
within the environmental chamber durina the test proaram.

The test arranaement consisted of five openings in the

environmental test chamber: two for strikers to excite the
blades. two for the transducers. and one for the vacuum draw-
off connections. The location of the transducers and the
Damping Properties of Steam Turbine Blades 523

strikers within the environmental test chamber is shown in

Figure 4.

The amount of test data obtained from this program was

substantial. Between ten and twenty-five blades of each type
were tested. over n unge of pre-load. initiating amplitude.
environment. etc.. conditions. and each test was repeated
five to ten times. to ensure that representative statistical
results bad been obtained. In anticipation, a computer aided
data acquisition system was developed and used to reduce the
test data. This system is shown in Figure S. A calibrated
decay signal fro111 the blade transducer was captured on a
spectrum analyzer. visually reviewed. and then processed.
Accepted traces were then transmitted by a data link to the
memory of a minicomputer. This data was then processed and
printed as a tabulation of log. dec. vs. decreasing vibration
amplitude. with test details.

11.5.4 Experimental Damping Resplts

Seven blade configurations were tested to determine their

damping propertie11. These included two high pressure blade
designs. three intermediate blade designs and two low
pressure blade desi2ns. Statistical analyses were performed
on the damping results for selected tests. The influence of
root coatings. of root modifications. and of several
operating environments on damping was also investigated. for
low pressure blades. The results obtained are described in
tlds section. a11d their significance is disctJssed in the
following sections.

High Pressure Blade Tests:

S em Vane. Straddle Mounted. Tangential Entry. Figure 6:

Sixteen blade pairs were tested in air for damping in the

tangential and axial directions. each under four centrifugal
load conditions (1000. 2000, 3000 and SOOO lb.). The
results. Figvres 7 and 8, show that the log. dec. values
increase with blade tip displacement in both axial and
tangential modes. Damping is approximately linearly related
to the vibration amplitude. Increasing the centrifugal load
progressively decreased the damping valtJe toward a lower
limit value. This result has been further confirmed in
subsequent tests.

10 em Vane. Ball a11d Shank. Axial Entry Blade. Fisure 9:

Damping values were obtained from the tests conducted in air

on two sets of eleven pairs of ball and shank blades. in the
tangential and axial mode11. Both long shank and short shank
blades with 4 inch vanes were tested. Typical results for
524 N.F.Rieger

log. dec. vs. amplitude are shown in Figure 10. Damping

increased witb increased blade tip displacement in the
tangential mode, for both long and short shank blade8. For
long shank blades, the axial mode damping increased only
slightly with increased load. For short shank blades, axial
damping increased by an order to magnitude over axial damping
of the long shank blades.

Comments on High Pressure Blade Results:

High pressure blade damping appears to be related to

vibration amplitude in a linear manner, and inversely related
to centrifugal load.

Intermediate Pressure Blade Tests:

15 em Vane Fir Tree Root, Axial Entry. Figure 3:

Damping data was obtained for twenty-three blade pairs tested

in air. Results for log.dec. values are shown in Figure 11
for the tangential and axial directions, as a function of
blade tip displacement and centrifugal load. Damping
increased linearly witb increased tip displacement, ar•d
decreased with increased centrifugal load.

23 em Vane, Fir Tree Root, Axial Entry Blade:

Twelve blade pairs were tested in air. Tests were conducted

in both the tangential and axial modes. Damping values
increased with blade tip displacement, and decreased with
increased centrifugal load.

Comments on Inte~ediate Pressure Blade Results:

Damping results for IP blades with fir tree roots again

indicate the trends observed with the high pressure blades,
i.e., damping increased with tip displacement, and decreased
with increase in centrifugal load toward a limit value.

Low Pressure Blades:

70 em Vane, Fir Tree Root, Cu~ed Axial Entry Blade. Figure

12:

Twenty-two pairs of blades were tested in their disk root

sections, for seven centrifugal load conditions. Trends for
both tangential and axial modes are shown in Figure 13. The
curve showed a small negative hook followed by a shallow
curve of positive slope. The spread of damping values for
centrifugal loads of SOOO, 9000, and 13,4000 lbs. were
statistically analyzed.
Damping Properties of Steam Turbine Blades 525

Comments on Low Pressure Blade Results:

Tests on the curved axial entry blades showed some interest-

ing features. Initially. the blade damping decreased to a
minimum value. from which it increased with centrifugal load.
These blades were much longer. the attachment was curved. and
it was oriented across the disk rim at an angle. These
results appear to be some combination of effects seen in
tangential and axial modes of previous tests. The two-part
curve may be an expanded version of the characteristic seen
for stiffer blades. e.g •• Figure 11. Damping increases with
larger tip amplitudes of vibration. The LP vane section is
generally more flexible than other blade types. and so the
damping which results from flexure occurs mostly toward the
vane tip.

Environmental Effects: Low Pressure Blades. Piaure 14:

Ten pairs of 25 inch low pressure blades were tested in air.

in vacuum. and in steam. Statistical data was compiled from
the multiple testing performed on each pair. Negligible
differences were detected between the mean damping values
obtained for the three environments.

Root Coating Effects. Figure 15:

The results from the root coating test on 25 inch low

pressure blades indicate that attachment coatiaas consistina
of thin films of structural adhesives on the hook faces have
only a small effect on the overall blade damping in the
normal operating range.
Influence of Root Modifications. Figure 16:
Six 25 inch vane low pressure blades with the followin&
relieved hook conditions were tested in their disk attach-
ments. Two sets of conditions applied: (a) the two upper
hooks were relieved (reduced contact loads) and the four
lower hooks were therefore equally loaded. and (b) the two
middle hooks were relieved. and the upper and lower hooks
were equally loaded. Little difference was found between
results obtained from these two test conditions. Under
circumstances where the relieved upper hooks could slide.
case (b). the dampin& values were hiaher. This condition is
aot expected to occur under normal conditions.

11.5.5 Discussion of Test Results

The blade types aad test conditions involved ia this proaram

are 1 is ted ia Tables 1 aad 2. The hiah pressure blades
s~owed taaaeatial loa. dec. values raagiaa between 0.012 aad
0.18. aad axial log. dec. values between 0.024 aacl 0.21.
526 N.F. Rieger

Corresponding values for the IP blades range from 0. 008 to

0.09 in the tangential direction, and from 0.015 to 0.15 in
the axial direction. Log. dec. values for the low pressure
blades range from 0.005 to 0.030 in the tangential direction,
and from 0.005 to 0.035 in the axial direction. The range of
variation of log. dec. in each instance is due to variation
of dynamic amplitude, i.e., to dynamic stress range values,
and to test conditions, in particular the applied pre-load
values.

Damping data for short high pressure blades are shown in

Figures 7 and 8, for a tangential entry root, and in Figure
10 for a ball type axial entry root. The general form of the
curve in each instance is similar. For tangential entry
blades, log. dec. increases with blade tip displacement in
both the tangential and axial directions. Centrifugal load
tends to decrease the damping for higher centrifugal load
values. This trend is more pronounced for tangential
vibrations than for axial vibrations. The ball type roots
repeat these trends, but for the short shank blades the axial
vibration damping is almost constant with blade tip dis-
placement and centrifugal force.

Results for 6 inch blades with axial entry fir tree roots are
shown in Figure 11, and these again show an increase in log.
dec. with blade tip displacement. Data is given for tangen-
tial vlbrations and for axial vibrations, in air. The
damping relationship is seen to be almost linear in these
instances, though the vibration test amplitudes were
relatively small. Later tests established that damping
continues to increase with higher vibration amplitudes.
Increased centrifugal load causes the log. dec. values to
decrease, as was found by Lazan [17] for damping which is
related to material hysteresis.

Results from tests on 25 inch blades with curved axial entry

attachments under a variety of conditions are shown in Figure
13. Log. dec. vs. blade tip displacement is shown for
several pre-load values, for axial vibrations, and for
tangential vibrations. ID this instance, a higher mean
stress appears to increase the blade damping, at loading
values beyond a certain lower 1 imit. The reason for the
difference between these results and, for example, those in
Figures 7 and 11 is thought to lie in the interaction between
the local pre-load and the dyaamic stress. This causes local
plastic deformations to occur at higher pre-loads, which
would give higher material damping values. Higher amplitudes
of vibration were achieved for the tangential vibrations than
for the axial vibrations.

Figure 14 shows the influence of gaseous environment on the

test results, for tests involving new roots (assembled for
Damping Properties of Steam Turbine Blades 527

the first time). and old roots (which had been used in
previous tests). Generally speaking there is little
difference between the values obtained under vacuum and in
air. Damping data with used roots in air gave somewhat lower
values than air tests with new roots. These tests were
conducted using a 25 inch blade loaded lengthwise with a
force of 13,400 lbs. which is smaller than typical operating
attachment loads carried by such blades.

Figure 15 shows the influence of root coatings on log. dec.

for various amplitudes of vibration. Results for uncoated
roots were compared with those obtained from roots coated
with epoxy, and with loctite. It was found that the epoxy
coatings would smear throughout the attachment upon
insertion, and were only present over portions of the root
surface when the blades were removed. The influence of
loctite appeared to decrease the overall dampins by a small
amount, probably because of its effectiveness in securing the
blade into the disk attachment.

Figure 16 demonstrates the effect of severe mal-distribution

of root tolerances in LP blade attachments. For similar
applied loads, relief of the middle hook contact was found to
increase the damping more than for relief of the upper hook.
This is because the middle hook relief applied a greater load
to the upper hook, where damping and movement is most likely
to occur. The test data is consistent for loads between
5,000 lbs. and 13,400 lbs. These results may be compared
with similar data for unmodified roots in Figures 13 and 14.

Variation of log. dec. with centrifugal load increase is

shown in Figure 17. The vertical bars show in the data
scatter limits for one standard deviation. -~is data applies
to vibration amplitudes up to 3.7 -~ 10 inches in the
tangential direction, and to 0.7 x 10 inches in the axial
direction. It is seen that the log. dec. is almost constant
for centrifugal load variation in the axial direction, but it
decreases with load increase for vibrations in the tangential
direction. These results are again in keeping with results
obtained by Lazan [17] for material hysteresis.

Figure 18 shows addi tiona! damping data for a 30 inch LP

blade, where log. dec. is plotted against vibratory stress.
The blade pairs used had curved axial entry roots~ Dynamic
stress values were obtained with a strain gage attached on
the vane near the root. It is seen that log. dec. decreases
in the lower stress region, and then increases as the dynamic
stress increases further. This is again due to the complex
interaction between the dynamic stress, the mean stress under
test conditions, and the material properties, as mentioned
above.
528 N.F. Rieger

11.5.6 Copclusions

0 Blade damping increased with increase of vibration

amplitude in most instances, for both tangential
vibrations and a:r.ial vibrations. This trend was most
evident for HP and IP blades.

o Damping in low pressure blades increased somewhat with

increase in vibration amplitude, but at a lower rate
than in HP and IP blades. Average log. dec. values were
moderately constant with vibration amplitude.

o Damping values decrease toward a constant limiting value

with increased attachment centrifugal load effect. The
limit appears to be the rigid attachment condition.

o No significant changes in damping values were observed

with LP blades where root coatings were applied.

o No change in damping was found from tests involving the

presence of stationary air, steam, or vacuum surrounding
the blades. Gasdynamic damping evidently derives from
doing work on the local pressure distribution. This
only becomes significant with a moving gas stream under
operating conditions.

11.5.7 References

1) Grady, R. F., 'Investigation of Dovetail Damping Contri-

bution of Propulsion Steam Turbine Buckets,' General
Electric Company Report NOBS94390, Submitted to Bureau
of Ships, Department of the Navy, November 1967.

2) Grady, R. F., 'Investigation of Material Damping Proper-

ties of Propulsion Turbine Blade Material,' General
Electric Company Report NOBS-94390, Submitted to Bureau
of Ships, Department of Navy, December 1967.

3) Wagner, J. T., 'Blade Damping Tests,' Westinghouse

Engineering Report EC-401, NOBSN00024-67-C-S494, May
1969.

4) Rieger, N. F., Beck, C. M., 'Damping Tests on Steam

Turbine Blades, 'EPRI Project RP-1185-1, Palo Alto,
California, 1980.

S) Brown, W. G., 'Determination of Damping Value. for

Turbine Blades,' ASME Design Engineering Technical
Conference, 81-DET-131, Hartford, 1981.
Damping Properties of Steam Turbine Blades 529

6) Gotoda, Hidemi, 'An Analysis on Resonant Stresses in

Steam Turbine Blades,' Technical Designing Department,
Kawasaki Heavy Industries, Ltd., Kobe, Japan, 1974.

7) DiTaranto, R. A., 'Blade Vibration Damping Device, •

Journal of Applied MechaDics, Trans. ASME, Vol. 80, pp.
21-27, 1958.

8) Srinivasan, A. V., Cutts, D. G., Sridhar, S., 'Turbojet

Engine Blade Damping,' United Technologies Report No.
R81-91441031, Submitted to NASA (Report No. CR-165406).
July 1981.

9) Jones, D. J .G., Jluszynska, A. 'Design of Turbine Blades

for Effective [Slip Damping,' The Shock and Vibration
Bulletin, Part 2, Washington, D.C., September 1979.

10) Muszynska, A., Jones, D.I.G., Lagnese, T., and Whitford,

L., 'On Non-linear Resronse of llultiple Blade Systems,'
Shock and Vibration Symposi\UII, San Diego, California,
October 1980.

ll) Hanson, M. P., 'A Vibration Damper for Axial Flow

Compressor Blading,' Proceedings Society Experimental
Stress Analysis XIV, pp. 155-162, 1955.

12) Hanson, M. P., Meyer, A. J., and Manson, S. S. 'A Method

for Evaluating Loose Blade Mountings as a Means of
Suppressing Turbine and Compressor Blade Vibrations,'
Proceedings Society Experimental Stress Analysis, 10 (2)
PP• 103-116, 1953.

13) Beards, J. F.., 'Damping in Structural Johlts,' Shock

Vibration Digest, (8). pp. 35-41, 1979.

14) Goodman, L. E., 11\UIIpp. J. R., 'Analysis of Slip Damping

-.·ith Reference to Turbine Blade Vibration,' Journal of
Applied Mechanics, Trans. ASME, 23, (3), pp. 421-429,
September 1956.

15) Jones, D.I.G., lluszynska, A., 'Vibrations of a Compres-

sor Blade with Slip at the Root,' The Shock and Vibra-
tion Bulletin, Washington, D.C., September 1978.

16) Muszynska, A., Jones, D.I.G., 'On Discrete Modellisation

of Response of Blades with Slip and Hysteretic Dampins,'
Proc. of the Fifth World Congress on Theory of Machines
and Mechanisms, 1979.

17) Lazan, B. J., 'Damping of Materials and Members in

Structural Mechanics,' Pergammon Press, Inc., New York,
1968.
530 N.F. Rieger

18) Rowett. F. E.. 'Elastic Hysteresis in Steel,' Proceed-

ings, Royal Society of London, Vol. 89, Londor., 1914.

19) Kimball, A. L.. 'Vibration Problems Part S: Friction

and Damping in Vibrations,' Trans. ASME, 63, pp. A-135
A-140, 1941.

20) Robertson. J. M•• Yorgiadis, A. L., 'Internal Friction

in Engineering Materials,' Journal of Applied Mechanics.
Vol. 13, No. 3, p. Al3182, September 1946.
Damping Properties of Steam Turbine Blades 531

Figure 1 Blade Damping Test Apparatus

Figure 2 Principle of T·e st Rig

532 N.F. Rieger

..·

-
··~-.....~· .~-!r•• ----
•:f.,· .. _
- :;i-_'·

Figure 3 Welded Blade Pair Showing Flexure Link

Figure 4 Test Rig with Environmental Chamber

Damping Properties of Steam Turbine Blades 533

466l Digital
Axlol Transducer Interactive
Transducer Power S..pply Ploller

• - hu
Filter
,.,. R•O T,_ r-D
SFMCtrulll An.lyzerl .

crCJ---CJ-o
MP 100
Tangential Transducer O.t. Ceneral
Transducer Power Supply Computer

Figure 5 Computer Aided Data Acquisition System

Figure 6 2" Straddle Mounted Tangential

Entry High Pressure Blade Pair
534 N.F. Rieger

o. 50- --.. ,--- -,r- ---r ---,

c
".
E

¥
0

·eu
z
..
·;:

o.oo
0
0
0
0
"'
N

0
0
0

0"'
..
0
"'
0
0
0
0
.:
0
Blade Tip Displacement (Mils)

Figu re 7 Loga rithm ic Decr emen t vs. Tip Disp lacem ents
and Cent rifug al Load . Tang entia l Mode 2"
sure Blad e
Strad dle Mount. Tang entia l Entry High Pres

o.so

[7"u.
~E Centrif ugal l.oad
.. 0. 30

~
0

'/
u
e .o.2
=
.
·;:

j ~30001b.
o. 1ov.....,.....,..
_soo o lb.
t:- - roo lb.

o.o0
0
0
0
0
0

"'
N
0 ..
0
0
"'
.."'
0

0
0
0

Blade Tip Displac ement (Mils)

Figu re 8 Loga rithm ic Decr emen t vs. Tip Disp lacem ents
and Cent rifug al Load (Axi al). 2" Strad dle
Mount. Tang entia l Entry High Pres sure Blad e
Damping Properties of Steam Turbine Blades 535

Figure 9 4" Ball and Shank - Axial Entry

High Pressure Blade Pairs

AXIAL
0.~0

17SO, 2SOO Axial

0. )0
1000 Axial

o.oo
c.
E TANGENTIAL
~ 0 . 029
~
c
"
.E

.8'
~ 0.021
\:

..J

O.OIQ

0. 000 L---...L...---!----:!::-----:!
c
0
::!
.,;
c 0

:::
0

.,;
0
c
..,c
.;
.
0

.,;

Figure 10 Logarithmic Decrement vs. Blade Tip

Displacement and Centrifugal Load -
Short Shank 4" High Pressure Blade Pairs
 U1
l,.o.)
a..

0.060 I I 1- '•

c.. o.tooJ-,..;.-,--r--r:::;=;r~
E
I 0.0, ..
~
Q
u
..v
p e
£
:i }
~ O. Ol

. 6 I 10 12
'
Bl•de Tip Displacement (In x to•')
Blede Tip Dlspl•cement (In x to•'l

Figure 11 Logarithmic Decrement vs. Blade Tip Displacement

and Centrifugal Load - 6" Axial Entry Fir Tree
Root, Intermediate Pressure Blade
z
~
:;tl
(b "
~
....
Damping Properties of Steam Turbine Blades 537

Figure 12 25" Curved Axial Entry Fir Tree

Root - Low Pressure Blade Pair

AXIAL
o.ou

c 0.0)0

I...
~ \. 22, 00 lb.
... 0.020
'E \
:; 1l.-r lb.
'l:
j 0.010
'-- !--"' t.oo:o lb.
0.000
0 0.25
t
0.50 0.75 1.00
Blede Tip Dlaplac-t (Mile)

.000 ~--"":"":-----+----::-': ---~~

0 2.5 s 7.5 10.1
Blade Tip Dl~t IMilel

Figure 13 Logarithmic Decrement vs. Blade

Tip Displacement and Centrifugal
Load - 25" Fir Tree Root - Curved
Axial Entry - Low Pressure Blade
538 N.F. Rieger

0.0)0 ---.,....---r----,~---,

c..
..
E
0.020
Q
~
.!l
E

=
.
·;:
g'
..J 0.01U

0 .ooo __________...____
0 0.003 0.006 0.009 0.012

Mean Tip Blade Displacement (In)

Figure 14 Environmental Effects - Logarithmic

Decrement vs. Tip Displacement -
Tangential Air/Vacuum/Steam - 9000 lbs.-
25" Curved Axial Entry - Fir Tree LP Blade

..... Axlol

••••
I ..-- !!5'".....
~-
! ,.....-:
t.tzt
E;:;.y
Ii:

J
t.tlt

.....
I
.,;
i".: i.
- · - T i p D l . , . _ t (IN,)

Figure 15 Root Coatings Comparison -

Logarithmic Decrement vs. Tip
Displacement - Axial - 13,400 lbs. -
25" Curved Axial Entry - Fir Tree
LP Blade
Damping Properties of Steam Turbine Blades 539

0 • 03 or-----~------~--M-o-d-if-ie-d~M-id_d_l_e--,
Hook
I

c
II
0.020~----~----t--.rl~~---t-------1
E
f
~
0

..
u
"j§
.t:
·5en 0 • 010 ~-------1.1-
• •-.-.-:-1~34~0:-::0~I:-b-::C:-e-n7
tr-:;i7fu~g::a:;l1L-::oa::-::~rl
.3 ----- 9000 lb Centrifugal Load
- 5000 lb Centrifugal Load

0.000 ' - - - - - - - - - . . __ _ _ _ _ ___,

0 0.003 0.006 0.009 0.012

Mean Tip Displacement (In)

Figure 16 Modified Root Load Study - Logarithmic

Decrement vs. Tip Displacement Tangential
25" Curved Axial Entry - Fir Tree Root -
Low Pressure Blade

o. 100 ,...,..--,~-.,.--,--....--..,........

..i 0. 080
1 Axial Deflection at 0. 0007"
.] 1 Standard Deviation

f 1 Tangential Deflection at 0. 0037"

1 Standard Deviation
~ 0.060

.
J!
E
.t: 0.040

} 0.020

o.ooo
0 3000 5000 7000 9000 11000 13000
Centrifuqal Load. Lbs.

Figure 17 Variation of Damping with Preload Showing

Limits of the Scatter. 25" LP Curved Axial
Fir Tree Entry Blade Pair. Specified
Vibratory Amplitudes.
540 N.F. Rieger

-
....
0.070

• lO Inch Blode Test - Axiol Load

---
0.060
I
• •
,.. ••.... ·~·
..~·
Logarithmic 0.050 I'
Decrement
6 o.on
~.
, ••
0.030 • • •
0.020

0.010

o.ooo
250 750 1250 1750 2250
0
Stress (PSI)

Figure 18 Variation of Damping with Vibratory Stress.

30" LP Blade Pair - Axial Vibration - Curved
Axial Entry Root - Preload 22,500 lbs.

Table 1 Details of Turbine Blades and

Root Attachments Tested

ROOT TYPE YANE LENGTH ~ STAGE

Straddle Mount 2.00 Tangential First
Ball and Shank 3.50 Axial L-3
Fir Tree 6.00 Axial L-5
Fir Tree 9.50 Axial L-1
Fir Tree 25 .oo Curved Axial Last
 Table 2 Range of Logarithmic Decrement Values 0
Ill
s
"'0
Range of Range of
Vane Root Axial Tangential s·
(1!:1
Stage Size .Il:'.E_e Log. Dec. Log. Dec. Comment '"C
-- ....
0
"'0
HP 2" Tangential o. 05-0. 12 0. 07-0. 18 Air environment I~
straddle mount I~·
"'0
......
HP 4" Ball & shank 0.12-0.21 0.012-0.021 Air environment
short I;Ill
s
HP 4" Ball & shank 0.024-0.041 0.012-0.024 Air environment
long I~s·
(1)

IP 6" Axial fir tree 0.015-0.035 0.008-0.055 Air environment I~

Ill
p..
(1)

IP 9. 5• Axial fir tree 0.015-0.15 0.015-0.090 Air environment ·"'

LP 25• Curved axial 0.005-0.035 0.008-0.030 Air environment

fir tree

LP 25• Cuved axial o.o1o-o.o2o 0.008-0.010 AirNaccum/Steam

entry

LP 25• Curved axial 0.02o-0.024 o. oos-o. 010 L.octite ' epoxy

entry coatings

LP 25• Curved axial o. 02o-o. 025 0. 02o-0. 030 Adjacent root study
entry

LP 25. Curved axial o. 025-0. 030 0.015-0.028 Modified root study

1-'
I~
 CHAPTER 12.1

MAGNETIC BEARINGS

G. Sebweltzer
Institute for Meebanies, ETH Zurieb, Switzerland

ABSTRACT
Magnetic bearings have some distinct advantages. They do not
generate wear and they do not need 1ubri cation. These features
make them at tractive for vacuum applications. And their dynamic
behaviour can be adjusted in a wide range, which allows active
vibration damping and control.
This chapter presents the state of the art for the design of an
electromagnetic bearing system. It introduces first the main
elements and then discusses control and system aspects. Models
for describing an elastic rotor and its active vibration control
are included. The characteristics and the losses of such a
suspension system are detailed.
Several applications are demonstrated, and future trends are
indicated.

CONTENTS

1. Introduction
2. Functional principle
3. Design goals
4. Elements of the magnetic bearing system
4.1 Model for the rotor
4.2 Sensor, controller, amplifier
4.3 Magnetic actuator
5. System aspects
5.1 Control of rigid and elastic rotors
5.2 Characteristics of the magnetic bearing system
5.3 Losses
6. Applications
7. Conclusions
8. References
544 G. Schweitzer

1. INTRODUCTION
Magnetic bearings can support a rotor in such a way that it
levitates freely without any contact. Furthermore the dynamics of
this suspension can be easily adjusted in a wide range for
various applications. These two main properties already make the
magnetic bearing a very attractive device for solving the
classical bearing problem. On the other side the magnetic bearing
is complex, expensive, usually not readily available from the
shelf, and up to now only used for some advanced machinery. In
the following the state of the art is presented so that future
trends can be derived.
Magnetic forces are generated either by permanent magnets,
electrodynamically or electromagnetically. In the constant field
of permanent magnets, however, a ferromagnetic body cannot hover
in a stable way /BR 39/, and electrodynamic forces are usually
too small or still too difficult to generate to be of actual
technical interest. They are used where small forces are
sufficient, for example in space applications for the support of
flywheels or for a micro-g-platform in a near zero-g-environment.
Or they are used where the high currents necessary for large
forces are generated by means of cryogenics as in prototypes for
an electrodynamically levitated high-speed vehicle.
It is the electromagnetic force that is used most efficiently.
For the l~vitation of guided vehicles a technology of its own has
developed /GO 84/, that basically of course has some connection
to magnetic bearings for rotors, too. Rotors have been supported
magnetically at first for physical experiments. Spectacular
2.10E7 rpm have been reached while testing the strength of small
steel balls under a centrifugal field of l.OE8 g /BY 46/. Since
then the electromagnetic rotor bearing has been applied to solve
a number of different technical problems, and therefore the
construction and the properties of the bearings differ
remarkably. Numerous patents in this field have been issued or
are pending. And a few types of bearings are already available
commercially for use in machine tools and in turbomachinery /HB
85/. More details on applications will be given in chapter 6. Now
at first the functional principle, some design goals, and the
elements of the bearing system will be discussed.

2. FUNCTIONAL PRINCIPLE
Fig. shows a hovering shaft S. Any deviation from this
reference position is measured by a sensor, the measured signal
is transformed into a control signal according to a suitable
control law. The control signal is amplified and fed as a control
current to the coils of the e 1ectromagnet, the actuator within
this control loop. When the shaft for example starts to fall down
it produces a measuring signal which leads to an increase in the
Magnetic Bearings 545

Fig. 1: Magnetic suspension

contro 1 current and thus the increasing magnetic force attracts

the shaft again. Without the feedback the shaft would either fall
down or be attracted by the magnet. The control law has to take
care of the stability and of the dynamic properties of the
hovering state. The magnetic forces can be made to be a function
of the rotor motion in such a way that the actuator usually has
spring and damping characteristics which suitably depend on the
excitation frequency.
For fully suspending a complete rotor the simple loop explained
above will not be sufficient. Fig. 2 shows part of the radial
suspension of a rigid rotor with four degrees of freedom. For
each degree of freedom a magnetic actuator has to be controlled
individually. The control signals, however, depend on one
another, i.e. each bearing force will depend on all sensor
signals, leading to a typical multivariable control. The axial
suspension of the rotor is not shown here. Its control is
decoupled from the radial one and can be dealt with separately.
Figure 3 shows the hardware setup for such a bearing system. The
rotor has a length of about 1 m, a diameter of 120 mm and a mass
of 12 kg. The air gap of 10 mm is extremely large for technical
purposes. The device was used for the exposition "Phanomena" in
Zurich as a demonstration object. For measuring the rotor
displacement within the large gap optical CCD-sensors are
applied. Their signals are directly processed by a microprocessor
and fed to switched power amplifieres. The block in between the
two bearings is a simple asynchronous motor drive.
546 G. Schweitzer

Fig. 2: Block-diagram for the radial suspension of a rotor in one

plane

Fig. 3: Magnetic bearing system, front view with control unit,

power supply and drive /TM 84/

3. DESIGN GOALS
Primary design goals have been to support a rigid rotor, because
the rigidity of the rotor facilitates the control design
essentially. Magnetic bearings are quite capable of supporting
even a heavy rotor. The considerable freedom, however, in
assigning dynamic characteristics to the actuator can be used not
Magnetic Bearings 547

only for supporting a rigid rotor or even an elastic rotor but

for its vibration control as well. Some efforts have already been
undertaken to control various kinds of vibrations. Pietruszka and
Wagner /PW 82/ show how an unbalanced rigid rotor can be made to
rotate about its principal axis of inertia, thus avoiding
vibrational unbalance forces on the bearing foundations.
Gondhalekar et al. /GH 84,SA 84/ investigated vibrational control
problems of an elastic rotor. Even the active damping of self-
excited vibration caused by internal friction and of parametric
vibration due to rotor asymmetry have been looked into.
This survey, however, will restrict itself to the basic features
of the magnetic bearings and to their current main applications.

4. ELEMENTS OF THE MAGNETIC BEARING SYSTEM

The magnetic suspension of a rotor in reality is a well planned
interconnection of various elements, a system. Its design is
based on three steps, characterizing the different levels of
expertise and teamwork necessary for a good overall solution. On
the first level the elements of the bearing system have to be
dealt with, the rotor, the sensors, the controller, the
amplifiers and the magnetic actuators. Each of it has to be
specified with respect to its tasks within the control loop.
Their main characteristics will be presented in this chapter,
emphasizing the most prominent element, the magnetic actuator. On
the second level aspects of the control loop have to be
discussed, and the control law has to be chosen according to the
design goals and the desired overall characteristics. These
design goa 1s most often are determined by the specific
application, and this has to be taken account of on the third
level, where an efficient team work between different specialists
will be essential. These steps, which of course have to be
iterated, will be outlined in the next chapters.

4.1 Model of the rotor

The rotor obviously is the central part of the suspension system,
and it has to be modeled mathematically as the plant within the
control loop. At first let us describe a rigid rotor and then we
will model an elastic rotor, which will be somewhat more
complicated.
Model for rigid rotor: The rotor with a vertical axis (fig. 4)
shall be kept in this reference position by two radial bearings
and of course an axial bearing. The axial suspension with one
degree of freedom is decoupled from the radial one and will not
be dealt with here. The radial displacements are expressed either
by the "analytical" coordinates for the translation of the center
of gravity x,y and the inclination of the rotor a,S, or by the
548 G. Schweitzer

;- 1 _
· Sensor plane "d"

~-- ~ Bearing plane "b"

+--G_~_- _!_a--I-c .. X
Bearing pla{le "a"

----- ~- - - - - - - - - Sensor plane "c"

cb2
I
Fig. 4: Rigid rotor model /BL 84/

"technical" coordinates, the displacements x , ~d' ~c' yd as

measured by the sensors. The geometry is gi v~n 1n f1 g. 4, the
mass of the rotor is m, the rotor is symmetric, any unbalance is
sma 11. Then the linearized equation for the radi a 1 motion z is
obtained as

~ i + E. i + ~ ~ = l!f .!l.f + Y.f~ , ~ = [ S , x, - J

a,y T ( 4. 1)

The vector .!l.f contains the bearing forces

-uf = [ f ax , fb x , f ay , fb y] T
(4.2)
The harmonic disturbances by unbalance are introduced by the
vector

~= [sin ~t. cos ~t] T (4.3)

The structura 1 matrices for mass, gyroscopic effects and

I
stiffness are

0 0: 1 0
p = I ~ _o- ~ ~- ~ -~- ' s = 0 (4.4)
Z -1 0 0 0I

0 0: 0 0
I
Magnetic Bearings 549

The gyroscopic matrix P is proportional to the rotational

speed ~and the axial moment of inertia I • It is the only matrix
coupling the two lateral directions (x, ~) and (y, a.) with each
other. The matrix S describes all position dependent forces
except the bearing forces, and it contains pendulous forces and
small coupling forces to the axial suspension which may be
neglected here. Matrix ~f introducing the bearing forces consists
mainly of the sign-valued coordinates a and b for the bearing
locations. And the matrix Y._f for the unbalance effects contains
the products of inertia I , I and the excentri city e of the
centre of gravity C from t~€ (m~~etic) rotor axis.

a b :0 0 -I I
1 1 I0 0 Oyz mexz
~f --- -'----
I y_f = ~2 (4.5)
0 0 1a b I xz I
0 0 II 1 1 me Oyz

The rotor displacements z are expressed in terms of the measured

displacement signals -s
z from the sensors by
C 1I 0 0
d 1: 0 0
I=----+--- (4.6)
.o 0 I c 1
0 0 : d 1

This can be substituted into the equations of motion in order to

express them in the measured variables z as well. Furthermore
for contra 1 purposes it is appropriate t~ use the state space
representation

-x =[ -s i T] T
z T -s ( 4. 7)

l [ l [ l
E I- 1
, B=
-
-
I !1.-0 1 ~f
V=
-
-
I !1.-0 1 Y..f

In this representation the input of the rotor, the controlled

bearing forces ~f are connected to th~ output, the measured
displacement signals ~ • Using control techniques the input has
now to be generated su~h that the output behaves as desired, for
example by letting the control forces act as damping and
restoring forces.
550 G. Schweitzer

Model for an elastic rotor: In principle the objective is the

same-as for~he rigid rotor. We will establish a relation between
the input, the control forces, and the output, the motion at
distinct locations of the rotor. The model has to be of low
order, so that it can be incorporated into the control loop more
easily. For the derivation of such a model /BU 85/ let us start
with a finite element model of the elastic rotor

(4.8)
where g is a n x 1 vector of generalized displacements, £ is the
vector of generalized forces, and the A. are structural matrices,
characterizing inertia and elasticity -Js well as gyroscopic and
nonconservative properties. Equ. (4.8) may be transformed into
the complex frequency domain and expressed by

[s 2 ~2 + s ~1 + ~] Q(s) = f(s) (4.9)

The matrix in brackets is termed dynamical stiffness matrix and
its inverse ~(s) is the dynamical flexibility matrix

~(s) = [s 2 ~2 + s ~1 + ~]- 1 (4.10)

The elements of H(s), the dynamical flexibility transfer
functions, are well-known in measurement and modal analysis
techniques. A general way of reducing the large set of finite
element equations is to truncate the modal representation of equ.
(4.8). In order to find that representation we have to solve the
eigenvalue problem of equ. (4.8). For the sake of simplicity -
a 11 the subsequent derivations can be extended to the genera 1
system (4.8) as well /BU 85/- let us assume a simple elastic
structure
( 4. 11)

From the so 1uti on of the ei genva 1ue prob 1em we obtain a set of
e~genvalues w1,._.. w0 and the c?rresponding real normalized
e1genvectors ~k w1th the modal matr1x ~ =[~ 1 •••• ~k····~n]·
Then the modal expansion of g

g = l: ~k z k = ~ ~ ' k = 1, ••• ,n (4. 12)

k
leads to the modal representation of equ. (4.11)
Mi + K z = ~T £ (4. 13)

with
M= ~T Me ~ = diag (mk)

! = ~T !a ~ = diag (kk)
Magnetic Bearings 551

This modal representation is a time domain representation of the

partial fraction expansion of the corresponding flexibility
matrix H(s). Translating equations (4. 12, 4. 13) into the s-
domain, yields

T
2 + K) -1 UT = I:n ~k ~k
H(s)
- = --
U(M s - - k=1 k 2 2
(4. 14}
m (s + wk )

For the design of the controller, a model of the flexible

structure, that relates the displacements and velocities at the
location of the sensors to the excitation forces of the
actuators, is required. By analyzing the closed loop system with
respect to the actuator and sensor coup 1i ng points, it can be
checked whether the design objectives have been reached, i.e.
whether the system is stable and shows up with good damping
performance. Also, test measurements for the closed loop system
are easily carried out by exciting the structure with the
avail ab 1e actuators. Hence, for both contra 1 design and c 1osed
loop analysis, we may confine the selection of coordinates to the
coupling points of the actuators and sensors with the structure.
A modal representation of the dynamical flexibility matrix of the
elastic structure, with respect to the cited coordinates,
provides the desired description, as will be shown subsequently.
The "A actuator forces form the vector !• the n measured
displacements are given by 1o• and the n measured ve~ocities by
1 . These vectors are related by approp~iate incidence matrices
t~ the nodal forces and displacements by

T
! = IA £ • 1 0 = I 0 s. 1v = Iv s. 1 T = [ 1o·
T 1vT] (4. 15)

The elements of the incidence matrices are zero or one, when

discrete actuators and sensors are 1ocated at nod a 1 points. We
assume that we can truncate the number of modes retaining only m
essential modes, so that instead of equ. (4. 12) we now have
n~
.:1.
I:• -1
u. Z.
1
= -m
U Z
-m i=1, ••• m; m<n (4.16)
1

with the truncated n x m modal matrix U and the truncated modal

vector -m
z • This leads to the truncated SWt of modal equations
M
-m -m
z
+ -m
K z
-m
= -m
UT n
J:.
( 4. 17)

With the subset of retained coordinates (4.15) we have in the s-

domain a transfer maxtrix representation
 l
552 G. Schweitzer

=~
lo !!m( s 2 !:\n
+ ~) -1 !!mT !A
.Y_(s) !!(s) ~ =[ 2 -1 T (4.18)
s !v !!m<s !:\n + ~) !!m !A
Here the trans f-er matrix H is a generalized dynami ca 1
flexibility matrix, since it rifates a force excitation (actuator
forces) to nodal displacements and velocities (sensor signals).
By introducing the state vector
T
~ =[ ~.
T • T]
~
an equivalent state space representation is obtained
= -e
C

l [l
V X
-e -e (4. 19)

A=- -
E ,B= .0- , c =[ !o0!!m
-e
0
r.
-M-1 K 0
-m -m -
-e U1 T
-111 -A
-e

All of the representations (equs. 4. 17, 4. 18, 4. 19) will be

suitable for the control design. The method can be extended to
include other structural elements as well, i.e. foundation
dynamics, by applying the building block approach. The rotor
dynamics program MADYN /MA 82/ has been readily adopted to model
magnetically controlled elastic rotors. Problems left over and to
be discussed later are that of truncation and the control design
itself.
4.2 Sensor, controller, amplifier
The sensors have to measure the rotor motion. The quality of this
measurement is essential for the quality of the suspension. The
main requirement is that there is no drift of the zero-output.
The resolution is directly connected to the positioning accuracy
of the rotor, i.e. when the rotor has to be positioned within
1 micron the sensor output has to be at least as sensitive as
that. The frequency response of the sensors is part of the
dynamic properties of the suspension. The sensors should be
insensitive to magnetic fields and to temperature variations.
Good results have been obtained with inductive sensors, eddy-
current pick-offs and with optical sensors. In the latter case
CCD-arrays with subsequent direct digital control have been used.
The controller has to process the measured signals. It may
consist of filters, of the controller in the strict sense, and of
a linearization unit. It may be built as an analogue network, or
it is a microprocessor where the different tasks are taken over
by algorithms. Section 5.1 will show more details on that. The
Magnetic Bearings 553

control law itself has to be chosen according to the requirements

of the specific application, leading to performance values, which
are discussed later on.
The power amplifiers have to amplify the control signals and feed
them to the bearing coils. The input of an amplifier, the control
voltage, can generate an output voltage or an output current, and
consequently the input/output- characteristics of the bearing
have to be described accordingly. The bearing coils represent an
essentially inductive load. Therefore the capability of the
amplifier to drive a high-frequency current through this load
mainly determines the frequency response, i.e. the dynamics, of
the whole suspension system. Furthermore we distinguish between
analogue DC-amplifiers and switched amplifiers. The first ones
are usually applied when the load is small (less than 0.5 kVA)
and when the amplifier losses are not important. Switched
amplifiers work at switching frequencies outside the audio range.
They do not yet seem to be commercially and easily a·.... !lable. In
the following the properties of two amplifiers for two different
bearings are listed:
Bearing load 180 N 160 N
Bearing diameter 80 mm 120 mm
Bearing gap 0.7 mm 10 mm
Type of amplifier DC switched with 28 kHz
Ampl. voltage/current/losses ! 50/1 A/1 OOW !:300V/2.6A/30W

4.3 Magnetic actuator

The electromagnetic actuator is the element within the control
loop, that transforms a voltage or a current input into the most
desired output: the bearing force. Fig. 5 shows the magnetic

a b

Fig. 5: Radial magnetic Fig. 6: Two different configurations for

actuator as a radial bearing
input/output
element
554 G. Schweitzer

actuator as such an input/output element, whereas fig. 6 already

shows constructional details. The two typical constructions for a
radial actuator differ mainly in the path for the magnetic flux
with respect to the rotor axis, and as a consequence thereof the
magnetic losses are different. The characteristics of the
actuator have to be calculated from geometrical and electrical
data, or they have to be measured.

For calculating the bearing forces the radial bearing of fig. 6b

is assumed to consist of U-shaped elements as shown in fig. 7.

Fig. 7: Single U-shaped electromagnet as part of a radial

bearing a) real geometry b) simplified geometry

Each of them exerts forces in the positive or negative X- or Y-

direction. The force of each magnet is controlled by a control
current. The force resu 1t i ng from a certain current depends on
the geometry of the magnet and on the used ferromagnetic
material. The nonlinear behaviour of the ferromagnetic material
makes the analytical calculation of the forces difficult. There-
fore, the U-shaped magnets are modeled with a simplified geometry
and then the forces are calculated numerically.' Figure 7 shows
the relation between the real iron path and the simplified one.
The cross-section a 1 area Afe = a x b of the ferrous core is
constant along the whole length lfe of the iron path. The width c
of the pole shoes is used only for the calculation of the cross-
sectional area Al of the airgap lo. With the simplified geometry,
the measured magnetisation-curv~ B = B(H) can be modified to a
socalled sheared magnetisation-curve which relates the flux <t> to
the ampere-turns 8. This relation is given by the two equations

(4.20)

2Af e 1o
8 =H lfe + B (4.21)
Al ~o
Magnetic Bearings 555

where 0 is the product of the current and the number of

windings n
0 = n (4.22)
With equations (4.20) and (4.21) we can find the flux <I> for a
given ampere-turn 0. From the flux <I> we can calculate the force f
(4.23)
The four equations connect the input current i, the airgap lo,
and the force f on the rotor with the electrical and geometrical
characteristics of the bearing. The variables can be time-varying
as well. As long as no saturation occurs within the iron the
force wi 11 be proportion a1 to the square of the current i (fig.
8a) and it will be inversely proportional to the square of the
airgap lo (fig. 8b). The deviation between calculated and
measured values is mainly due to stray effects which have not
been considered in the simplified mathematical model. The
influence of stray fields, however, is small as has been shown by
numerical flux-line investigations /TR 85/.
The experimenta 1 resu 1ts on the forces were obtai ned by using a
piezo-electric dynamometer for static loads and for load
variations up to 400 Hz. Above that until 1400 Hz indirect
measurements have been used by measuring the acce 1erat ion of a
test rotor in order to identify the magnetic forces exciting the
rotor motion. The amplitude and phase response for the
force/current factor of fig. 9 is flat within the measuring range

FORCE UN
300.

200.

i0~.

21111.. HZ 1.4- K
FORCE f/N
300.

-- :r-t------~----.jl
.s ..... , '' t
IIIRGIIP lotmm
l
Fig. 8: Calculated (----) and Fig. 9: Force/current factor de-
measured (---,o) bea- pending on the excitation
ring force of a radial frequency
bearing (diameter = 80 mm,
a = 40 mm, n = 720)
556 G. Schweitzer

up to 1400 Hz and has to be corrected only at about 900 Hz where

the test rotor had a bending resonance /TR 85/. That means that
for this specific bearing the ratio of force to current for small
signals is constant at least up to 1400 Hz.
The maximally permissible current depends on the copper losses in
the windings of the coil and its cooling capacity.The simplest
model for the cooling capacity is given by

pmax =A h T (4.24)
where Pmax is the cooling power for the temperature-difference
T, A is the surface of the magnetic bearing and h is a cooling
factor, depending on the cooling medium. For a given application
one can of course use much more sophisticated models. In most
cases, the eddy-current 1asses and the hysteres i s-1 asses in the
stator are negligible compared to the copper losses. Therefore we
can calculate the maximum ampere-turns for one magnet by

P
max Aw f a
e max = n ; max = -4- PT (4.25)
w

where Aw is the cross-sectional area of the coil, fa is the

copper filling factor which gives together with Aw the copper 1 s
cross-section, p is the specific resistance of copper and lw is
the 1ength of one winding of the coil. In order to ca 1cu 1ate the
maximum forces for a magnetic bearing we proceed in the following
steps:
1. Calculation of the cooling power, i.e. the maximum copper loss
power , for a given geometry and cooling medium.
2. Calculation of maximal ampere-turns 0max for the given
geometry, where for example the space for the coils Aw and the
length of one winding lw may be given.
3. With the equations (4.20) and (4.21) and the relation B = B(H)
for the magnetisation-curve we get the flux ~ belonging to the
given 0 •
4. Calculation of the bearing force with (4.23).
Of course this procedure may have to be iterated in order to
obtain an optimal configuration.

A bearing may be termed optimal, for example, when for a given

bearing space the magnetic force has a maximal value. Assuming
the geometry to be fixed, except for one free parameter - the
Magnetic Bearings 557

core width b - the abovE' design rules are iteratively applied,

leading to the results shown in fig. 10. For each air-gap lo
there is an optimal core width b where the force f is maximal.
The cooling power has been kept constant in this example.

FORCE f'/N
500.0

400. 0

300.0
POLEWIDTH b

200. 0

100. 0

"· " ........._._................_._................_._................_._...._.......

_._..........

2."
"· " •5 1. " 1. 5
AIRGAP lo/"'"'

Fig. 10: Optimization of the force, depending on the airgap lo

and the core width b

The force/current relation of fig. 8 is nonlinear. For the use of

the bearing magnet within the control loop, however, a linear
current-force characteristic is most desired. This can be
achieved in two different ways. The nonlinear characteristic of
fig. 8 can be linearized by a linearization network, or by a
suitable algorithm in a microprocessor. Another very convenient
method uses differentia 1 winding of the coi 1s and
premagnetization /SL 76/. In this case the magnetic force is
given by

F(x,i) = kX x + k. i
1
(4.26)
where the force/displacement factor k and the force/current
factor k. depend on the premagnetization~ the nominal air-gap and
the geom~try of the actuator. For small displacements and small
control currents linear characteristics as in fig. 11 are
obtained. A disadvantage of this linearization is that the
premagnetizing current causes higher copper losses: the advantage
is that this bearing is especially well applicable to control
purposes.
558 G. Schweitzer

t.O
Bearing Force ~
0
0.8

-o. a 0.8

Fig. 11: Theoretical (--) and measured (av a• •) bearing force

characteristics for a premagnetized bearing
(x ~ relative displacement)
The load capacity of the bearing depends on its size and
geometry, its magnetic material, cooling power, and on th~
control current. The specific load capacity amounts to 50 N/cm
for conventional transformer Si-alloys under a magnetic field of
1.5T. It can be increased by 50 % by using special Co-alloys. By
increasing the bearing size obviously high loads can be
supported. The potentia 1 of the actuator of course can on 1y be
used when it is suitably controlled and incorporated in the total
rotor-bearing-system. This will be shown more explicitly in the
following chapters.

5. SYSTEM ASPECTS
5.1 Control of rigid and elastic rotors
Any control requires information about the motion of the rotor.
Here the rotor position is measured by distance sensors. The
signals for the displacement speed can be produced in three
different ways: by electronically differentiating the
Magnetic Bearings 559

displacement signal, by directly measuring the velocity with

sui tab 1e sensors, or by using a reduced Luenberger observer, an
algorithm for estimating not mea,sured data from measured ones.
The various control schemes differ in how the available signals
are used to derive control signals. In the following the control
for a rigid rotor and for an elastic rotor will be outlined.
The control of ~ rigid rotor is based on linear multivariable
contro 1 theory. A robust feedback, in the sense of low
sensitivity to parameter changes, can be computed by minimizing
the performance index J

00

J = f ( 5. 1)
0

where~ and ~fare motion and control variables (eq. 4.7). When
all the state variables x, displacements z and their time
i are used, we have a complete -tor central) state
derivatives -s
feedback. Then the control ~f depends linearly on the state
(5.2)

where K is the gain matrix. In our case (fig. 12a), with 8 state
variables and 4 control variables for the two radial bearings,
the gain matrix has 32 coefficients. Obviously this large number
of coefficients makes any realization of the controller somewhat
cumbersome.
a) b)

Fig. 12: Decentralized vs. central feedback structure

Therefore a decentralized control has been designed /BL 84/ where
the control signals for each bearing only depend on the motion of
the rotor at or near this bearing (fig. 12b). Thus the central
control is broken down into smaller subsystems. Their
coefficients are still determined by minimizing a quadratic error
integra 1 as in eq. ( 5. 1), and the contro 1 retains to some extent
560 G. Schweitzer

the robustness properties of the complete state feedback. Now due

to the simplified structure of the controller it is feasible to
use direct digital control with a microprocesor. Cycle times of
less than 1 ms have been achieved /TM 84/. The application of a
microprocessor has the addition a 1 advantage that it can perform
operational tasks and redundancy management at the same time much
more versatile than an analogue network.
In any case, whether we use a central or a decentralized
feedback, the dynamic behaviour of the rotor is determined by
choosing the weighting matrices Q and R in eq. (5.1). The matrix
R influences the amount of control power to be used, and the
elements of Q weigh the displacements and velocities of the
rotor. Another possibility to influence the dynamics is to assign
the poles of the system behaviour, i.e. to assign natural
frequencies and damping. Theoretically, for complete state
feedback and unrestricted control forces even arbitrary dynamics
can be assigned. Other methods, too, for designing a control
loop, including integral feedback and disturbance compensation,
have been applied successfully. Practical values that have been
attained for system characteristics will be discussed in section
5. 3.

The control of the elastic rotor can be accomplished by the same

methods as for the rigid rotor. There are pragmatic solutions to
the control design, a systematic approach, however, is still
under investigation. The problems arise from the following
reasons.
The control design has to be based on a reduced-order model of
the rotor (equ. 4.17), containing only the most important modes
of the elastic rotor. The question is whether the real rotor will
indeed be controlled correctly by such a simplified controller.
Actually there may be detrimental effects, and they will be
explained subsequently by using the elastic rotor model.
Let us partition the high-dimensional modal coordinate vector z
of equ. (4.13) into two parts. The important part with the low
dimension n will be the one to be controlled, the other residual
part of dirlrension n is the one to be neglected. Then equation
(4.13) and the meas~rement equations (4.15) can be arranged in
the following form

( 5. 3)

~0 = -T0 -m
u u -r
z + -T0 -r
-m z , ~v = -Tv -m i
u -m i
u -r
+ -Tv -r
Magnetic Bearings 561

For design purposes it will be assumed that the controller is

based on the reduced-order mode 1 only, being the m-subset of
equations (5.3):

T T
M
-m-m z K z = -mJ:.
+ -m-m U n =-Um w
TA- -

(5.4)
~0m = -T0 -m-
u z , ~v m = -Tv -m-m
u z

The relation between the control vector w and the measurement

follows from the chosen law, which preferably will be a linear
one and can be determined in the usua 1 ways. Such a contro 1
design will result i. n the desired and "good" centro 1 for the
reduced-order mode 1 ( 5. 4), but when it is app 1i ed to the rea 1
full-order-system (5.3) the system qualities can alter
essentially (fig. 13). In reality the measurements~ do not only
consist of the modeled part z but they also depend on the
residual vector -r
z , causing the 1'ocalled "observation spillover".

IM

Actuator Rotor Sensor 100

M, z, + K, Zr

cs OS

Control
y

Fig. 13: Reduced order centro 1 Fig. 14: Typical full-order-

applied to the real full- closed-loop eigenvalue
order system, demonstra- curves for an elastic
ting the observation spill- rotor with low stiff-
over (OS) and the control ness magnetic bearings
spill-over (CS) and increasing damping
D assigned to the modes
562 G. Schweitzer

Furthermore the control vector w does not only act on the modeled
part but on the real full-order system itself which obviously
contains the unmodeled part as well. This influence is called
"contro 1 spi 11 over". These spi 11 over terms can change and
deteriorate the behaviour of the real system and even destabilize
it.
The objective of the control design now is to derive the control
on the basis of the reduced-order system without exact knowledge
about the residual, unmodelled part.
It is advisable to use a direct output feedback, i.e. the control
forces depend linearly on the displacement and velocity
measurements
(5.5)
with the yet unkown gain matrices G.• Introducing this control
law and the measurement equations -~4. 15) into the full-order
system equations (4. 13) leads to the full-order closed-loop
system
0
!1 ••
~ + !!T IA §v Iv !! • (
! + ~ + !!
T
IA §o Io !! ) ! =- ( 5. 6)

In order to avoid spillover effects the gain matrices G. have to

be determined such that the solutions of (5.6) are stable and
show some desired and specified behaviour. In /SA 84/ the gain
matrices G. are derived so that the resulting reduced-order
control sh6ws robustness qualities with respect to parameter
errors and truncation effects and 1eads to a stab 1e full-order
closed-loop system of even infinite dimension, if
- the rotor-bearing system is not unstable from the beginning
(rigid body modes allowed),
- the reduced-order model contains at least all rigid-body
modes,
- the sensors and actuators are collocated and their number is
at least equal to the number of rigid-body modes.
This control approach includes the technically most interesting
case where a real flexible rotor is suspended by actuators which
at the same time have to control its elastic vibrations. As an
example a flexible homogeneous beam supported at its ends will be
considered. The two bearings will have to control the two rigid
body modes, and as a design goal, two elastic modes should be
simultaneously damped with these bearings, too. The gain matrices
for the direct output control are determined such that a desired
bearing stiffness is obtained, characterized by the natural
frequencies of the two rigid body modes. Then the damping D
assigned to the two rigid body modes and the two elastic modes is
varied between 0 and 1, and the resulting root locus curves are
represented in fig. 14. When we increase the damping the
Magnetic Bearings 563

eigenvalues of the originally elastic modes are going back to the

imaginary axis along semi-circles, ending there at the modal
frequencies of a "fixed-fixed" beam.
5.2 Characteristics of the magnetic bearing system
Typical specification terms for rotor bearings are the maximal
load capacity, the specific load capacity, stiffness, damping,
frequency response, maximal angular velocity, and losses.
The maximal load capacity, as derived in chapter 4.3, depends on
the bearing size, the magnetic material and the control current,
and there are real bearings where it amounts up to 50 kN~ When
the actual load is larger than the allowable maximal load the
rotor can not hover any more and touches upon the bearing.
The specific load capacity relates the maximum load capacity to
the cross sectional area of the bearing hole. Characteristic
values ffr usual Fe/Si-alloys with a saturation of 1.5 2esla are
50 N/cm • With special Fe/Co alloys up to 80 N/cm can be
obtained.
The stiffness of the bearing is typically frequency dependent.
For static loads - the frequency of the loading is zero - an
integrating feedback of the displacement signals leads to a
theoretically infinite stiffness. For load frequencies below the
cut-off frequency of the bearing control, which usually is
between 100 Hz and 1500 Hz, the stiffnesss characteristics can be
shaped according to requirements. This property can be used to
counteract disturbance forces. For load frequencies above the
cut-off frequency the stiffness gradually decreases and finally:
becomes very small. Very high frequency forces are thus not
actively transmitted by the bearing, which makes it quite useful
for vibration isolation.
The damping that can be obtained by the bearing depends on the
control laws and on the available force. This available force is
given by the maxima 1 load capacity, and it can be used as a
restoring force, leading to the bearing stiffness, or as a
damping force. The frequency dependence typically is the same as
that for the stiffness. Of course any combination of damping and
stiffness can simultaneously be generated as long as the total
force does not exceed the load capacity.
The frequency response characterizes the dynamic behaviour of the
bearing. The actuator itself corresponds to a series of R-L-
elements. And a harmonic voltage input, with the highest
amplifier voltage-amplitude possible, generates a current through
the coils with a frequency response as in fig. 15. The cut-off
frequency f corresponds to a time-constant of T=1/f for a step-
input, whi~ can be easily measured. Most often, %owever, the
power-amplifier is internally controlled as a voltage-current
564 G. Schweitzer

converter. This reduces the order of the overall-control-loop and

reduces the time constant of coil and amplifier by the factor of
the open-loop-gain of the amplifier. Thus the current of the
bearing -and consequently the magnetic force- follows the input
voltage practically without delay. Of course this behaviour is
only valid within the current-frequency response of fig. 15, i.e.,
the transfer function for amplifier and bearing refers to small

a)
b)
imi'IH
imaH

0.1

0.01
f/Hz 1/ms
: fg

10 100 1000 0 2 4 6 8

Fig. 15: a) Frequency response of the current through a bearing

coil, with full amplifier voltage applied
b) step response
signals within the limits of the frequency response of fig.l5.
And within these limits the transfer function of the closed loop
itself, determining the dynamics of the rotor suspension, can be
shaped almost arbitrarily by a suitable control design.
Further dynamic characteristics and their relation to the size of
the power amplifier are derived in /TR 85/.
The maximal rotor velocity depends on two factors. One is the
strength of material of the rotor. Because of the laminated
ferromagnetic sheets on the rotor circumferential velocities of
more than 200 m/s are difficult to obtain. The second factor is
the power of the rotor drive necessary to overcome rotor braking
torques. As shown in the next section these 1osses are sma 11
compared to conventionally supported rotors, and they have to be
considered only at high speeds.
The losses of the magnetic suspension mainly occur in the rotor,
the actuator and the power amp 1ifi er /TR 85/. The rotor 1osses
are the equivalent to the friction in hydrodynamic- or
ro 11 erbeari ngs as these 1osses have to be overcome by the rotor
drive, too. The magnetic 1osses in the rotor are caused by the
modulation of the magnetic flux in the ferromagnetic part of the
turning rotor, when it passes the poles of the stator magnets.
This modulation is much lower in the bearing of fig.6a than in
the bearing of fig.6b, because the rotor parts pass only poles
with the same polarity. These magnetic losses are caused by
hysteresis and by eddy currents. Hysteresis losses depend on the
Magnetic Bearings 565

materia 1, used for the ferromagnetic part of the rotor in the

bearing area. They are proportional to the rotation frequen-
cy ~ and the square of the flux density B:

Plh = kh *~* B2 (5.7)

with kh being a material constant. The eddy current losses also
depend on the material used. The ferromagnetic part of the rotor
should be laminated and built from thin discs or rings to reduce
the eddy currents. The losses are proportional to the square of
the rotation frequency, and the square of the thickness s of the
ferromagnetic material and they can be approximated by
2
Ple = ke * s * ( 5. 8)

The other rotor losses, the air losses, depend on the geometry of
the rotor and its surrounding parts. For an example the run-down
curve of a magnetically suspended rotor has been measured in air
and in vacuum (fig. 16). The braking torques caused by the various
effects are shown in fig. 17.
The stator losses are dominated by the copper losses due to the
resistance in the coils. According to equations (4.24, 4.25) they
are proportional to the square of the current i in the coils,
plc = kc * i2 (5.9)

As a consequence, considering (4.23), the stator losses typically

increase proportional with the bearing force f.
The amplifier losses are the main source for the electrical
losses. In the conventional analogue amplifiers the losses depend
on the maximal output voltage, that has to be kept available for
a good dynamic performance even when it is not required by the
quiescent load. Switched amplifiers, however, work much more
economi ca 11 y as their 1os ses are dependent on the actua 1 1oad
only. In this case the amplifier loss increases proportional to
the load.

6. APPLICATIONS
Magnetic bearings have been used for rotating machines in vacuum-
techniques because there wi 11 be no contamination by wear or
lubricants. One of the first industrial applications was for a
turbomolecular pump. An extremly low vibration level ( o.o5 mm/s
at the pump), an operational speed of 5oo Hz, and an ultrahigh
vacuum of 10 E-10 mbar free of any hydrocarbon contamination are
outstanding specifications. Now several companies are developing
and building such pumps.
566 G. Schweitzer

U/min

51!11!10..1!1

41!11!10..1!1

31!11!10..1!1

21!11!11!1.1!1

ll!ll!li!J.I!l

1!1.1!1 L---........__~~ .........~~~__.__.~~----

0.0 21!10.0 400.0 600.0 801!1.0 t/s

Fig. 16: Run-down curve for a test-rotor

1 - in air, 2 - in vacuum)
Bremsmoment/Nm

.l?l225

• l?l2

.l?l175

.l?llS

• IZI125 ___ _..._...

---
• IZil
__,....-
2 ...-....-

.IZIIZI75
....-
~....-
.IZIIZIS
-·-------------
3

.IZIIZI25

121. 121
Drehfrequenz/Hz
l?l. l?l 2IZI. 1Z1 4IZI. 1Z1 60. 0 80. 0 10IZI. l?l

Fig. 17: Braking torques (---measured,--- calculated,

1 - in air, 2 - in vaccum, 3 - only hysteresis losses)
Magnetic Bearings 567

A centrifuge for epitaxy experiments on semiconductor material

under UHV-conditions is using magnetic bearings. The bearings are
outside a vacuum chamber for suspending the rotor within the
vacuum /ST 83/. The set-up of the centrifuge is shown in fig. 18.

Uocuum Housing
Rotor
Motor

Rodiol Beorlng
A•dol Beorlng
Rodiol Sensor

Frome

AKiol Sensor

Ouen
Furnoce

Fig. 18: Set-up of the magnetically supported epitaxy-centrifuge

An eight stage high speed centrifugal compressor on magnetic
bearings has been built and tested recently /HK 85/. The rotor
weight is about 37oo N, the bearing span 1.3 m, and three
cri t i ca 1 speeeds are be 1ow the operating speed of 13, ooo rpm.
He 1i urn compressors, compressors for submarines, and centrifuges
for chemical industry are other examples.
Machine tools, too, have been designed with magnetic bearings: a
lathe, replacing a polishing machine for heavy cylinders in the
graphical industry, high speed milling machines in aerospace
industry, and spindles for high speed grinding of threads.
A future application cou 1d be the vibration contro 1 of e 1ast i c
rotors with magnetic bearings or dampers. For a rotor, modeled in
568 G. Schweitzer

fig. 19 and shown on a test-rig in fig. 20, the vibration

response is given in fig. 21. The reduction of the resonance
amplitude at the critical speed by using an active damper is
obvious.

Fig. 19: Magnetic damper for a Fig. 20: Test rig

rotor on an elastic shaft

a
u.t
c
:::>
1-
...J
ll.
~
<
b

FREQUENCY
Fig. 21: Frequency response Fig. 22: Trajectory of rotor
a - undamped, shaft center for a
b - with active damper rotor touching down
on the bearing
Magnetic Bearings 569

7. CONCLUSIONS

Magnet1c bearings offer some features that allow to tackle some

of the problems of classical rotor dynamics in a new way: no
mechanical wear, no lubrication, low maintenance effort, low
losses, adjustable dynamic performance. Of course there are some
drawbacks, too, at least up to now: the bearing works within a
closed loop system and thus for its design it typically
requires system engineering and mechatronics. This knowledge is
not yet w1dely spread, and therefore the acceptance level in
industry is st i 11 low, except for some advanced machinery. Part
of the current scepticism comes from the question of reliability.
It has not yet been solved in a sytematic way even though for
space applications and other ones the specified reliability has
been demonstrated. Further effort therefore will have to go into
this direction. For example, the dynamic behaviour of a rotor
touching down on an emergency bearing wi 11 have to be
investigated. Fig. 22 shows a typical trajectory of the shaft
center of such a rotor /SC 85/. Another area of interest is the
active control of rotor vibrations, where future theoretical and
practical results can be expected.

Research activities of various companies indicate that these

specific features of magnetic bearings are going to lead to new
designs and a new generation of rotating machinery.

References

/BL 84/ Bleuler, H.: Decentralized control of magnetic rotor-

bearing systems. Diss. ETH Zurich No. 7573, 1984.

/BR 39/ Braunknecht, W.: Freischwebende Kerper im elektrischen

und magnetischen Feld. Zeitschr. f. Physik, 1939,
s. 753-763.
/BU 85/ Bucher, Ch.: Contributions to the modeling of flexible
structures for vibration control. Oiss. ETH Zurich,
No. 7700, 1985.

/BY 46/ Beams, J.W., Young, J.l., and J.W. Moore: The production
of high centrifugal fields. J. Appl. Phys., 1946,
pp. 886-890.

/GH 84/ Gondhalekar, V., and R. Holmes: Design of electro-

magnetic bearing for vibration control of flexible
transmission shaft. NASA Conf. Publ. 2338, 1984.
570 G. Schweitzer

/GO 84/ Gottzei n, E.: Das 11 Magnet i sche Rad 11 a 1s autonome

Funktionseinheit modularer Trag- und Fuhrsysteme fur
Magnetbahnen. Fortschr.Ber. VDI-Z., R.B. Nr. 68,
Dusseldorf VDI-Verlag, 1984.
/HB 85/ Habermann, M., and M. Brunet: The active magnetic
bearing enables optimum control of machine vibrations.
ASME Paper 85-GT -22, Gas Turbine Conf. Houston, March
1985.
/HK 85/ Hustak, J., Kirk, G.R., and K.A. Schoeneck: Active
magnetic bearings for optimum turbomachinery design.
Symp. on Instability in Rotating Machinery, Carson City,
June 1985, NASA Conf. Publ. 2409.
/MA 82/ MADYN-Handbuch, Kramer/Klement, TH-Darmstadt, 1982.
/PW 82/ Pietruszka, W. D., and N. Wagner: Akt i ve Beei nfl us sung
des Schwingungsverhaltens eines magnetisch gelagerten
Rotors. VDI-Bericht Nr. 456, 1982.
/SA 84/ Salm, J., and G. Schweitzer: Modeling and control of a
flexible rotor with magnetic bearings. Conf. on
Vibrations in Rotating Machinery, Inst. of Mech. Eng.,
York, 1984, pp. 553-561.
/SC 85/ Szczygielski, W., and G. Schweitzer: Dynamics of a high-
speed rotor to~ching a boundary. IUTAM/IFToMM-Symp. on
Multibody Dynamics, CISM Udine, Sept. 1985, Springer-
Verlag, 1986.
/SL 76/ Schweitzer, G., and R. Lange: Characteristics of a
Magnetic Rotor Bearing for Active Vibration Control.
Conf. on Vibrations in Rotating Machinery, Inst. of
Mech. Eng., Cambridge, U.K., 1976.
/ST 83/ Schweitzer, G., Traxler, A., Bleuler, H., Sauser, E.,
und P. Koroknay: Magnetische Lagerung einer
Epitaxiezentrifuge bei Hochvakuumbedingungen.
Vakuumtechnik 32 (1983), S. 70-74.
/TM 84/ Traxler, A., Meyer, F., and H.P. Murbach: Fast digital
control of a magnetic bearing with a microprocessor.
11. Internat. Congr. Microelectronics, Munich,
Nov. 1984.
/TR 85/ Traxler, A.: Eigenschaften und Auslegung beruhrungs-
freier elektromagnetischer Lager. Diss. ETH Zurich No.
7851, 1985.
 CHAPTER 12.2

VIBRATIONS IN VARIABLE SPEED MACHINES

H. Inetier
University of Kassel, Kassel, FRG

ABSTRACT

In a wide range of dynamically loaded machines variable

speeds occur which lead to time-dependent excitation
spectra of forced vibrations, passages through resonances
and other phenomena. Some examples of the basics of these
instationary vibrations are discussed and the most impor-
tant mathematical relations for instationary forced vi-
brations of linear systems are presented.

1• INTRODUCTION
An important field in practical mechanical engineering is the
construction of dynamically loaded machines e.g. compressors,
combustion engines, gas and steam turbines and other machines
in which rotating or accelerated components are used /1/, /2/,
/3/. These components are loaded by the forces of the machine
process and/or their own inertia forces. Thus, high stress levels
may occur in particular when the frequency spectrum of the load
excites the component or the machine in resonance.
In a lot of practical cases, these exciting frequency spectra are
time-independent. This is valid for instance for those machines
which run with constant angular velocity and with no change in
the machine process. However, important cases exist where the
exciting frequencies change with time /4/. This non-stationary
case occurs for instance in turbines which run up or run down, in
compressors with external load increase or decrease or in electro-
572 H. lrretier

notor-driven systems. I~ these cases, it is mostly unavoi~a~le

to :Jrotect the r.1achine from resonance vibrations.
The purpose of this pa~er is to consider the snecial effects
which occur in linear systems which are sub:ected to r~sonance
vibrations by excitations ~ith time-denendent spectra. In nar-
ticular the run-u~ and run-down of various machines an~ their
com~onents are consi~ered. Pll of them ~elan~ to t~e class of
r.~ec:-tanical problems which can ~e described ~Y the e0uation of
motion
.
!!~+Q~+!~=.e. ( 1.1)

for a discretized system. In this equation w denotes the dis-

placement vector of the system, ~1. D and K are the mass, damoing
and stiffness matrix, respectively,-and p~is the vector of the
external forces. In a lot of practical cases, for instance in
rotating machinery, these forces are periodic in time and can be
described by the Fourier-series
00

.E. = L £vexp(jv~t) ( 1.2)

v=-oo
in which :! is the angular speed of the machine, t is the time and
p are the Fourier-coefficients. The stiffness matrix K in equa-
tVon (1.1) often is of the form -
! = ~ + ~ 2!~ + ~!(t) (1.3)

in which ~ is a time-independent part ~hich describes the stiff-

ness properties of the system itself, ~ ~ is an additional stiff-
ness term which comes into consideration 1n those cases where the
stiffness properties of the machine change with rotational speed.
Very important examples for this effect are rotating turbine
blades and discs. Finally, ~K(t) describes a time-dependent part
of the stiffness matrix which must be included when parametric
excitation occurs. This is true when the stiffness properties of
the system change periodically. Important examples for such types
of parametric excitations are the vibrations of cracked shafts
where the bending stiffness changes periodically with rotation
and beam-like structures which are loaded in longitudinal direc-
tion by periodic forces.
The solution of the equations (1.1) to (1.3) has been an impor-
tant subject in the past. Many results are reported in the lit-
erature. However, in the most cases they are restricted to con-
stant rotational speeds ~ of the machine which yields spectra
with time-independent frequencies. Time varying excitations as
they occur in variable speed machines were considered much less
Vibrations in Variable Speed Machines 573

and - because of mathenatical difficulty - mostly for some prin-

cipal studies.
In the following chapters some examples are discussed in which
instationary vibrations and passages through resonances occur due
to changing excitation frequencies. The examples stand for the
large amount of related problems in practical mechanical engi-
neering.
2. EXAt 1PLES,
1 i~ECHANICAL i·10DELS AND ~1ATHEt1ATICAL DESCRIPTION
2.1 SPRING-MASS-SYSTEM EXCITED BY FORCE
The simplest model of a mechanical system which is able to per-
form vibrations is the single-degree-of-freedom system shown in
Fig. 1 which consists of a spring with a spring-constant k, a
viscous damper with the coefficient d
and a mass m. If this system is excited
by a harmonic force F(t)=Fcos~~ t its
behaviour is described by the 8ifferen-
tial equation
k d
mw + dw + kw = F"cos~~ 0 t. (2.1)
"
IfF, the exciting force amplitude, and
~ 0 , the exciting frequency, are constan~
equation (2.1) is a standard oroblem of
vibration theory and there is· no need to
repeat the results here. However, when
instationary conditions, for instance by
an increasing or decreasing exciting
frequency, are considered, ~ t must be
replaced by the phase 0
Fig. 1: Spring-mass-
system excited by force t
q>(t) = fi:(t)dt (2.2)
0

where );(t) now is a variable exciting frequency. Thus, the equa-

tion of motion of the spring-mass-system reads
n~ + dw + kw = Fcosq>(t) (2.3)
where w(t) is a more or less complicated function of time. This
differential equation will be solved in chapter 3 for a linear
decreasing and increasing excitation frequency.
2.2 SPRING-MASS-SYSTEM EXCITED BY UNBALANCE
A classical problem of mechanical engineering is the study of the
behaviour of a spring-mass-system which is excited by a rotating
unbalance. A corresponding mechanical model is sho\'m in Fig. 2.
If time-varying angular speeds and, thus, non-zero angular
accelerations
574 H. Irretier

~ = ~(t) + ~ = n(t) = a(t) (2.4)

are included in the consideration, the
equation of motion is (/3/, /5/)
(m+m )w+dw+kw=m u(<P 2coscp+(psin<P)
u u (2.5)
in which <P, ~ and 0 are functions of
time. We notice from this equation of
motion that now besides the phase <P
also the excitation amplitude can be a
complicated time-dependent function.
2.3 UNBALANCED FLEXISLE ROTOR
The understanding of the dynamics of a
flexible rotor, which is excited by
its rotating unbalance, is a classical
problem of mechanical engineering
sciences. In the most casessuch prob-
lems have been considered for a con-
Fig. 2: Spring-mass stant rotational speed of the rotor.
system excited by However, from the practical point of
unbalance view for some types of rotating
machines it is of importance to study
the behaviour of the rotor also for varying rotational speeds
e.g. a run-up or a shut-down.
The most simple model of a rotor is the well-known Laval-rotor
which is shown in Fig. 3. If Yc and zc denote the coordinates of
the centre of mass of the dis~
m its mass, d a viscous damp-
ing coefficient, k the spring
constant of the shaft, £ the
unbalance, J the polar mass
moment of the disc, and T the
torque driving the rotor, the
e~uations of motion of the de-
y fleeted Laval-rotor are(/5/,
/6/' /7/)
myc+dyc+kyc=k£sin<P
mzC+dzC+kzc=k£COS<P (2.6)
z JQ;-kE:(y CCOS<.p-ZCS i n(l)) =T.
Fig. 3: Deflected Laval-rotor We notice from these equa-
tions that generally the rotor
vibrations are described by non-linear differential equations.
However, for the most practical cases for rotors with small un-
balance in comparison to the mass radius of ~yration of the disc,
the non-linear term in equation (2.6) can be neglected, the prob-
Vibrations in Variable Speed Machines 575

lem is linear and the acceleration <Pis proportional to the

driving torque T. Thus, an increasing or decreasing torque yields
a proportional increasing or decreasing angular velocity n and
again the angle ~which describes the angular position of the
disc, is a time-dependent function according to equation (2.2).
In this way, for this instationary case,the differential equa-
tions (2.6) are of similar type as equation (2.3) for the spring-
mass-system excited by an external force.
2.4 ROTATING TURBINE BLADE
In gas and steam turbines the blades are one of the most signifi-
cant components from the structural mechanics point of view. In
Fig. 4 a typical turbine rotor with its shaft, its discs and some
blades is shown. In view of the fact that the rotor blades run

0° 360°
L_ {7 {7 {7 __j
=.I~---------~~--...!_ stator blades
~- rotor blade
P rotation

Fig. 4: Turbine rotor and nozzle excitation of the rotor blades

behind several stages of fixed stator blades,they are loaded pe-
riodically by aerodynamic forces which can be described by the
Fourier-series
p = r p cos(vznt-~v )
v=O v
(2. 7)

where p are the amplitudes of the various harmonics, z is the

number ~f nozzles in the stage, and n is the angular speed of the
blade. If the very simple model of an Euler-Bernoulli-beam is
introduced to describe the behaviour of the blade, the corre-
sponding equation of motion reads
2 2
a (EI a w) + (2.8)
~ ~
where EI is the bending stiffness of the blade, pA its mass den-
sity per unit length and w = w(x,t) describes the deflection. We
notice from this equation of motion that - for varying angular
speeds n of the turbine - again a problem of instationary excita-
tion of a mechanical system must be solved to find the response
576 H. Irretier

of the rotating blade during a run-uo or shut-down.

In general, the rotating blade vibration problem is more compli-
cated than described by the equation of motion (2.8). Refined
models have been developed in the past /8/, /9/. All these mod-
els can be set up in the form of equation (1.1) where it is par-
ticularly important to mention that the stiffness matrix K de-
pends on the angular speed of the blade because the centrTfugal
force field increases the stiffness of the rotating blade consid-
erably. Thus, the stiffness matrix follows the equation (1.3)
with 6K = O,so that an additional effect occurs that is the fact
of changing eigenfrequencies of the system during a run-up or
run-down of the blade. /10/, /11/.
2.5 TORSIONAL VIBRATIONS OF ELECTRO-MOTOR-DRIVEN SYSTEMS
A final example for a mechanical system in which variable speed
problems occur is an electro-motor-driven compressor system. A
sufficient mechanical model for such types of systems is a dis-
cretised spring-mass-model of several degrees of freedom as shown
in Fig. 5. During the run-up or the shut-down of this ~otor-com­
pressor-system one or more torsional critical speeds of the ma-
chine can be passed which yield strongly increasing stress level~
A typical plot of the shaft torque is shown in Fig. 6 for a motor
start-up /12/.

Js

Ks SPEED
.__,:-+--=-+~~1--......;--t INCREASER

JlO J,

MOTOR
l~RPM
20000 HP
COMPRESSOR
5500 RPM

Fig. 5: Typical spring-mass-model for a torsional vibrating

system /12/
This plot of shaft torque shows - and by this means represents a
transition to the following chapters - the typical response of a
linear mechanical system running through resonance. Consequently,
the next considerations deal with these phenomena in application
to the examples discussed here.
Vibrations in Variable Speed Machines 577

2
al
....1

~
w· 0
:::>
0
a:
0
1-

·2
FIRST TORSIONAL
FREQUENCY EXCITED BY
TWICE SLIP FREQUENCY
COMPONENT
·4
0 2 4 6 8 10 12
TIME tSECI

Fig. 6: Shaft tor~ue as function of time during motor start-up

/12/
3. BASIC PHENQii1ENA IN ~1ECHANICAL SYSTEMS RUNNING THROUGH
RESONANCES
The examples in the previous chapter stand for all similar prob-
lems in machines with variable speed. Next, the vibrational be-
haviour of the systems during passages through resonances is con-
sidered.
A typical response of a linear system running through a resonance
is plotted in Fig. 7 which shows the deflection of a Laval-rotor

zc
1 ~=o.o31
10
la=0,02 w~l
,,
5

-5
'' ~/

Fig. 7: Deflection of a Laval-rotor running through resonance as

function of time /7/
578 H. Irretier

(Damping ratio ~ = 0,03) as function of time during a run-up com-

pared with the response curve for stationary resonance which is
synonymous withaninfinite-slow run-up. All other comparible ex-
amples and problems show a similar behaviour during the passage
through resonance. This behaviour is characterized by the follow-
ing phenomena:
i) The maximum of the amplitude of the instationary response
did not occur when the exciting frequency equals the eigen-
frequency of the system but shifts to later times during a
run-up and to earlier times during a run-down .of the system,
respectively. This shifting increases with increasing veloc-
ity of changing of the exciting frequency and vice versa.
ii) The magnitude of the amplitude maximum depends on the veloc-
ity of resonance passing and proves to be smaller than the
amplitude during a stationary resonance. The more the veloc-
ity of frequency changing is, the less is the maximum of the
response amplitude.
iii) After running through resonance typical beat vibrations oc-
cur because the response of the system consists of the ex-
cited and the natural motion, and shortly after resonance
both are vibrations with frequencies closed to each other.
The next step of consideration is to discuss numerical results
for the various examples considered before to find the dependence
of the different parameters of the system and its response during
a passage through resonance. In view of the shortness of the
present paper, it is only possible to give the fundamental re-
sults; for details on the ¢alculations the reader is referred to
the literature.
4. EXN~PLES AND NUMERICAL RESULTS OF SYSTEMS RUNNING THROUGH
RESONANCES
The most considered problem with respect to non-stationary reso-
nance vibration are the spring-mass-systems shown in the Figures
1 and 2. A lot of numerical and even analytical results had been
given in t~e literature. The most extensive reference is /4/
where the mathematics of solution is discussed in detail for a
lot of problems. Recent papers like /5/ and /11/ to /14/ deal to
some extend with applications to special problems and only a few
results can be quoted here.
4.1 SPRING-fo1ASS-SYSTEMS
The spring-mass-system of Fig. 1 is considered first. Intro-
ducing the eigenfrequency, the damping ratio and the dimension-
less time
k • d '[ .. w t (4.1)
w =-' r,;=--
0 m 2/riik 0

the equation of motion (2.3) can be transformed into

W11 + 2tw• + w = coscp(-r) (4.2)
Vibrations in Variable Speed Machines 579

where a unit force for which F/k = "1" is assumed to simplify

matters. If now a linear increasing (a > 0) or decreasing
(a < 0) exciting circular frequency according to the equation
n(t) = n0 + at (4.3)
is assumed, it is possible to solve equation (4.2) analytically.
For this case, the phase ~(t) in equation (4.2) follows from
equation (2.2) to
1 2 1 2
~(t) = n0 t + 2 at or ~(T) = n0 T + ~T , (4.4)
where
and (4.5)

are the exciting circular frequency for t=O related to the

eigenfreguency and the parameter which relates the angular accel-
eration n = a to the square of the eigenfrequency. Introducing
equation (4.4) into (4.2) the ~athematical model of a force ex-
cited linear single-degree -of-freedom system with linear in-
creasing or decreasing exciting frequency becomes
w" + 2r,;w' + w = cos(n 0 1 + ~/) (4.6)
As shown in /4/ the solution of the equation (4.6 e found
in the form
(4. 7)

Where W(T) describes the time-dependent amplitude Ot the response

of the system and ~(1) is a time-dependent phase angle. The am~
plitude function W(T) iS a COmplex function
(4.8)
which can be calculated with the help of probability integrals/4/.
The physical interpretation of the equation (4.7) is very simple
if the dynamical behaviour of a mechanical system excited under
stationary conditions with time-dependent exciting frequency is
considered for comparison. First, the system follows the phase~
given in equation (4.4) but with a phase difference ~which is
time-dependent. Second, the amplitude of the system is time-de-
pendent, too, so that an amplitude-modulated vibration occurs
which depends on the damping ratio and the actual ratio of the
exciting frequency to the eigenfrequency of the system.
The amplitude function lw(T)I and the phase angle ~(T) are calcu-
lated in the references /4/, /5/, /13/ and /14/ for spring-mass-
systems subjected to various types of excitation. Here, there-~
sponse of a spring-mass-system shown in Fig. 1 during a passage
580 H. Irretier

through its resonance frequency is considered first. Fig. 8 shows

the amplitude function \Q(T)! as a function of the dimensionless
angular velocity <P 1 = n +aT for a damping ratio r.;; = 0,02. Para-
meter on the curves is ~he quantity a given in equation (4.5)
which is the most important parameter to decide the behaviour of
the system during the passage through resonance in comparison to
the stationary case. Both cases, a run-up (a> 0) as well as a
run-down (a< 0), are plotted in the diagram in Fig. 8. We notice

!: = 0,02
15
11
_ j 0 (a>O)
0 l100 (a<O)

10

Fig. 8: Amplitude functions for a spring-mass-system excited by

force during a passage through resonance by linear in-
Greasing and decreasing exciting frequency /14/
from the curves that the more a shifts from 0 to positive or
negative values the more the amplitude maximum shifts from the
stationary resonance point and the more its magnitude decreases
in comparison to the stationary case.
As an second example, in Fig. 9 a running through resonance is
plotted for the spring-mass-system under unbalance excitation
(Fig. 2). The curves show the amplitude of the force at the
foundation of the system related to the centrifugal force of the
Vibrations in Variable Speed Machins 581

unbalance. Again, the typical behaviour of the passage through

resonance is visible. In addition, we notice that the maxima of
amplitudes which occur are different for the run-up and the run-
down. ~ore details and parameter studies are given in /5/ and
further investigations are presented in /4/, /13/ and /14/.

72 a.=O

IFF(-dl ~ = 0,02
.2
muU<P

c:o
10 (a>O)
=
(a<O)

2 <P' = n0 -+a:r
Fig. 9: Force at the foundation of a spring-mass-system excited
by unbalance during a passage through resonance by linear
increasing and decreasing exciting frequency /5/
4.2 RUN-UP AND SHUT-DOWN OF A FLEXIBLE ROTOR
To continue our consideration of an unbalanced flexible rotor
in chapter 2.3, now its response is discussed for the case of a
passage through resonance due to a constant angular acceleration.
The describing, linearized equations of motion (2.6) are of
similar type as for a force excited spring-mass-system. Thus, the
solution is possible in a way comparable to that one described in
the previous chapter. Fig. 10 shows the amplitude of a Laval-
rotor running through its critical speed during a run-up (a. > 0)
and a shut-down (a.< 0). The damping ratio is~ = 0,02 and curves
are plotted for various values of the constant angular accelera-
tion a. Again, the typical behaviour of a linear system running
through resonance is visible. The maximum of response occurs
after the critical speed for stationary conditions is reached,
and this maximum is less than in the case of stationary rotation
of the rotor. For acceleration lal > 0,1 no typical resonance re-
582 H. Irretier

sponse is observed even for the small damping ratio of 0,02.

Fig. 10: Response amplitude during a run-up and a shut-down

with constant anqular acceleration /7/
In the discussion of the dynamical behaviour of a flexible rotor
running through critical speeds, an important phenomena may not
be ignored. The plots shown in Fig. 10 are only valid, if the
driving torque T in equation (2.6) exceeds a certain value to
insure that the system is able to overcome the critical resonance
rotation. Details on that are reported in /4/ and /7/.
Here,only some basic considerations are performed.
If the problem of sufficient driving torque is considered mathe-
matically, the complete non-linear equation (2.6) must be solved.
For a constant driving torque T and a large excentricity £ it
proves during the passage through resonance, where y and z in-
crease to large values, that the deviation of the an6ular ve~oci­
ty ~from a linear time-dependent function can increase to a con-
siderable amount. This effect is shown in the left diagram of
Fig. 11 which shows the angular velocity as a function of time
during the passage through resonance for the case of a sufficient
driving torque to exceed the resonance vibration. The response of
the rotor shows the typical form. In contrast, if the driving
torque does not suffice to accelerate the rotor through the
critical speed, the angular velocity reaches the critical value
and then oscillates below this value as shown in the right dia-
gram of Fig. 11. The system does not pass the critical resonance
vibration and the amplitude rests on the resonance level pre-
scribed by the value of damping.
Vibrations in Variable Speed Machines 583

1: ; 0,02
'""

Fig. 11: Angular velocity and deflection of a Laval-rotor

accelerated through the critical speed and blocked at
the critical speed /7/
4.3 RUN-UP OF ,fl. TURBINE BLADE LOADED BY PARTIAL AD:USSIOt'
An outlook on more complicated ~roblems, in which transient
vibration caused by a variable speed of the machine can occur, is
discussed now. A rotating turbine blade is considered which runs
up from one rotational speed to another while it passes several
resonance points. This oroblem differs from the example regarded
in the previous chapters in two sorts of respect . The first
point is, that the turbine blade must be described by a continu-
ous mechanical model which basically requires a formulation in a
partial differential equation as it was shown for a simple model
by equation (2.3). Thus, more than one eigenfrequency must be
considered. The second fact is that.due to the rotation of the
blade,its spectrum ofeigenfrequencies increases with rotational
speed. In this way, complicating effects occur when the accel-
eration of the blade through critical speeds is considered.
As one example, which is discussed in detail in /11/ a rotating
blade is considered which is loaded by partial admission. This
partial admission is modelled by two 90°-sectors (Fig. 12) while
it is unloaded in the two other sectors. The blade force is dis-
tributed over the blade length as shown in Fig. 12, too. During
a run-up of the blade with such tyoe of partial admission it is
mostly unavoidable to pass through several resonance points as
indicated in the Campbell-diagram in Fig. 13 which is taken from
/11/. The corresponding response of the blade is simulated by a
computer program which is described in /11/, too. A typical
result is shown in Fig. 14. The simulation starts with a rota-
tional frequency of fo = 10 Hz and ends at a rotational frequency
of 60 Hz which is reached within 25 s (Fig. 13). During this run-
up the blade passes several resonance points as indicated in the
584 H. lrrcticr

t:O

blade force

Fig. 12: Rotating blade and load due to two arcs of partial
admission /11/

10.0 15.0 t [s] 25.0

I

: -resonance points
I

600.0
f 1[Hz]

400.0
300.0
200.0 2
100.0
.0
.o 10 25.0 50.0 60 75.0 fofHzl 100.0

Fig. 13: Campbell-diagram of a simulated run-up of a turbine

blade loaded by partial admission /11/
Vibrations in Variable Speed Machines 585

Campbell-diagram in Fig. 13.

The response of the blade is shown in Fig. 14 where the blade
root stress is plotted as function of time durin9 the run-up. He
notice again the more or less time difference due to the non-
stationary conditions between the resonance points indicated in
the Campbell-diagra~ and the maxima of stress in Fig. 14.

30
dma. (I) 7 - i ---- 3
2 -1- 2
20

10

00

- 10

-2 0

-3 0 U1=const) - - 9.41
00 25 5.0 7.5 t[s] 10.0

Fig. 14: Response of a turbine blade loaded by ~artial admission

during a run-uo /11/
~loreover,the plot shows that for the type of load distribution
along the blade only the resonances between the first engine
order and the first eigenfrequency and third engine order and
second eigenfrequency are of significance. This effect occurs
because of decreasing intensity of the exciting s~ectrum for in-
creasing engine order.
5. CONCLUSIONS
The presented examples of dynamical problems occuring in machines
with variable speed show that a lot of additional phenomena
exist in com~arison to stationary conditions. The most important
results are the decrease of the resonant amolitude and the shif-
ting of the resonance frequency. As to more· complicated effects
for instance the influences of non-linearities or parametric
resonances the reader is referred to the literature in particular
to /4/ in the following list of references.
586 H. lrretier

6. REFERENCES
/1/ !MechE: Proc. 3rd. Int. Conf. on Vibrations in Rotating
Machinery, York (England), Sept. 1984
i21 IFToMM: Proc. Int. Conf. on Rotordynamic Problems in Power
Plants, Rom (Italien), Sept./Oct. 1982
/3/ Holzweissig, R.; Dresig, H.: Lehrbuch der 11aschinendynamik.
Springer-Verlag Wien/~ew York, 1979
/4/ Goloskokow, E. G.; Filippow, A. P.: Instationare Schwin-
gungen mechanischer Systeme. Akademie-Verlag, Berlin, 1971
(translated from Russian)
/5/ i~arkert,R.; PfUtzner, H.: An- und Auslaufvorgange einfacher
Schwinger. Forsch. Ing.-Wes. 47,4 (1981),117-125
/5/ Gasch, R.; PfUtzner, H.: Rotordynamik. Springer-Verlag
Berlin/Keidelberg/New York, 1975
171 ~arkert, R.: Biegeschwingungsverhalten unwuchtiger elasti-
scher Rotoren bei der Resonanzdurchfahrt. VDI-Bericht 381
(1980) '155-160
/8/ Traupel, W.: Thermische Turbomaschinen. Springer-Verlag
Berl in/Heidelberg/rlew York, 1968
/9/ Rao, J.S.: Turbine Blade Vibration. Shock and Vibration
Digest 12 (1980), 19-26

/10/ Irretier, H.: Turbine blade dynamics. CISM-Courses and

Lectures - Rotor Dynamics II, Udine (Italien), Oct. 1985
/11/ Irretier, H.: Computer Simulation of the Run-up of a Turbine
Blade Subjected to Partial Admission. ASME Design Eng. Techn.
Conf.; Cincinnati (USA), Sept. 1985; ASt-1E Publication
85-DET-128 (1985), 1-12
/12/ Evans, B.F.; Smalley, A.J.: Simmons, H.R.: Start-up of
Synchronous Motor Driven Trains: The Application of Transient
Torsional Analysis to Cumulative Fatigue Assessment. ASME
Design Eng. Techn. Conf., Cincinnati (USA), Sept. 1985;
ASME Publication 85-DET-122 (1985), 1-9
/13/ Dittrich, G.; Sommer, H.: Das instationare Verhalten des
linearen Schwingers mit einem Freiheitsgrad. VDI-Zeitschrift
118, 8 u. 10 (1976) 375-378 u. 477-482
/14/ f-1arkert, R.: An- und Auslaufvorgange einfacher Schwinger im
Kriechgrenzfall f)= 1. Forsch. Ing.-Ues. 48,1 (1982), 11-14