Journalof

Vibration,
Acoustics, Stress
and
Reliability in
Design

Downloaded from http://asmedigitalcollection.asme.org/vibrationacoustics/article-pdf/111/3/354/5436232/354_1.pdf by guest on 26 January 2024

Random Data-Analysis and Measurement Procedures, J. S. density function, («) frequency response function, and (o)
Bendat and A. G. Piersol, John Wiley & Sons, Inc., New coherence function. We conclude by introducing a number of
York, NY 1986, 566 pages. different aspects which are discussed later in the book.
Chapter 2 clarifies some pertinent definitions related to
dynamic behavior of linear systems. The chapter opens with a
description of constant parameter linear systems, their basic
Reviewed by H. Saunders dynamic characteristics and frequency response functions.
Next, we illustrate the frequency response function of com-'
mon physical systems with reference to mechanical and elec-
trical systems. The undamped related frequency of the former
Wow! This book leads the pack! Random data progressed is considered with discussions of gain and phase factors,
from a laboratory tool to a full fledged subject. The magnification function, resonant system and transmissibility
emergence, research, and applications become an indispen- function. The authors furnish a table of gain factors for sim-
sable tool to the data analyst, vibration, structural analyst, ple mechanical systems plus one for analagous terms for elec-
and acoustic engineers in their quest for unravelling some of trical and mechanical systems.
the mysteries concerning our present-day surroundings and Chapter 3 dwells upon probability fundamentals. For one
environment. We meet random (stochastic) data in design of random variable, these are (a) probability distribution func-
tall buildings, wind pressure, earthquake motion, fluid flow, tion, (b) mean (expected value), (c) change of variable (ex-
ocean waves, and turbulent jet streams. Research activities in ample of sine wave distribution), (d) moment generating and
this impose a challenge to the investigators. New materials characteristic function, and (e) Chebychev inequality. Swing-
constantly emerge. The advent of digital data processing ing over to two random variables, we encounter (J) expected
points to a new slant in collecting and analyzing random data. values and definition of correlation function, (g) distribution
Now, the mechanical engineer must possess a deeper and sum of the two random variables, and (h) moment
knowledge of the workings of the equipment rather than just a generating and characteristic function. The Gaussian (normal)
cursory knowledge. The reviewer has seen this progress in distribution is revamped and expanded from the 1st edition.
leaps and bounds from the time of simple vibration gathering This covers (i) central limit theorem, (J) joint Gaussian
information to the more exotic and interesting aspects of distribution, (k) moment generating and characteristic func-
present-day investigation and application. This book updates tion, and (/) N-dimensional Gaussian distribution. The
and revises the previous edition and includes material which chapter includes a neat table on special probability density
reflects recent changes in model formulation, statistical error functions. A good chapter!
evaluations, data collection procedures, and computational Chapter 4 reviews and illustrates a number of statistical con-
algorithms. As stated by the authors, "The current book has cepts that have far-reaching uses in present-day data evalua-
been written primarily to provide convenient references for tion. The mean and variance samples are discussed. This
practicing engineers and scientists." follows with the important probability distribution functions,
This version consists of 13 chapters and 2 important appen- i.e., normal (Gaussian), chi-square distribution, student's t
dices. Appendix A consists of a number of important distribution and F distribution. Sampling distribution con-
statistical tables. Appendix B is an excellent section on defini- siders (a) distribution of sample mean and known variance,
tions and mathematical symbols for random data analysis. and (b) distribution of sample variance. This follows with
Chapter 1 provides basic descriptions of deterministic and confidence intervals, hypothesis tests, chi-square goodness of
random data. The deterministic consists of the following fit tests, run test, and reverse arrangement test. All are ex-
periodic data (a) sinusoidal, (b) complex, and (c) almost plained by means of vivid illustrations. The chapter ends with
periodic plus transient nonperiodic data. The authors classify the subject of linear correlation and linear regression analysis.
random data by (d) stationary, (e) ergodic, and (/) nonsta- All are expressed mathematically and accompanied by vivid
tionary. The latter considers these properties computed over illustrations.
short time intervals. They don't vary significantly from one in- Chapter 5 furnishes a comprehensive discourse on sta-
terval to the following ones. We continue with the static basic tionary random processes. It expands the basic matters and in-
statistical properties. These are important in describing single cludes a new topic (derivative random process). The basic con-
stationary random records. These are (g) mean and square cepts consider auto and cross correlation (covariance func-
values, (h) probability density function, (/) autocorrelation tions), correlation coefficient functions and spectral density
function, and (J) autospectral density functions. If two ran- (auto and cross) functions. This continues with spectra via
dom records of different stationary random exist, the impor- Finite Fourier transforms, coherence functions and uncertain-
tant joint statistical properties are (k) joint probability densi- ty relations. Ergodic processes with arbitrary probability
ty function, (/) cross-correlation function, (m) cross spectral structure and Gaussian (ergodic and nonergodic) are the next

354/Vol.111,JULY1989 Transactions of the ASME

Copyright © 1989 by ASME
 subjects. The new topic, derivative random process, includes authors derive FFT in a lucid mathematical fashion. The
correlation function and unexpected numer of zero crossings. Cooley-Tukey version is a special case of the FFT algorithm
The topics are accompanied by extensive mathematical and one easily adapts it to binary digital computers. The
derivations. Winograd Fourier transform involves a new approach to FFT
Chapter 6, a decided expansion of the original edition, and significantly reduces computer time. The book provides a
points out the relationships of single input/output models. short discourse on probability density and autocorrelation
This covers correlation and spectral relations, ordinary estimates via FFT computations. Presently, FFT computes
coherence functions, models containing extraneous (input and autospectral density functions with a decided time advantage
output) noise and the optimum frequency response functions. over discrete Fourier transforms. The most prominent side
This leads us into various forms of single-input/multiple- lobe suppression can be accomplished by tapering window
output modes. Chapter 7 extends the previous chapter to (Hamming and cosine squared). An additional procedure is
multiple-input/output relations. Multiple-input/single-output the zoom transform which permits one to use fewer data
and two input/one output initiates this chapter. We then en- points in calculating the autospectra. Joint probability density
counter the procedures of optimum response functions, or- functions, cross-correlation and cross-spectral density func-
dinary and multiple coherence functions plus partial tions, frequency response, and correlation functions are other
coherence functions. We further investigate (a) general and quantities often desired. The chapter concludes with iterative

Downloaded from http://asmedigitalcollection.asme.org/vibrationacoustics/article-pdf/111/3/354/5436232/354_1.pdf by guest on 26 January 2024

conditioned multiple-input models, i.e., conditional Fourier algebraic procedures to solve multiple-input/single-output
transforms, conditioned spectra, partial and multiple problems.
coherence functions. We close this chapter by reporting on Chapter 12 leads us into the topic of nonstationary data
matrix representation for multiple-input/multiple-output analysis. The chapter begins with probability structure of
models with no noise or output noise. nonstationary data (higher order and time averaged), nonsta-
The next chapter defines statistical errors in basic estimates. tionary mean value and mean square values. The following
This encompasses estimates of (a) mean value and square topics cover nonstationary data of correlation structure (dou-
values, (b) variance, (c) probability density functions, (d) ble time correlation plus various other types of output rela-
bias effect, and (e) variance and joint probability density. We tions for nonstationary data). This chapter contains a great
then meet correlation function estimates with an associated deal of material rarely found in other books on this subject.
bandwidth limited white noise, noise-to-signal considerations The final chapter discusses Hilbert transform which is a
and estimated location of peak correlation values. The casual function (contains no singularities at the origin). It is
autospectral density function (ADF) estimates are extended similar to Fourier transform. The Hilbert transform is derived
from previous discussions with a clear sighted explanation mathematically with its properties closely stated as to lineari-
employing a time history record. Estimates of variance, bias, ty, inverse transform, orthogonality and modulation proper-
Finite Fourier transforms explain ADF. The chapter closes ties. The book delves into correlation and envelope definition
with a table and discussion of record length and averages for for correlation function and its corresponding analytical
basic estimates of ADF in order to achieve a decided error E in signals. Useful table of Hilbert transforms accompany this
terms of other parameters in the data analysis. section. Dispersive and nondispersive properties complete the
picture. The chapter concludes with envelope detection
Chapter 9 points out the magnitude and phase estimates of followed by correlation employing the Hilbert transform. An
cross-spectral density functions (variance and covariance for- excellent example of exponential cosine cross-correlation
mulas, phase angle). Detailed discourses on single input/out- function illustrates the above comments.
put model estimates, bias in response function estimates, In summary, this is an excellent book and has little equal.
coherence output spectrum estimates, gain factor estimates The reviewer would have preferred seeing sections on Ibrahim
plus phase factor estimates comprise this section. This con- Time Domain (ITD), cepstrum analysis, Z transforms and a
tinues with multiple-input/output model estimates, i.e., multi- larger discussion of ARMA plus maximum entropy spectral
ple coherence function estimates, partial coherence function analysis and maximum likelihood spectral analysis. The
estimates and output spectrum estimates, gain factor plus reviewer recommends this book to engineers interested in ran-
phase factor for conditional models. dom vibration analysis and its associated data processing.
Chapter 10, the lengthiest, unveils the topic of data acquisi- Bouquets to the authors!
tion and processing. This operation may be divided into five
primary categories. They are (a) data collection, (b) data
recording, (c) data preparation and its errors (quantization,
aliasing), (d) ADC errors (includes aperture error, jitter, Proceedings of Noise-Con 87—High Technology for Noise
nonlinearities), (e) data quantification tests (stationarity, Control, Noise Control Foundation, P.O. Box 2469, Ar-
periodicities, normality), and (/) data analysis. The latter lington Branch, Poughkeepsie, NY, J. Tichy and S. Hayek,
covers various procedures for analysis of autocorrelation, Eds., 1987, 774 pages.
mean and mean squares, probability density, nonstationarity
and transient data plus periodic and nonperiodic data. Pro-
cedures for analyzing a set of records encompass (g) analysis
of single set, (h) correlation, (/) equivalence of uncorrelated The advent of each conference makes the reading more in-
data, 0) pooling of equivalent uncorrelated data, (k) cross- formative and apropos to the present status of noise control.
correlation, (/) cross-spectral, (m) coherence function, and Little by little we are striving to reach the goal of succeeding in
(n) test for equivalence of autospectra. furthering ourselves in noise control. The present administra-
Chapter 11 focuses upon digital data analysis. Since the tion hampers our activities but we are making great strides in
data consists of an analog recording, it must be converted to spite of this. As in each preceding conference, the papers
digital time series by means of ADC process. The data needs become "meatier," pro fused with information concerning
to be standardized for fixed point and any trend must be various phases of noise control. The joint sponsorship of the
removed. Prior to more detailed analysis, the data must be Research Center for Acoustics and Vibration Engineering of
filtered in order to remove periodic components. It can be per- Pennsylvania State University and the Institute of Noise Con-
formed in either frequency or time domains. Fast Fourier trol Engineering (INCE) is a perfect match.
transforms (FFT), due to their speed, can be employed to fur- The conference consists of the following topics and its
nish estimates of desired spectral density and correlation. The associated papers for a grand total of 127 papers.

Journal of Vibration, Acoustics, Stress, and Reliability in Design JULY 1989, Vol. 111/355