NONLINEARITY IN STRUCTURAL DYNAMICS

NONLINEARITY IN
STRUCTURAL DYNAMICS
Detection, Identification and Modelling

K Worden and G R Tomlinson

University of Sheffield, UK

Institute of Physics Publishing

Bristol and Philadelphia
c IOP Publishing Ltd 2001

All rights reserved. No part of this publication may be reproduced, stored

in a retrieval system or transmitted in any form or by any means, electronic,
mechanical, photocopying, recording or otherwise, without the prior permission
of the publisher. Multiple copying is permitted in accordance with the terms
of licences issued by the Copyright Licensing Agency under the terms of its
agreement with the Committee of Vice-Chancellors and Principals.

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

ISBN 0 7503 0356 5

Library of Congress Cataloging-in-Publication Data are available

Commissioning Editor: James Revill

Production Editor: Simon Laurenson
Production Control: Sarah Plenty
Cover Design: Victoria Le Billon
Marketing Executive: Colin Fenton

Published by Institute of Physics Publishing, wholly owned by The Institute of

Physics, London
Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK
US Office: Institute of Physics Publishing, The Public Ledger Building, Suite
1035, 150 South Independence Mall West, Philadelphia, PA 19106, USA

Typeset in TEX using the IOP Bookmaker Macros

Printed in the UK by MPG Books Ltd, Bodmin
For Heather and Margaret
‘As you set out for Ithaka
hope your road is a long one,
full of adventure, full of discovery.
Laistrygonians, Cyclops,
angry Poseidon—don’t be afraid of them:
You’ll never find things like that in your way
as long as you keep your thoughts raised high,
as long as a rare sensation
touches your body and spirit.
Laistrygonians, Cyclops,
wild Poseidon—you won’t encounter them
unless you bring them along inside your soul,
Unless your soul sets them up in front of you.’
C P Cavafy, ‘Ithaka’
Contents

Preface xv
1 Linear systems 1
1.1 Continuous-time models: time domain 1
1.2 Continuous-time models: frequency domain 10
1.3 Impulse response 13
1.4 Discrete-time models: time domain 17
1.5 Classification of difference equations 21
1.5.1 Auto-regressive (AR) models 21
1.5.2 Moving-average (MA) models 21
1.5.3 Auto-regressive moving-average (ARMA) models 22
1.6 Discrete-time models: frequency domain 22
1.7 Multi-degree-of-freedom (MDOF) systems 23
1.8 Modal analysis 29
1.8.1 Free, undamped motion 29
1.8.2 Free, damped motion 35
1.8.3 Forced, damped motion 37
2 From linear to nonlinear 41
2.1 Introduction 41
2.2 Symptoms of nonlinearity 41
2.2.1 Definition of linearity—the principle of superposition 41
2.2.2 Harmonic distortion 46
2.2.3 Homogeneity and FRF distortion 49
2.2.4 Reciprocity 51
2.3 Common types of nonlinearity 52
2.3.1 Cubic stiffness 52
2.3.2 Bilinear stiffness or damping 55
2.3.3 Piecewise linear stiffness 55
2.3.4 Nonlinear damping 56
2.3.5 Coulomb friction 57
2.4 Nonlinearity in the measurement chain 57
2.4.1 Misalignment 58
viii Contents

2.4.2 Vibration exciter problems 59

2.5 Two classical means of indicating nonlinearity 59
2.5.1 Use of FRF inspections—Nyquist plot distortions 60
2.5.2 Coherence function 62
2.6 Use of different types of excitation 65
2.6.1 Steady-state sine excitation 66
2.6.2 Impact excitation 67
2.6.3 Chirp excitation 68
2.6.4 Random excitation 68
2.6.5 Conclusions 69
2.7 FRF estimators 69
2.8 Equivalent linearization 72
2.8.1 Theory 72
2.8.2 Application to Duffing’s equation 76
2.8.3 Experimental approach 78
3 FRFs of nonlinear systems 81
3.1 Introduction 81
3.2 Harmonic balance 81
3.3 Harmonic generation in nonlinear systems 88
3.4 Sum and difference frequencies 90
3.5 Harmonic balance revisited 91
3.6 Nonlinear damping 93
3.7 Two systems of particular interest 95
3.7.1 Quadratic stiffness 95
3.7.2 Bilinear stiffness 98
3.8 Application of harmonic balance to an aircraft component ground
vibration test 101
3.9 Alternative FRF representations 105
3.9.1 Nyquist plot: linear system 105
3.9.2 Nyquist plot: velocity-squared damping 107
3.9.3 Nyquist plot: Coulomb friction 108
3.9.4 Carpet plots 109
3.10 Inverse FRFs 111
3.11 MDOF systems 112
3.12 Decay envelopes 122
3.12.1 The method of slowly varying amplitude and phase 122
3.12.2 Linear damping 124
3.12.3 Coulomb friction 125
3.13 Summary 125
 Contents ix

4 The Hilbert transform—a practical approach 127

4.1 Introduction 127
4.2 Basis of the method 128
4.2.1 A relationship between real and imaginary parts of the FRF128
4.2.2 A relationship between modulus and phase 132
4.3 Computation 132
4.3.1 The direct method 133
4.3.2 Correction methods for truncated data 135
4.3.3 Fourier method 1 142
4.3.4 Fourier method 2 149
4.3.5 Case study of the application of Fourier method 2 153
4.4 Detection of nonlinearity 156
4.4.1 Hardening cubic stiffness 160
4.4.2 Softening cubic stiffness 161
4.4.3 Quadratic damping 161
4.4.4 Coulomb friction 163
4.5 Choice of excitation 165
4.6 Indicator functions 168
4.6.1 NPR: non-causal power ratio 168
4.6.2 Corehence 170
4.6.3 Spectral moments 170
4.7 Measurement of apparent damping 173
4.8 Identification of nonlinear systems 175
4.8.1 FREEVIB 180
4.8.2 FORCEVIB 189
4.9 Principal component analysis (PCA) 190
5 The Hilbert transform—a complex analytical approach 202
5.1 Introduction 202
5.2 Hilbert transforms from complex analysis 202
5.3 Titchmarsh’s theorem 205
5.4 Correcting for bad asymptotic behaviour 207
5.4.1 Simple examples 209
5.4.2 An example of engineering interest 211
5.5 Fourier transform conventions 215
5.6 Hysteretic damping models 217
5.7 The Hilbert transform of a simple pole 223
5.8 Hilbert transforms without truncation errors 224
5.9 Summary 228
6 System identification—discrete time 230
6.1 Introduction 230
6.2 Linear discrete-time models 232
6.3 Simple least-squares methods 233
6.3.1 Parameter estimation 233
x Contents

6.3.2 Parameter uncertainty 235

6.3.3 Structure detection 237
6.4 The effect of noise 237
6.5 Recursive least squares 242
6.6 Analysis of a time-varying linear system 246
6.7 Practical matters 249
6.7.1 Choice of input signal 249
6.7.2 Choice of output signal 251
6.7.3 Comments on sampling 252
6.7.4 The importance of scaling 253
6.8 NARMAX modelling 255
6.9 Model validity 257
6.9.1 One-step-ahead predictions 258
6.9.2 Model predicted output 258
6.9.3 Correlation tests 259
6.9.4 Chi-squared test 260
6.9.5 General remarks 260
6.10 Correlation-based indicator functions 260
6.11 Analysis of a simulated fluid loading system 261
6.12 Analysis of a real fluid loading system 273
6.13 Identification using neural networks 277
6.13.1 Introduction 277
6.13.2 A linear system 282
6.13.3 A nonlinear system 283
7 System identification—continuous time 285
7.1 Introduction 285
7.2 The Masri–Caughey method for SDOF systems 286
7.2.1 Basic theory 286
7.2.2 Interpolation procedures 290
7.2.3 Some examples 292
7.3 The Masri–Caughey method for MDOF systems 305
7.3.1 Basic theory 305
7.3.2 Some examples 310
7.4 Direct parameter estimation for SDOF systems 315
7.4.1 Basic theory 315
7.4.2 Display without interpolation 319
7.4.3 Simple test geometries 322
7.4.4 Identification of an impacting beam 325
7.4.5 Application to measured shock absorber data 334
7.5 Direct parameter estimation for MDOF systems 341
7.5.1 Basic theory 341
7.5.2 Experiment: linear system 346
7.5.3 Experiment: nonlinear system 350
 Contents xi

7.6 System identification using optimization 355

7.6.1 Application of genetic algorithms to piecewise linear and
hysteretic system identification 356
7.6.2 Identification of a shock absorber model using gradient
descent 367
8 The Volterra series and higher-order frequency response functions 377
8.1 The Volterra series 377
8.2 An illustrative case study: characterization of a shock absorber 380
8.3 Harmonic probing of the Volterra series 386
8.4 Validation and interpretation of the higher-order FRFs 394
8.5 An application to wave forces 404
8.6 FRFs and Hilbert transforms: sine excitation 405
8.6.1 The FRF 405
8.6.2 Hilbert transform 411
8.7 FRFs and Hilbert transforms: random excitation 416
8.7.1 Volterra system response to a white Gaussian input 418
8.7.2 Random excitation of a classical Duffing oscillator 421
8.8 Validity of the Volterra series 431
8.9 Harmonic probing for a MDOF system 434
8.10 Higher-order modal analysis: hypercurve fitting 438
8.10.1 Random excitation 440
8.10.2 Sine excitation 444
8.11 Higher-order FRFs from neural network models 450
8.11.1 The Wray–Green method 452
8.11.2 Harmonic probing of NARX models: the multi-layer
perceptron 455
8.11.3 Radial basis function networks 458
8.11.4 Scaling the HFRFs 460
8.11.5 Illustration of the theory 462
8.12 The multi-input Volterra series 466
8.12.1 HFRFs for a continuous-time MIMO system 467
8.12.2 HFRFs for a discrete-time MIMO system 473
9 Experimental case studies 477
9.1 An encastré beam rig 477
9.1.1 Theoretical analysis 478
9.1.2 Experimental analysis 481
9.2 An automotive shock absorber 493
9.2.1 Experimental set-up 494
9.2.2 Results 501
9.2.3 Polynomial modelling 507
9.2.4 Conclusions 510
9.3 A bilinear beam rig 511
9.3.1 Design of the bilinear beam 512
xii Contents

9.3.2 Frequency-domain characteristics of the bilinear beam 518

9.3.3 Time-domain characteristics of the bilinear beam 523
9.3.4 Internal resonance 526
9.3.5 A neural network NARX model 530
9.4 Conclusions 531
A A rapid introduction to probability theory 533
A.1 Basic definitions 533
A.2 Random variables and distributions 534
A.3 Expected values 537
A.4 The Gaussian distribution 541
B Discontinuities in the Duffing oscillator FRF 543
C Useful theorems for the Hilbert transform 546
C.1 Real part sufficiency 546
C.2 Energy conservation 546
C.3 Commutation with differentiation 547
C.4 Orthogonality 548
C.5 Action as a filter 549
C.6 Low-pass transparency 550
D Frequency domain representations of Æ (t) and (t) 552
E Advanced least-squares techniques 554
E.1 Orthogonal least squares 554
E.2 Singular value decomposition 560
E.3 Comparison of LS methods 562
E.3.1 Normal equations 562
E.3.2 Orthogonal least squares 563
E.3.3 Singular value decomposition 563
E.3.4 Recursive least squares 563
F Neural networks 566
F.1 Biological neural networks 566
F.1.1 The biological neuron 567
F.1.2 Memory 569
F.1.3 Learning 570
F.2 The McCulloch–Pitts neuron 570
F.2.1 Boolean functions 571
F.2.2 The MCP model neuron 573
F.3 Perceptrons 579
F.3.1 The perceptron learning rule 581
F.3.2 Limitations of perceptrons 582
F.4 Multi-layer perceptrons 583
F.5 Problems with MLPs and (partial) solutions 586
F.5.1 Existence of solutions 586
 Contents xiii

F.5.2 Convergence to solutions 586

F.5.3 Uniqueness of solutions 586
F.5.4 Optimal training schedules 587
F.6 Radial basis functions 587
G Gradient descent and back-propagation 590
G.1 Minimization of a function of one variable 590
G.1.1 Oscillation 591
G.1.2 Local minima 592
G.2 Minimizing a function of several variables 592
G.3 Training a neural network 595
H Properties of Chebyshev polynomials 601
H.1 Definitions and orthogonality relations 601
H.2 Recurrence relations and Clenshaw’s algorithm 602
H.3 Chebyshev coefficients for a class of simple functions 604
H.4 Least-squares analysis and Chebyshev series 605
I Integration and differentiation of measured time data 607
I.1 Time-domain integration 608
I.1.1 Low-frequency problems 608
I.1.2 High-frequency problems 614
I.2 Frequency characteristics of integration formulae 616
I.3 Frequency-domain integration 619
I.4 Differentiation of measured time data 622
I.5 Time-domain differentiation 624
I.6 Frequency-domain differentiation 626
J Volterra kernels from perturbation analysis 627
K Further results on random vibration 631
K.1 Random vibration of an asymmetric Duffing oscillator 631
K.2 Random vibrations of a simple MDOF system 633
K.2.1 The MDOF system 633
K.2.2 The pole structure of the composite FRF 634
K.2.3 Validation 636
Bibliography 641
Index 655
Preface

Nonlinearity is a frequent visitor to engineering structures which can modify—

sometimes catastrophically—the design behaviour of the systems. The best laid
plans for a linear system will often go astray due to, amongst other things,
clearances and interfacial movements in the fabricated system. There will be
situations where this introduces a threat to human life; several illustrations
spring to mind. First, an application in civil engineering. Many demountable
structures such as grandstands at concerts and sporting events are prone
to substantial structural nonlinearity as a result of looseness of joints, this
creates both clearances and friction and may invalidate any linear-model-based
simulations of the behaviour created by crowd movement. A second case comes
from aeronautical structural dynamics; there is currently major concern in the
aerospace industry regarding the possibility of limit cycle behaviour in aircraft,
i.e. large amplitude coherent nonlinear motions. The implications for fatigue
life are serious and it may be that the analysis of such motions is as important
as standard flutter clearance calculations. There are numerous examples from
the automotive industry; brake squeal is an irritating but non-life-threatening
example of an undesirable effect of nonlinearity. Many automobiles have
viscoelastic engine mounts which show marked nonlinear behaviour: dependence
on amplitude, frequency and preload. The vast majority of engineers—from all
flavours of the subject—will encounter nonlinearity at some point in their working
lives, and it is therefore desirable that they at least recognize it. It is also desirable
that they should understand the possible consequences and be in a position to take
remedial action. The object of this book is to provide a background in techniques
specific to the field of structural dynamics, although the ramifications of the theory
extend beyond the boundaries of this discipline.
Nonlinearity is also of importance for the diagnosis of faults in structures. In
many cases, the occurrence of a fault in an initially linear structure will result in
nonlinear behaviour. Another signal of the occurrence of damage is the variation
with time of the system characteristics.
The distinction between linear and nonlinear systems is important; nonlinear
systems can exhibit extremely complex behaviour which linear systems cannot.
The most spectacular examples of this occur in the literature relating to chaotic
systems [248]; a system excited with a periodic driving force can exhibit an

xv
xvi Preface

apparently random response. In contrast, a linear system always responds to a

periodic excitation with a periodic signal at the same frequency. At a less exotic
level, but no less important for that, the stability theory of linear systems is well
understood [207]; this is emphatically not the case for nonlinear systems.
The subject of nonlinear dynamics is extremely broad and an extensive
literature exists. This book is inevitably biased towards those areas which the
authors are most familiar with and this of course means those areas which the
authors and colleagues have conducted research in. This review is therefore as
much an expression of personal prejudice and taste as anything else, and the
authors would like to sincerely apologise for any inadvertent omissions. This is
not to say that there are no deliberate omissions; these have good reasons which
are explained here.
 There is no real discussion of nonlinear dynamical systems theory, i.e. phase
space analysis, bifurcations of systems and vector fields, chaos. This is a
subject best described by the more mathematically inclined and the reader
should refer to many excellent texts. Good introductions are provided by
[79] and [12]. The monograph [125] is already a classic and an overview
suited to the Engineer can be found in [248].
 There is no attempt to summarize many of the developments originating
in control theory. The geometrical approach to nonlinearity pioneered by
Brockett has led to very little concrete progress in mainstream structural
dynamics beyond making rigorous some of the techniques adopted lately.
The curious reader is directed to the introduction [259] or to the classic
monograph [136]. Further, there is no discussion of any of the schemes
based on Kalman filtering—again the feeling of the authors is that this is
best left to control engineers.
 There is no discussion of some of the recent approaches based on spectral
methods. Many of these developments can be traced back to the work
of Bendat, who has summarized the background admirably in his own
monograph [25] and the recent update [26]. The ‘reverse-path’ approach
typified by [214] can be traced back through the recent literature survey
[2]. The same authors, Adams and Allemang, have recently proposed an
interesting method based on frequency response function analysis, but it is
perhaps a little early to judge [3].
 There is no discussion of nonlinear normal modes. Most research in
structural dynamics in the past has concentrated on the effect of nonlinearity
on the resonant frequencies of systems. Recently, there has been interest in
estimating the effect on the modeshapes. The authors here feel that this has
been dealt with perfectly adequately in the monograph [257]. There is also a
useful recent review article [258].
So, what is in this book? The following is a brief outline.
Chapter 1 describes the relevant background in linear structural dynamics. This
is needed to understand the rest of the book. As well as describing
 Preface xvii

the fundamental measured quantities like the impulse response function

(IRF) and the frequency response function (FRF) it serves to introduce
notation. The backgrounds for both continuous-time systems (those based on
differential equations of motion) and discrete-time (those based on difference
equations) are given. The chapter begins by concentrating on single-degree-
of-freedom (SDOF) linear systems and finally generalizes to those with
multiple-degrees-of-freedom (MDOF) with a discussion of modal analysis.
Chapter 2 gives essentially the ‘classical’ approaches to nonlinearity which have
longest been within reach of structural dynamicists. This basically means
approaches which can make use of standard dynamic testing equipment
like frequency response analysers. Ideas like FRF distortion and coherence
are discussed here. The chapter also discusses how nonlinearity can enter
the measurement chain and introduces some of the more common types of
nonlinearity. Finally, the idea of linearization is introduced. This chapter
is not just of historical interest as most of the instrumentation commonly
available commercially is still extremely restricted in its ability to deal with
nonlinearity.
Chapter 3. Having discussed FRF distortion, this chapter shows how to compute
FRFs for nonlinear systems. It describes how each type of nonlinearity
produces its own characteristic distortions and how this can lead to
qualitative methods of analysis. The chapter also discusses how nonlinear
systems do not follow certain behaviour patterns typical of linear systems.
It shows how nonlinear systems subject to periodic forcing can respond at
harmonics and combination frequencies of the forcing frequencies. The
chapter concludes with an analysis of IRF distortion.
Chapter 4 introduces more modern methods of analysis, in particular those
which cannot be implemented on conventional instrumentation. The subject
of this chapter is the Hilbert transform. This versatile technique can not only
detect nonlinearity but also, in certain circumstances, estimate the equations
of motion, i.e. solve the system identification problem. All the basic theory is
given, together with detailed discussion of how to implement the technique.
Chapter 5 continues the discussion of the Hilbert transform from a completely
different viewpoint; namely that of complex analysis. Although this chapter
does give some extremely interesting results, it places rather more demands
on the reader from a mathematical point of view and it can be omitted on
first reading. A background in the calculus of residues is needed.
Chapter 6 provides the first discussion of system identification, i.e. the vexed
question of estimating equations of motion for systems based only on
measurements of their inputs and outputs. The particular viewpoint of this
chapter is based on discrete-time equations, more specifically the powerful
and general NARMAX method. This chapter also provides the most
xviii Preface

complete description in this book of the effects of measurement noise and

the need for rigorous model validity testing. Finally, the chapter introduces
the idea of neural networks and shows how they can be used to identify
models of systems.

Chapter 7 balances the discussion of system identification by giving the

continuous-time point of view. The approach is not at all general but follows
a class of models devised by Masri and Caughey and termed here restoring
force surfaces (RFS). The development of MDOF approaches are addressed
and a simpler, more powerful, variant of the idea is discussed. The chapter
concludes with a discussion of how the system identification problem can be
posed in terms of optimization and how this makes available a number of
powerful techniques from mathematics.

Chapter 8 shows one approach to generalizing the idea of the FRF from linear
systems to nonlinear. The method—based on a type of functional power
series—defines an infinite set of impulse response functions or FRFs which
can characterize the behaviour of a class of nonlinear systems. The
interpretation of the higher-order FRFs is discussed and it is also shown how
the approach can give a means of identifying equations of motion of general
MDOF systems—essentially a multi-dimensional version of modal analysis.

Chapter 9 is most concerned with practical matters. The object was to describe
some simple (and one not-so-simple) laboratory rigs which can be used to
illustrate and validate the techniques developed in the earlier chapters.

A substantial set of appendices contain useful material which would

otherwise interrupt the flow of the discussion. Amongst other things these discuss:
basic probability theory, neural networks and the integration and differentiation
of measured time data.
Having discussed the contents, it is important to identify the potential
readership. If the reader has leafed through the remaining pages of this book,
it is possible that the number of equations has appeared daunting. This is actually
rather deceptive. The mathematics required of the reader is little more than a
capability of dealing with matrices, vectors, linear differential equations and
Fourier analysis. Certainly nothing which would not be covered in a degree in
a numerate discipline: mathematics, physics or some flavour of engineering. The
exceptions to this rule come in chapter 5 and in one section of chapter 8. There,
the reader is required to know a little complex analysis, namely how to evaluate
integrals using the calculus of residues. These sections can be omitted on a first
reading—or omitted altogether for that matter—without losing the thread of the
book. This means that the book is accessible to anyone who is in the later stages
of a degree in the disciplines previously identified. It is also suitable for study at
a beginning postgraduate level and also as a survey of the field of nonlinearity for
an expert structural dynamicist.
 Preface xix

A book like this does not spring into being without a lot of help from a
lot of people. It is a pleasure to thank them. First of all, much of this material
is the result of collaboration with various colleagues and friends over the years;
(in roughly chronological order) the authors would like to thank: Matthew Simon,
Neil Kirk, Ian Kennedy, Ijaz Ahmed, Hugh Goyder, Steve Billings, Steve Gifford,
Khalid Mohammad, Mike Reid, Tunde Oyadiji, David Storer, Roy Chng, Jan
Wright, Jonathon Cooper, Wieslaw Staszewski, Qian Chen, Nigel King, Mike
Hamilton, Steve Cafferty, Paul Holmes, Graeme Manson, Julian Chance, Brian
Deacon, Robin Wardle, Sophoclis Patsias and Andreas Kyprianou. In many cases,
the authors have shamelessly lifted figures from the PhD theses and publications
of these collaborators and they would like to offer thanks for that. A special
mention must go to Professor Tuong Vinh who, as a close friend and valued
colleague, provided continuous inspiration and guidance to Geof Tomlinson in
his early career; without his encouragement, the road may have been a linear one.
In terms of producing the manuscript, the authors are grateful to: Steve
Billings, Steve Gifford and particularly Graeme Manson and Heather Worden
for their critical readings of portions of the manuscript. Also Julian Chance
and (predominantly) Jonny Haywood did a valiant job of translating a mass of
disorganized sketches and photocopies into a beautiful sequence of postscript
files. The book would certainly not exist in this form without the efforts of these
people; nonetheless, any mistakes or omissions which exist are entirely the fault
of the authors (who would be grateful if the readers could bring them to their
attention).
Thank you for reading this far, the authors sincerely hope that it will be
useful and illuminating to carry on further.

K Worden
G R Tomlinson
Sheffield 2000