An Introduction to Random Vibrations, Spectral & Wavelet Analysis: Third Edition

Copyright

Copyright © 1975, 1984, 1993 by D. E. Newland All rights reserved.

Bibliographical Note

This Dover edition, first published in 2005, is an unabridged republication of the 1997 printing of the third (1993) edition of the work originally published by Longman, London and New York, in 1975.

International Standard Book Number: 0-486-44274-8

9780486136967

Manufactured in the United States of America

Dover Publications, Inc., 31 East 2nd Street, Mineola, N.Y. 11501

Table of Contents

Title Page

Copyright Page

Preface to the first edition

Preface to the second edition

Preface to the third edition

Acknowledgements for the first edition

Acknowledgements for the second edition

Acknowledgements for the third edition

List of symbols

Chapter 1 - Introduction to probability distributions and averages

Chapter 2 - Joint probability distributions, ensemble averages

Chapter 3 - Correlation

Chapter 4 - Fourier analysis

Chapter 5 - Spectral density

Chapter 6 - Excitation - response relations for linear systems

Chapter 7 - Transmission of random vibration

Chapter 8 - Statistics of narrow band processes

Chapter 9 - Accuracy of measurements

Chapter 10 - Digital spectral analysis I: Discrete Fourier transforms

Chapter 11 - Digital spectral analysis II: Windows and smoothing

Chapter 12 - The fast Fourier transform

Chapter 13 - Pseudo random processes

Chapter 14 - Application notes

Chapter 15 - Multi -dimensional spectral analysis

Chapter 16 - Response of continuous linear systems to stationary random excitation

Chapter 17 - Discrete wavelet analysis

Appendix 1 - Table of integrals used in the calculation of mean square response (Chapter 7)

Appendix 2 - Computer programs

Appendix 3 - The Gaussian probability integral

Appendix 4 - Distribution of

Appendix 5 - Random numbers

Appendix 6 - Distribution of reverse arrangements for a random series

Appendix 7 - Wavelet programs

Problems

Answers to problems

References

Index

Preface to the first edition

The importance of random vibrations has increased in recent years. Whereas ten years ago undergraduate courses in engineering and applied science barely mentioned the subject, now a well-educated engineer needs at least some familiarity with the concepts and methods involved. Furthermore the applications of random process theory extend far beyond traditional engineering fields. The same methods of analysis which apply to the time-varying pressure in an aircraft’s turbulent boundary layer apply also to the daily fluctuations of an economic index. Because of this growing importance, the theory of random vibrations is now being introduced to undergraduate students, and the methods of measurement and analysis are being increasingly used in research laboratories and in industry.

This book has two main objectives: first, to introduce the fundamental ideas of random vibrations, and, secondly, to deal in some depth with digital spectral analysis, which involves the measurement and analysis of random vibrations. Chapters 1 to 9, which take up about half the book, try to meet the first objective and provide the background for an introduction to the subject at undergraduate level. Chapters 10, 11 and 12 are then more specialized and cover digital spectral analysis and the fast Fourier transform. They deal in detail with the application of discrete Fourier transforms to the analysis of sampled data. The emphasis is on the so-called direct method of spectrum estimation based on the FFT (fast Fourier transform) and the approximations involved in this method are discussed at length. In the author’s experience, many people who find themselves using the FFT for spectral analysis have an incomplete understanding of the nature of the approximations involved, and it is hoped that these chapters will provide the necessary explanation. The last two chapters of the book deal with the properties of pseudo-random processes and a variety of more specialized subjects including the measurement and applications of coherence functions and a brief introduction to the analysis of nonstationary processes.

An important feature of the book is the set of carefully selected problems and answers; most of these have been specially prepared to illustrate points in the text. They serve the dual purpose of allowing readers to test their understanding while at the same time allowing the author to include alternative developments of the theory and different applications which could not otherwise have been discussed because of lack of space.

Although there is a good deal of mathematics, this has been kept as simple as a reasonably complete treatment of the subject allows. The knowledge assumed is limited to that normally possessed by final year undergraduates in the applied sciences, and no additional statistical background is required. This slightly restricts the mathematical scope of the book, but the text is supported by references to more specialized books and to the original literature where appropriate, and mathematically minded readers can pursue these sources if they wish.

The author will be pleased to hear from readers who spot errors and misprints, or who find points where amplification is desirable or see other ways of improving the book. All such suggestions will be gratefully received and most carefully studied.

Sheffield, 1974

D. E. NEWLAND

Preface to the second edition

For the second edition, two additional chapters have been added. Chapter 15 introduces multi-dimensional random processes and discusses multi-dimensional spectral analysis in detail. This is important when a random process depends on more than one independent variable. For example, the height of the sea depends on time and on position. The logic of the multi-dimensional discrete Fourier transform is explained and Appendix 2 now contains a two-dimensional DFT program as well as the original one-dimensional FFT program, which it uses. Chapter 16 extends the previous theory to the random vibration of a continuous linear system subjected to distributed random excitation. This introduces a specialized but growing field of random vibration analysis.

A new section on the Weibull distribution of peaks has been added to Chapter 14 and a number of other small additions made.

The computer programs in Appendix 2 are now listed in both Fortran and Basic languages. Readers with suitable mini-computers will be able to use the latter programs for some of the exercises. Because HP-enhanced Basic has been used, some modifications may be necessary for computers using other versions of Basic, but it is hoped that this will not present serious difficulties.

As before, the author will be glad to hear from readers and will be most grateful for the notification of errors and for suggestions for improvement.

Cambridge, 1984

D. E. NEWLAND

Preface to the third edition

Since completing the second edition in 1984, the most important development in signal analysis has been the wavelet transform. This overcomes a fundamental disadvantage of the Fourier transform. When applied to non-stationary processes, the Fourier transform gives frequency data averaged over the record length being analysed. The local position of particular features is lost. In contrast, wavelet analysis allows a general function of time to be decomposed into a series of orthogonal basis functions, called wavelets, of different lengths and different positions along the time axis. A particular feature can be located from the positions of the wavelets into which it is decomposed. Results of the wavelet transform can be presented in the form of a contour map in the frequency-time plane. This display, which has been compared to the way that notes are shown on a musical score, allows the changing spectral composition of non-stationary signals to be measured and compared.

Therefore it has seemed desirable to add a chapter on the wavelet transform with three objectives: (i) to include enough of the basic theory of wavelet analysis to establish the background for practical calculations; (ii) to explain how the discrete wavelet transform (DWT) works; (iii) to indicate some of the potential applications of the DWT. In addition it has been necessary to prepare computer programs for the DWT and its inverse, and for a two-dimensional DWT and inverse, and also for displaying their results. Considerable thought has been given to the language for these programs and MATLAB®a has been chosen as currently the most appropriate. To use the programs in Appendix 7 it will therefore be necessary to have MATLAB software. Although the program listings will not be usable by readers without MATLAB, the program logic is fully described in Chapter 17, and it is possible for readers who wish to recode the listings given in the language of their choice.

This edition includes all the material in the second edition (subject to minor editing and corrections) together with the following additions: a new section for Chapter 14 on local spectral density calculations; an extensive new Chapter 17, Discrete wavelet analysis; a new Appendix 7 with computer programs for wavelet analysis; and a set of supporting problems for Chapter 17. The objective of the new material is to put readers into the same position vis-à-vis wavelet analysis as readers of earlier editions have been in respect of spectral analysis. The program listings in Appendix 7 allow readers to make their own wavelet calculations, just as readers can use the programs in Appendix 2 to make their own spectral analysis calculations.

To recognize the importance of wavelet analysis in the overall scope of the new book, it has been decided to change the title from Random Vibrations and Spectral Analysis to Random Vibrations, Spectral and Wavelet Analysis. Although wavelets have been studied widely in the mathematical world, their application to vibration analysis is still new. The DWT provides an important procedure for analysing non-stationary vibration records, and it is hoped that this third edition will help vibration practitioners to understand and use the wavelet transform.

As for the earlier editions, the author would be glad to hear from readers with corrections and suggestions for improvement.

Cambridge, 1993

D. E. NEWLAND

Acknowledgements for the first edition

During the early 1960s I had the privilege of working with Professor J. P. Den Hartog and Professor Stephen H. Crandall at the Massachusetts Institute of Technology. My professional interest in vibrations dates from that period and it is a pleasure to acknowledge my debt to both these famous teachers. The textbook Random Vibration in Mechanical Systems by S. H. Crandall and W. D. Mark (Academic Press, 1963) has been a valuable reference during preparation of the first half of the present book.

I have been most fortunate to have had the untiring assistance of my father-in-law, Philip Mayne, M.A., formerly scholar of King’s College, Cambridge. Beginning with the first hand-written draft, he has read every page, checked every formula and worked out every problem. Serious readers will appreciate the enthusiasm needed to undertake such a task and the resolution needed to pursue it to a conclusion. As a result of his help I have made many corrections and alterations to the original manuscript and I believe that many obscurities have been removed. I am more than grateful.

Completion of the manuscript coincided with a visit to Sheffield by Professor Philip G. Hill of Queen’s University, Kingston, Ontario, who has been kind enough to read and comment helpfully on a major part of the book. Again I have been pleased to incorporate many of the changes suggested. I am also indebted to my colleague Dr H. K. Kohler, who provided information and advice on a wide range of topics, and to Professor R. E. D. Bishop, who reviewed the manuscript for Longman and gave encouragement at a critical stage. My publishers have always been enthusiastic and I would like to thank them for the patience and care with which this book has been prepared for publication. Lastly I must mention my secretary, Elaine Ibbotson, who toiled painstakingly with successive drafts, cheerfully accepting my determination to alter every section after, but never before, it had been typed.

Only an author who has written a book in his spare time knows the tremendous extra demands that this inevitably places on those nearest to him. Without the support and understanding of my wife and family this work could not have been brought to completion.

D.E.N.

Acknowledgements for the second edition

During the preparation of the second edition, I have been glad to have had the help of my doctoral research student David Cebon, formerly of Melbourne University. He has read and commented on the two new chapters and tested the Fortran version of the new computer program. A first draft of the new chapters was used as lecture notes at a short course at Monash University in 1983 and this allowed a number of errors and misprints to be identified. Professor J. D. C. Crisp organized this course and I am very glad to have had that opportunity to try out the new material for the second edition. I am grateful also for discussions at Monash with Professor W. H. Melbourne, as a result of which I decided to add a section on the Weibull distribution of peaks to Chapter 14.

I have been encouraged by the comments I have received from readers of the first edition. As a result of readers’ suggestions, I have been able to make a number of corrections and improvements to the text, and I am most grateful to all those who have helped in this way.

D.E.N.

The author and publishers are indebted to the following for permission to reproduce copyright material; the McGraw-Hill Book Company for Fig. 12.7(a) which is based on Fig. 6.23 of Gold and Rader, Digital Processing of Signals, 1969; the Institute of Electrical and Electronics Engineers for the computer program in Appendix 2 which is based on that listed on p. 29 of the IEEE Transactions on Education, Vol. 12, No. 1, March 1969 (paper by Cooley, Lewis and Welch, The Fast Fourier Transform and its Applications), and to the authors and publishers of Fisher and Yates, Statistical Tables for Biological, Agricultural and Medical Research (6th edition), Longman, 1974 (previously published by Oliver and Boyd), from which Tables II(i), IV and XXXIII(i) are reproduced here as Appendices 3, 4 and 5 respectively.

Acknowledgements for the third edition

In preparing the new chapter on wavelets for this edition, I have studied many papers on the mathematics of wavelets. I am indebted to all their authors, but particularly to Professors Ingrid Daubechies and Stéphane Mallat whose pioneering papers have been immensely important in the subject’s development. For me, a most valuable contribution has been the review paper by Professor Gilbert Strang [¹¹⁵]; this made clear many ideas that I had only gleaned roughly from other publications.

Many colleagues have helped in various ways, but I would like to mention particularly Dr Bill Fitzgerald who suggested key publications and commented on a first draft of Chapter 17, and Mrs Margaret Margereson who word-processed the text with care and good humour. My thanks are due to both of them.

There would not have been time to prepare this new edition without the opportunity afforded by sabbatical leave from the furious pace of university teaching, research and administration in the 1990s. I am grateful to Cambridge University for that opportunity.

D.E.N.

List of symbols

The principal symbols in chapters 1 to 14 are defined below with, where appropriate, reference to the chapter or equation in which each symbol first appears.

Chapter 1

Introduction to probability distributions and averages

A system is vibrating if it is shaking or trembling or moving backwards and forwards in some way. If this motion is unpredictable then the system is said to be in random vibration. For instance the motion of a leaf fluttering in the breeze is unpredictable. The leaf is subjected to random excitation as the wind’s direction and strength change and as a result it moves backwards and forwards in random vibration. However the rate and amount of movement of the leaf are dependent not only on the severity of the wind excitation, but also on the mass, stiffness and inherent damping in the leaf system. The subject of random vibrations is concerned with finding out how the statistical (or average) characteristics of the motion of a randomly excited system, like a leaf, depend on the statistics of the excitation, in that case the wind, and the dynamic properties of the vibrating system, in that case the mass, stiffness and damping of the leaf system.

Figure 1.1 shows part of a possible time history for a system in random vibration. The displacement x from an arbitrary datum is plotted as a function of time t. Since motion is random, the precise value of x at any chosen time t = to cannot be precisely predicted. The best we can do is to find the chance, or probability, that x at to will lie within certain limits. The subject of probability is therefore at the heart of random vibration theory and we begin by considering some of the fundamental ideas of probability theory.

Fig. 1.1 Possible time history for a system in random vibration

Probability density function

Suppose, first, that we are dealing with a time history which is non-random or deterministic and is, in fact, a sine wave, Fig. 1.2. In this case we can exactly predict the value of x for any given value of t. We can therefore calculate the proportion of time that the waveform spends between any two levels of x. With reference to Fig. 1.3, during one complete cycle, x(t) lies in the band x to x + dx for two periods of duration dt each. If

(1.1)

then

dx = x0ω cos ωt dt

so that

Fig. 1.2 Waveform for steady state deterministic vibration x(t) = X0 sin ωt

Fig. 1.3 Illustrating calculation of time for which x≤ x(t) ≤ x+dx

(1.2)

Now substituting for

from (1.1) into (1.2) gives

(1.3)

The proportion of time per cycle that x(t) spends in the band x to x + dx is therefore

which, putting

for the period of the sine wave, gives

(1.4)

For any complete number of cycles, equation (1.4) gives the proportion of the total elapsed time for which x(t) lies within the x to x + dx band.

Now consider a situation in which we have to choose an instant of time, t = t0 (say), and find the value of x at this instant. Since x(t) is a deterministic sine wave, as soon as we specify t0, we immediately know x(t0). But suppose that t0 is not precisely specified. Instead we are just told that it may lie anywhere along the time axis. In this case we cannot say what x(t0) will be; the best we can do is to predict what it may be. If t0 is chosen perfectly arbitrarily, t0 may lie anywhere during a complete cycle (the record is assumed to exist for ever with no beginning and no ending so that there is no question of t0 falling in an unfinished cycle). The chance or probability that x(t0) lies in the band x to x + dx will then depend only on how long per cycle x(t) lies between x and x + dx. The probability that x≤ x (t⁰≤x+dx) is therefore given by

Prob (x x(t0) ≤ x + dx) = Fraction of time per cycle for which x(t) lies within the x to x + dx band

(1.5)

since x0 is the amplitude of motion.

The first-order probability density function p(x) is defined so that

(1.6)

therefore in this case, from (1.5),

(1.7)

This function is plotted in Fig. 1.4. Notice that this is a probability density curve so that it is the area under the curve which gives probability. Hence the probability that x(t0) lies in the band x to x + dx is given by the shaded area shown in Fig. 1.4 and the probability that x(t0) has any value between —x0 and + x0 is the total area under the curve which is

(1.8)

Fig. 1.4 Probability density function p(x) for the sine wave of Fig. 1.2

Since, for the sine wave shown in Fig. 1.2, any value of x, chosen at random, must lie in the range —x0 to +x0, the integral on the r.h.s. of (1.8) must be unity, since the probability of a certainty is 1 or 100 per cent. This fact may easily be verified by reference to any standard list of integrals. The