Professional Documents
Culture Documents
Wave Propagation in Structures Mechanical Engineering Series James F Doyle Online Ebook Texxtbook Full Chapter PDF
Wave Propagation in Structures Mechanical Engineering Series James F Doyle Online Ebook Texxtbook Full Chapter PDF
Wave Propagation in Structures Mechanical Engineering Series James F Doyle Online Ebook Texxtbook Full Chapter PDF
https://ebookmeta.com/product/radio-wave-propagation-
fundamentals-second-edition-artem-saakian/
https://ebookmeta.com/product/advanced-electromagnetic-wave-
propagation-methods-1st-edition-gonzalez/
https://ebookmeta.com/product/wave-fields-in-real-media-wave-
propagation-in-anisotropic-anelastic-porous-and-electromagnetic-
media-4th-edition-jose-m-carcione/
https://ebookmeta.com/product/nonlinear-wave-and-plasma-
structures-in-the-auroral-and-subauroral-geospace-1st-edition-
evgeny-mishin/
Electromagnetic Vortices: Wave Phenomena and
Engineering Applications 1st Edition
https://ebookmeta.com/product/electromagnetic-vortices-wave-
phenomena-and-engineering-applications-1st-edition/
https://ebookmeta.com/product/aircraft-structures-for-
engineering-students-aerospace-engineering-7th-edition-megson/
https://ebookmeta.com/product/fear-no-evil-alex-cross-29-1st-
edition-james-patterson/
https://ebookmeta.com/product/mastering-uncertainty-in-
mechanical-engineering-1st-edition-peter-f-pelz/
Mechanical Engineering Series
James F. Doyle
Wave Propagation
in Structures
Third Edition
Mechanical Engineering Series
Series Editor
Francis A. Kulacki
Department of Mechanical Engineering
University of Minnesota
Minneapolis, MN, USA
The Mechanical Engineering Series presents advanced level treatment of topics
on the cutting edge of mechanical engineering. Designed for use by students,
researchers and practicing engineers, the series presents modern developments
in mechanical engineering and its innovative applications in applied mechanics,
bioengineering, dynamic systems and control, energy, energy conversion and energy
systems, fluid mechanics and fluid machinery, heat and mass transfer, manufacturing
science and technology, mechanical design, mechanics of materials, micro- and
nano-science technology, thermal physics, tribology, and vibration and acoustics.
The series features graduate-level texts, professional books, and research mono-
graphs in key engineering science concentrations.
Wave Propagation
in Structures
Third Edition
James F. Doyle
School of Aeronautics & Astronautics
Purdue University
West Lafayette, IN, USA
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2021
2nd edition: © Springer 1997
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
This book is dedicated to
my father and mother,
Patrick and Teresa Doyle.
Thanks for the dreams.
The study of wave propagation seems very remote to many engineers, even to those
who are involved in structural dynamics. One of the reasons for this is that the
examples usually taught in school are either so simple as to be not applicable
to real world problems or so mathematically abstruse as to be intractable. This
book contains an approach, spectral analysis or frequency domain synthesis, that
I have found to be very effective in analyzing waves. What has struck me most
about this approach is how I can use the same analytical framework to examine
the experimental results as well as to manipulate the experimental data itself.
As an experimentalist, I had found it very frustrating having my analytical tools
incompatible with my experiments. For example, it is experimentally impossible to
generate a step function wave and yet that is the type of analytical solution often
available.
Spectral analysis is very encompassing—it touches on analysis, numerical
methods, and experimental methods. I want this book to do justice to its versatility,
so many subjects are introduced. As a result, some areas may seem a little thin, but I
do hope, nonetheless, that the bigger picture, the unity, comes across. Furthermore,
spectral analysis is not so much a solution technique as it is a different insight into
the wave mechanics; consequently, in most of the examples an attempt is made to
make the connection between the frequency domain and time domains.
In writing the second edition, I strived to keep what was good about the
first edition—that is, the combination of experimental and analytical results—but
incorporated more recent developments and extensions at that time. The question
not fully articulated in the first edition is: What should be different about a book
on waves in structures ? This is the question that guided my reorganization of the
material as well as the selection of new topics. It had become clearer to me that
the essence of a structure is the coupling of systems and, this should be the central
theme of a book on waves in structures. There are two readily recognized forms
of coupling: mechanical coupling “at the ends” such as when two bars are joined
at an angle, and differential coupling as when two bars are connected uniformly
along their lengths by springs. Both couplings are intimately related to each other as
seen from the example of a curved beam: it can be modeled as a collection of small
vii
viii Preface
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1 Spectral Analysis of Wave Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1 Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Continuous Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2 Discrete Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.3 Fast Fourier Transform Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.1.4 Space Distributions Using Fourier Series . . . . . . . . . . . . . . . . . . . . 20
1.2 Applications Using the FFT Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.2.1 Explorations of the Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.2 Experimental Aspects of Wave Signals. . . . . . . . . . . . . . . . . . . . . . . 26
1.3 Spectral Analysis of Wave Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.3.1 General Functions of Space-Time and Spectrum Relations . 31
1.3.2 Some Wave Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3.3 Group Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.3.4 Summary of Wave Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.4 Propagating and Reconstructing Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.4.1 Basic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.4.2 Integration of Signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2 Longitudinal Waves in Rods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1 Elementary Rod Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1.1 Equation of Motion and Spectral Analysis . . . . . . . . . . . . . . . . . . . 50
2.1.2 Basic Solution for Waves in Rods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 Dissipation in Rods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.1 Distributed Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.2 Viscoelastic Rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.3 Coupled Thermoelastic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
ix
x Contents
Afterword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
A.1 Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
A.1.1 Bessel Equations and Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
A.1.2 Limiting Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Notation
Roman letters
a Radius
A Cross-sectional area
Â, B̂, Ĉ, D̂ Frequency dependent coefficients
b Thickness, depth
B Bulk modulus
cg Group speed
√
co Longitudinal wave speed, EA/ρA
cP , cS , cR Primary, Secondary, and Rayleigh wave speeds
CE Specific heat
D, D̄ Plate stiffness, D = Eh3 /12(1 − ν 2 ), D̄ = GH 3 /12
E, E ∗ , Ê Young’s modulus, E ∗ = E/(1 − ν 2 ), viscoelastic modulus
EI Beam flexural stiffness
F Member axial force
ĝi (x) Element shape functions
G(t), Ĝ Structural unit response, frequency response function
h Beam or rod height, plate thickness
Hn Hankel function, Hn = Jn ± iYn
√
i Complex −1
I Second moment of area, I = bh3 /12 for rectangle
In Modified Bessel functions of the first kind
Jn Bessel functions of the first kind
k, k1 , k2 Waveguide spectrum relations
kx , ky , kz 2-D wavenumbers
K Stiffness, thermal conduction
[ k ], [ K ] Stiffness matrix
Kn Modified Bessel functions of the second kind
xv
xvi Notation
Special Symbols
Rn Random noise
∂2 ∂2
∇2 Differential operator, ∂x 2
+ ∂y 2
V Volume
[ ] Square matrix
{ } Vector
Subscripts
a Acoustic medium
m Space wavenumber counter
n Frequency counter
1, 2 Sensors, modes
P , S, R Primary, Secondary, and Rayleigh waves
, (Comma) derivative with respect to indicated variable
Superscripts
∗ Complex conjugate
¯ Bar, local coordinates
˙ Dot, time derivative
ˆ Frequency domain (transformed) quantity
˜ Wavenumber domain (transformed) quantity
Prime, derivative with respect to argument
Abbreviations
Reference
1. Yeats, W.B.: Collected Poems of W.B. Yeats. Macmillan and Co., London (1960)
Introduction
We must gather and group appearances, until the scientific imagination discerns their hidden
laws, and unity arises from variety; and then from unity we must re-deduce variety, and force
the discovered law to utter its revelations of the future.
W.R. Hamilton [4]
© The Editor(s) (if applicable) and The Author(s), under exclusive license to 1
Springer Nature Switzerland AG 2021
J. F. Doyle, Wave Propagation in Structures, Mechanical Engineering Series,
https://doi.org/10.1007/978-3-030-59679-8_1
2 Introduction
Over the years many analytical techniques have been developed for treating wave
propagation problems. Central among these is the method of Fourier synthesis (or
spectral analysis), where the behavior of a signal is viewed as a superposition of
many infinitely long wave trains of different periods (or frequencies). The actual
response is synthesized by a judicious combination of these wave trains. Thus the
problem of characterizing a signal is transformed into one of determining the set of
combination coefficients. These coefficients are called the Fourier transform of the
signal. While the problem being tackled invariably simplifies when it is expressed
in terms of the Fourier transform (Sneddon’s book [6], for example, shows a wide
range of applications to both static and dynamic problems) the last step in the
analysis involves performing an inverse transform (reconstructing the signal) and
this, generally, is very difficult to do in an exact analytical manner. Consequently,
many approximate and asymptotic schemes have usually been resorted to. These are
quite adequate for determining the remote behavior (as is required in seismology,
say) but can lose too much information when applied to structural impact problems.
It should also be pointed out that analytical transforms are feasible only if the
function to be transformed is mathematically simple—unfortunately this is not the
case in any situation of practical interest and is certainly not true when dealing
with experimental data. This inversion problem is the biggest impediment to a more
widespread use of the transform methods.
Spectral analysis forms the basis of this book, but the approach is different
from the classical method in that, from the outset, the transforms are approximated
by the discrete Fourier transform (DFT). In contrast to the continuous transform,
this represents the signal by a finite number of wave trains and has the enormous
advantage in that the fast Fourier transform (FFT) computer algorithm can be
used for economically computing the transforms. Being able to do transforms
and inversions quickly adds great heuristic value to the tool in which the waves
can be actually “seen” and iterated on, and realistic signals (even ones that are
experimentally based) can be treated. It should be pointed out that while the method
uses a computer, it is not a numerical method in the usual sense, because the
analytical description of the waves is still retained. As a consequence, the very
important class of problems called inverse problems can be tackled.
The approach presented in this book takes advantage of many of the techniques
already developed for use in time series analysis and for the efficient numerical
implementation of them. (Chatfield’s book [1], for example, is a readable introduc-
tion to the area of signal processing.) In fact, this aspect of spectral analysis is really
part of the more general area of digital processing of the signals and herein lies
one of the unifying advantages of the present approach—the programming structure
is then already in place for the subsequent post-processing of the data. This is
especially significant for the manipulation of experimental data.
Introduction 3
A major aspect of this book is the persistent treatment of the effect of boundaries and
discontinuities on the waves because real structures have many such terminations.
This can be done efficiently because the quantities used in the analysis of the
waveguide are also used to set up the connectivity conditions. As a result, wave
solutions for structures more interesting than simple rods and beams can be pieced
together successfully. Moreover, the way to solve problems of structures with many
members and boundaries is then available.
A connection not often investigated when studying structural dynamics is the
relationship between wave propagation and vibrations. For many engineers, these
are two separate areas with quite distinct methods of analysis. However, another
advantage of the spectral approach to dynamics is that the close connection between
waves and vibrations becomes apparent. Even the same language can be used,
terms such as power spectral density, filtering, spectral estimation, convolution,
and sampled waveforms have the same meaning. Consequently, many of the
technologies developed over the last four decades for vibrations and modal analyses
are directly applicable to the spectral analysis of waves.
An exciting possibility (and one of the motivations for writing this book) is
to facilitate the reverse process, that is, to transfer many of the wave ideas into
vibration analysis. This should lead to a richer understanding of such topics as
impulse testing, transient vibrations, and filtering in periodic structures. A number of
examples throughout the book demonstrate the “evolution” of resonance as multiple
reflections are included in the analysis.
This book concentrates on wave propagation in the basic structural elements of rods,
beams, and plates. These form a rich collection of problems and the intent is to
show that they all can be analyzed within the same framework once the spectral
analysis approach is adopted. Because of the structural applications (and because
all structures are finite in extent), a primary theme is the interaction of the wave
with discontinuities such as boundaries, junctions, and attachments. Supplemental
themes involve the construction of the mechanical models, and the duality between
the time and frequency domains.
While the material of each chapter is reasonably self-contained, Graff’s book [3]
can be used as an excellent supplemental reference on elastic wave propagation.
The book by Elmore and Heald [2] gives a broader and simpler introduction to
waves. Chapter 1 recapitulates the essence of the continuous Fourier transform and
its approximation in the form of the discrete Fourier series. The factors affecting the
quality of the approximation (or spectral estimate) are elucidated. It also discusses,
in a general way, how spectral analysis can be used to solve differential equations
Introduction 5
and especially those associated with wave motion. Two concepts of significance
emerge from this analysis. One is the idea of the spectrum relation (that unique
relation between the frequency and the wavenumber) and is essentially the transform
equivalent of the space-time differential equation. The other is that of multi-mode
solutions. These are shown to play a fundamental role in the solution of actual
boundary value problems even though all are not necessarily propagating modes.
The following two chapters deal, respectively, with rods and beams, in nearly
the same format. First, the governing differential equations are derived and then,
by spectral analysis, the kernel solutions and spectrum relations are obtained.
Having initiated a wave, how it interacts with structural discontinuities is then
investigated. Each chapter includes an example of coupling both at the differential
and mechanical levels. Chapter 4 addresses the question of the adequacy of the
waveguide models and a foundation for the construction of higher-order waveguides
is provided that readily allows extensions to be made to other specialized problems.
Chapter 5 introduces the spectral element method as a matrix method approach
to structural dynamics that combines aspects of the finite element method with
the spectral analysis method. In a way, this chapter is the culmination of the
connected waveguide approach and clearly makes the study of wave propagation
in complicated frame structures practical.
Chapter 6 expands the analysis to multiple dimensions to cover flexural wave
propagation in plates and cylinders. Chapter 7 begins the development of a matrix
methodology for plated structures including cylindrical shells.
The final two chapters extend the spectral analysis to some specialist areas.
Chapter 8 deals with the problem of plate/fluid interactions; Chap. 9 is new to this
edition and focusses on discrete systems (atoms, for example), discretized systems
(FE models, for example), and periodic structures (grilles and frame, for example).
The examples are drawn from metamaterials, nanotechnology and focus on filtering
actions.
A number of threads run through all the chapters. The spectral methodology is an
ideal companion for experimental analysis, and throughout the chapters summaries
of some of its experimental applications are given. The emphasis is on how spectral
analysis can extend the type of information extracted from the experimental data—
for example, how a structural response can be used to infer the force history causing
it. Another thread is the integration of finite element (FE) methods into the spectral
analyses. They are used to enhance the examples through what are called computer
experiments and vice versa how spectral analysis can be used to enhance and explain
FE results. An important thread is the role played by elastic constraints. This is
first introduced in connection with the simple rod encased in an elastic medium
but reappears with increasing complexity in the subsequent chapters. Its ultimate
expression is found in the coupled modes of the curved plate.
Admittedly, a large range of problems have been left out even though most
of them are treatable by the spectral methods. Consequently, an effort is made
to supplement each chapter with a collection of pertinent problems plus specific
references that indicate extensions of the modeling and the applications.
6 Introduction
A Question of Units
The choice of any particular set of units is bound to find disfavor with some
readers. Nondimensionalizing all of the plots is an unsatisfactory solution especially
because experimental results are included. Furthermore, working with dimensional
quantities help give a better “feel” for the results. The resolution of this dilemma
adopted here is to do most of the examples using a nominal material with nominal
properties. In the two major systems of units, these are given in the following table:
The choice of shear modulus results in a Poisson’s ratio of ν = 0.25. One of the
recurring material parameters is the ratio
E
co = = 200 × 103 in./s = 5000 m/s
ρ
References
1. Chatfield, C.: The Analysis of Time Series: An Introduction. Chapman and Hall, London (1984)
2. Elmore, W.C., Heald, M.A.: Physics of Waves. Dover, New York (1985)
3. Graff, K.F.: Wave Motion in Elastic Solids. Ohio State University Press, Columbus (1975)
4. Hamilton, W.R.: The Mathematical Papers of Sir W.R. Hamilton. Cambridge University Press,
Cambridge (1940)
5. Redwood, M.: Mechanical Waveguides. Pergamon Press, New York (1960)
6. Sneddon, I.N.: Fourier Transforms. McGraw-Hill, New York (1951)
Chapter 1
Spectral Analysis of Wave Motion
It has long been known that an arbitrary time signal can be thought of as the
superposition of many sinusoidal components, that is, it has a distribution or
spectrum of components. Working in terms of the spectrum is called spectral
analysis. In wave analysis, the time domain for a motion or response is from minus
infinity to plus infinity. Functions in this domain are represented by a continuous
distribution of components which is known as its continuous Fourier transform
(CFT). However, the numerical evaluation and manipulation of the components
require discretizing the distribution in some manner—the one chosen here is by
way of the discrete Fourier transform (DFT). This has the significant advantage
that it allows the use of the very efficient fast Fourier transform (FFT) computer
algorithm.
The goal of this chapter is to introduce the discrete Fourier transform for the
efficient computation of the spectral content of a signal. Possible sources of errors
in using it on finite samples and ways of reducing their influences are discussed. The
Fourier transform is then applied to wave analysis. The crucial step is to set up the
connection between the spectral responses at different space locations—we do this
through the governing differential equations. In doing so, certain key ideas emerge
which recur throughout this book; central among these ideas are that of a wave
mode, its spectrum relation, and its phase and group speeds. In Fig. 1.1, movement
of the amplitude corresponds to the group speed, movement of the zero crossings
correspond to the phase speed.
© The Editor(s) (if applicable) and The Author(s), under exclusive license to 7
Springer Nature Switzerland AG 2021
J. F. Doyle, Wave Propagation in Structures, Mechanical Engineering Series,
https://doi.org/10.1007/978-3-030-59679-8_2
8 1 Spectral Analysis of Wave Motion
Fig. 1.1 Response showing the difference between the group and phase speeds. Movement of
the envelope corresponds to group behavior while movement of the zero crossings corresponds to
phase behavior
properties are given here; a more complete account can be found in Refs. [20, 21].
Background information on time series analysis can be found in Ref. [5].
The continuous Fourier transform is a powerful technique but has the drawback
that the functions (signals) must be known analytically over the complete domain.
This occurs in only rare cases making it unsuitable for practical situations, especially
if the signals are experimental in origin. Discrete Fourier transforms are more
suitable for our purposes.
The continuous Fourier transform pair of a function F (t), defined on the time
domain from −∞ to +∞, is given as
∞ ∞
2π F (t) = Ĉ(ω)e+iωt dω , Ĉ(ω) = F (t)e−iωt dt (1.1)
−∞ −∞
The spectrum Ĉ(ω) usually has nonzero real and imaginary parts. We use the
overhead “hat” to indicate the frequency domain spectrum of a function.
By way of a simple example of the application of Fourier transforms, consider a
rectangular pulse where the time function is given by
and zero otherwise. Substituting into the forward transform and integrating give
+a/2
sin (ωa/2)
Ĉ(ω) = Fo e−iωt dt = Fo a ≡ Ĉo (ω)
−a/2 ωa/2
F (t) = Fo t o ≤ t ≤ to + a
real
imag 0 to to+a
to=-a/2
to=-a/5
to=+a/10
sin(ωa/2) −iω(to +a/2)
Ĉ(ω) = Fo a e = Ĉo (ω)e−iω(to +a/2) (1.2)
ωa/2
which has both real and imaginary parts and is not symmetric with respect to ω = 0.
On closer inspection, however, we see that the magnitudes of the two transforms
are the same; it is just that the latter is given an extra phase change of amount
ω(to + a/2). Figure 1.2 shows the transform for different amounts of shift. We
therefore associate phase changes with shifts of the signal along the time axis.
If a pulse is visualized to be at different positions relative to the time origin,
then the amplitude of the spectra will be the same, but each will have a different
phase change. That is, movement in the time domain causes phase changes in the
frequency domain. Investigating these phase changes is the fundamental application
of spectral analysis to wave propagation and is pursued later in this chapter.
For completeness, we now summarize some of the major properties of Fourier
transforms; more detailed accounts can be found in Ref. [21]. In all cases, the
results can be confirmed by taking the example of the rectangular pulse and working
through the transforms long hand. To aid in the summary, we refer to the transform
pair
F (t) ⇔ Ĉ(ω)
where the symbol ⇔ means “can be transformed into.” The double arrowheads
reinforce the idea that the transform can go in either direction and that its properties
are symmetric.
If the functions FA (t) and FB (t) have the transforms ĈA (ω) and ĈB (ω),
respectively, then the combined function [FA (t) + FB (t)] has the transform
[ĈA (ω) + ĈB (ω)]. That is,
1
F (at) ⇔ Ĉ(ω/a) (1.4)
|a|
indicating a reciprocal scaling relationship for the arguments. That is, time domain
compression corresponds to frequency domain expansion (and vice versa). For
example, if the width of a pulse is made narrower in the time domain, then its extent
in the frequency domain is made broader. It should also be noted that the amplitude
decreases because the energy is distributed over a greater range of frequencies.
If the function F (t) is time shifted by to , then it has the transform pair
1.1 Fourier Transforms 11
F (t − to ) ⇔ Ĉ(ω)e−iωto (1.5)
This property was already seen in the last example where the rectangle was
displaced from the origin. Of course, the property refers to any change at any
position. There is a corresponding relation for frequency shifting.
The transform pairs, Eqs. (1.1), are valid for both F (t) and Ĉ(ω) being complex,
but the functions of usual interest in wave analysis are when F (t) is real. To see the
effect of this, rewrite Eq. (1.1) in terms of its real and imaginary parts as
2π FR = [ĈR cos ωt − ĈI sin ωt] dω , ĈR = [FR cos ωt + FI sin ωt]dt
2π FI = [ĈR sin ωt + ĈI cos ωt]dω , ĈI = − [FR sin ωt − FI cos ωt]dt
From the above it is apparent that when F (t) is real-only, ĈR is even and ĈI is odd.
Mathematically, this is expressed as
ĈR (−ω) = ĈR (ω) , ĈI (−ω) = −ĈI (ω) or Ĉ(−ω) = Ĉ ∗ (ω) (1.7)
which says that the functions are symmetrical and antisymmetrical, respectively,
about the zero frequency point. This can also be expressed as saying that the negative
frequency side of the transform is the complex conjugate of the positive side.
A very interesting property arises in connection with the products of functions.
Consider the transform of two time functions
Ĉ(ω) = FA (t)FB (t)e−iωt dt
This shows that a time domain convolution can be performed as frequency domain
multiplication. Even though this involves both a forward and inverse transform it is
computationally faster than performing the convolution directly. All the mechanical
systems considered in the later chapters can be represented in the frequency
domain as products of the input times the system response. In fact, the reason
why the frequency domain is so useful for analyzing these systems is because the
complicated convolution relations become simple algebraic relations.
We discretize the CFT in two steps: first the frequency integrals are discretized
giving the Fourier series transform (FST), and then the time integrals are discretized
giving the Discrete Fourier transform (DFT).
Let the time function F (t) be known only over a period T ; to apply the
continuous Fourier transform to it, we must extend it somehow to the infinity limits.
In the Fourier series representation, the function is assumed extended to plus and
minus infinity as a periodic function with the period T . We can view this either as
separate functions of duration T placed one after the other or as the superposition
of separate functions of infinite duration but with nonzero behavior only over the
period T . We adopt this latter view as shown in Fig. 1.3.
We saw from the rectangular pulse example that if Ĉ∞ (ω) is the transform of a
pulse, the transform of the same pulse shifted an amount T is Ĉ∞ (ω)e−iωT . It is
obvious therefore that the transform of the periodic signal can be represented as
This transform shows an infinite peak whenever the frequency is one of the discrete
values ωn = 2π n/T . Under this circumstance, each of the exponential terms is
unity and there is an infinity of them. (This result should not be surprising since the
Ĉ(0) component is the area under the curve, and for the periodic rectangle of Fig. 1.3
we see that it is infinite.) At other frequencies, the exponentials are as likely to be
positive as negative and hence their sum will be relatively small. Therefore our first
conclusion about the transform of a periodic extended signal is that it shows very
sharp spectral peaks. We can go further and say that the transform is zero everywhere
except at the discrete frequency values ωn = 2π n/T where it has an infinite value.
We represent this behavior by use of the delta function, δ(x); this special function
is zero everywhere except at x = 0 where it is infinite. It has the very important
additional property that its integral over the whole domain is unity. Thus
Interchanging the summation and the integration, and using the properties of the
delta function, gives
2π F (t) = A Ĉ∞ (ωn )e+iωn t
n
Integrate both sides of this equation over a time period T , and realizing that all terms
on the right-hand side are zero except the first, gives that
T
2π F (t) dt = AĈ∞ (0)T
0
Because Ĉ∞ (0) is the area under the single pulse, we conclude that A = 2π/T .
14 1 Spectral Analysis of Wave Motion
Except for a normalizing constant, this is the complex Fourier series representation
of a periodic signal.
Consider the transform of the rectangular pulse of the last section. The coeffi-
cients are given by
to +a to +a
e−iωn t sin(ωn a/2) −i(to +a/2)ωn
Ĉn = Fo e−iωn t dt = Fo = Fo a e
to −iωn to ωn a/2
This result is the product of three terms. The first, Fo a, is the size of the pulse
as represented by the area. The third, the exponential, is a phase change due to the
shifting of the pulse relative to the time origin. For example, if the pulse is symmetric
about t = 0, then to = −a/2 and there is no phase change of the transform. The
second term, sinc function, is the core of the transform and its amplitude is shown
plotted in Fig. 1.4 for various periods. Notice that the spacing of the coefficients is
at every 1/T hertz in the frequency domain. For the periods T = 500, 200, 100 µs
this gives spacings of f = 2, 5, 10 kHz, respectively.
Notice that in all the cases of Fig. 1.4, the Fourier series gives the exact
values of the continuous transform, but it does so only at discrete frequencies. The
discretization of the frequencies is given by
ωn = 2π n/T
Fourier series
Continuous
Fig. 1.4 Fourier series coefficients for a rectangular pulse of nonzero duration a = 50 µs extended
with different periods T
1.1 Fourier Transforms 15
Therefore, for a given pulse, a larger period gives a more dense distribution of
coefficients, approaching a continuous distribution in the limit of an infinite period.
This, of course, is the continuous transform limit. The finite time integral causes the
transform to be discretized in the frequency domain.
The discrete coefficients in the Fourier series are obtained by performing
continuous integrations over the time period. These integrations are now replaced
by summations as a further step in the numerical implementation of the continuous
transform.
In reference to Fig. 1.5, let the function F (t) be divided into M, piece-wise
constant, segments whose heights are Fm and base T = T /M. The coefficients
are now obtained from
M−1 tm + T /2
Ĉn ≈ D̂n = Fm e−iωn t dt
m=0 tm − T /2
sin ωn T /2
= T Fm e−iωn tm
ωn T /2 m
We see that this is the sum of the transforms of a series of rectangles each shifted in
time by t = tm + T /2. The contribution of each of these is now examined more
closely.
First look at the summation term. If n > M, that is, if n = M + n∗ , then the
exponential term becomes
∗ω t ∗ω t ∗ω t
e−iωn tm = e−inωo tm = e−iMωo tm e−in o m = e−i2π m e−in o m = e−in o m
M−1
∗ω t
Fm e−in o m
m=0
16 1 Spectral Analysis of Wave Motion
sin(x) ωn T n T n
sinc(x) ≡ , x= =π =π
x 2 T M M
The sinc function is such that it decreases rapidly with increasing argument and is
small beyond its first zero. The first zero occurs where x = π or n = M; if M
is made very large, that is, the integration segments are made very small, then it is
the higher order coefficients (i.e., large n) that are in the vicinity of the first zero.
Let it be further assumed that the magnitude of these higher order coefficients are
negligibly small. Then an approximation for the coefficients is
M
D̂n ≈ T {1} Fm e−iωn tm
m
on the assumption that it is good for n < M and that Ĉn ≈ 0 for n ≥ M.
Because there is no point in evaluating the coefficients for n > M − 1, the
approximation for the Fourier series coefficients is now taken as
N −1 N −1
1 +iωn tm 1
Fm = F (tm ) ≈ D̂n e = D̂n e+i2π nm/N
T T
n=0 n=0
N −1
N −1
D̂n = D̂(ωn ) ≈ T Fm e−iωn tm = T Fm e−i2π nm/N (1.11)
m=0 m=0
where both m and n range from 0 to N − 1. These are the definition of what
is called the discrete Fourier transform (DFT). It is interesting to note that the
exponentials do not contain dimensional quantities; only the integers n, m, N
appear. In this transform, both the time and frequency domains are discretized,
and as a consequence, the transform behaves periodically in both domains. The
dimensional scale factors T , 1/T have been retained so that the discrete transform
gives the same numerical values as the continuous transform. There are other
possibilities for these scales found in the literature.
The discrete transform enjoys all the same properties as the continuous trans-
form; the only significant difference is that both the time domain and frequency
domain functions are now periodic. To put this point into perspective consider the
following: A discrete Fourier transform seeks to represent a signal (known over a
finite time T ) by a finite number of frequencies. Thus it is the continuous Fourier
transform of a periodic signal. Alternatively, the continuous Fourier transform itself
1.1 Fourier Transforms 17
can be viewed as a discrete Fourier transform of a signal but with an infinite period.
The lesson is that by choosing a large signal sample length, the effect due to
the periodicity assumption can be minimized and the discrete Fourier transform
approaches the continuous Fourier transform.
To help better illustrate the properties of the discrete transform, we consider an
eight-point sampled signal. This can also serve as the test case for any numerical
implementation of the DFT. Let the real-only function be given by the following
sampled values:
F1 = F2 = 1, F0 = F3 = F4 = F5 = F6 = F7 = 0
7
D̂n = Fm e−i2π nm/8 = F1 e−iπ n/4 + F2 e−iπ n/2
m=0
D̂0 = 2.0
D̂1 = 0.707 − 1.707i
D̂2 = −1.0 − 1.0i
D̂3 = −0.707 + 0.293i
D̂4 = 0.0
D̂5 = −0.707 − 0.293i
D̂6 = −1.0 + 1.0i
D̂7 = 0.707 + 1.707i
D̂8 = 2.0
D̂9 = 0.707 − 1.707i
The obvious features of the transform are that it is complex and that it begins to
repeat itself beyond n = 7. Note also that D̂4 [the ( 12 N + 1)th value] is the Nyquist
value. The real part of the transform is symmetric about the Nyquist frequency,
while the imaginary part is antisymmetric. It follows from this that the sum 12 [D̂n +
D̂N −n ] gives only the real part, that is,
18 1 Spectral Analysis of Wave Motion
These two functions are the even and odd decompositions, respectively, of the
transform. Also note that D̂0 is the area under the function.
The final step in the numerical implementation of the CFT is the development
of an efficient algorithm for performing the summations of the discrete Fourier
transform on a computer. The fast Fourier transform (FFT) is simply a very efficient
numerical scheme for computing the discrete Fourier transform. This is not a
different transform—the numbers obtained from the FFT are exactly the same in
every respect as those obtained from the DFT. The intention of this section is to just
survey the major features of the FFT algorithm and to point out how the great speed
increase is achieved. More detailed accounts can be found in the Refs. [4, 7] and
FORTRAN code is given in Ref. [19].
Consider the generic forward transform written as
N −1
Sn = Fm e−i2π nm/N , n = 0, 1, . . . , N − 1
m=0
S0 = {F0 + F1 + F2 + · · · }
S1 = {F0 + F1 e−i2π 1/N + F2 e−i2π 2/N + · · · }
S2 = {F0 + F1 e−i2π 2/N + F2 e−i2π 4/N + · · · }
..
.
Sn = {F0 + F1 e−i2π n/N + F2 e−i2π n2/N + · · · }
and so on. For each sum Sn , there are (N − 1) complex products and (N − 1)
complex sums. Consequently, the total number of computations (in round terms) is
on the order of 2N 2 . The purpose of the FFT is to take advantage of the special form
of the exponential terms to reduce the number of computations to less than N 2 .
The key to understanding the FFT algorithm lies in seeing the repeated forms
of numbers. This is motivated by considering the special case of N being 8. First
consider the matrix of the exponents −i2π( mn N ):
1.1 Fourier Transforms 19
⎡ ⎤
0 0 0 0 ··· 0
⎢0 ··· (N − 1) ⎥
⎢ 1 2 3 ⎥
⎢ ⎥
−i2π ⎢ 0 2 4 6 ··· 2(N − 1) ⎥
⎢ ⎥
N ⎢0⎢ 3 6 9 ··· 3(N − 1) ⎥
⎥
⎢ .. .. .. .. .. .. ⎥
⎣. . . . . . ⎦
0 (N − 1) 2(N − 1) 3(N − 1) · · · (N − 1)(N − 1)
then the effective number of different integers in the matrix is decreased. Thus if
N = 8 we get
⎡ ⎤
0 0 0 0 0 0 0 0
⎢0 1 2 3 4 5 6 7 ⎥
⎢0 8+2 8+4 8+6 ⎥
⎢ 2 4 6 0 ⎥
−i2π ⎢
⎢0 3 6 8+1 8+4 8+7 16 + 2 16 + 5 ⎥
⎥
8 ⎢ ⎢0 4 8+0 8+4 16 + 0 16 + 4 24 + 0 24 + 0 ⎥
⎥
⎢0 5 8+2 8+7 16 + 4 24 + 1 24 + 6 32 + 3 ⎥
⎣ ⎦
0 6 8+4 16 + 2 24 + 0 24 + 6 32 + 4 40 + 2
0 7 8+6 16 + 5 24 + 4 32 + 3 40 + 2 48 + 1
This comes about because the exponentials take on the simple forms
The regularity is enhanced even more if (N/2 = 4) is added to the latter part of the
odd rows, that is, if it is written as
20 1 Spectral Analysis of Wave Motion
⎡ ⎤
0 0 0 0 (0 0 0 0) + 0
⎢0 1 2 3 (0 1 2 3) + 4 ⎥
⎢0 6) + 0 ⎥
⎢ 2 4 6 (0 2 4 ⎥
−i2π ⎢
⎢0 3 6 1 (0 3 6 1) + 4 ⎥
⎥
8 ⎢ ⎢0 4 0 4 (0 4 0 4) + 0 ⎥
⎥
⎢0 5 2 7 (0 5 2 7) + 4 ⎥
⎣ ⎦
0 6 4 2 (0 6 4 2) + 0
0 7 6 5 (0 7 6 5) + 4
We see that many of the computations used in forming one of the summations
is also used in the others. For example, S0 , S2 , S4 , S6 all use the sum (F0 + F4 ).
Realizing that e−i2π 4/8 = −1, then we also see that all the odd summations contain
common terms such as (F0 −F4 ). This re-use of the same computations is the reason
a great reduction of computational effort is afforded by the FFT. The algorithm sets
up the bookkeeping so that this is done in a systematic way.
The number of computations with and without the FFT algorithm is given by
3
2N log2 N versus 2N 2
When N = 8, this gives a speed factor of only 3.5:1, but when N = 1024, this jumps
to over 100:1. It is this excellent performance at large N that makes the application
of Fourier analysis feasible for practical problems. On a benchmark machine of 1
GFlops, say, a 1024-point transform takes less than one millisecond.
The fast Fourier transform algorithm is so efficient that it has revolutionized the
whole area of spectral analysis. It can be shown quite simply that it enjoys all the
same properties of the continuous transform. Therefore, in the subsequent analyses,
we assume that any time input or response can be represented in the spectral form
F (t) = Ĉn e+iωn t
and the tasks of forward and inverse transforms are accomplished with a computer
program.
1 M 1 M
f (x) = a0 + am cos(2π mx/W ) = a0 + am cos(ξm x)
W 1 W 1
Multiply both sides by cos(ξn x) and integrate over the complete window. Because
the cosine functions are orthogonal on the window, and the integrals of the squares
are 12 W , the right-hand side is nonzero only if n = m and we get
+W/2
f (x) cos(ξm ) dx = {a0 , 12 am }
−W/2
If the integrals are computed numerically (through quadratures), then for efficiency,
both need only be computed on the half window and subsequently multiplied
by two. A general function is represented as the superposition of these two
representations.
As an example, consider a rectangular function of width a symmetrically
positioned at x = 0 as shown in Fig. 1.6. The coefficients evaluate to
Fig. 1.6 Fourier series representation of space distributions. (a) Rectangular distribution. Effect of
window size and number of terms. (b) Representing a concentrated load on a large space window
W = 800a
22 1 Spectral Analysis of Wave Motion
2
{a0 , 2 am }
1
= a, sin(ξm a/2) fo
ξm
The figure shows the reconstructions. Doubling the number of terms improves the
representation; the same effect is achieved by halving the window size. In our later
developments, the window size is chosen, thus if the window is too large for a given
number of terms accuracy suffers, if it is too small, we get the equivalent of wrap-
around problems which we interpret as reflections or the effect of image loads.
To elaborate on this last point, in some of our later analyses we have a need
to model a large space subjected to a relatively narrow distributed load (think of
an infinite beam subjected to a point load). Figure 1.6a shows that a rectangular
distribution is represented reasonably well with M = 40 when the window is W =
5a which implies that a window size of 800a requires M = 6400 which is a large
number. Therefore, ways of doing this efficiently and rationally are useful.
A Gaussian distribution has the representation
M 2 −(αmπ/W )2
e−(x/α) = 12 a0 +
2
am cos(ξ x) , am = e
1 W
A function of somewhat similar shape is the sine-squared function of extent a and
a reasonable connection sets a = 3.32α. The heavy and thin lines in Fig. 1.6b
for M = 2048 compare the Gaussian and sine-squared functions, they differ only
at their base where the latter has a definite delimination of its compact support.
Staying with the example of W = 800a, Fig. 1.6b shows the effect of the number
of terms in the summation. Superficially, there is a significant deterioration as the
number of terms is decreased. What is surprising is that in each case shown the area
under the curve is the same. Thus, if these plots represented a traction distribution,
then the resultant force would be the same in each case. Consequently, especially if
relatively remote responses are of interest, the resultant load is the significant factor.
In other words, while the M = 256 case does not represent the actual distribution,
it does represent the resultant effect and therefore could accurately predict remote
behaviors. The example problems in Chaps. 6 and 7 use this extensively and we
refer to the approach as periodically extended loads (pEL).
The following examples serve to show the basic procedures used in applying the
FFT to transient signals. The FFT is a transform for which no information is gained
or lost relative to the original signal. However, the information is presented in a
different format that often enriches the understanding of the information. In the
examples to follow, we try to present that duality of the given information.
1.2 Applications Using the FFT Algorithm 23
The examples are divided into two groups: one group explores the properties of
the FFT, the other shows applications to experimental data.
1
fNyquist =
2 T
FFT
CFT
Fig. 1.7 FFT transform of a rectangular pulse of width a = 50 µs. (a) Time domain signals.
Circles are sampled data. (b) Frequency domain transforms
24 1 Spectral Analysis of Wave Motion
This range is increased only by decreasing T . Thus, for a fixed number of points,
fine resolution in the time domain (small T ) means course resolution in the
frequency domain. Finer resolution in the frequency domain is achieved only by
increasing the sample length T .
It is also seen from the figure that the match between the FFT amplitude and the
continuous Fourier transform is very good at low frequencies but gets worse at the
higher frequencies. As mentioned before, the discrete and the continuous transforms
match closely only if the highest significant frequency in the signal is less than
the Nyquist. The rectangular pulse can be represented exactly only by using an
infinite number of sinusoids but the contributing amplitudes get smaller as frequency
increases and therefore a finite number of sinusoids can suffice.
For a given sample rate T , the number of samples only determines the density
of transform points. Thus the top two FFTs in Fig. 1.7 are numerically identical at
the common frequencies. Increasing the sample rate for a given number of samples
increases the Nyquist frequency, and therefore the range of comparison between the
discrete and continuous transforms. It is seen that by the fourth plot, the amplitudes
in the vicinity of the Nyquist are negligible. Using the discrete transform puts
an upper limit on the maximum frequency available to characterize the signal. If
the signal is not smooth, the amplitudes of the high-frequency sinusoids used to
describe the signal are high. Any attempt at capping the high-frequency sinusoids
introduces a distortion in the amplitudes of the lower-frequency sinusoids. This is
called aliasing and is discussed in more detail presently.
For future reference, we now summarize some of the inter-relationships among
the various parameters. Consider a signal of duration T sampled as N points. The
discretization rates in the two domains are
T 1
T = , f =
N T
Various forms for the Nyquist frequency are
N 1 N f
fNyquist = = =
2T 2 T 2
The complementarity of information between the time and frequency domains
is illustrated in Fig. 1.8. On comparing the top plots, we notice that the longer
duration pulse has a shorter main frequency range; however, both exhibit side lobes
which extend significantly along the frequency range. The second trace shows a
smoothed version of the second triangle compared to a rectangular pulse of the same
duration; in the frequency domain it is almost identical to the shorter triangle, the
only difference being the reduction in the high-frequency side lobes. Interestingly,
however, the rectangular pulse has a shorter main significant frequency range but the
side lobes are very significant extending beyond what is shown. Thus time domain
smoothing (using moving averages, say) acts as a high band filter in the frequency
1.2 Applications Using the FFT Algorithm 25
0. 100. 200.
. 300. 400. 0. 10. 20. 30. 40. 50.
Fig. 1.8 Comparison of some pulse-type signals. (a) Time domain signals. (b) Frequency domain
transforms
domain. In the wave analyses of the later chapters, we need to control the frequency
range of the inputs and we do this by using signals without sharp edges.
As a rule of thumb, a triangular pulse has a frequency content of about 2/Tp
where Tp is the duration of the pulse. This is confirmed in the figure for the
smoothed triangular pulse.
The remaining two plots are for modulated signals—in this case a sinusoid
modulated (multiplied) by a triangle. The sinusoid function on its own would give
a peak at 20 kHz, whereas the triangular modulations on their own are as shown for
the top traces. The product of the two in the time domain is a convolution in the
frequency domain. The convolution tends to distribute the effect of the pulse but it
is worth noting that while both waveforms have the same time domain amplitudes,
their transforms have different amplitudes indicating different energy levels. A point
of interest here is that if a very narrow-banded signal in the frequency domain is
desired, then it is necessary to extend the time domain signal as shown in the bottom
example.
The modulated waves are called wave packets or wave groups because they
contain a narrow collection (group) of frequency components. It might have been
thought that the triangular pulse is a more natural group because in the time domain
it is highly localized, but what is also important to us is the frequency content of the
wave and we see that the modulated wave is localized in both the space and time
domains. We use this wave packet idea regularly throughout the remaining chapters
to interrogate complex systems.
The signals we deal with often have multiple reflections. A sense of their effect
can be gaged from Fig. 1.9 where the traces are arranged so as to have a different
number of reflections.
In the time domain, we recognize the multiple reflections as localized distur-
bances in time; in fact, in this simple example we can cut the trace so as to isolate
each of the reflections separately. No such separation is possible in the frequency
domain because any small time-localized portion of signal contributes over the
total frequency range. Consequently, we get an interference effect—the adding and
26 1 Spectral Analysis of Wave Motion
Fig. 1.9 Effect of the superposition of multiple reflections. (a) Time domain signals. (b) Fre-
quency domain transforms
subtracting of phases result in amplifying certain components. We can see from the
figure the evolution of spectral peaks as the number of reflections is increased. In
vibration analyses, which would correspond to an infinite number of reflections, the
spectral peaks become very sharp and this situation would then be called resonance.
Force Force
Transducer Transducer
PCB 303A03
accelerometers Power
Supplies
Tek AM502
computer waveform recorder pre-amps
Fig. 1.10 Typical experimental setup for a plate (left) or frame (right)
Most types of electrical resistance strain gages are usually adequate and because
the events are of short duration, temperature compensation is usually not a serious
concern. The largest gage size allowable is dependent on the highest significant
frequency (or shortest wavelength) of the signal, but generally gages equal to or less
than 3 mm (0.125 in.) are good for most situations. A rule of thumb for estimating
the maximum length is
1 E
L= (1.12)
10f ρ
where L is the gage length, f is the highest significant frequency, and E and ρ are
the Young’s modulus and density, respectively, of the structural material. A typical
gage of length 3 mm (0.125 in.) is good to about 160 kHz. Other aspects of the use
of strain gages in dynamic situations are covered in Refs. [2, 8, 18]. Either constant-
current, potentiometer or Wheatstone bridge gage circuits can be used. Nulling of
the Wheatstone bridge is not necessary for dynamic problems because only the ac
portion of the signal is recorded.
A wide variety of small accelerometers are available. There is a problem with
accelerometers, however, in that if the frequency is high enough, ringing is observed
in the signal. This can often occur for frequencies as low as 20 kHz. Generally
speaking, accelerometers do not have the necessary frequency response of strain
28 1 Spectral Analysis of Wave Motion
gages, but they are movable, thus making them more suitable for larger, complex
structures. Also, note that if the stress is of interest, then integration of the signal is
required.
It is impossible to be categorical about the minimum setup necessary for
waveform recording but the following pieces of equipment definitely seem essential.
Good preamplifiers are essential for dynamic work; they not only provide the gain
for the low voltages from the bridge but also perform signal conditioning. The
former gives the flexibility in choice of gage circuit as well as reducing the burden
on the recorder for providing amplification. A separate preamplifier is needed for
each strain gage circuit and a typical one should have up to 100K gain, dc to 1
MHz frequency response, and selectable band pass filtering. Generally, the analog
filtering (which is always necessary) is performed at or below the Nyquist frequency
associated with the digitizing.
The most significant advance in recent years is the availability of digital
recorders. A typical digital waveform recorder would have 4 channels, 2048 8-
bit words of memory, and a sample rate of 0.2 µs to 0.1 s. It is common for
high-speed recorders to digitize to 8-bit words (although newer ones use 12-bit
words). This means it has a resolution of 1/256 of the full scale. If the signal
can be made to fill at least half full scale, then this resolution of about 1% of
the maximum signal is adequate. This situation is best achieved by having good
preamplifiers. Many of the newer digital recorders combine the features of an
oscilloscope with those of a small stand-alone computer. The DASH-18 [1] is an
example of the newer type of recorders; it is like a digital strip chart recorder and
notebook computer rolled into one. The DASH-18 can record up to 18 channels
and has a very flexible interface for manipulating the recorded data. There are also
relatively inexpensive recorders available that can be installed in desktop computers.
The Omega Instruments DAS-58 data acquisition card is an example; this is a 12-bit
card capable of 1 Mhz sampling rate and storing up to 1M data points in the on-
board memory. A total of 8 multiplexed channels can be sampled giving the fastest
possible sampling rate of 1M/8 = 125 K samples per second.
The measured signal can contain unwanted contributions from at least two
sources. The first is the ever present electrical noise. Analog filters can be used to
remove much of this, although care must be taken not to remove some of the signal
itself. The second may arise from unwanted reflections. This is especially prevalent
for dispersive signals as the recording is usually extended to capture as much as
possible of the initial passage of the wave. There are many ways of smoothing
digital data but the simplest is to use moving averages of various amounts and points
of application. Figure 1.11 shows an example of smoothing a signal that has an
exaggerated amount of noise and reflections present. The following sequence was
used on data values sampled at every 1 μs:
Another random document with
no related content on Scribd:
exposure to the wildest winter night. For two-and-fifty years he
was the subject of a supernatural palpitation, which kept his bed
and chair, and everything moveable about him, in a perpetual
tremble. For that space of time his breast was miraculously
swollen to the thickness of a fist above his heart. On a post-
mortem examination of the holy corpse, it was found that two of
the ribs had been broken, to allow the sacred ardour of his heart
more room to play! The doctors swore solemnly that the
phenomenon could be nothing less than a miracle. A divine hand
had thus literally ‘enlarged the heart’ of the devotee.[320] St. Philip
enjoyed, with many other saints, the privilege of being
miraculously elevated into the air by the fervour of his
heavenward aspirations. The Acta Sanctorum relates how Ida of
Louvain—seized with an overwhelming desire to present her
gifts with the Wise Men to the child Jesus—received, on the eve
of the Three Kings, the distinguished favour of being permitted to
swell to a terrific size, and then gradually to return to her original
dimensions. On another occasion, she was gratified by being
thrown down in the street in an ecstacy, and enlarging so that her
horror-stricken attendant had to embrace her with all her might to
keep her from bursting. The noses of eminent saints have been
endowed with so subtile a sense that they have detected the
stench of concealed sins, and enjoyed, as a literal fragrance, the
well-known odour of sanctity. St. Philip Neri was frequently
obliged to hold his nose and turn away his head when
confessing very wicked people. In walking the streets of some
depraved Italian town, the poor man must have endured all the
pains of Coleridge in Cologne, where, he says,
VI.
VII.
VIII.
IX.
X.
XI.
In the summer of 1681, Madame Guyon, now thirty-four years of
age, quitted Paris for Gex, a town lying at the foot of the Jura,
about twelve miles from Geneva. It was arranged that she should
take some part in the foundation and management of a new
religious and charitable institution there. A period of five years
was destined to elapse before her return to the capital. During
this interval, she resided successively at Gex, Thonon, Turin,
and Grenoble. Wherever she went, she was indefatigable in
works of charity, and also in the diffusion of her peculiar
doctrines concerning self-abandonment and disinterested love.
Strong in the persuasion of her mission, she could not rest
without endeavouring to influence the minds around her. The
singular charm of her conversation won a speedy ascendency
over nearly all with whom she came in contact. It is easy to see
how a remarkable natural gift in this direction contributed both to
the attempt and the success. But the Quietest had buried nature,
and to nature she would owe nothing,—these conversational
powers could be, in her eyes, only a special gift of utterance from
above. This mistake reminds us of the story of certain monks
upon whose cloister garden the snow never lay, though all the
country round was buried in the rigour of a northern winter. The
marvellous exemption, long attributed by superstition to miracle,
was discovered to arise simply from certain thermal springs
which had their source within the sacred inclosure. It is thus that
the warmth and vivacity of natural temperament has been
commonly regarded by the mystic, as nothing less than a fiery
impartation from the altar of the celestial temple.
At Thonon her apartment was visited by a succession of
applicants from every class, who laid bare their hearts before
her, and sought from her lips spiritual guidance or consolation.
She met them separately and in groups, for conference and for
prayer. At Grenoble, she says she was for some time engaged
from six o’clock in the morning till eight at evening in speaking of
God to all sorts of persons,—‘friars, priests, men of the world,
maids, wives, widows, all came, one after another, to hear what I
had to say.’[332] Her efforts among the members of the House of
the Novitiates in that city, were eminently successful, and she
appears to have been of real service to many who had sought
peace in vain, by the austerities and the routine of monastic
seclusion. Meanwhile, she was active, both at Thonon and
Grenoble, in the establishment of hospitals. She carried on a
large and continually increasing correspondence. In the former
place she wrote her Torrents, in the latter, she published her
Short Method of Prayer, and commenced her Commentaries on
the Bible.[333]
But alas! all this earnest, tireless toil is unauthorized. Bigotry
takes the alarm, and cries the Church is in danger. Priests who
were asleep—priests who were place-hunting—priests who were
pleasure-hunting, awoke from their doze, or drew breath in their
chase, to observe this woman whose life rebuked them—to
observe and to assail her; for rebuke, in their terminology, was
scandal. Persecution hemmed her in on every side; no
annoyance was too petty, no calumny too gross, for priestly
jealousy. The inmates of the religious community she had
enriched were taught to insult her—tricks were devised to
frighten her by horrible appearances and unearthly noises—her
windows were broken—her letters were intercepted. Thus,
before a year had elapsed, she was driven from Gex. Some
called her a sorceress; others, more malignant yet, stigmatized
her as half a Protestant. She had indeed recommended the
reading of the Scriptures to all, and spoken slightingly of mere
bowing and bead-counting. Monstrous contumacy—said, with
one voice, spiritual slaves and spiritual slave-owners—that a
woman desired by her bishop to do one thing, should discover
an inward call to do another. At Thonon the priests burnt in the
public square all the books they could find treating of the inner
life, and went home elated with their performance. One thought
may have embittered their triumph—had it only been living flesh
instead of mere paper! She inhabited a poor cottage that stood
by itself in the fields, at some distance from Thonon. Attached to
it was a little garden, in the management of which she took
pleasure. One night a rabble from the town were incited to terrify
her with their drunken riot,—they trampled down and laid waste
the garden, hurled stones in at the windows, and shouted their
threats, insults, and curses, round the house the whole night.
Then came an episcopal order to quit the diocese. When
compelled subsequently, by the opposition she encountered, to
withdraw secretly from Grenoble, she was advised to take refuge
at Marseilles. She arrived in that city at ten o’clock in the
morning, but that very afternoon all was in uproar against her, so
vigilant and implacable were her enemies.