Professional Documents
Culture Documents
SCH Merr JR 2016
SCH Merr JR 2016
Lester W. Schmerr Jr.
Fundamentals
of Ultrasonic
Nondestructive
Evaluation
A Modeling Approach
Second Edition
Springer Series in Measurement Science
and Technology
Series Editors
Markys G. Cain, Surrey, UK
Giovanni Battista Rossi, Genova, Italy
Jiřı́ Tesař, Prague, Czech Republic
Marijn van Veghel, JA Delft, The Netherlands
Fundamentals of Ultrasonic
Nondestructive Evaluation
A Modeling Approach
Second Edition
Lester W. Schmerr, Jr.
Ames, Iowa
USA
It has been over 15 years since the first edition of this book was published. In that
period, the use of models in the ultrasonic NDE field has seen a steady growth.
Model-based ultrasonic NDE simulation packages were in their infancy when the
first edition appeared, but now both research-grade and commercial software
packages are available. Thus, it is more important than ever to understand the
fundamental physics and mathematics behind ultrasonic NDE models. Another
notable change has been the development of models to aid in the evaluation of
NDE reliability and flaw detectability. This application of models has been termed
model-assisted probability of detection or “MAPOD.” Also, starting in 2001 a
series of ultrasonic benchmarks were sponsored by the World Federation of NDE
Centers and the results presented at the annual Review of Progress in Quantitative
NDE meetings. The aim of the benchmark studies was to validate modeling
assumptions and compare the results of different modeling approaches to experi-
ments on well-characterized samples. Such efforts, which continue today, are
extremely important for models to gain a much wider acceptance and use by the
NDE community.
Two books by the author have appeared since the first edition that can be
considered companions and extensions of this book. The first book, Ultrasonic
NDE Systems—Models and Measurements, coauthored with Prof. Sung-Jin Song,
extended significantly the linear time-shift-invariant (LTI) system approach that
appeared in the first edition of this book and showed how all the electrical and
electromechanical parts of an ultrasonic NDE system (the pulser/receiver, cabling,
transducers) could be characterized by models and the parameters of those models
obtained with simple electrical measurements. These system elements were then
combined to show how they produce the system function (or system efficiency
factor) discussed in this book. The book with Prof. Song also provided a detailed
discussion of Gaussian beam theory, parts of which have been modified and
incorporated into this second edition because Gaussian beams have proven to be
a highly effective ultrasonic beam model. That book also provided a new derivation
of ultrasonic measurement models that has been adopted in this second edition
v
vi Preface to the Second Edition
1
MATLAB® is a registered trademark of the Math Works, Inc.
Preface to the Second Edition vii
ix
x Preface to the First Edition
measurement systems and to demonstrate the use of such models in some funda-
mental calibration and sizing applications. Although an ultrasonic NDE system is a
complicated collection of many electromechanical processes, only a relatively
small number of fundamental concepts are needed to develop a complete model
of those processes. These concepts include linear time-shift-invariant systems and
the Fourier transform, reciprocity and fundamental solutions, and plane wave
theory and the stationary phase method. These concepts are described in a relatively
self-contained fashion in both the main text and appendices.
To make this material accessible to those without a background in elasticity and
elastic wave propagation theory, the book uses a two-tier approach: we first
describe models in terms of a scalar-based theory using acoustic waves, followed
by the more complete elastic wave models. Since acoustic (fluid) models in many
cases capture the essence of the physics involved, this approach has the added
advantage of illustrating most of the important results in a simple framework
uncluttered by the complex tensorial aspects inherent in general elastic wave
models.
Major portions of this book come from a graduate-level course on the funda-
mentals of ultrasonic NDE that I taught for a number of years at Iowa State
University. The general approach taken here also draws heavily on the work of
my colleagues at the Center for NDE at Iowa State University and elsewhere. The
book is a mixture of wave propagation and scattering fundamentals, well-
established research results, and new developments. Many of the topics covered
are scattered throughout the literature, so that another objective of this book is to
bring these pieces together for the first time in a coherent framework.
Since the book stresses mathematical models involved in simulating an ultra-
sonic NDE measurement system, sufficiently detailed derivations are given, so that
assumptions (and approximations) on which models are based can be appreciated.
To aid the learning process, problems at the end of each chapter apply and extend
the concepts presented. Although all the material in the text cannot be covered in a
semester course, the organization is such that several different semester course
structures at the advanced undergraduate and graduate levels can be constructed.
I would like to acknowledge specifically a number of individuals whose own
research led me to write this book. They include my friend and colleague, Alexan-
der Sedov; colleagues at the Center for NDE—D.O. Thompson, R. B. Thompson,
J. Rose, T. Gray, R. Roberts, D. Hsu, and M. Garton; and my students—T. P. Lerch,
S. J. Song, C. P. Chiou, and J. S. Chen.
1 An Ultrasonic System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Elements of an Ultrasonic NDE System . . . . . . . . . . . . . . . . . . 1
1.2 The Pulser-Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Ultrasonic Transducers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Ultrasonic Digitizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Ultrasonic Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 About the Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Linear Systems and the Fourier Transform . . . . . . . . . . . . . . . . . . 15
2.1 Linear Time-Shift Invariant Systems . . . . . . . . . . . . . . . . . . . . 15
2.2 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 LTI Systems and the Impulse Response Function . . . . . . . . . . . 20
2.4 An Ultrasonic NDE Measurement System as an LTI System . . . 22
2.5 About the Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Wave Motion Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Governing Equations for a Fluid . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.2 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.3 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.4 Interface/Boundary Conditions . . . . . . . . . . . . . . . . . . 36
3.2 Governing Equations for an Elastic Solid . . . . . . . . . . . . . . . . . 38
3.2.1 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.2 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.3 Navier’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.4 Interface/Boundary Conditions . . . . . . . . . . . . . . . . . . 42
xi
xii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753
Chapter 1
An Ultrasonic System
Figure 1.1 shows a sketch of the basic ingredients that make up an ultrasonic
measurement system that can be used to evaluate materials for flaws [1–3]. The
“driver” of the system is the pulser section of a pulser-receiver. A spike type of
pulser typically puts out very short (approximately 0.1 μsec in duration) repetitive
electrical pulses (approximately 1 ms apart) having amplitudes on the order of
several hundred volts (Fig. 1.2). These electrical pulses drive a transducer in contact
with a part being examined. The transducer is normally made from a piezoelectric
material which converts the electrical pulses into mechanical pulses that then
propagate as a beam of ultrasound into the part.
If a flaw is present in this beam, then a portion of the incident ultrasonic energy is
scattered as additional mechanical waves throughout the part and can be picked up,
either by the same transducer or, as shown in Fig. 1.1, a second receiving trans-
ducer. The receiving piezoelectric crystal transducer transforms the scattered pulses
from the flaw back into electrical pulses, using the fact that piezoelectric crystals
are reciprocal in nature, i.e. they can convert electrical energy into mechanical
energy and vice-versa.
These electrical signals are then amplified in the receiver portion of the pulser-
receiver and displayed as a voltage versus time trace on an oscilloscope (Fig. 1.1),
which is triggered in synchronization with the pulse repetition frequency of the
pulser-receiver. Although the pulser puts out a repetitive signal, what is viewed on
the oscilloscope screen is the response from only a single excitation, since the time
scale over which the flaw signals are viewed is normally much smaller than the time
scale characteristic of the repetitive pulses (Fig. 1.3). Note that at the very begin-
ning of the time trace on the oscilloscope one normally sees a very large “main
1.2 The Pulser-Receiver 3
bang” signal. This signal is associated with some of the driving voltage that leaks
into the receiver portion during the excitation of the transmitter. Because this large
signal can mask very early arriving flaw (or other) signals, this early portion of the
time trace is often referred to as a “dead zone.”
In modern ultrasonic NDE systems, it is necessary to capture the received signals
so that they can be processed further and quantitatively evaluated. This is done
through an analog-to-digital conversion process which can take place inside the
oscilloscope itself, if it is a digital scope, as shown in Fig. 1.1, or via an external
digitizer. Once in digital form, the signal can then be easily transferred to a
computer for analysis.
The elements shown in Fig. 1.1 describe in very general terms what constitutes a
basic ultrasonic NDE measurement system. Now, we will examine in more detail
some of the individual elements—the pulser-receiver, the transducers, and the
digitizers.
Ultrasonic pulser-receiver units that are designed for use in the field or on the
manufacturing shop floor (see Fig. 1.4) typically are combined with a display screen
in a single, portable instrument. These instruments may have many microprocessor-
controlled options for calibration and signal interpretation purposes that are impor-
tant for effective use of the instrument but are not part of its basic function—namely
to drive the transducer on the pulser end and receive and amplify the signals on the
receiving end. Figure 1.5 shows a commercially available square wave pulser-
receiver that contains only a minimum number of essential control features.
Through the pulser repetition setting (PRF) the user can control the frequency at
which pulses are generated to drive the transducer. For this pulser repetition rates
range between 100 and 5000 pulses per second or at a rate that can be controlled by
an external source. The pulser output voltage is also adjustable, which in this
instrument can be set over a 100–400 V range. The characteristics of the driving
square wave pulse can also be set to match the transducer frequency in millions of
cycles per second (MHz).
When a receiving transducer that is attached to the pulser-receiver is excited by
scattered waves, a transient charge appears across the transducer faces and gener-
ates an electrical pulse that is amplified and, if necessary, filtered in the receiver
section. The front panel gain settings (Fig. 1.5) control the amount of amplification
needed on reception, and the low/high pass filters control the amount of high
frequency and low frequency “noise” that is filtered out, if any.
Finally, the pulser-receiver shown has a front panel switch to control whether a
single transducer is being used as both transmitter and receiver (T/R output used
only) or separate transmitting and receiving transducers are present (both T/R and R
outputs used for the sending and receiving transducers, respectively). On the back
panel of the pulser-receiver, an additional output is available to drive the scope
display with the amplified received signal, and a signal in synchronization with the
pulser is available as a scope trigger.
A pulser-receiver is an electrical network whose behavior can be modeled in
terms of an equivalent circuit [4]. Similarly, an ultrasonic piezoelectric crystal
transducer that is driven by the pulser-receiver is a complex electromechanical
device that must be modeled as a rather complicated equivalent circuit. Several
authors have described such transducer circuit models so we refer the reader to
those sources [5–7]. We will not discuss in more depth here either those transducer
circuit models or give a more detailed description of the actual pulser-receiver
circuits, since as we will show later, the effects of these parts of the measurement
process can be determined experimentally without an explicit knowledge of their
characteristics.
1.3 Ultrasonic Transducers 5
A P-wave transducer of the contact type is normally used with a thin layer of
couplant between the wear plate and the part. This coupling layer, which can be
water, oil, glycerin, or one of a number of commercially available fluids, allows
efficient transfer of the mechanical motion of the transducer into the part as the
transducer is scanned across the surface. An S-wave contact transducer, in contrast,
requires a couplant that can transfer shearing motions. Since “ordinary” fluids do
not support shear waves, special highly viscous shear-wave couplants or permanent
glues are needed to allow transfer of this type of energy. In either case, the S-wave
transducer is essentially immobile and scanning is not possible. In addition to
contact transducers there are a number of other commonly used types of ultrasonic
transducers. Figure 1.8 shows the internal construction of a transducer typically
used for immersion testing, where the transducer and the part to be tested are placed
in a water bath. In this configuration, a P-wave is launched into the fluid and then
propagates through the fluid and into the part. Immersion testing has a distinct
advantage over contact testing in that the water bath provides a source of constant
coupling to the part. Thus, the input energy can be controlled precisely while
scanning the transducer. In contact testing it is very difficult to maintain a constant
coupling to the part while moving the transducer, so that there may be a large
variability in the input energy. However, in contact testing the efficiency of energy
transfer into the part is normally considerably better than in immersion testing since
a large percentage of the energy that strikes the interface between the water and the
part being inspected in immersion testing is reflected back into the fluid (approx-
imately 80 % of the incident energy is reflected when going from water to steel, for
example).
The internal construction of immersion transducers (Fig. 1.8) is very similar to
that of contact probes. Generally, however, the external connector is of the UHF
type which can attach to a search tube (wand). This tube is then typically connected
to a mechanical scanning assembly to allow precise control over the transducer
motion.
1.3 Ultrasonic Transducers 7
Another difference between the contact and immersion probes is in the wear
plate. In the immersion case, this plate is called a quarter wavelength plate and is
designed specifically to allow the efficient transfer of energy from the crystal to the
water.
Other transducers used in ultrasonic testing include focused immersion trans-
ducers (Fig. 1.9), angle beam contact transducers (Fig. 1.10), delay line contact
transducers (Fig. 1.11), and phased arrays (Fig. 1.12). A focused probe uses an
acoustic lens between the crystal and the water to concentrate the beam of energy
produced in a narrow region. As a consequence, focused probes have a higher
sensitivity and spatial resolution than unfocused transducers. Angle beam trans-
ducers (Fig. 1.10) have the crystal mounted internally at an angle on a plastic
wedge. This configuration allows, through the process of mode-conversion, the
generation of either shear waves traveling into the part at an angle (Chap. 6) or
Rayleigh waves (Chap. 7) which travel along the part surface. In either case, the
transducer is coupled to the part with a thin fluid couplant, as with ordinary P-wave
contact probes, and can be scanned along the surface in the same manner. The delay
line contact probe (Fig. 1.11) internally contains a thick plastic shoe between the
crystal and the part. This shoe provides sufficient material between the crystal and
the part so that pulse reflected from the front surface of the part is removed from the
“dead region” associated with the “main bang” of the transducer. By measuring the
time of arrival of both this front surface pulse and that of the pulse received from the
back surface of a part, one can easily perform thickness measurements, even on
very thin stock. The use of phased array transducers (Fig. 1.12) is now becoming
commonplace in ultrasonic NDE applications. In a phased array a series of small
piezoelectric elements send and receive sound independently. Figure 1.12 shows
the individual waves generated by elements (dashed lines) that are driven identi-
cally. These waves coalesce to form an overall propagating beam similar to that of a
single element transducer the size of the overall array. The array sound beam,
however, can also be electronically steered and focused by varying the time delays
of excitation of the elements, resulting in a very versatile tool for conducting
inspections. The characteristics of arrays also makes them ideal for generating
images using methods such as the synthetic aperture focusing technique (SAFT)
and others [11].
There also are a wide variety of other types of ultrasonic transducers used in
special applications that we will not discuss here. In choosing an ultrasonic trans-
ducer for a particular purpose, one is usually faced with the specification of
characteristics such as housing and connector type, frequency (wide or narrow
bandwidth), focal length (if focused), and angle desired in a particular material
(for an angle beam probe). The effects of many of these choices on the measure-
ment process will be discussed in later chapters.
The voltage versus time trace on the oscilloscope screen is an analog signal that
needs to be captured in digital form to allow further processing and manipulation
via computer. If a digital oscilloscope is being used, then provided the sampling
frequency of the scope is adequate for the very short duration pulses characteristic
of ultrasonic signals, this digital conversion process is taken care of automatically.
Otherwise, some form of external digitizer is necessary. Both digital scopes and
stand-alone digitizers can often operate in either of two modes. In a real time
1.4 Ultrasonic Digitizers 9
sampling mode, a very fast (approximately 100 MHz or greater sampling rate)
analog to digital converter is used to capture an ultrasonic wave form signal during
one repetition cycle, as illustrated in Fig. 1.13 for a simple pulse shape being
sampled at a frequency f ¼ 1=Δt, where Δt is the time spacing between samples.
Because the total time required to capture the signal in a real time mode is very
small, most digitizers operating in this mode can also do averaging of the signals
received over many repetition cycles to reduce electronic noise. In an equivalent
time sampling mode, on the other hand, advantage is taken of the fact that the
ultrasonic signal received is actually a repetitive signal. This allows us to capture
only a portion of the signal during each repetition. For example, as shown in
Fig. 1.14, if the first point on the wave form is captured at time t ¼ 0 during one
repetition cycle and subsequent single points captured at times T þ nΔt during the
next n cycles, where T is the pulser repetition period and Δt is a small time shift,
then the entire wave form can be built up, one sample at a time, using an analog-to-
digital converter operating at a frequency of only 1/T samples/s. The equivalent
sampling rate of such a process is still, however, 1/Δt since Δt is the time separation
between successively sampled point on the wave form. Of course, it is not necessary
to acquire only a single point at each repetition if the analog-to-digital converter
present can operate at a much higher rate than 1/T samples/s. Signal averaging is
also possible in equivalent time sampling, although it will be inherently slower than
in real time mode because of the lower real sampling rates involved.
Once a wave form signal is digitized, transfer to a computer allows further
processing and analysis of the signal. A common processing step used in many
systems is to compute the frequency components of the signal using Fourier
analysis. Many modern digital scopes now have internal processing capabilities
so that Fourier analysis can be performed in the scope itself without the assistance
of a computer. In Chap. 2 we will discuss this type of analysis and why it is so
important in modeling ultrasonic systems.
10 1 An Ultrasonic System
All fields have special terms that they use to define particular setups and processes
and the field of ultrasonic NDE is no exception. Here, we will briefly describe some
of the commonly used ultrasonic terminology.
Figures 1.15a–c show three different types of ultrasonic contact testing config-
urations. In Fig. 1.15a, a single transducer is used both as a transmitter and as a
receiver of ultrasound from some scattering source such as a flaw. This is referred to
as a pulse-echo testing setup. In the field of electromagnetic scattering, a similar
setup would be called a mono-static scattering configuration. When separate trans-
mitting and receiving transducers are used instead, as shown in Fig. 1.15b, the setup
is referred to as a pitch-catch setup. In electromagnetics, a similar configuration
would be called a bi-static scattering setup. Finally, a third type of NDE setup is a
through-transmission setup where the transmitting and receiving transducers are
placed facing one another, as shown in Fig. 1.15c. This case differs from either the
pulse-echo or pitch-catch setups in that the flaw signal is seen as a perturbation of a
directly transmitted wave pulse that is always present. Although the pulse-echo,
pitch-catch and through-transmission setups are shown in Fig. 1.15 for contact
testing, the same terminology is used to describe similar immersion testing setups.
In ultrasonic NDE, there are three types of displays often used for presenting
flaw information. The voltage versus time display on an oscilloscope screen that we
have described previously is called an A-scan. This is the type of display most
commonly used in NDE testing. In Fig. 1.16, we show a different type of display,
called a B-scan. Here, a position sensor is used to monitor the distance of the
transducer along a linear scan path. When a signal is received from a source, such as
a flaw or a surface of the part, the time of arrival of the signal is converted to a depth
location for the signal and the position sensor gives the horizontal location. A point
on a display screen can then be displayed to show the origin of the signal. Collecting
the display points generated during the scan, a two-dimensional cross-sectional side
view of the flaw (and other returns) is obtained. An ultrasonic C-scan, in contrast, is
obtained from a two-dimensional scanning procedure (Fig. 1.17). In this case a
transducer moves in a raster-like fashion and both its x- and y- coordinates are
recorded along with the received signal. A time gate is often set so as to only
consider signals within a given time range (and, hence, corresponding depth range)
within the specimen. The peak signal amplitudes at points along the raster scan are
recorded and often plotted versus x and y on a color scale according to their
magnitudes. The result is a two-dimensional “downwards” view of the flaw
responses. Figure 1.17 shows a typical C-scan setup and associated output. If one
is using a phased array transducer, it is also possible to generate B- and C-scans as
well as a variety of other scans by manipulating the sequence of firing and time
delays of the array.
Other types of displays of ultrasonic data are used in NDE testing besides the A,
B, and C-scans described above. However, the terms used to describe such displays
have not been standardized [3, 8] and the reader is advised to consider other setup
displays carefully to understand what is actually being presented.
12 1 An Ultrasonic System
1.7 Problems
References
Journals
Materials Evaluation
Research in Nondestructive Evaluation
Journal of Nondestructive Evaluation
Nondestructive Testing and Evaluation
NDT & E International
British Journal of NDT
Soviet Journal of NDT
Ultrasonics
Journal of the Acoustical Society of America
Journal of Applied Physics
Wave Motion
Books
In Chap. 1 we saw that an ultrasonic system has many components. Those components
individually can be complex electromechanical systems such as, for example, the
ultrasonic transducers. To model each of the elements that go into an ultrasonic
system and how they work together to produce a measured response is a
challenging task, indeed. In this Chapter we will present a very general modeling
framework of linear time-shift invariant (LTI) systems which we will use in order
to describe a complete ultrasonic NDE measurement system. Many of the
remaining chapters will fill in the details of this general framework and ultimately
produce an explicit LTI models of the entire ultrasonic measurement process.
Figure 2.1 shows the general schematic of a system which takes some input, i(t),as a
function of time, t, and produces an output o(t). For example, the system might
represent the entire ultrasonic measurement process itself, where i(t) is the driving
voltage pulse generated from the pulser and o(t) is the voltage versus time trace on
the oscilloscope screen. Alternatively, the system may be only a particular compo-
nent of the entire measurement process, such as a transducer, where i(t) is the
voltage driving the transducer and o(t) is the resulting mechanical velocity or force
that is used to launch waves into the surrounding medium. In all such cases, we will
assume that the system being described can be modeled as a linear time-shift
invariant (LTI) system where
where, i1 and i2 are two arbitrary inputs and c1 and c2 are constants. Thus, LTI
systems obey the principle of superposition. The time-shift invariance property of
LTI systems requires
which says that a delay in the input produces an identical delay in the output. Most
ultrasonic NDE systems can be characterized as LTI systems. However, in some
other ultrasonic applications where extremely high power is used, such as in
ultrasonic cutting, for example, nonlinear behavior may invalidate the use of
Eq. (2.1).
There are several concepts that play key roles in LTI systems. One of these
concepts is convolution, where by definition the convolution integral of two
functions, f(t) and g(t), f ∗g, is given by
ð
þ1 ð
þ1
We will see shortly where convolution appears in LTI systems. Another important
concept for LTI systems is that of the Fourier transform.
Ultrasonic NDE deals primarily with pulses of various types: voltage pulses,
pressure pulses in fluids, elastic wave pulses in solids, etc. These pulses are
transient time disturbances that characterize the behavior of an ultrasonic system
component in the time domain. It is often desirable, however, to consider other
domains for describing the response of the components. One domain that is
particularly useful for LTI systems is the frequency domain. In the frequency
domain, one describes responses in terms of the decomposition of a pulse into a
distribution of sinusoids of different frequencies and amplitudes. A time domain
2.2 The Fourier Transform 17
pulse is transformed into the frequency domain through the Fourier transform,
defined as
ð
þ1
pffiffiffiffiffiffiffi
where F(ω) is the Fourier transform of f(t) and i ¼ 1. This transformation is
reversible so that given F(ω) we can recover f(t) through the inverse Fourier
transform, defined by:
ð
þ1
1
f ðt Þ ¼ FðωÞexpðiωtÞ dω: ð2:6Þ
2π
1
Note that while f(t) is a real function, F(ω) is complex. Thus we can write F in
general in terms of its magnitude,jF(ω)j, and phase, ϕ(ω), as
f ¼ ω=2π; ð2:9Þ
where the 2π factor appears since there are 2π radians in one cycle. Note that in
recovering f(t) one needs to integrate over both positive and negative frequency
components of F(ω) as shown in Eq. (2.6). However, these negative frequency
components are present only as a mathematical requirement to guarantee that the
function f(t) that is recovered from Eq. (2.6) is indeed real. In fact, all the informa-
tion actually needed in the frequency domain is contained in the behavior of F(ω)
for the positive frequency components only. This can be seen from the definition of
F(ω) in Eq. (2.5), where if f(t) is real we can easily show that
where ( )* indicates complex conjugation, i.e. (a + i b)* ¼ ai b. Sections A.1 and
A.2 in Appendix A list some of the other important properties of the Fourier
transform and give some example transforms for specific functions.
18 2 Linear Systems and the Fourier Transform
The magnitude of F is plotted in Fig. 2.3. As can be seen from Fig. 2.3, most of the
frequency content of F is contained in the main “lobe” of frequencies between
2π=Δt and 2π/Δt. If Δt is small, this main lobe will be very wide and we say the
pulse has a very broad band response in the frequency domain. Conversely, when
Δt is large, the pulse has a narrow band response, mostly centered in this case
about its zero frequency (d.c.) component.
An important limiting case of this function is obtained if we set AΔt ¼ 1 and
consider f(t) and F(ω) as Δt ! 0. In this case, f(t) becomes a spike of infinite
amplitude (but containing unit area) at t ¼ 0 while FðωÞ ! 1 for all frequencies
(Fig. 2.4a, b). This limiting “function” is called a delta function, δ(t). As can be seen
from Fig. 2.4b, the delta function has infinite bandwidth and excites all frequency
components in the frequency domain equally. Appendix B lists some of the
important properties of delta functions.
In practice, Fourier transforms often cannot be obtained analytically in this
fashion. However, the numerical calculation of Fourier transforms is now
2.2 The Fourier Transform 19
commonplace, through the use of the concepts of the discrete Fourier transform
and its efficient calculation by the Fast Fourier Transform (FFT) algorithm, as
discussed in Sects. A.3 and A.4 in Appendix A.
One property of Fourier transforms which is very useful for LTI systems is
embodied in the following theorem:
Thus, in the frequency domain the frequency components of two convolved func-
tions is just the product of their individual frequency components. Note that the
multiplication involved here is a complex multiplication since in general both F and
G are themselves complex.
Proof of this theorem is not difficult. From the definition of the Fourier transform
and convolution we have
ð
ð þ1
þ1
ð þ1
þ1 ð
Fð f ∗gÞ ¼ f ðuÞexpðiωuÞgðτÞexpðiωτÞ dτ du ð2:13Þ
1 1
In this section, we will show that by combining the concepts of LTI systems,
convolution, Fourier transforms, and the delta function, one can arrive at a result
which will allow us to model complex systems such as an ultrasonic measurement
system.
Consider a LTI system as shown in Fig. 2.5 where the input is a delta function,
i.e. iðtÞ ¼ δðtÞ. The output, g(t), of this system is called the unit impulse response
function. As we saw in the last section, g(t) is the response to an ideal infinitely wide
band input. The impulse response function is important for LTI systems because the
response of such a system to an arbitrary input, i(t), is given by the convolution
integral of g(t) with that input:
ð
þ1
Thus, knowing the impulse response function of a LTI system completely charac-
terizes the output of that system to any input.
To prove this result, we first break i(t) into small rectangles (Fig. 2.6). At t ¼ τ
consider the highlighted rectangle shown in Fig. 2.6. This rectangle approximates a
delta function input at time τ of strength (area) i(τ)Δτ so that the output, Δo(t), at
time t, using the time-shift invariance property of an LTI, is
ΔoðtÞ ffi iðτÞΔτgðt τÞ. By superposition we can then add up all the contributions
from all the rectangular areas of the input to obtain the total output, o(t), as
X
oðtÞ ffi iðτÞΔτgðt τÞ
ð
þ1
Ultrasonic systems are examples of causal systems [1], i.e. we can assume gðtÞ ¼ 0
and iðtÞ ¼ 0 for t < 0. In this case Eq. (2.16) can also be written equivalently as
ð
τ¼t
From the theorem of the last section on the relationship between the convolution
integral and Fourier transforms, it follows from Eq. (2.16) that
O i ð ωÞ
Gi ðωÞ ¼ ; ð2:21Þ
I i ð ωÞ
where (Oi(ω), Ii(ω)) are the frequency domain values of the output and input,
respectively, of the ith LTI system. Unlike the convolution relationship of
Eq. (2.18), however, deconvolution by the direct division of Eq. (2.21) is unstable
in the presence of noise. This problem is inherently present in ultrasonic systems
when the input and output signals are measured experimentally since those signals
are typically weak at frequencies outside the bandwidth of the transducer(s)
present where an application of Eq. (2.21) would simply be dividing noise by
noise. In ultrasonic NDE a Wiener filter is typically used to replace Eq. (2.21) by
a form that makes the deconvolution process stable without significantly affecting
the end result. Details of that replacement are given in problems discussed in
Chaps. 9 and 14.
Figure 2.8 shows the components of an ultrasonic NDE testing situation where a
flawed part is being interrogated in a pitch-catch immersion arrangement. This type
of setup will be used here, and in later Chapters, to represent a generic NDE
measurement system. Contact systems, however, can be treated in a very similar
manner, as we will see in Chap. 12. First, we will consider the ultrasonic system in
three parts (Fig. 2.9). During sound generation a driving voltage pulse, vi(t),
generated by the pulser travels to the transmitting transducer and is transformed
to a velocity on the face of the transducer. We can represent this part of the system
by an impulse response function, tg(t), whose Fourier transform is the sound
generation transfer function, Tg(ω). The motion of the sending transducer face
produces a sound beam that travels through the water and is transmitted into the
part where it interacts with a flaw. The waves that are incident on this flaw generate
scattered waves, some of which propagate to the receiving transducer, generating a
transmitting receiving
transducer transducer
flaw
d4
d1 d2 d3
2.4 An Ultrasonic NDE Measurement System as an LTI System 23
force, on the face of that transducer. This part of the system, which involves the
acoustic and elastic ultrasonic waves present, we will characterize by an impulse
response, ta(t) whose corresponding acoustic/elastic transfer function is Ta(ω). The
force on the receiving transducer is transformed into an electrical pulse that travels
to the receiver, where it is amplified and produces an output voltage, v0(t). This
reception part of the system can be characterized by the impulse response function,
tr(t), and its corresponding sound reception transfer function, Tr(ω). Thus, the entire
system can be characterized by the three LTI systems shown in Fig. 2.9, where
Vi(ω) is the Fourier transform of the driving voltage and V0(ω) is the Fourier
transform of the output voltage. Obviously, for the entire ultrasonic system we have
measured output voltage spectrum, V0(ω), which can then be transformed back
into the time domain to obtain the A-scan time domain flaw response measured by
the system, V0(t). Since there are many elements that need to be measured in the
sound generation and reception transfer functions, it would be advantageous to also
have a simpler approach to characterize the ultrasonic system. Fortunately, this is
possible by combining the sound generation and reception transfer functions and
the input voltage spectrum into a single function, s(ω), that we will call the system
function, i.e.
Then an LTI model of the entire ultrasonic system becomes simply (Fig. 2.10)
This representation is useful, since, as shown in Chap. 9, we can model the acoustic/
elastic transfer function, Tref
a (ω), explicitly in a simple calibration setup and mea-
sure the output voltage spectrum, Vref 0 (ω), in that same setup. Then by
deconvolution we have formally
V ref
0 ðωÞ
sðωÞ ¼ : ð2:25Þ
a ð ωÞ
T ref
able to model the acoustic/elastic transfer function, Ta(ω). In Chap. 12, we will
obtain a very general model of this transfer function that uses a reciprocity
relationship originally developed by Auld [4]. Thus, with this transfer function
expression we will call Eq. (2.24) Auld’s ultrasonic measurement model. We will
also show in Chap. 12 that with some additional assumptions we can decompose the
acoustic/elastic transfer function into a product of simpler terms where the flaw
response appears separately from the beam propagation terms in a form similar to
that originally developed by Bruce Thompson and Tim Gray [5]. This separation of
the beam propagation and flaw response terms is extremely valuable, since in a flaw
measurement one is primarily interested in the type of flaw present and its proper-
ties. In the Thompson-Gray model of the entire ultrasonic system, we can write the
acoustic/elastic transfer function as a product of LTI components in the form:
where P(ω) and M(ω) account for the propagation time delays and attenuation,
respectively, of the ultrasound in going from the sending transducer to the flaw and
back to the receiving transducer, T1(ω) and T2(ω) are plane wave transmission
coefficients associated with the propagation of the sound beam through the fluid-
solid interfaces for the incident and scattered waves. The C1(ω) term is a diffraction
correction that characterizes the sound beam incident on the flaw, and C2(ω) is a
diffraction correction that accounts for the sound beam of the waves scattered from
the flaw and the averaging of those waves (over the face of the receiving transducer)
during sound reception. The A(ω) term is the far field scattering amplitude of the
flaw, a quantity that is discussed extensively in Chap. 10. If we place Eq. (2.26) into
Eq. (2.24) we then have a model of the entire ultrasonic system in the form
Linear time-shift invariant systems are often used to describe some closely related
optical problems of light propagation and transmission so that the books by Gaskill
[6] and Papoulis [1, 7] are particularly good references. However, LTI systems are
also frequently used elsewhere, particularly in the field of electrical engineering so
that many electrical engineering texts can also provide substantial background
information on LTI systems. The Fourier integral also is widely used in many
fields and there are numerous references available. The books by Sneddon [8],
Bracewell [9], and Papoulis [10] contain most of the results needed for our
ultrasonic NDE applications. As mentioned previously, calculating Fourier inte-
grals numerically typically involves the use of Fast Fourier Transform (FFT)
routines. Burrus and Parks [11] describe the basic FFT algorithm and give some
explicit implementations in software.
One of the earliest uses of a combination of LTI system and Fourier concepts
specifically for ultrasonic NDE problems is the work of Frederick and Seydel
[12]. That same theme also appears in applications of “ultrasonic spectral analysis”
considered by a number of authors, as described, for example, in Fitting and Adler
[13]. The most complete representation of an entire ultrasonic system with LTI
system concepts is in the book of Schmerr and Song [2] which showed how all of
the electrical and electro-mechanical elements (pulser/receiver, cabling, transducer
(s)) of an ultrasonic measurement system can be explicitly modeled and also
measured with purely electrical measurements.
2.6 Problems
2.1. Which of the following systems are linear time-shift invariant systems?
d2 d
S½iðtÞ ¼ a 2 þ b þ c iðtÞ
dt dt
ðt
S½iðtÞ ¼ iðtÞ dt
1
S½iðtÞ ¼ afiðtÞg2 þ bfiðtÞg
ð
þ1
ð
þ1
f ∘g ¼ f ðtÞ gðt τÞ dt
1
ð
þ1
¼ f ðt þ τÞ gðtÞ dt,
1
where note that f ∘g 6¼ g∘f . Show that the Fourier transform of the cross
correlation of f and g, F½f ∘g, is related to the Fourier transforms of the
functions themselves through
ð
þ1 ð
þ1
∗
f ðτÞg ðτÞ dτ ¼ FðωÞG∗ ðωÞ dω:
1 1
Note that when f ¼ g this reduces to Rayleigh’s theorem (also called the
energy theorem, Parseval’s theorem, or Plancherel’s theorem):
ð
þ1 ð
þ1
∗
f ðτÞg ðτÞ dτ ¼ FðωÞG∗ ðωÞ dω:
1 1
2.4. Show that the area under the convolution is equal to the product of the areas
under the functions being convolved, i.e. if h ¼ f ∗g then
ð
þ1 ð
þ1 ð
þ1
2.5. Show that the convolution h ¼ f ∗g of two functions f and g has the following
properties
(a) scaling
(b) derivatives
½vðτÞ∗wðτÞ∗gðτÞ ¼ vðτÞ∗½wðτÞ∗gðτÞ
ð
þ1
mk ¼ τk f ðτÞ dτ;
1
FðkÞ ð0Þ
mk ¼
ð2πiÞk
ðk Þ dk FðωÞ
where F ð 0Þ ¼ :
dωk ω¼0
h i
For the Gaussian function f ðtÞ ¼ exp π ðt t0 Þ2 =b determine its centroid,
mean-square abscissa, and variance.
2.7. Consider the LTI system shown in Fig. P2.1. If we give this RC circuit an
input voltage shown in Fig. P2.2, determine by any means (except the use of
convolution) the output voltage, V0(t). Taking the appropriate limit of this
answer, find the impulse response function of this system. Using this impulse
response function and the convolution theorem, show that the output pro-
duced by the input function of Fig. P2.2 agrees with your original answer.
2.8. All ultrasonic systems are band limited due to the limited frequency response
range of real transducers, attenuation effects, etc. Thus, all measured ultra-
sonic responses are to some extent filtered responses. If we represent a
component of our ultrasonic system as a black box LTI system with a band
limited frequency response given by Fig. P2.3, determine
t
t0
f (Hz)
–fmax –fmin fmin fmax
30 2 Linear Systems and the Fourier Transform
ð
1
1
uð t Þ ¼ FðωÞexpðiωtÞ dω;
2π
0
ð
þ1
1
uðtÞ ¼ lim f1 þ sgn ωexpðεjωjÞgexpðiωtÞFðωÞdω:
ε!0 4π
1
2.11. Consider a spring-mass system subjected to a force F(t) as shown in Fig. P2.4.
If the force is given as a unit step function FðtÞ ¼ HðtÞ, where
8
<1 t>0
HðtÞ ¼ 1=2 t¼0 ;
:
0 t<0
what is the resulting unit step response, xH(t), of the system? How is the unit
step response related to the impulse response of the system? Using convolu-
tion, write the response of the system, x(t), to an arbitrary force input F(t) in
terms of the unit step response.
References
where xs is an arbitrary point on the surface S of V whose outward normal is the unit
vector n. The vector quantities f and a are the body force (force/unit volume) and
acceleration of the fluid, respectively, and ρ(x, t) is the fluid density.
If we apply the divergence theorem (see Appendix C) to the first term in Eq. (3.1)
to convert it to a volume integral, we obtain
ð
½∇pðx; tÞ þ f ðx; tÞ ρaðx; tÞdV ðxÞ ¼ 0; ð3:2Þ
V
with ∇ being the vector gradient operator. The integral balance expressed by
Eq. (3.2) must be true for an arbitrary volume V so the integrand itself must vanish
at every point, x, in the fluid leading to the differential equation:
Since all of the waves in NDE applications involve very small displacements and
velocities, we can assume the density is the same as its value, ρ0, in the undisturbed
2
state, i.e. ρ ¼ ρ0 ðxÞ, and the acceleration a ¼ ∂ u=∂t2 where u is the displacement
vector. Under these conditions Eq. (3.3) becomes
2
∇p þ f ¼ ρ0 ∂ u=∂t2 : ð3:4Þ
For economy of notation, we will not continue to show the explicit dependency of
the field variables on x and t in most subsequent equations and use more abbreviated
forms such as given in Eq. (3.4).
To turn the differential equation of motion (Eq. (3.4)) into an equation involving
only a single variable, such as the pressure, we need to relate the pressure, p, to the
motion, u, through the material properties of the fluid. Such a relationship is called a
constitutive equation. For an ideal compressible fluid we have
3.1 Governing Equations for a Fluid 35
p ¼ λ∇ u; ð3:5Þ
i.e. the pressure is proportional to the divergence of the displacement vector, also
known as the dilatation of the fluid. The proportionality constant, λ, is the bulk
modulus of the fluid.
The dilatation term in Eq. (3.5) can be given an explicit physical meaning. To
see this, consider a small element of fluid as shown in Fig. 3.2a. If this element is
given a displacement component in the x direction, ux, only and this displacement is
spatially varying, then the volume change, ΔVx, produced by this x-displacement is,
to first order, given by
or
where ΔV ¼ ΔxΔyΔz is the original volume of our small element. Similarly, the
displacements (uy, uz) in the (y, z) directions, respectively, produce volume changes
(ΔVy, ΔVz) given by
When all three displacements are present simultaneously, then to first order we have
ΔV x þ ΔV y þ ΔV z ¼ ∂ux =∂x þ ∂uy =∂y þ ∂uz =∂z ΔV
so that
∇ u ¼ ΔV x þ ΔV y þ ΔV z =ΔV; ð3:9Þ
i.e. the dilatation is just the relative change in volume of the compressible fluid due
to its motion and Eq.(3.5) is just a statement that the pressure is proportional to
those changes. The minus sign in Eq. (3.5) is present since positive pressures cause
the volume of the fluid to decrease.
36 3 Wave Motion Fundamentals
If we take the divergence of both sides of Eq. (3.4), we can use the constitutive
relationship (Eq. (3.5)) to obtain
2
∇2 p þ f ¼ ρ0 =λ ∂ p=∂t2 ; ð3:10Þ
In later chapters we will use Eq. (3.11) to obtain a variety of solutions of wave
motion problems. To obtain those solutions explicitly, however, it will be necessary
to specify interface/boundary conditions as well as satisfying Eq. (3.11). For
example, consider first the case where there is an interface between two fluid
media (Fig. 3.3) whose densities and wave speeds are (ρ1, cf1) and (ρ2, cf2), respec-
tively. If we remove a small “pillbox” volume ΔV of height ε, as shown in Fig. 3.3,
whose ends are normal to that volume, we have
ð
p1 ns ΔS þ p2 ns ΔS pnε dS
ð ΔSε
where pi is the pressure in medium i (i ¼ 1,2) on the end areas ΔS, ns is the unit
normal to the surface ΔS pointing into medium one and nε is the unit normal vector
acting on the side area ΔSε. As ε ! 0, both the surface and volume integrals go to
zero in Eq. (3.12) so that we find
p1 ¼ p 2 ; ð3:13Þ
i.e. the pressure must be continuous across the interface. If we assume that the two
fluids do not separate from one another along the interface, then it follows that the
normal component of the velocities vi in medium i (i ¼ 1, 2) must also be
continuous:
v1 ns ¼ v2 ns ð3:14aÞ
or
Note that since we are modeling the fluid here as an ideal fluid (i.e. viscosity-free),
the tangential component of the velocities need not be continuous at the interface.
Interface conditions for the normal velocity (Eqs. (3.14a and 3.14b)) can also be
expressed in terms of a normal derivative of the pressure. To see this let f ¼ 0 in
Eq. (3.4) and take the vector dot product of both sides with the unit normal to the
interface, ns. We find
2
∇p ns ¼ ρ0 ∂ ðu ns Þ=∂t2 ð3:15Þ
¼ ρ0 ∂ðv ns Þ=∂t
Since Eq. (3.17) must hold for all times t, Eq. (3.14) is equivalent to the condition
on the pressure p.
Equations (3.13, 3.14a, and 3.14b) or Eqs. (3.13 and 3.18) represent the conditions
that need to be satisfied at a general interface between two fluid media. In the special
38 3 Wave Motion Fundamentals
p1 ¼ 0: ð3:19Þ
Similarly, when medium two becomes a very dense and immobile, v2 ¼ 0 and the
interface is a rigid boundary where
v1 n ¼ v1n ¼ 0 ð3:20Þ
In this section we will derive the basic equations of motion for an elastic solid.
Appendix C provides an introduction to the index notations, integral theorems, and
concepts of stress and strain that will be used throughout this section.
x S
r f(x,t)
V
O
3.2 Governing Equations for an Elastic Solid 39
where t(n) is the traction vector acting on S, f(x,t) is the body force (force/unit
volume) acting in V, ρ(x, t) is the density, and a(x,t) is the acceleration. Similarly, if
we apply the global balance of angular momentum to V with respect to an arbitrary
fixed origin O (Fig. 3.4), we find instead
ð ð ð
r tðnÞ dS þ r f dV ¼ ρ r a dV; ð3:23Þ
S V V
Since the volume of integration in Eq. (3.24) is arbitrary, the integrand must vanish
at every point, x, leading to the differential equation
If we write the tractions in terms of the stresses, i.e. set tk ¼ τkl el , and express the
body force, f, and the acceleration, a, in terms of their Cartesian components as
f ¼ f l el
a ¼ al e l
Similarly, if we express the traction vector, t(n), in terms of the tractions, tk, and
apply Gauss’ theorem to the surface integral in Eq. (3.23), we obtain
ð
∂ r tk =∂xk þ r f ρr a dV ¼ 0 ð3:27Þ
V
which, by the chain rule and the relation ∂r=∂xk ¼ ek can be rewritten as
ð
½ek tk þ r ð∂tk =∂xk þ f ρaÞ dV ¼ 0: ð3:28Þ
V
40 3 Wave Motion Fundamentals
which, since the volume V is arbitrary, implies that at every point, x, we have
ek tk ¼ 0: ð3:30Þ
Writing the tractions tk ¼ τkl el in terms of the stresses, and vector cross product,
ek el as ek el ¼ εikl el in terms of the alternating tensor, εikl, gives εikl τkl el ¼ 0
or, equivalently
1
u ¼ A þ Bij eij þ Cijkl eij ekl ð3:34Þ
2
where, if the reference state is taken to be an unstressed and unstrained state we can
set A ¼ 0, and Bij ¼ 0. Then, from Eq. (3.33) we find the stress–strain constitutive
relationship
in terms of the fourth order tensor of elastic constants, Cijkl. Because of the
symmetry of the stress tensor and the existence of the strain energy density
function, the elastic constants tensors can be shown to have the symmetries
which reduces the number of independent elastic constants to 21 in the most general
case. For all of the applications considered in this book the material will be assumed
to be isotropic, i.e. at a given point in the materials, the stress–strain relationship
will be independent of the orientation of the coordinate axes chosen. In this
isotropic case the Cijkl tensor can be written as
Cijkl ¼ λδij δkl þ μ δik δjl þ δil δjk ð3:37Þ
in terms of the two Lame’ constants, λ, and μ. Then the stress–strain relations also
reduce to
Note that some authors prefer to write these relations instead in terms of Young’s
modulus, E, and Poisson’s ratio, ν, which are related to λ and μ through
Assuming the displacement and velocity gradients are all small (see Appendix C),
we can write the strain, velocity, acceleration, and density, as
1
eij ¼ ∂ui =∂xj þ ∂uj =∂xi
2
vi ¼ ∂ui =∂t ð3:40a dÞ
2
ai ¼ ∂vi =∂t ¼ ∂ ui =∂t 2
ρ ¼ ρ0 ;
respectively. Using these relations and placing the stress strain constitutive equa-
tions (Eq. (3.38)) into the equations of motion, Eq. (3.26), gives those equations
explicitly in terms of the displacement components as
42 3 Wave Motion Fundamentals
2 2 2
μ ∂ ui =∂xj ∂xj þ ðλ þ μÞ ∂ uj =∂xj ∂xi þ f i ¼ ρ0 ∂ ui =∂t2 ; ð3:41Þ
assuming λ and μ are both independent of x, i.e. the body is homogeneous. Equation
(3.41), which is called Navier’s equations, can also be written in vector notation as
2
μ∇2 u þ ðλ þ μÞ∇ð∇ uÞ þ f ¼ ρ0 ∂ u=∂t2 : ð3:42Þ
For a general anisotropic, inhomogeneous body, the equations of motion can also be
expressed in terms of the displacement as
2
∂ Cijkl ∂uk =∂xl =∂xj þ f i ¼ ρ0 ∂ ui =∂t2 ; ð3:43Þ
where Cijkl ¼ Cijkl ðxÞ and ρ0 ¼ ρ0 ðxÞ. Although we will consider only isotropic,
piece-wise homogeneous materials in all our later discussions, in some cases, for
notational convenience, it may be desirable to use Navier’s equations as given in
Eq. (3.43) in place of the more explicit form of Eq. (3.41).
If the surface of an elastic material is in (welded) contact with another solid that is
immobile and significantly stiffer than the elastic solid, at that surface we can
assume the displacements (and velocities) vanish, i.e.
u¼v¼0
or
uk ¼ vk ¼ 0: ð3:44Þ
Since most structural materials are themselves very stiff, unlike the fluid case this
boundary condition is generally not encountered in NDE problems.
3.2 Governing Equations for an Elastic Solid 43
For a free surface of an elastic solid the traction vector must vanish so that we have
t(n) ¼ 0. These conditions can be written in terms of the traction vector components
normal and tangential to the surface at any point (see Fig. 3.5) as
or, in terms of the stresses with respect to a fixed set of Cartesian axes as
τkl nk ¼ 0: ð3:46Þ
In this case it is assumed that two elastic solids, which we will call media one and
two, respectively, are in intimate (welded) contact so that both the tractions and
displacements are continuous. Thus
ðτkl nk Þ1 ¼ ðτkl nk Þ2
ð3:47Þ
ð u k Þ 1 ¼ ð uk Þ 2
ðτnn Þ1 ¼ ðτnn Þ2
ðτns Þ1 ¼ ðτns Þ2
ðτnt Þ1 ¼ ðτnt Þ2
ð3:48Þ
ð un Þ 1 ¼ ð un Þ 2
ð u s Þ 1 ¼ ð us Þ 2
ð ut Þ 1 ¼ ð ut Þ 2 :
44 3 Wave Motion Fundamentals
If a thin fluid layer separates two elastic solids where the layer thickness can be
neglected, then at the interface the normal component of both the traction and
displacement (or velocity) vectors must be continuous and the tangential (shear)
components of the traction vector must vanish:
ðτnn Þ1 ¼ ðτnn Þ2
ðτns Þ1 ¼ ðτnt Þ1 ¼ 0
ð3:49Þ
ðτns Þ2 ¼ ðτnt Þ2 ¼ 0
ð u n Þ 1 ¼ ð un Þ 2 :
If we have a boundary between a fluid and solid (where the fluid is medium one and
the solid medium two) the interface conditions are very similar to that of the smooth
interface since here also the normal stress and displacement (or velocity) must be
continuous, and the shear stresses must vanish. In terms of the pressure in the fluid
and the displacement in the solid, then these conditions become
p1 ¼ ðτnn Þ2
ð τ Þ
ðð ns 2 ¼ ð τnt Þ2 ¼ 0
ð3:50Þ
1=ρ1 ∂p1 =∂n ∂t∂t ¼ ðun Þ2 :
Unlike the governing equations for the pressure field in a fluid, Navier’s equations
(Eqs. (3.41) or (3.42)) are in general not wave equations for the displacement.
However, Navier’s equations do in fact implicitly describe the behavior associated
with two distinct wave equations. To see this, we decompose the displacement
vector into a scalar potential, ϕ, and a vector potential, ψ, through the Helmholtz
decomposition theorem [2] given by:
3.2 Governing Equations for an Elastic Solid 45
u ¼ ∇ϕ þ ∇ ψ: ð3:51Þ
Placing this decomposition into Navier’s equations, setting the body force f ¼ 0
(this is done merely for convenience and is not essential), and using the fact that
gradient of the curl of a vector is zero, we find
h i h i
2 2
∇ ðλ þ 2μÞ∇2 ϕ ρ0 ∂ ϕ=∂t2 þ ∇ μ∇2 ψ ρ0 ∂ ψ=∂t2 ¼ 0; ð3:52Þ
which can be satisfied if the potentials ϕ and ψ satisfy the homogeneous wave
equations
2
∇2 ϕ 1=c2p ∂ ϕ=∂t2 ¼ 0
2 ð3:53Þ
∇2 ψ 1=c2s ∂ ψ=∂t2 ¼ 0
having the two distinct wave speeds cp cl c1 , and cs ct c2 , given in terms of the
Lame’ constants as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cp ¼ pffiffiffiffiffiffiffiffiffi
ðλ þffi 2μÞ=ρ0
ð3:54Þ
cs ¼ μ=ρ0 :
Thus, in an elastic solid, disturbances associated with the scalar potential ϕ, travel
with the wave speed cp. These disturbances are called P-waves, compressional
waves, primary waves, dilatational waves, longitudinal (L ) waves, or irrotational
waves. In contrast, disturbances associated with the vector potential, ψ travel with
the wave speed cs, and are called S-waves, shear waves, secondary waves, tangen-
tial (T ) waves, distortional waves, equivoluminal waves, or rotational waves. Since
both the P- and S-waves represent disturbances propagating in the “bulk” of the
elastic solid, these waves are also called P- and S- type bulk waves. Note that if we
consider the ratio of these two wave speeds, κ ¼ cp =cs , this ratio can be written as a
function of Poisson’s ratio only:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
κ¼ 2ð1 νÞ=ð1 2νÞ: ð3:55Þ
Table 3.1 lists the wave speeds of both P- and S-waves for some common materials.
In most structural materials we see that κ ffi 2, so that P-waves travel roughly twice
as fast as S-waves.
These bulk waves are the ones most frequently used in NDE testing so they will
play a major role in our modeling efforts. However, other waves, such as surface
(Rayleigh) and plate (SH and Lamb) waves can exist in an elastic solid and be used
for NDE testing. In Chap. 7 we will describe some of the properties of surface and
plate waves.
46 3 Wave Motion Fundamentals
Table 3.1 Wave speeds, density, and specific acoustic impedance (for longitudinal waves) of
some common materials
cp cs ρ zp
Material (m/s 103 ) (m/s 103 ) kg=m3 103 kg=m2 s 106
Air 0.33 – 0.0012 0.0004
Aluminum 6.42 3.04 2.70 17.33
Beryllium 12.89 8.90 1.87 24.1
Brass 4.70 2.10 8.64 40.6
Cadmium 2.80 1.50 8.60 24.0
Copper 5.01 2.27 8.93 44.6
Fused quartz 5.96 3.76 2.20 13.1
Glass 5.64 3.28 2.24 13.1
Gold 3.24 1.20 19.7 63.8
Inconel 5.70 3.00 8.28 47.2
Iron 5.90 3.20 7.69 46.4
Iron (cast) 4.60 2.60 7.22 33.2
Lucite 2.70 1.10 1.15 3.1
Magnesium 5.77 3.05 10.0 5.3
Mercury 1.45 – 13.53 19.6
Molybdenum 6.30 3.40 10.0 63.1
Monel 5.40 2.70 8.82 47.6
Nickel 5.60 3.00 8.84 49.5
Nylon 2.60 1.10 1.12 2.9
Oil (SAE30) 1.70 – 0.88 1.5
Platinum 3.26 1.73 21.4 69.8
Polyethylene 1.95 0.54 0.92 1.79
Polystyrene 2.40 1.15 1.05 2.52
Polyurethane 1.90 – 1.00 1.90
Steel, mild 5.90 3.20 7.90 46.0
Steel, stainless 5.80 3.10 7.83 45.4
Teflon 1.39 – 2.14 2.97
Tin 3.30 1.70 7.3 24.2
Titanium 6.10 3.10 4.48 27.3
Tungsten 5.20 2.90 19.4 101.0
Uranium 3.40 2.00 18.5 63.0
Water 1.48 – 1.00 1.48
Zinc 4.20 2.40 7.0 29.6
If we set the body force term f ¼ 0 again in Navier’s equations (Eq. (3.42)) and take
the divergence (∇) of this entire equation, then we find
3.2 Governing Equations for an Elastic Solid 47
2
∇2 Δ 1=c2p ∂ Δ=∂t2 ¼ 0; ð3:56Þ
where Δ ¼ ∂uk =∂xk is the dilatation of the solid. Thus, the dilatation travels with
the P-wave speed and is the reason that this type of bulk wave is also called a
dilatational wave. In fact, as the shear modulus goes to zero, the dilatational wave
pffiffiffiffiffiffiffi
speed simply becomes cp ¼ λ=ρ, where λ is the bulk modulus, so we see that in
this limit we recover the same wave equation for these disturbances as found for a
compressible fluid.
If, instead, we take the curl (∇) of Navier’s equation with f ¼ 0, we find
2
∇2 ω 1=c2s ∂ ω=∂t2 ¼ 0; ð3:57Þ
where ω ¼ ð∇ uÞ=2 is the local rotation of the elastic solid (see Appendix C).
Since this rotation travels with the same wave speed as S-waves, these bulk waves
are also called rotational waves, as mentioned previously. Since S-waves require
the existence of a non-zero shear modulus, we see that there is no comparable
disturbance of this type in the motion of an ideal compressible fluid.
Although index notation is convenient for expressing and manipulating the equa-
tions governing an elastic solid, in solving some problems we will find it useful to
resort to a more direct notation. In this section we will describe the governing
equations in Cartesian coordinates in terms of both the displacements directly
(displacement formulation) or in terms of the potentials (potential formulation)
since both formulations are useful in particular applications.
If we express the three Cartesian coordinate axes (x1, x2, x3) as (x, y, z) and the three
displacement components (u1, u2, u3) with respect to those axes as (u, v, w), then
Navier’s equations become
2
μ∇2 u þ ðλ þ μÞ∂Δ=∂x þ f x ¼ ρ ∂ u=∂t2
2
μ∇2 v þ ðλ þ μÞ∂Δ=∂y þ f y ¼ ρ ∂ v=∂t2 ð3:58Þ
2
μ∇2 w þ ðλ þ μÞ∂Δ=∂z þ f z ¼ ρ ∂ w=∂t2 ;
where ( fx, fy, fz) are the components of the body force along the (x, y, z) axes,
respectively, and ∇2 and Δ are given by
48 3 Wave Motion Fundamentals
2 2 2
∇2 ¼ ∂ =∂x2 þ ∂ =∂y2 þ ∂ =∂z2
and
τxx ¼ λ Δ þ 2μ∂u=∂x
τyy ¼ λ Δ þ 2μ∂v=∂y
τzz ¼ λΔ þ 2μ∂w=∂z
ð3:60Þ
τxy ¼ τyx ¼ μð∂u=∂y þ ∂v=∂xÞ
τxz ¼ τzx ¼ μð∂u=∂z þ ∂w=∂xÞ
τzy ¼ τyz ¼ μð∂w=∂y þ ∂v=∂zÞ:
Here, we let the vector potential components along a set of Cartesian axes be given
by ðψ 1 ; ψ 2 ; ψ 3 Þ ¼ ψ x ; ψ y ; ψ z , respectively, In this case the Helmholtz decompo-
sition can be written as
3.2 Governing Equations for an Elastic Solid 49
In the general three-dimensional case just described, we see that the potential
formulation can become algebraically quite complex, particularly where stresses
are involved. However, in some applications we can make the assumption of plane
strain conditions, i.e. u ¼ uðx; y; tÞ, v ¼ vðx; y; tÞ, w ¼ 0. In this case we can obtain
the solution of Navier’s equations with potentials given by
u ¼ ∂ϕ=∂x þ ∂ψ=∂y
ð3:65Þ
v ¼ ∂ϕ=∂y ∂ψ=∂x
2 2
where now ∇2 ¼ ∂ =∂x2 þ ∂ =∂y2 is the two-dimensional Laplacian operator.
Similarly, the stresses can be written as
h
i
2 2
τxx ¼ μ κ2 ∇2 ϕ þ 2 ∂ ψ=∂x∂y ∂ ϕ=∂y2
h
i
2 2
τyy ¼ μ κ2 ∇2 ϕ 2 ∂ ψ=∂x∂y þ ∂ ϕ=∂x2
h i ð3:67Þ
2 2 2
τxy ¼ μ 2 ∂ ϕ=∂x∂y þ ∂ ψ=∂y2 ∂ ψ=∂x2
τzz ¼ ν τxx þ τyy
τxz ¼ τyz ¼ 0
Another special case that will arise occasionally in some applications is that of anti-
plane strain where we can assume that u ¼ v ¼ 0 and w ¼ wðx; y; tÞ. In this case it is
not necessary to introduce potentials since, directly from Navier’s equations, the
displacement w satisfies the wave equation itself, i.e.
2
∇2 w 1=c2s ∂ w=∂t2 ¼ f z =μ; ð3:68Þ
2 2
where again here ∇2 ¼ ∂ =∂x2 þ ∂ =∂y2 . In this case the only non-zero stresses
associated with the (x, y, z) axes are
τxz ¼ μ∂w=∂x
ð3:69Þ
τyz ¼ μ∂w=∂y:
3.4 Problems 51
3.4 Problems
x
z
V (x,t ) V (x,t ) + dV
dx
neutral
surface
3.4. Consider an element of a beam whose cross sectional area is A and the moment
of inertia about the z-axis is I (Fig. P3.3). If M is the bending moment in the
beam, V is the shear force, and w is the distributed force/ unit length acting on
the beam, using F ¼ ma determine the equation of motion for the vertical
deflection, y, using Bernoulli-Euler bending theory, i.e. assume plane sections
2
remain plane so that from the moment-curvature relation M ¼ E I ∂ y=∂x2 ,
where E is Young’s modulus. Also assume we can neglect rotary inertia, i.e.
X
Mi ¼ 0 (net moment equals zero).
3.5. If the body force f and the initial values of the velocity, u_ , and the displace-
ment, u, are expressed in terms of Helmholtz potentials via
f ¼ ∇Φ þ ∇ Ψ ∇Ψ¼0
u_ ðx; 0Þ ¼ ∇A þ ∇ B ∇ B ¼ 0
uðx; 0Þ ¼ ∇C þ ∇ D ∇ D ¼ 0
then Lame’s theorem says that there exists potentials ϕ and ψ for the displace-
ment u(x, t) with the following four properties:
uðx; tÞ ¼ ∇ϕ þ ∇ ψ
∇ψ¼0
€ ¼ Φ=ρ þ c2 ∇2 ϕ
ϕ p
€ ¼ Ψ=ρ þ c2s ∇2 ψ:
ψ
References 53
Show that the following representations satisfy all of the four properties listed
above:
ðt n o
ϕðx; tÞ ¼ 1=ρ ðt τÞ Φðx; τÞ þ ρc2p ∇ uðx; τÞ dτ þ t A þ C
0
ðt
ψðx; tÞ ¼ 1=ρ ðt τÞ Ψðx; τÞ ρc2s ∇ uðx; τÞ dτ þ t B þ D:
0
3.6. Prove the work-kinetic energy theorem for an elastic solid, i.e.
dW dU dK
¼ þ ;
dt dt dt
where dW/dt is the rate of work being done by the traction t and body force f:
ð ð
dW
¼ ti vi dS þ f i vi dV:
dt
S V
References
1. J.A. Hudson, The Excitation and Propagation of Elastic Waves (Cambridge University Press,
New York, 1980)
2. J.D. Achenbach, A.K. Gautesen, H. McMaken, Ray Methods for Waves in Elastic Solids
(Pitman, Boston, 1982)
3. P.M. Morse, K.V. Ingard, Theoretical Acoustics (McGraw-Hill, New York, 1968)
4. A.D. Pierce, Acoustics (McGraw-Hill, New York, 1968)
5. K.F. Graff, Wave Motion in Elastic Solids (Dover, New York, 1991)
6. J.D. Achenbach, Wave Propagation in Elastic Solids (American Elsevier, New York, 1973)
7. A.C. Eringen, E.S. Suhubi, Elastodynamics, vols. 1 and 2 (Academic, New York, 1975)
54 3 Wave Motion Fundamentals
8. J. Miklowitz, The Theory of Elastic Waves and Waveguides (North Holland, Amsterdam,
1978)
9. A.H. Harker, Elastic Waves in Solids (Adam Hilger, Philadelphia, 1988)
10. K.J. Langenberg, R. Marklein, K. Mayer, Ultrasonic Nondestructive Testing of Materials—
Theoretical Foundations (CRC, Boca Raton, 2012)
11. K. Aki, P.G. Richards, Quantitative Seismology, vols. 1 and 2 (W.H. Freeman, San Francisco,
1980)
12. H. Kolsky, Stress Waves in Solids (Dover, New York, 1963)
13. R.J. Wasley, Stress Wave Propagation in Solids (Marcel Dekker, New York, 1973)
14. R.B. Lindsay, Mechanical Radiation (McGraw-Hill, New York, 1960)
Chapter 4
Propagation of Bulk Waves
This Chapter describes the propagation characteristics of bulk waves in fluid and
solid media. Plane waves, spherical waves, and Gaussian beams will be considered.
In Chap. 8 these types of waves will be used to describe the sound beams generated
by ultrasonic transducers. The spherical waves will also be useful in describing the
waves scattered by flaws considered in Chap. 10.
The general 3-D wave equation for the pressure, p, in a fluid was given by
Eq. (3.11). If we assume that the disturbances traveling in the fluid vary only in
one spatial dimension, x, i.e. p ¼ pðx; tÞ, then the wave equation reduces to
2 2
∂ p=∂x2 1=c2 ∂ p=∂t2 ¼ 0; ð4:1Þ
where f and g are arbitrary functions and c ¼ cf is the fluid wave speed. The solution
corresponding to f represents a wave traveling in the +x direction while the solution
corresponding to g represents a wave traveling in the x direction (Fig. 4.1). Both
of these solutions are one-dimensional plane wave solutions since we see that at a
fixed time t0 the pressure is constant on the planes x ¼ ct0 and x ¼ ct0 , respec-
tively, for f and g.
x x
c t1 – c t1
If we take the Fourier transforms of the solutions f ðt x=cÞ and gðt þ x=cÞ in
Eq. (4.2), we find
ð
þ1
f ðt x=cÞexpðiωtÞdt ¼ FðωÞexpðiωx=cÞ
1
ð4:3Þ
ð
þ1
where
ð
þ1
F ð ωÞ ¼ f ðtÞexpðiωtÞdt
1
ð4:4Þ
ð
þ1
GðωÞ ¼ gðtÞexpðiωtÞdt
1
are the Fourier transforms of f(t) and g(t). From the inverse Fourier transforms, then
ð
þ1
1
f ðt x=cÞ ¼ FðωÞexp½iωðx=c tÞdω
2π
1
ð4:5Þ
ð
þ1
1
gðt þ x=cÞ ¼ GðωÞexp½iωðx=c tÞdω:
2π
1
Equations (4.3) and (4.5) show that the effects of propagation of a plane wave
through a distance d in the plus or minus x-directions will introduce complex
exponentials in their Fourier transforms given by expðiωd Þ, respectively.
4.1 Plane Waves in a Fluid 57
Equation (4.5) shows that an arbitrary plane wave traveling in one-dimension in the
+x-direction can be considered to be the superposition (over all frequencies) of a
harmonic wave of the form
where A ¼ FðωÞ=2π is its amplitude and ωðx=c tÞ is its phase. Thus, in many
problems it is only necessary to consider the response to harmonic waves and then
obtain the solution to an arbitrary pulse by performing the inverse Fourier trans-
form. Alternate forms for the harmonic wave of Eq. (4.6) are
p ¼ Aexp½ikðx ctÞ
or ð4:7Þ
p ¼ Aexp½2πiðx ctÞ=λ
where k ¼ ω=c is called the wave number and λ ¼ 2π=k is the wave length. Since ω
is related to the frequency, f, in Hertz through ω ¼ 2πf , we see that f and λ are
related directly through the wave speed by:
f λ ¼ c: ð4:8Þ
Figure 4.2 shows a chart of frequency and corresponding wave length for both
pressure waves in water (c ¼ 1480 m/s) and P-waves in a typical structural
material such as steel (c ¼ 5900 m/s). Most NDE testing is done with pulses
containing significant frequency components ranging roughly from 1 to 20 mega-
Hertz (MHz). Thus, as Fig. 4.2 shows, the corresponding wavelengths in steel
range from about 0.25 in. to 0.01 in., respectively. The reason why the NDE
testing range is limited to approximately 1 MHz at the low frequency end is
because it is difficult to detect flaws that are significantly smaller than a wave-
length. Thus, detecting small flaws on the order of, say, 1 mm in length, will be
difficult at frequencies much less than 1 MHz. In contrast, the NDE testing range
is limited at the high frequency range to approximately 20 MHz for most struc-
tural materials since, although the wavelengths are very small, material attenua-
tion due to the scattering of the waves by the grains of the material is large at such
high frequencies, limiting the penetration of the ultrasonic waves. In other
materials, such as ceramics, for example, material attenuation due to grain
scattering is insignificant at these frequencies, and testing frequencies into the
gigaHertz (GHz) region can often be considered.
58 4 Propagation of Bulk Waves
ultrasonic
common .584⫻10–2
107 2.3⫻10–2
ultrasonic NDE
ultrasonic
The generalization of our results for one-dimensional plane waves to three dimen-
sions is directly possible if we recognize that the equation x n ¼ D defines a fixed
plane in 3-D space (Fig. 4.3) at a distance D from the origin in the direction n,
where n is the unit normal to the plane. It follows that x n ¼ ct defines a plane
moving in the direction n with wave speed c, where the plane passes through the
origin at time t ¼ 0.
4.2 Plane Waves in an Elastic Solid 59
p ¼ f ðx n ctÞ
or ð4:9Þ
p ¼ f ðt x n=cÞ
is indeed a solution of the 3-D wave equation and thus represents a general plane
wave traveling in space in the n direction. Similarly, for 3-D harmonic plane waves
we have
which is the ordinary homogeneous 1-D wave equation for P-waves, with solutions
given by u ¼ f ðt x=c1 Þ for disturbances traveling in the +x-direction. For
this P-wave, the motion is entirely in the direction of propagation, as shown in
Fig. 4.4a.
If, instead we assume the displacement vector has components only in the y- or
z-directions respectively, i.e. u ¼ vðx; tÞey or u ¼ wðx; tÞez we find that Navier’s
equation reduce to
2 2
∂ v=∂x2 1=c22 ∂ v=∂t2 ¼ 0
or ð4:12Þ
2 2
∂ w=∂x2 1=c22 ∂ w=∂t2 ¼ 0
Navier’s equations (Eq. (3.42)) can also be written in terms of the two wave speeds,
c1 and c2, as
2
c22 ∇2 u þ c21 c22 ∇ð∇ uÞ ¼ ∂ u=∂t2 ð4:13Þ
∇ 2 u ¼ ∇ð ∇ u Þ ∇ ð ∇ u Þ ð4:14Þ
is used, becomes
2
c21 ∇ð∇ uÞ c22 ∇ ð∇ uÞ ¼ ∂ u=∂t2 : ð4:15Þ
0 00
If we now let u ¼ uðt n x=cÞ and define u ¼ duðαÞ=dα and u ¼ d2 uðαÞ=dα2 ,
where α ¼ t n x=c, then it follows that
0
∂u=∂t ¼ u
0 ð4:16Þ
∂u=∂xk ¼ ðnk =cÞu :
There are two 3-D plane wave solutions to this equation. The first of these is
00
u ¼ An f ðt x n=cÞ; ð4:18Þ
which will satisfy Eq.(4.17) for an arbitrary function f if c ¼ c1 . The second solution
is of the form
00
u ¼ Aðn dÞf ðt x n=cÞ; ð4:19Þ
where d is an arbitrary unit vector in the plane whose normal is n (Fig. 4.5). Placing
this expression into Eq. (4.17) shows that in this case we must have c ¼ c2 . Since f is
an arbitrary function in both Eqs. (4.18) and (4.19), these equations can be inte-
grated twice to yield identical forms for the displacement vector u itself, i.e.
and
u ¼ Anexp½ik1 ðx n c1 tÞ
and ð4:22Þ
u ¼ Aðn dÞexp½ik2 ðx n c2 tÞ;
where k1 ¼ ω=c1 and k2 ¼ ω=c2 are the wave numbers for P- and S-waves,
respectively. Alternatively, we could have used the potentials, ϕ and ψ, to describe
these harmonic plane waves. For P-waves, we would have
where t is a unit vector in the plane whose normal is n (Fig. 4.5). In Eqs. (4.23) and
(4.24) Φ and Ψ are amplitudes of the potentials, which are not physical quantities.
However, we can easily relate these potential amplitudes to the displacements and
stress amplitudes. To see this, first consider the case of P-waves:
where Un is the displacement amplitude of the wave in the n direction. Thus, Un and
Φ are related through
U n ¼ ik1 Φ ð4:26Þ
and the in-plane displacement amplitudes Ut and Us in the t and s directions (see
Fig. 4.5), respectively, are zero. For this P-wave, the normal stress along the
n direction is given by
T nn ¼ ρω2 Φ
¼ ik1 ρc21 U n
¼ ρc1 V n : ð4:28Þ
4.3 Spherical Waves in a Fluid 63
U s ¼ ik2 Ψ ð4:30Þ
and the other displacement amplitudes, Un and Ut, are zero. A comparison of
Eqs. (4.29) and (4.21) shows that we have taken d ¼ t in the present case. The
only non-zero stress associated with this wave in the (n, t, s) coordinates is the shear
stress component, τns, given by
so that if we let Tns be the amplitude of this component we see that this stress
amplitude is related to the potential amplitude, Ψ, the displacement amplitude, Us,
and the velocity amplitude V s ¼ iωUs , by
T ns ¼ ρω2 Ψ
¼ ik2 ρc22 U s ð4:32Þ
¼ ρc2 V s :
Consider the equation of motion for a fluid again where a point source of pressure
acts at a point, y, causing a spherical wave to propagate (Fig. 4.6). In this case, we
model the point source as a body force term given by a time dependent three
dimensional Dirac delta function f ¼ f ðtÞδðx yÞ so that the equation of motion
for this source is
2
∇2 p 1=c2 ∂ p=∂t2 ¼ f ðtÞδðx yÞ: ð4:33Þ
64 4 Propagation of Bulk Waves
The three dimensional delta function, δðx yÞ like its one dimensional counterpart
(see Appendix B) can be defined in terms of its sampling properties as:
δðx yÞ ¼ 0 x 6¼ y
8
ð < h ð xÞ x inside V
hðyÞδðx yÞdV ðyÞ ¼ hðxÞ=2 x on S ; ð4:34Þ
:
V 0 x outside V
where the volume V is enclosed by a (smooth) surface S. Since we expect this point
source to produce a pressure disturbance that is independent of the angles (θ, ϕ) in a
spherical coordinate system (r, θ, ϕ) whose origin is at y (Fig. 4.6), for x 6¼ y the
equations of motion in spherical coordinates become
2
1=r2 ∂=∂r r2 ∂=∂r 1=c2 ∂ p=∂t2 ¼ 0: ð4:35Þ
If we let p ¼ Pðr; tÞ=r then P(r, t) satisfies simply the one-dimensional wave
equation in r, i.e.
2 2
∂ P=∂r2 1=c2 ∂ P=∂t2 ¼ 0 ð4:36Þ
p ¼ P1 ðt r=cÞ=r: ð4:38Þ
To complete the solution, of course, we must find the unknown function P1. To do
this first write the wave equation for p in integral form as
4.3 Spherical Waves in a Fluid 65
x
Ve y
e
ð ð ð
2
∂p=∂ndS 1=c2 ∂ p=∂t2 dV ¼ f ðtÞδðx yÞdV ðxÞ ð4:39Þ
S V V
for an arbitrary volume V whose surface is S, which follows directly from Gauss’
theorem, and then apply it to a small spherical volume of radius Vε and surface Sε
whose center is at point y (Fig. 4.7). From Eq. (4.39) and the sampling properties of
the delta function we have
ð ð
2
∂p=∂rdS 1=c2 ∂ p=∂t2 dV ¼ f ðtÞ: ð4:40Þ
Sε Vε
Since p is a constant (in space) on Sε, the surface integral in Eq. (4.40) can be
performed, giving
h 0
i
4πε2 P1 ðt ε=cÞ=ε2 P1 ðt ε=cÞ=cε ð4:41Þ
0
(where P1 ¼ dPðuÞ=du) which goes to the limit 4πP1 ðtÞ as ε ! 0. The remaining
2
volume integral in Eq. (4.40) vanishes as ε ! 0 (since ∂ p=∂t2 1=r and the
volume element is dV ¼ r 2 dr sin θdθdϕ) leaving, finally
which, when placed back into Eq. (4.38) gives the explicit solution for the pressure
from the point source as
p ¼ f ðt r=cÞ=4πr: ð4:43Þ
In the special case when the time function, f, itself is a delta function, this solution is
for an impulsive point source and is called a fundamental solution, g(r, t), given by
As we will see, this fundamental solution serves as an important building block for
constructing many other more general solutions for wave propagation in a fluid.
Taking the Fourier transform on time of g, we find
Gðr; ωÞ ¼ expðiωr=cÞ=4πr
ð4:45Þ
¼ expðikr Þ=4πr:
Note that G is also the fundamental solution for a point source having har-
monic time dependency, i.e. f ¼ expðiωtÞδðx yÞ, where if we set p ¼ G
ðr; ωÞexp ðiωtÞ, from the wave equation it follows that G satisfies the Helm-
holtz equation with a point source:
Equation (4.45) represents spherical waves (of harmonic time dependency expðiωtÞ)
traveling from the source at y to an arbitrary point x in the fluid, where
r ¼ jx yj ¼ ½ðxk yk Þðxk yk Þ1=2 . As in the case of harmonic plane waves we
see that propagation through a distance r causes a phase term to appear in the complex
exponential of the form ωr/c, or equivalently, kr. However, unlike the plane wave
case, propagation of the spherical wave also causes the amplitude to decrease like 1/r.
Other types of waves will propagate with different amplitude variations but will
contain the same type of phase terms as found here for plane waves and spherical
waves. Thus, in an LTI model of an ultrasonic system, we will take the propagation
term to represent only the phase changes and account for the amplitude changes in the
diffraction terms (see the summary box near the end of this chapter).
When considering fundamental solutions for elastic media in the next section, we
will see that spherical waves also appear in that case but in a more complicated
fashion. To aid in the interpretation of the elastic wave fundamental solutions, an
integral representation of the spherical harmonic wave solution, G, in terms of its
spatial Fourier components will prove to be particularly useful. To obtain this
representation, we note that since G satisfies the Helmholtz equation (Eq. (4.46))
for a point source, if we write that equation in Cartesian coordinates, where x ¼ xm em
and y ¼ ym em , and apply the three-dimensional spatial Fourier transform defined as
þ1 ð þ1
ð þ1 ð
G¼ Gexp½iðp1 x1 þ p2 x2 þ p3 x3 Þdx1 dx2 dx3 ð4:47Þ
1 1 1
4.3 Spherical Waves in a Fluid 67
we find
ð þ1
þ1 ð þ1
ð
ðpm pm ÞG þ k2 G ¼ δðx1 y1 Þδðx2 y2 Þδðx3 y3 Þexpðipm xm Þdx1 dx2 dx3
1 1 1
ð4:48Þ
so that G is given by
G ¼ expðipm ym Þ= pn pn k2 : ð4:50Þ
ð þ1
þ1 ð þ1
ð
1
Gðr; ωÞ ¼ 3
exp½ipm ðym xm Þdp1 dp2 dp3 = pn pn k2 ; ð4:51Þ
ð2π Þ
1 1 1
which shows that a spherical harmonic wave can be expressed in spatial Fourier
transform form as
ð þ1
þ1 ð þ1
ð
expðikr Þ 1
¼ exp½ipm ðym xm Þdp1 dp2 dp3 = pn pn k2 : ð4:52Þ
4πr ð2π Þ3
1 1 1
ð þ1
þ1 ð
expðikr Þ i
¼ 2 expfi½ p1 ðx1 y1 Þ þ p2 ðx2 y2 Þ þ γ jx3 y3 jgdp1 dp2 =γ;
4πr 8π
1 1
ð4:53Þ
where
( 1=2
k2 p21 p22 f or p21 þ p22 < k2
γ¼ 1=2 ; ð4:54Þ
i p21 þ p22 k2 f or p21 þ p22 > k2
68 4 Propagation of Bulk Waves
which is called Weyl’s integral. If we further change the ( p1, p2) integration vari-
ables to polar coordinates ( pr, pθ) given by
p1 ¼ pr cos ðpθ Þ
ð4:55Þ
p2 ¼ pr sin ðpθ Þ
ð
1
expðikr Þ i
¼ J 0 ðpr ρÞpr expðiγ jx3 y3 jÞdpr =γ; ð4:56Þ
4πr 4π
0
The fundamental solution G(x, y, ω) in Eq. (4.45) is a two point function since it
depends on both the location point Q(y) of the source and the point P(x) in the
fluid where the pressure field is being evaluated (Fig. 4.8). In later applications,
we will need to determine the behavior of G and its derivatives when the distance
from a given origin O to Q is much larger than the distance from O to P. In this
case, we have jyj >> jxj (Fig. 4.8) so that we can expand the distance r in G as
follows
r ¼ ½ðx yÞ ðx yÞ1=2
h i1=2
¼ jxj2 þ jyj2 2x y
h i1=2
¼ jyj 1 þ jxj2 =jyj2 2x y=jyj2
ð4:57Þ
ffi jyj½1 x ^y =jyj;
where ^y ¼ ^y m em is the unit vector from O to Q along y (Fig. 4.8). If we only keep
the first order approximation for r in the amplitude term of G and the first two terms
in the phase of G, we obtain, approximately
Keeping two terms in the approximation for the phase is necessary, since in order to
neglect the remaining terms in a phase expression, not only must the neglected
terms be less than the terms retained, the neglected terms themselves must also be
much less than 2π to be negligible. Thus, the phase term is much more sensitive to
such approximations. Similarly, when the derivative of G is taken with respect to
the coordinates of x, we have
The fundamental solution for an elastic medium is the solution of Navier’s equa-
tions for a delta function body force term. However, since the body force is now a
vector, we can consider point sources where the body force can be oriented along
any one of the three coordinate axes (Fig. 4.9). Thus we write
ðjÞ
f i ¼ f i ¼ δji δðtÞδðx yÞ; ð4:61Þ
ð jÞ
where f i signifies the ith component of a unit point load which acts in the jth
direction. If we write Navier’s equations as
70 4 Propagation of Bulk Waves
g32
x2
g31
x
g33
y x1
f (3)
x3
2 2 2
c22 ∂ ui =∂xm ∂xm þ c21 c22 ∂ um =∂xm ∂xi ∂ ui =∂t2 ¼ f i =ρ ð4:62Þ
where gji is the ith component of the displacement vector due to a body function
delta function acting in the jth direction. To solve this equation we first take the
Fourier transform on time of gji of Eq. (4.63) to obtain
2 2
c22 ∂ Gji =∂xm ∂xm þ c21 c22 ∂ Gjm =∂xm ∂xi þ ω2 Gji ¼ δji δðx yÞ=ρ; ð4:64Þ
where Gji ¼ Gji ðx; y; ωÞ. Then taking the three dimensional spatial Fourier trans-
form (see Eq. (4.47)) of Eq. (4.64) and using the sifting property of the delta
function, we find
p2 c22 Gji c21 c22 pm pi Gjm þ ω2 Gji ¼ δji expðipm ym Þ=ρ; ð4:65Þ
where we have defined p2 ¼ pm pm . To solve Eq. (4.65) for Gji we first multiply this
equation by pi and sum over i to obtain
pj expðipm ym Þ=ρ
pi Gji ¼ : ð4:66Þ
c21 p2 ω2
4.4 Spherical Waves in an Elastic Solid 71
Placing this result into the second term of Eq. (4.65) (replacing i by m in Eq. (4.66)
which is permissible since i is summed out anyway) then gives
2
2 2 c1 c22 pi pj expðipm ym Þ=ρ
c2 p ω Gji ¼ δji expðipm ym Þ=ρ
2
; ð4:67Þ
c21 p2 ω2
However, a more convenient form for Gji can be obtained by expanding Eq. (4.68)
in terms of partial fractions of the form
Aji Bji
Gji ¼ þ expðipm ym Þ=ρc22 : ð4:69Þ
p2 k21 p2 k22
Cross multiplying this expression and Eq. (4.68) by p2 k21 p2 k22 and
equating the coefficients of the like powers of p, we obtain
Aji ¼ pi pj =k22
ð4:70Þ
Bji ¼ δji pi pj =k22
Formally inverting the three dimensional spatial Fourier transform then gives
ð þ1
þ1 ð
ð þ1
1 k22 δji pi pj p i pj
Gji ¼ 2 k2
þ 2 2
ð2π Þ 3
p 2 p k 2
1 p k22 ð4:72Þ
1 1 1
exp½ipm ðym xm Þ
dp1 dp2 dp3 :
ρω2
From the representation of a spherical wave in terms of its spatial Fourier transform,
(recall (Eq. (4.52)) we have
ð þ1
þ1 ð þ1
ð
expðikn r Þ 1 exp½ipm ðym xm Þdp1 dp2 dp3
¼ ð4:73Þ
4πr ð2π Þ3 p2 k2n
1 1 1
72 4 Propagation of Bulk Waves
2 ð þ1
ð þ1
ð
þ1
∂ expðikn r Þ 1 pi pj exp½ipm ðym xm Þdp1 dp2 dp3
¼ :
∂xi ∂xj 4πr ð2π Þ 3
p2 k2n
1 1 1
ð4:74Þ
where ^ν i are the components of a unit vector pointing from Q to P (Fig. 4.8), the
derivatives in Eq. (4.75) can be performed and Gji written as
^νi ^νj expðik1 r Þ δij ^νi ^νj expðik2 r Þ
Gji ¼ 2
þ
4πρc
1 r 4πρc22 r
δij 3^ν i ^ν j expðik1 r Þ expðik2 r Þ ik1 expðik1 r Þ þ ik2 expðik2 r Þ
þ :
4πρω2 r3 r2
ð4:77Þ
Gji has both spherical P- and S-waves contained in it. In contrast to the fluid case,
however, Gji is not simply a function of r only as can be seen from Eq. (4.77).
To obtain the fundamental solution as a function of time from Eq. (4.77), direct
Fourier inversion of the first two terms gives
^νi ^νj δðt r=c1 Þ δij ^νi ^νj δðt r=c2 Þ
þ : ð4:78Þ
ρc21 4πr ρc22 4πr
ðb
iðtÞ ¼ τδðt τÞdτ: ð4:79Þ
a
ð ðb
þ1
ðb
I ðωÞ ¼ τexpðiωτÞdτ; ð4:81Þ
a
If we let a ¼ r=c1 and b ¼ r=c2 then the last term in Eq. (4.77) can be recognized
simply as
δij 3^ν i ^ν j I ðωÞ
ð4:83Þ
4πρ r 3
τ¼r=c
ð 2
δij 3^νi ^νj
τδðt τÞdτ ð4:84Þ
4πρ r 3
τ¼r=c1
τ¼r=c
ð 2
δij 3^νi ^νj
gji ðx; y; tÞ ¼ τ δðt τÞdτ
4πρr 3
τ¼r=c1 ð4:85Þ
^νi ^νj δðt r=c1 Þ δij ^νi ^νj δðt r=c2 Þ
þ 2 þ ;
ρc1 4πr ρc22 4πr
which is a result due to Stokes [2]. For the more general case where the body force
term has an arbitrary time dependency f(t), Eq. (4.85) still holds if we make the
replacement δðtÞ ! f ðtÞ.
When the point Q is at a large distance from the origin (Fig. 4.8), again we can use
the far field results of Eq. (4.60) directly in Eq. (4.75) for Gji to obtain this
fundamental solution approximately as
74 4 Propagation of Bulk Waves
es
P (x)
ev
v
O
^yi^yj expðik1 jyj ik1 x ^y Þ δij ^yi^yj expðik2 jyj ik2 x ^y Þ
Gji ¼ þ ð4:86Þ
4πρc21 jyj 4πρc22 jyj
and its derivatives with respect to the coordinates of x are similarly given approx-
imately by
∂Gji ik1^yi^yj^yk expðik1 jyj ik1 x ^y Þ ik2 δij ^yi^yj ^yk expðik2 jyj ik2 x ^
yÞ
¼ þ :
∂xk 4πρc1 jyj
2 4πρc2 jyj
2
ð4:87Þ
We will use these expressions later when we describe scattered waves in the far
field of a flaw. In that case it can also be convenient to define a set of mutually
orthogonal “scattering” coordinates (s, t, v) as shown in Fig. 4.10 and unit vectors
along those coordinates (es, et, ev). In this case we see ^y ¼ es so ^yi ¼ esi . We can
write the P-wave and S-wave terms in Eqs. (4.86) and (4.87) in similar forms by
noting that for any vector, u, we have
or, in component form with respect to a set of xk coordinates having unit vectors ik
along their axes:
All the terms in Eq. (4.89) have the common term uiij so that we must have
and the fundamental solution can be written as the sum of three components, all of
similar forms:
esiP esjP exp ik1 jyj ik1 x esP
GjiP ¼
1 jyj
4πρc 2
ð4:91Þ
S S S
S eηiη eηjη exp ik2 jyj ik2 x es η
Gjiη ¼ ðη ¼ t, vÞ ;
4πρc22 jyj
where GPji is the P-wave part of the fundamental solution, traveling in the ePs
direction and containing the P-wave polarization terms ePsi ePsj . Similarly,
GSjit ; GSjiv are S-waves traveling in the eSs t ; eSs v directions, respectively, with
polarization terms eStit eStjt , eSviv eSvjv , respectively. [Note: in some cases, we may want
to consider propagation of these waves in the same scattering direction so that
esP ¼ eSs t ¼ eSs v but in other cases we may want to consider scattering directions
that are different for the components so we have included that possibility here].
In Eq. (4.91) the S-wave part of the fundamental solution has been decomposed
into St and Sv waves, which are generalizations of the common decomposition of
polarized shear waves into “shear-vertical” (SV) and “shear-horizontal” (SH)
components. The derivatives of the fundamental solution, Eq. (4.87) can also
be placed into similar forms as:
∂GjiP ik1 esiP esjP eskP exp ik1 jyj ik1 x esP
¼
∂xk 4πρc21 jyj
S
ð4:92Þ
∂Gjiη S S S S
ik2 eηiη eηjη eskη exp ik2 jyj ik2 x es η
¼ ðη ¼ t, vÞ :
∂xk 4πρc2 jyj2
In later Chapters we will use both the original forms of Eqs. (4.86) and (4.87) as
well as the alternative forms of Eqs. (4.91) and (4.92).
The pressure in a fluid satisfies the wave equation as do the scalar and vector
potentials for waves in an elastic solid. In this section we want to consider
approximate solutions to these wave equations when the waves are all traveling
in essentially the same direction. This case is important for modeling ultrasonic
transducers since, as we will see in Chap. 8, an ultrasonic bulk wave transducer
typically produces a beam of sound that is well collimated, i.e. it is highly
76 4 Propagation of Bulk Waves
and write the unit vector, e, in spherical coordinates (θ, ϕ) (see Fig. 4.11) where θ is
the angle of the wave with respect to the z-axis, giving
Let us consider the angle θ to be small and write this wave as a “quasi-plane wave”
traveling in the z-direction of the form
This is a quasi-plane wave since for a true plane wave traveling in the z-direction,
the amplitude P would necessarily be a constant. For any exact solution of the wave
equation in the form of Eq. (4.95), we have, placing that form into the wave
equation
!
2 2 2
∂ P ∂ P ∂ P ∂P
þ þ þ 2ik expðikz iωtÞ ¼ 0 ð4:97Þ
∂x2 ∂y2 ∂z2 ∂z
4.5 Propagation of Waves in the Paraxial Approximation 77
However, let us consider the derivatives present in Eq. (4.98) for our quasi-plane
wave of Eq. (4.96). We have, for θ small
2
∂ P
¼ k2 ½ sin θ cos ϕ2 P ffi k2 cos 2 ϕP θ2
∂x 2
2
∂ P
¼ k2 ½ sin θ sin ϕ2 P ffi k2 sin 2 ϕP θ2
∂y 2
ð4:99Þ
∂P
2ik ¼ 2k2 ½ cos θ 1P ffi k2 P θ2
∂z
2
∂ P
¼ k2 ½ cos θ 12 P ffi k2 P=4 θ4
∂z 2
so that we see the last term (second derivative on z) is much smaller than the other
terms appearing in Eq. (4.98), all of which are of the same order of magnitude in θ.
Generalizing this result, we will say that any solution of the form of Eq. (4.95)
which satisfies
2 2 2
∂ P ∂ P ∂ P ∂P
<< 2 , , 2ik ð4:100Þ
∂z 2 ∂x ∂y 2 ∂z
is an (approximate) paraxial solution to the wave equation and satisfies the paraxial
wave equation:
2 2
∂ P ∂ P ∂P
þ þ 2ik ¼ 0: ð4:101Þ
∂x2 ∂y2 ∂z
expðikr iωtÞ
p¼A ð4:102Þ
r
78 4 Propagation of Bulk Waves
and its approximate behavior in a small region about the z-axis, i.e. where x, y << z.
Since
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ y2
r ¼ x þy þz ¼z 1þ
2 2 2
z2
x2 þ y2
ffiz 1þ
z2
we find
Aexp½ikðx2 þ y2 Þ=2z
p¼ expðikz iωtÞ; ð4:103Þ
z
A
P¼ exp ik x2 þ y2 =z : ð4:104Þ
z
It can be shown that P does indeed satisfy the paraxial wave equation.
Paraxial solutions for waves in an elastic solid are treated very similar to the fluid
case since the P-waves and S-waves written in terms of the potentials (ϕ, ψ) satisfy
wave equations, and paraxial solutions of the form
ϕ ¼ Φðx; y; z; ωÞexp ikp z iωt
ð4:105Þ
ψ ¼ Ψβ ðx; y; z; ωÞtβ expðiks z iωtÞ
and show that in the paraxial approximation we have the paraxial wave equation for
P-waves given by
2e 2e e3
∂ U 3 ∂ U 3 ∂U
þ þ 2ik p ¼0 ð4:108Þ
∂x2 ∂y2 ∂z
e1 ¼ U
with U e 2 ¼ 0 while for S-waves
2e 2e eβ
∂ U β ∂ U β ∂U
þ þ 2iks ¼ 0 ðβ ¼ 1, 2Þ ð4:109Þ
∂x2 ∂y 2 ∂z
e 3 ¼ 0. We will not give the lengthy details of the derivation of Eqs. (4.108)
with U
and (4.109) here but refer the reader to [2] instead. Note that since the velocity, vi, is
just vi ¼ iω ui , the velocity field V e i ¼ iωU e i also satisfies the paraxial wave
equations in the paraxial approximation.
Plane waves and spherical waves are important building blocks for constructing
more general wave fields. However, in order to construct models of the sound
beams of bulk wave ultrasonic transducers, generally one needs to superimpose
many plane waves or spherical waves to accurately reproduce the entire transducer
wave field. However, because the paraxial approximation is an appropriate approx-
imation to consider when modeling the highly directional beams of ultrasonic
transducers, one can look at the possibility of combining relatively few simple
propagating beams that satisfy the paraxial wave equation to form the more
complex beams of ultrasonic transducers. Gaussian beams, as we will see in
Chap. 8, are ideal building blocks of this type that are particularly well-suited to
this task. Gaussian beams have been considered for many years in laser applications
as they are suitable for modeling the optical waves present in those cases [3]. For
ultrasonic NDE applications they are also powerful modeling tools because they
can simulate ultrasonic transducer wave fields in complex geometries and
for complex materials (anisotropic, inhomogeneous cases). In this section and the
following one we will not treat Gaussian beams in their most general form but will
restrict ourselves to the propagation of elliptical-shaped cross section Gaussians. In
Chap. 6 we will see that these types of Gaussian beams are sufficient for describing
the transmission and reflection of ultrasonic bulk wave transducer wave fields in
geometries involving plane (and some curved) interfaces and for media that are
either fluids or isotropic elastic solids. One can find details on the use of Gaussian
beams in more complex geometries in [4] and in more general media in [2].
80 4 Propagation of Bulk Waves
where (qx, qy) and P are complex functions that we will determine in order to satisfy
e where
the paraxial wave equation for P
" !#
e ðx; y; z; ωÞ ¼ Pðz; ωÞexp ik x2 y2
P þ : ð4:111Þ
2 qx ðz; ωÞ qy ðz; ωÞ
dqx dqy
¼ 1, ¼1
dz dz
ð4:113Þ
1 dP 1 1
þ þ ¼ 0:
P dz 2qx 2qy
where (qx0, qy0) are complex constants. The solution for P can also be verified to be
given by
sffiffiffiffiffiffiffiffiffiffiffiffi
qx0 qy0
P ¼ P0 ð4:115Þ
qx qy
where P0 is the value of P at z ¼ 0. Placing these results into Eq. (4.110) we find
sffiffiffiffiffiffiffiffiffiffiffiffi " ! #
qx0 qy0 ik x2 y2
p ¼ P0 exp ikz þ þ iωt : ð4:116Þ
qx qy 2 qx q y
parameters that have a more physical meaning. We can do this by letting the
complex constants (qx0, qy0) be expressed as
!
iπw2x0 iπw2y0
qxo ¼ zx0 þ , qy0 ¼ zy0 þ ; ð4:117Þ
λ λ
where (zx0, zy0) and (wx0, wy0) are all parameters with the dimensions of a length and
λ is the wavelength. In these expressions we see the so-called confocal parameters
(zRx, zRy) (which also have the dimensions of a length) given by
πw20x πw20y
zRx ¼ , zRy ¼ ð4:118Þ
λ λ
so that
qx0 ¼ ðzx0 þ izRx Þ, qy0 ¼ zy0 þ izRy : ð4:119Þ
1 1 λ
¼ þi 2
qx ðzÞ Rx ðzÞ πwx ðzÞ
ð4:121Þ
1 1 λ
¼ þi 2
qy ðzÞ Ry ðzÞ πwy ðzÞ
and the phase terms in Eq. (4.116) can be rewritten in these terms as
sffiffiffiffiffiffiffiffiffiffiffiffi " 2 ! #
qx0 qy0 ik x y2 x2 y2
p ¼ P0 exp ikz þ þ þ 2 iωt :
qx qy 2 Rx ð z Þ Ry ð z Þ w x ðzÞ w y ðzÞ
2
ð4:122Þ
The exp(ikz) term in Eq. (4.122) shows that indeed this is a quasi-plane wave
traveling in the z-direction, but it has an elliptical-shaped Gaussian amplitude
profile in the x- and y-directions because of the terms
82 4 Propagation of Bulk Waves
" !#
x2 y2
exp 2 þ 2 :
w x ðzÞ w y ðzÞ
The terms (wx(z), wy(z)) are the widths of this Gaussian beam in the x- and
y-directions that define the distances in those directions from the central z-axis of
the beam at which the beam amplitude drops to a factor e1 from its on-axis value.
From Eq. (4.120) we see that these beam widths are always positive and have their
smallest values at (zx0, zy0) which define the location of the “waists” of the beam in
x- and y-directions, respectively. Figure 4.12a shows schematically a cross-
sectional view of the beam in the x-z plane and its waist in the x-direction, with a
similar cross-sectional view for the y-z plane in Fig. 4.12b. Now consider the other
phase terms
2
ik x y2
exp þ :
2 Rx ðzÞ Ry ðzÞ
We saw a similar term in Eq. (4.104) for the paraxial approximation of a spherical
pffiffiffiffiffiffiffiffiffiffiffiffi
wave traveling in the z-direction. In fact if we let P0 qx0 qy0 ¼ A and set zx0 ¼ zy0
pffiffiffiffiffiffiffiffiffi
¼ 0 and wx0 ¼ wy0 ¼ 0 we find qx qy ¼ z, Rx ¼ Ry ¼ z and we just recover that
spherical wave solution where z is the radius of the spherical wave front at a
4.6 Gaussian Beams in Fluids and Elastic Solids 83
distance z from the origin of the spherical wave. Thus, in the general case the
parameters (Rx(z), Ry(z)) are the corresponding radii of the Gaussian beam wave
front in the x- and y-directions (see Fig. 4.13a, b) and from Eq. (4.120) we see these
radii are negative for distances z smaller than the location of the beam waist and
positive for distances z larger than the location of the beam waist, with infinite
values exactly at the waist locations. Negative Rx, Ry-values represent a beam that is
converging towards the beam waist while positive Rx, Ry-values represent a beam
that is diverging and Rx ¼ Ry ¼ 1 values denote a planar wave front, a condition
which Eq. (4.120) shows occurs at the beam waists.
Although expressing a Gaussian beam in terms of its beam width and wave front
curvature parameters allows us to examine that beam in more physical terms, the
algebraic complexity of the Gaussian beam in terms of those parameters (see
Eq. (4.120)) means that it is usually more convenient to write the propagating
Gaussian beam in terms the complex (qx, qy) and (qx0, qyo) parameters, as seen in
Eqs. (4.114) and (4.116), where the beam widths and wave front curvatures are not
seen explicitly. Note that in either case in order to define a Gaussian beam explicitly
we must choose P0 (which can be complex) and (1) either the two complex
parameters, (qx0, qyo), or equivalently, (2) four real parameters such as (zx0, wx0,
zy0, wy0). We will see in Chap. 8 if this choice is made judiciously for a small
number (~15) of Gaussian beams, then a superposition of those beams can give us a
very fast and versatile beam model of an ultrasonic bulk wave transducer.
84 4 Propagation of Bulk Waves
Since the Cartesian displacement (or velocity) components satisfy paraxial wave
equations for paraxial wave propagating in an isotropic elastic solid, we can use all
of our previous results in a solid. The only differences we must account for in the
elastic wave case are that (1) the wave speeds and Gaussian beam parameters are
different for P- and S-waves and (2) there are differences in polarization of the
P- and S-wave Gaussian beams. Accounting for these differences, we can write the
velocity of an elliptical cross section Gaussian beam of type m ¼ ðp; sÞ, vm(x, ω),
at a point x ¼ ðx; y; zÞ and traveling in the z-direction as
pffiffiffiffiffiffiffiffiffiffiffiffi " ! #
m m
q x0 q yo ikm x2 y2
vm ðx; ωÞ ¼ V m dm pffiffiffiffiffiffiffiffiffiffiffiffi exp ikm z þ þ iωt ðm ¼ p, sÞ;
qxm qym 2 qxm qym
ð4:123Þ
qxm ¼ z þ qx0
m
ð4:124Þ
qym ¼ z þ qyo
m
:
PðωÞ ¼ expðiωz=cÞ
If several distances and wave speeds are present (on both sound generation
and reception), this result generalizes directly to:
More detailed discussions of the propagation of plane and spherical waves in fluids
can be found in the text by Morse and Ingard [5]. Jones [6] also discusses acoustic
waves in conjunction with electromagnetic wave problems. For elastic wave prop-
agation problems see Auld [7] and Pollard [8]. The Los Alamos report by Gubernatis,
Domany, and Krumhansl [9] describes the elastic wave fundamental solutions
considered here and summarizes NDE-related wave modeling work as of the late
1970s, but volume one in the handbook series edited by Varadan, Lakhtakia, and
Varadan [10] is more accessible and also contains a wide range of the basic relation-
ships needed to describe both propagation and scattering for acoustic, elastic, and
electromagnetic problems. A good source for describing the interrelationship
between plane wave, cylindrical wave, and spherical wave solutions, and for 3-D
elastic wave solutions in various related seismology problems is Ben-Menahem and
Singh [1]. Gaussian beams for laser applications are discussed by Siegman [3] and for
fluid and elastic solids by Schmerr and Song [4] and Cerveny [2].
We have limited our discussion in this chapter to theoretical models of wave
propagation in isotropic media. Both Auld [7] and Musgrave [11] discuss the
influence of anisotropy on modeling plane and spherical wave propagation and
Cerveny [2] considers Gaussian beams traveling in anisotropic and inhomogeneous
solids. Pollard [8] considers many of the experimental issues associated with the use
of these waves in ultrasonics.
4.8 Problems
4.1. Prove that Eq. (4.2) does represent a general solution to the one-dimensional
wave equation.
4.2. Consider the one-dimensional motion u ¼ uðx; tÞ of an elastic solid. If at time
t ¼ 0 the initial conditions uðx; 0Þ ¼ u0 ðxÞ and u_ ðx; 0Þ ¼ v0 ðxÞ for the
displacement u and velocity u_: are given, show that the solution to this initial
value problem is
2 3
ð
xþct
14
uðx; tÞ ¼ u ðx ctÞ þ u0 ðx þ ctÞ þ 1=c v0 ðuÞdu5:
2 0
xct
This is known as the D’Alembert solution. Consider the case when v0 ðxÞ ¼ 0
and
A a < x < a
u0 ð x Þ ¼ ;
0 otherwise
86 4 Propagation of Bulk Waves
where A is a constant. Sketch the solution for u at times t ¼ 0, a/2c, a/c, and
3a/2c.
4.3. Consider a 3-D harmonic plane wave traveling in a fluid in the n direction.
What is the relationship between (1) the pressure p and the displacement un,
(2) the pressure p and the velocity vn in this wave?
4.4. For an outward propagating spherical harmonic wave in a fluid, determine the
relationship between the pressure amplitude of this wave and the amplitude
of the velocity in the radial direction. At high frequencies (kr
1) what is
this relationship?
4.5. Consider the one-dimensional harmonic motion of a fluid, i.e. let p ¼ pðx; ωÞ
expðiωtÞ for the pressure and u ¼ ½ux ðx; ωÞex expðiωtÞ for the displace-
0
ment. Show that the fundamental solution (where f ¼ expðiωtÞδ x x ) is
0 0
expðikjxx jÞ
given by G x; x ; ω ¼ 2ik .
4.6. (a) Consider 2-D harmonic motion of a fluid, i.e. p ¼ p ðx; y; ωÞexpðiωtÞ
for the pressure and u ¼ ux ðx; y; ωÞex þ uy ðx; y; ωÞey expðiωtÞ for the
displacement. Show that the fundamental solution for this
two-dimensional case (the solution due to a line source f ¼ expðiωtÞ
0 0
δ x x δ y y ) is given by p ¼ GðR; ωÞexpðiωtÞ where
ð1Þ
iH 0 ðkRÞ
GðR; ωÞ ¼
4
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1Þ
and where R ¼ ðx x0 Þ2 þ ðy y0 Þ2 and H0 is the Hankel function
(of type one) of order zero. How does your answer change if we had
assumed expðþiωtÞ time dependency instead?
(b) How is the fundamental harmonic wave solution for the fluid obtained in
(a) related to the fundamental solution for two-dimensional SH-waves,
0 0
i.e. where f z ¼ expðiωtÞδ x x δ y y in Eq. (3.68)?
4.7. Determine the far field form of both the fundamental solution and its deriv-
atives for the solution obtained in problem 4.6. Assume expðiωtÞ time
dependency.
4.8. Determine the fundamental solution in two-dimensions considered in prob-
lem 4.6 but now where the line source is of arbitrary time dependency.
4.9. Consider
the two-dimensional harmonic
motion of an elastic solid, i.e.
u ¼ ux ðx; y; ωÞex þ uy ðx; y; ωÞey expðiωtÞ for the displacement. Show
that the fundamental solution due to the line source:
References 87
0 0
f α ¼ δαβ expðiωtÞδ x x δ y y ðα, β ¼ 1, 2Þ
f3 ¼ 0
is given by
i 2 ð1Þ ∂ ∂ h ð1Þ ð1Þ i
gαβ ¼ δαβ ks H0 ðks RÞ þ H ðk s R Þ H 0 k p R ;
4ρω2 ∂xα ∂xβ 0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1Þ
where R ¼ ðx x0 Þ2 þ ðy y0 Þ2 and H0 is the Hankel function (of type
one) of order zero.
4.10. Determine the far field form of both the fundamental solution and its deriv-
atives for the solution obtained in problem 4.9.
4.11. Determine the fundamental solution in two-dimensions considered in prob-
lem 4.9 but now where the line source is of arbitrary time dependency.
4.12. Prove that the quasi-plane wave amplitude of a propagating spherical wave
(Eq. (4.104)) does satisfy the paraxial wave equation.
References
1. A. Ben-Menahem, S.J. Singh, Seismic Waves and Sources (Springer, New York, 1981)
2. V. Cerveny, Seismic Ray Theory (Cambridge University Press, Cambridge, 2001)
3. A.E. Siegman, Lasers (University Science, Sausalito, 1986)
4. L.W. Schmerr, S.-J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and
Measurements (Springer, New York, 2007)
5. P.M. Morse, K.V. Ingard, Theoretical Acoustics (McGraw-Hill, New York, 1968)
6. D.S. Jones, Acoustic and Electromagnetic Waves (Oxford University Press, New York, 1986)
7. B.A. Auld, Acoustic Fields and Waves in Solids, Vols 1 and 2 (Wiley, New York, 1973)
8. H.F. Pollard, Sound Waves in Solids (Methuen, New York, 1977)
9. J.E. Gubernatis, E. Domany, J.A. Krumhansl, Elastic Wave Scattering Theory with Application
to Nondestructive Evaluation, LA-UR-79-2393 (Los Alamos Scientific Laboratory, Los
Alamos, New Mexico, 1979)
10. V.V. Varadan, A. Lakhtakia, V.K. Varadan (eds.), Field Representations and Introduction
to Scattering (North Holland Publishing, New York, 1992)
11. M.J.P. Musgrave, Crystal Acoustics (Holden Day, San Francisco, 1970)
Chapter 5
The Reciprocal Theorem and Other
Integral Relations
This Chapter will develop reciprocal theorems for both fluids and elastic solids and
will combine the fundamental solutions derived in the last Chapter with those
theorems to obtain general integral representation expressions and integral equa-
tions. We will also develop a more generalized electromechanical reciprocity
theorem which is applicable to all the elements (electrical, piezoelectric, and
mechanical) of an ultrasonic measurement system. These results will serve as the
foundation for many later applications, including transducer modeling (Chap. 8),
flaw scattering (Chap. 10), and a general model of the entire ultrasonic measure-
ment process (Chap. 12).
∇2 p1 þ k2 p1 ¼ f 1
: ð5:1Þ
∇2 p2 þ k2 p2 ¼ f 2
If we apply the divergence theorem to each of the terms in Eq. (5.2), we obtain
ð
I ¼ ½∇ ðp2 ∇p1 Þ ∇ ðp1 ∇p2 ÞdV ð5:3Þ
V
The first and third terms in Eq. (5.4) cancel and, using the Helmholtz equations for
p1 and p2, the second and fourth terms can be rewritten to give
ð
I ¼ p2 f 1 k2 p1 p1 f 2 k2 p2 dV; ð5:5Þ
V
Equating Eqs. (5.6) and (5.2), we finally obtain the general form of the reciprocal
theorem for a fluid as
ð ð
ðp1 f 2 p2 f 1 ÞdV ¼ ðp2 ∂p1 =∂n p1 ∂p2 =∂nÞdS: ð5:7Þ
V S
Consider again our general region V shown in Fig. 5.1, where now solution 1 is
taken as the solution to an arbitrary body force f:
∇2 p þ k2 p ¼ f ð5:8Þ
p1 ¼ p p2 ¼ G
∂p1 =∂n ¼ ∂p=∂n ∂p2 =∂n ¼ ∂G=∂n ; ð5:9Þ
f1 ¼ f f 2 ¼ δðx yÞ
which, using the sampling property of the delta function, can be rewritten as
ð
α pðy; ωÞ ¼ Gðx; y; ωÞf ðx; ωÞ dV x
V ð
þ Gðxs ; y; ωÞ∂pðxs ; ωÞ=∂nðxs Þ pðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs Þ dS xs ;
S
ð5:11Þ
where
8
<1 y inside V
α ¼ 1=2 y on S : ð5:12Þ
:
0 y outside V
The factor of one half appearing in Eq. (5.12) is valid only if the surface, S, is
smooth at point y. If y is a non-smooth point of S (edge or corner), this factor will be
a different constant [1].
Equation (5.11) is a general integral representation theorem for the pressure p at
any point in the fluid in terms of the surface values of p and its normal derivative
∂p=∂n. Thus, if the surface values of both p and its normal derivative, ∂p=∂n,
(or equivalently, the normal velocity component) can be obtained on S, then the
value of p is determined everywhere in V from Eq. (5.11).
92 5 The Reciprocal Theorem and Other Integral Relations
and let S be a surface which surrounds a finite volume V of that extended region.
Then if the point y is taken outside of the source region for these incident waves, the
integral representation theorem gives
ð
α p ¼ Gðxs ; y; ωÞ∂pinc ðxs ; ωÞ=∂n xs
inc
ð5:14Þ
S
p ðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs Þ dSðxs Þ;
inc
where
8
<0 y outside S
α ¼ 1=2 y on S :
:
1 y inside S
Up until now we have applied the integral representation theorem to finite regions V. In
many cases, however, it will be convenient to consider the surrounding region to be
infinite in extent and consider the resulting integral representation theorem in those
situations. For such infinite regions, application of the integral representation theorem
will require that we carefully consider the behavior of the solution, p, at infinity.
5.1 Reciprocal Theorem for a Fluid 93
To see this, consider a region V which is contained between some finite surface,
S, and the surface, SR, of a large sphere, of radius R, whose center is located at y.
Also let f be the body forces associated with some source region as shown in
Fig. 5.3. If we apply Eq. (5.11) to this region V, we obtain
ð
α pðy; ωÞ ¼ Gðx; y; ωÞf ðx; ωÞ dV x
V
ð
þ Gðxs ; y; ωÞ∂pðxs ; ωÞ=∂nI ðxs Þ pðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nI ðxs Þ dS xs
S
ðh
0
0
0
0
0 0
0
þ G xs ; y; ω ∂p xs ; ω =∂n xs p xs ; ω ∂G xs ; y; ω =∂n xs dS xs ;
SR
ð5:15Þ
Rð∂p=∂R ikpÞ ! 0 as R ! 1
ð5:19Þ
p ! 0 as R ! 1
or, equivalently,
Either Eqs. (5.19) or (5.20) are known as the Sommerfeld radiation conditions.
[Note: these forms are only for harmonic waves with expðiωtÞ time dependency
which is assumed throughout this book.] Physically, these conditions guarantee that
the waves at infinity are outgoing waves and decay sufficiently fast so that there are
no sources at infinity. Since under these conditions the integral at infinity vanishes
in the integral representation theorem, we have shown that if p satisfies the
Sommerfeld radiation conditions in an infinite region then
ð ð
α pðy; ωÞ ¼ Gðx; y; ωÞf ðx; ωÞ dV x þ ½pðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs Þ
V S
Gðxs ; y; ωÞ∂pðxs ; ωÞ=∂nðxs ÞdSðxs Þ;
ð5:21Þ
where nðxs Þ ¼ nI ðxs Þ is now the outward normal to the surface S.
S
Gðxs ; y; ωÞ∂pscatt ðxs ; ωÞ=∂nðxs ÞdSðxs Þ: ð5:22Þ
5.1 Reciprocal Theorem for a Fluid 95
However, recall from Eq. (5.14) for y outside the source region the pressure of the
incident wave, pinc, satisfies
ð
β pinc ¼ pinc ðxs ; ωÞ∂Gðxs ; y; ωÞ=∂n xs
S ð5:23Þ
Gðxs ; y; ωÞ∂pinc ðxs ; ωÞ=∂nðxs Þ dSðxs Þ;
S ð5:24Þ
Gðxs ; y; ωÞ∂pðxs ; ωÞ=∂nðxs Þ dSðxs Þ
which relates the surface values of p and ∂p=∂n and where p.v. denotes a principal
value integral [1]. For example, if S is a free surface, then by the boundary
conditions (3.19) p ¼ 0 on the surface, giving
ð
pinc ðys ; ωÞ ¼ Gðxs ; ys ; ωÞ∂pðxs ; ωÞ=∂nðxs Þ dSðxs Þ ð5:27Þ
S
and we can consider Eq. (5.27) to be an integral equation to solve for the surface
values of ∂p=∂n. [Note: we have dropped the principal value designation in
Eq. (5.27) since the fundamental solution itself is only weakly singular. The same
cannot be said for the normal derivative of the fundamental solution]. Similarly, for
a rigid boundary ∂p=∂n ¼ 0, and Eq. (5.26) reduces to
ð
pðys ; ωÞ=2 ¼ pinc ðys Þ þ p:v: pðxs ; ωÞ∂Gðxs ; ys ; ωÞ=∂nðxs Þ dSðxs Þ; ð5:28Þ
S
which is an integral equation for the unknown surface pressure, p. For a more
general inclusion problem it is necessary to also formulate an integral equation for
the fields within the flaw in a similar manner and solve a set of two integral
equations on the surface simultaneously (see problem 5.11).
Another case that needs special consideration is that of crack-like scatterer. If, in
our fluid model for a volumetric flaw we break up the surface S into two distinct
parts, Sþ and S (Fig. 5.5a), then from Eq. (5.25) we have
ð
þ þ
pðy; ωÞ ¼ p ðy; ωÞ þ
inc
p xs ; ω ∂G xs ; y; ω =∂nþ xþ s
þ þ
Sþ
þ þ
Gð x s ; y; ω ∂p x s ; ω =∂n xs dS
þ p xs ; ω ∂G xs ; y; ω =∂n x s ð5:29Þ
S
G x
s ; y; ω ∂p xs ; ω =∂n xs dS:
Thus, if we take the limit as the thickness, t, of the scatterer between the two
surfaces goes to zero, the geometry reduces to that of a crack-like open surface
(Fig. 5.5b) and Eq. (5.29) becomes, in this limit
ð
pðy; ωÞ ¼ p ðy; ωÞ þ ½Δpðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs Þ
inc
S
Gðxs ; y; ωÞΔð∂pðxs ; ωÞ=∂nðxs ÞÞdSðxs Þ; ð5:30Þ
where
Δpðxs ; ωÞ ¼ p xþ
s ; ω p xs ; ω
ð5:31Þ
Δð∂pðxs ; ωÞ=∂nðxs ÞÞ ¼ ∂p xþ þ þ þ
s ; ω =∂n xs ∂p xs ; ω =∂n xs
represent jumps in the pressure and its normal derivative, respectively, on the
surface and we have let S ¼ Sþ , n ¼ nþ in Eq. (5.30) for the open surface
(Fig. 5.5b). For a crack-like scatterer where the pressure vanishes on both faces,
Δp ¼ 0, and Eq. (5.30) becomes
ð
pðy; ωÞ ¼ pinc ðy; ωÞ ½Gðxs ; y; ωÞΔð∂pðxs ; ωÞ=∂nðxs ÞÞdSðxs Þ ð5:32Þ
S
which is an integral equation for the jump in the normal derivative of the pressure,
similar to Eq. (5.27) for the volumetric scatterer.
Solving integral equations such as Eqs. (5.27), (5.28), (5.33) must typically be
done numerically. One effective numerical method is boundary elements, where the
98 5 The Reciprocal Theorem and Other Integral Relations
surface S is broken up into small patches (elements) and the fields are approximated
as simple functions (constant, linear, quadratic, etc.) in terms of unknown param-
eters on these patches. The remaining surface integrals are then performed numer-
ically, paying due attention to the singularities present in the integrands. The
integral equations in this manner are reduced to a set of algebraic equations
which can be solved by standard methods [2]. Once the surface fields are entirely
known then the integral representation theorem Eq. (5.11) gives the fields every-
where in the region V. We will not develop such general numerical solutions in this
book. Instead, in Chap. 10 we will apply several approximations (the Kirchhoff
approximation and Born approximation) that give the surface fields directly from
the known incident wave fields. In these cases the integral representation theorem
can then be used directly to obtain the scattered waves.
If the volume V of Fig. 5.1 contains an elastic solid, then as in the fluid case we can
consider two displacement vector solutions for this medium, u1 and u2, which
satisfy the equations of motion (for harmonic wave disturbances):
where t1k ¼ τ1kl el and t2k ¼ τ2kl el are also the tractions and stresses associated with
these two solutions. In this case we consider the surface integral
ð
I ¼ t2 ðnÞ u1 t1 ðnÞ u2 dS
Sð
¼ t2k nk u1 t1k nk u2 dS ð5:35Þ
S
Using the equations of motion in the first and third terms of Eq. (5.37), we find
ð
I¼ ρω2 u2 f 2 u1 ρω2 u1 f 1 u2 dV
V ð
2
þ tk ∂u1 =∂xk t1k ∂u2 =∂xk dV; ð5:38Þ
V
Now, consider the two terms in the second integral of Eq. (5.39). In terms of the
stress and displacement components along the coordinate axes these terms become
τ2kl ∂u1l =∂xk τ1kl ∂u2l =∂xk ¼ Cklij ∂u2i =∂xj ∂u1l =∂xk Cklij ∂u1i =∂xj ∂u2l =∂xk
¼ Cklij ∂u2i =∂xj ∂u1l =∂xk Cijkl ∂u2l =∂xk ∂u1i =∂xj
¼ Clkji ∂u2i =∂xj ∂u1l =∂xk Cijkl ∂u2l =∂xk ∂u1i =∂xj ð5:40Þ
¼0
by using the stress-strain relations Eq. (3.35) and the symmetry properties of the
elastic constant tensor Eq. (3.36). Since the second integral of Eq. (5.39) vanishes,
equating the remaining integral in this equation to the original integral in Eq. (5.35),
we obtain the reciprocal theorem for an elastic medium:
ð ð
1 2
f u f 2 u1 dV ¼ t2k nk u1 t1k nk u2 dS; ð5:41Þ
V S
As in the case of the fluid we can apply the reciprocal theorem for an elastic solid to
a volume V where the first solution satisfies the equation of motion (Navier’s
100 5 The Reciprocal Theorem and Other Integral Relations
equations) with an arbitrary body force term f while the second solution is the
fundamental solution. In this case, then we set
If we apply Eq. (5.45) to the geometry shown in Fig. 5.2, where an incident wave
with displacement vector, uinc, is now propagating in an elastic medium through a
region of volume V, then outside of the source region, we have, as in the fluid case,
ð
α uinc
n ð y; ω Þ ¼ Cklij nk ðxs ÞGln ðxs ; y; ωÞ∂uinc
i ðxs ; ωÞ=∂xj
S ð5:46Þ
Cklij nk ðxs Þuinc
l ðxs ; ωÞ∂Gin ðxs ; y; ωÞ=∂xj dSðxs Þ;
Consider the infinite region shown in Fig. 5.3 again, now for the case where the
medium is an elastic solid. Applying the representation integral to the region
between the surface S and SR gives
ð
α un ðy; ωÞ ¼ f k ðx; ωÞGkn ðx; y; ωÞdV ðxÞ
V
ð
þ Cklij nIk ðxs ÞGln ðxs ; y; ωÞ∂ui ðxs ; ωÞ=∂xj
S
Cklij n Ik ðxs Þul ðxs ; ωÞ∂Gin ðxs ; y; ωÞ=∂xj dSðxs Þ ð5:47Þ
ð
0
0
0
0
þ Cklij nk xs Gln xs ; y; ω ∂ui xs ; ω =∂xj
SR
0 0 0 0 0
Cklij nk xs ul xs ; ω ∂Gin xs ; y; ω =∂xj dS xs ;
where nIk are the components of the inward unit normal vector nI (Fig. 5.3). To
guarantee that the integral over SR vanishes as R ! 1 we must again put some
radiation conditions on our solution at infinity. In an elastic medium, these radiation
conditions can be stated in several different forms. The form that is most similar to
the fluid case is as follows. First, assume that the displacement, u, is decomposed
into two parts as
u ¼ uP þ uS ð5:48Þ
where
∇ uP ¼ 0 ∇ uS ¼ 0: ð5:49Þ
uP ¼ ∇ϕ uS ¼ ∇ ψ ð5:50Þ
∇2 uP þ k21 uP ¼ 0
ð5:51Þ
∇2 uS þ k22 uS ¼ 0:
102 5 The Reciprocal Theorem and Other Integral Relations
Similar to the fluid case, the radiation conditions on uP and uS then are
lim uP ¼ 0 lim R ∂uP =∂R ik1 uP ¼ 0
R!1 R!1
ð5:52Þ
lim uS ¼ 0 lim R ∂uS =∂R ik2 uS ¼ 0:
R!1 R!1
[Note: again these are only valid for displacements with expðiωtÞ time depen-
dency, as assumed throughout this book.] It can be shown (for example, see Eringen
and Suhubi, [3]) that if the radiation conditions of Eq. (5.52) are satisfied, then the
integral over SR in Eq. (5.47) does indeed vanish as R ! 1, leaving
ð
α un ðy; ωÞ ¼ f k ðx; ωÞGkn ðx; y; ωÞdV ðxÞ
V
ð
þ Cklij nk ðxs Þul ðxs ; ωÞ∂Gin ðxs ; y; ωÞ=∂xj
ð5:53Þ
S
Cklij nk ðxs ÞGln ðxs ; y; ωÞ∂ui ðxs ; ωÞ=∂xj dSðxs Þ;
where n ¼ nI has now been changed to be the unit normal pointing outward
from the finite volume contained by the surface S (Fig. 5.3). The proof of this
result, however, is rather lengthy and will not be given here. The reader can find the
details in [3] where other equivalent forms of the radiation conditions are also
given.
Consider the scattering problem of Fig. 5.4 again, but now in an elastic medium
where the scattered displacements, uscatt, are produced when an incident wave with
displacements, uinc, interacts with the surface S. Since uscatt satisfies the radiation
conditions Eq. (5.52) and uinc satisfies Eq. (5.46), adding Eq. (5.46) for uinc and
Eq. (5.53) for uscatt we find for anywhere outside the source region that
ð
α unscatt ðy; ωÞ þ β uinc
n ð y; ω Þ ¼ Cklij nk ðxs Þul ðxs ; ωÞ∂Gin ðxs ; y; ωÞ=∂xj
S
Cklij nk ðxs ÞGln ðxs ; y; ωÞ∂ui ðxs ; ωÞ=∂xj dSðxs Þ
ð5:54Þ
for the total displacement u ¼ uinc þ uscatt , where α and β are defined as before
(see Eqs. (5.12) and (5.23)).
5.2 Reciprocal Theorem for an Elastic Solid 103
Equation (5.54) can serve as the basis for solving a number of scattering problems if
we take point y to the surface S so that we obtain an integral equation for the surface
stresses and displacements. In this case we find
where the integral is understood, as in the fluid case, as a principal value integral.
For the case of a void, for example, the surface tractions nk Cklij ∂ui =∂xj are zero on
the surface so that
which can be compared with the analogous result for the fluid model Eq. (5.32).
However, in this case, if we attempt to take point y to the surface we find
þ
un ðyðs ; ωÞ þ u n ðys ; ωÞ =2 ¼ un ðys ; ωÞ
inc
þ Cklij nþ
k ðxs ÞΔul ðxs ; ωÞ∂Gin ðxs ; ys ; ωÞ∂xj dS xs ;
ð5:60Þ
Sþ
where uþ
j and uj are the displacements on the plus and minus faces, respectively.
Unfortunately, Eq. (5.60) is not an integral equation that we can solve since both the
displacement discontinuity, Δul, and the displacements on the individual crack
faces are contained in Eq. (5.60). We can obtain an integral equation from
Eq. (5.57) by first taking the operator np Cpqnm ∂=∂ym on that equation and then
taking the limit as point y goes to the surface. Since the tractions on both faces of
the crack vanish, we obtain
nþ
p ðys ÞC ∂uinc ðy ; ωÞ=∂ym
ð hpqmn n s
Cpqmn Cklij nþ þ 2
¼ p ð y Þn
s k ð x s ÞΔu ð
l sx ; ωÞ∂ G ð x
in s s; y ; ω Þ=∂x j ∂x m dS xs ;
Sþ
ð5:61Þ
where we have used the fact that ∂Gin =∂ym ¼ ∂Gin =∂xm and now the highly
singular integral must be interpreted in the finite part sense [4]. Since the left hand
side of Eq. (5.61) represents the tractions on the crack due to the incident waves,
which are known, Eq. (5.61) is an integral equation for the unknown displacement
discontinuity Δuj. Because the second order derivatives of the fundamental solution
are present in Eq. (5.61), it is called a hypersingular integral equation. Numerical
methods such as boundary elements can also solve this type of equation. See [5], for
example, for further details.
In the case of an elastic inclusion, Eq. (5.55) by itself does not yield sufficient
information to solve for the unknown surface fields. Instead, as in the fluid case, we
need to also derive an equation similar to Eq. (5.55) for the fields within the
inclusion and solve both equations simultaneously (see problem 5.13).
where ρ0 is the mass density and ρ is the electrical charge density. For harmonic
disturbances (of expðiωtÞ time dependency) these equations can also be written, in
Cartesian tensor notation, as
t1k nk ¼ τ1kl el nk , v1 , f 1 , B1 , H1 , E1 , D1 , J1
ð5:66Þ
t2k nk ¼ τ2kl el nk , v2 , f 2 , B2 , H2 , E2 , D2 , J2 :
Converting this surface integral to a volume integral via Gauss’ theorem again and
writing temporarily the velocity components in terms of the displacements, we have
ð
I ¼ iω∂τ1kl =∂xk u2l iω τ1kl ∂u2l =∂xk þ iω∂τ2kl =∂xk u1l þ iω τ2kl ∂u1l =∂xk dV
V ð
þ εijk ∂E2j =∂xi H1k þ εijk E2j ∂H1k =∂xi εijk ∂E1j =∂xi H 2k εijk E1j ∂H2k =∂xi dV:
V
ð5:69Þ
However, the second and fourth terms in the first integral of Eq. (5.69) reduce, using
the constitutive equations and the symmetry of the elastic constants tensor
Eq. (5.65), to just
iω d ikl E1i ∂u2l =∂xk dikl E2i ∂u1l =∂xk
and the governing equations can be used in the remaining terms to obtain
5.3 An Electromechanical Reciprocal Theorem 107
ð
I¼ f 1l iωρ0 v1l v2l f 2l iωρ0 v2l v1l
V
þiω d ikl E1i ∂u2l =∂xk dikl E2i ∂u1l =∂xk dV
ð
þ iωB2k H1k iωB1k H 2k dV
ð5:70Þ
V
ðh
i
þ iωD1j þ J 1j E2j þ iωD2j þ J 2j E1j dV:
V
The second and fourth terms in the first integral of Eq. (5.70) cancel directly, as do
the two terms in the second integral since Bkn ¼ μH kn for n ¼ 1, 2. The integrand of
the third integral can be rewritten, using the constitutive relations, as
h i
iω djkl ∂u1k =∂xl þ cjk E1k E2j d jkl ∂u2k =∂xl þ cjk E2k E1j þ J 2j E1j J 1j E2j ;
where the above second and fourth terms cancel each other by the symmetry
property of cjk and the above first and third terms cancel the sixth and fifth terms,
respectively, in the first integral of Eq. (5.70) when the symmetry properties of dijk
are used, so collecting the only remaining terms, we have
ð
I ¼ f 2l v1l f 1l v2l þ J 2l E1l J 1l E2l dV: ð5:71Þ
V
Equating this result to that of Eq. (5.68), we have the reciprocal theorem for the
piezoelectric material given by
ð ð
1
τkl nk v2l τ2kl nk v1l dS þ εijk E2j H 1k ni εijk E1j H 2k ni dS
S ð S
¼ f 2l v1l f 1l v2l þ J 2l E1l J 1l E2l dV ð5:72Þ
V
The use of reciprocity relations and fundamental solutions for fluid, elastic solid,
and electrodynamic problems can be found in the reference handbook edited by
Varadan, Lakhtakia, and Varadan [7], where many other references to the literature
on related topics can also be found. There are also a number of texts [8–11] that give
more of the details needed to apply boundary element methods to solve the surface
integral equations that arise in flaw scattering problems.
5.5 Problems
5.1. (a) Consider the one-dimensional harmonic wave motion of a fluid medium
in a finite one-dimensional region (a, b). Let p1 and p2 be the solutions to the
following problems:
2
∂ p1 =∂x2 þ k20 p1 ¼ f 1 in ða; bÞ
2
∂ p2 =∂x þ2
k20 p2 ¼ f 2 in ða; bÞ;
where k0 is the wave number. Show that the one-dimensional form of the
reciprocal theorem (see Eqs. (5.7) and (5.42))
is given by
ðb
b
p1 ∂p2 =∂x p2 ∂p1 =∂xja ¼ ðp2 f 1 p1 f 2 Þdx:
a
where
8
< 1 x in ða; bÞ
α ¼ 1=2 x ¼ a or x ¼ b :
:
0 x not in ða; bÞ
(c) For the case where an incident wave pinc strikes a finite inclusion which
exists in the interval (x1, x2) in an infinite region ð1, þ 1Þ, as shown in
5.5 Problems 109
Fig. P5.1, show that if the scattered wave pscatt satisfies the 1-D radiation
condition for jxj ! 1 given by
∂p=∂jxj ikp ¼ 0
Fig. P5.1 One-dimensional scattering from a fluid “inclusion” in an otherwise homogeneous fluid
medium
and determine the analogous equation obtained when the scattered pressure
on the right side of the above equation is replaced by the total pressure.
5.2. Derive a reciprocal theorem analogous to Eq. (5.7) for the 2-D harmonic
wave motion of a fluid medium which occupies a two-dimensional region
S whose surrounding contour is C.
5.3. For two-dimensional harmonic wave motion (with expðiωtÞ time depen-
dency) of a fluid medium, show that the Sommerfeld radiation conditions in
2-D are given by
pffiffi
p ¼ Oð1= rÞ
pffiffi
∂p=∂r ikp ¼ oð1= rÞ
5.5. Derive a reciprocal theorem analogous to Eq. (5.42) for the 2-D (plane strain)
harmonic wave motion of an elastic medium which occupies a
two-dimensional region S whose surrounding contour is C.
5.6. Derive a reciprocal theorem analogous to Eq. (5.42) for the 2-D (anti-plane
strain) harmonic wave motion of an elastic medium which occupies a
two-dimensional region S whose surrounding contour is C. How does this
theorem differ from that for a fluid medium?
5.7. For two-dimensional (plane strain) harmonic wave motion (with expðiωtÞ
time dependency) of an elastic medium, what are the appropriate Sommerfeld
radiation conditions?
5.8. For two-dimensional (anti-plane strain) harmonic wave motion (with exp
ðiωtÞ time dependency) of an elastic medium, what are the appropriate
Sommerfeld radiation conditions? How do these conditions differ from those
for a fluid medium?
5.9. Consider the 2-D plane strain problem of a harmonic wave (compressional or
shear) incident on a homogeneous 2-D fluid inclusion in an infinite elastic
medium. Using the fundamental solutions for both the inclusion and the
surrounding medium derive a set of integral equations for the unknown fields
on the surface of the inclusion and show how they can be used to solve for the
wave field anywhere (in the inclusion or the surrounding medium).
5.10. Consider the 2-D plane strain problem of a harmonic wave (compressional or
shear) incident on a homogeneous 2-D elastic inclusion in an infinite elastic
medium. Using the fundamental solutions for both the inclusion and the
surrounding medium derive a set of integral equations for the unknown fields
on the surface of the inclusion and show how they can be used to solve for the
wave field anywhere (in the inclusion or the surrounding medium).
5.11. Consider the problem of a harmonic wave incident on a homogeneous 3-D
fluid inclusion in an infinite fluid medium. Using the fundamental solutions
for both the inclusion and the surrounding medium derive a set of integral
equations for the unknown fields on the surface of the inclusion and show
how they can be used to solve for the pressure wave field anywhere (in the
inclusion or the surrounding medium).
5.12. Consider the problem of a harmonic wave (compressional or shear) incident
on a homogeneous 3-D fluid inclusion in an infinite elastic medium. Using the
fundamental solutions for both the inclusion and the surrounding medium
derive a set of integral equations for the unknown fields on the surface of the
inclusion and show how they can be used to solve for the wave field anywhere
(in the inclusion or the surrounding medium).
5.13. Consider the problem of a harmonic wave (compressional or shear) incident
on a homogeneous 3-D elastic inclusion in an infinite elastic medium. Using
the fundamental solutions for both the inclusion and the surrounding medium
derive a set of integral equations for the unknown fields on the surface of the
References 111
inclusion and show how they can be used to solve for the wave field anywhere
(in the inclusion or the surrounding medium).
5.14. Derive an integral equation for a wave incident on an infinitesimally thin
rigid scatterer in a fluid by taking derivatives of the integral representation
theorem as done for a crack in an elastic medium.
References
When an incident bulk wave interacts with an interface, such as the boundary
between a fluid and a solid in immersion testing, both reflected and transmitted
bulk waves are generated. This Chapter will describe the general properties of those
reflected and refracted waves. We will first discuss extensively the case when a
plane wave is incident on a plane interface and then consider the transmission and
reflection of a Gaussian beam at a curved interface. In the process, a variety of
concepts will be introduced such as acoustic impedance, Snell’s Law, critical
angles, pulse distortion, and inhomogeneous waves. We will also show that at
high frequencies these plane wave and planar interface results are applicable to
more general cases with non-planar wave fronts and interfaces.
In this section and in Sect. 6.2 we will begin by analyzing the reflection and refraction
of pressure waves at an interface between two fluids. Although two-fluid problems of
this type are rarely present in ultrasonic NDE applications, dealing only with scalar
pressure waves will let us describe some of the basic physics of the reflection and
refraction processes without the burden of the extensive algebra of the fluid-solid and
solid-solid interface problems that will be discussed later.
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_6) contains supplementary material, which is available to authorized users.
Consider first the problem of a plane interface between two fluid media, whose
densities and wave speeds are ρm, cm, (m ¼ 1, 2), respectively. A plane harmonic
wave (of expðiωtÞ time dependency), pinc, having a known pressure amplitude Pi
and traveling in medium one, strikes the interface at normal incidence (Fig. 6.1).
We expect that part of this incident wave will be reflected back into medium one
and part will be transmitted (refracted) into the second medium. Since the reflected
wave will be traveling in the negative x-direction and the transmitted wave will be
traveling in the plus x-direction, the pressure in both these waves and the incident
wave can be written explicitly as
where the amplitudes Pr and Pt are unknown. At the interface both the pressure and
the normal (x-) component of the velocity must be continuous (recall Eqs. (3.13)
and (3.14)), so that on x ¼ 0
Pi þ P r ¼ P t
ð6:3Þ
ðik1 =ρ1 ÞPi ðik1 =ρ1 ÞPr ¼ ðik2 =ρ2 ÞPt :
2ρ2 c2
Pt =Pi ¼ T p ¼
ρ1 c 1 þ ρ2 c 2
ð6:4Þ
ρ c 2 ρ1 c 1
Pr =Pi ¼ Rp ¼ 2 ;
ρ1 c 1 þ ρ2 c 2
6.1 Reflection and Refraction at a Fluid-Fluid Interface (Normal Incidence) 115
where Tp and Rp are the transmission and reflection coefficients (the p subscript here
is to indicate these coefficients are based on pressure ratios). Although these
reflection and transmission coefficients were derived explicitly for an interface
between two fluid media, they are also applicable to fluid-solid and solid-solid
interface problems since, as shown in Chap. 4, 1-D plane waves in a solid also are
governed directly by wave equations and the boundary conditions are analogous to
those for a fluid. Note, however, that in the oblique incidence case, which will be
discussed later in this Chapter, there are significant differences between the bound-
ary conditions in the purely fluid case and fluid-solid and solid-solid cases, so that
reflection and transmission coefficients will have to be derived separately for those
problems.
Equation (6.4) shows that the amplitudes of the reflected and transmitted waves
are functions only of the products ρmcm, (m ¼ 1, 2), where zm ¼ ρm cm is called the
specific acoustic impedance. See Table 3.1 for acoustic impedance values of some
common materials. In terms of zm we have
2z2
Tp ¼
z1 þ z2
ð6:5Þ
z2 z1
Rp ¼ ;
z1 þ z2
which are plotted as a function of z2/z1 in Fig. 6.2. As Fig. 6.2 shows, at a fluid/
vacuum interface (z2 ¼ 0) the transmission coefficient is zero and the reflection
coefficient is minus one. If we transform these harmonic waves back into the time
domain as pulses via the inverse Fourier transform, i.e.,
preflt
preflt
ð
þ1
ð
þ1
¼ T p pinc ðt x=c2 Þ;
we see that a negative reflection coefficient merely means that the reflected pressure
pulse amplitude has the opposite sign of the incident wave everywhere (Fig. 6.3a).
It is evident from Fig. 6.2 this reversal of sign behavior occurs when the second
medium is of lower impedance (z2 < z1) than the first. When the second medium is
of higher impedance (Fig. 6.3b) both the reflected and transmitted wave pulses have
the same sign.
Also, from Table 3.1, we see that the acoustic impedance of air is so much
smaller than the impedance of the other common materials listed, that for all
practical purposes the air can be neglected and the interface treated as a material/
vacuum interface instead.
The normal incidence reflection and transmission coefficients derived in the last
section give the pressure amplitudes of the transmitted and reflected waves. If we
want to know how the energy of the incident wave is partitioned into these reflected
and transmitted waves, however, we need to obtain a different set of coefficients
based on “intensities” instead, which are defined in this section.
6.1 Reflection and Refraction at a Fluid-Fluid Interface (Normal Incidence) 117
Both p and vn in Eq. (6.7) must be real so that if we represent these quantities in
complex notation then we must write Eq. (6.7) in terms of their real (or imaginary)
parts explicitly. For example,
Π ¼ ReðpÞReðvn Þ; ð6:8Þ
or, explicitly
Pke
v¼ cos ðk x e ωtÞ; ð6:11Þ
ρω
so that we find
P2 ðe nÞ
Π¼ cos 2 ðk x e ωtÞ: ð6:12Þ
ρc
118 6 Reflection and Transmission of Bulk Waves
As can be seen from Eq. (6.12), the energy flux is a function of both x and t. The
acoustic intensity, I, of this wave is defined as this energy flux averaged over one
period of the wave:
ðT
1
I¼ Πdt; ð6:13Þ
T
0
ð
2π=ω
P2 ðe nÞω
I¼ cos 2 ðk x e ωtÞ∂t: ð6:14Þ
2πρc
0
But
1
cos 2 ðk x e ωtÞ ¼ ½1 þ cos f2ðk x e ωtÞg ð6:15Þ
2
and the integral of the cosine term in Eq. (6.15) vanishes, so that the constant first
term in Eq. (6.15) gives, finally
P2 ðe nÞ
I¼ : ð6:16Þ
2ρc
For the special case when the plane S is taken to be normal to the direction of the
wave, e ¼ n and we have
P2 P2
I¼ ¼ ð6:17Þ
2ρc 2z
in terms of the specific impedance, z. Now if we go back to the problem of the reflection
and refraction at normal incidence to an interface, and consider a cylindrical volume of
cross sectional area ΔA passing through the interface and normal to it (Fig. 6.5), we can
let Ii, Ir, and It be the intensities of the incident, reflected and transmitted waves,
respectively, across ΔA. By conservation of energy, therefore, we have
I i ΔA ¼ I r ΔA þ I t ΔA ð6:18Þ
or
P2i P2r P2
¼ þ t ð6:19Þ
2ρ1 c1 2ρ1 c1 2ρ2 c2
If we define reflection and transmission coefficients, RI and TI, in terms of the ratio
of the intensities as
I r P2r =2ρ1 c1
RI ¼ ¼ ¼ R2p
I i P2i =2ρ1 c1
ð6:21Þ
I t P2 =2ρ2 c2 ρ1 c1 2
T I ¼ ¼ t2 ¼ T
I i Pi =2ρ1 c1 ρ2 c2 p
RI þ T I ¼ 1: ð6:22Þ
Note that some authors prefer to use velocities (or displacements) to calculate the
reflection and transmission coefficients instead of the coefficients based on pressure
ratios used here. Since the pressure of the incident, reflected, or transmitted waves is
of the general form
p ¼ Pexpðikx iωtÞ ð6:23Þ
P
vx ¼ expðikx iωtÞ; ð6:24Þ
ρc
which would give vx ¼ P=ρc for a plane wave travelling in the + x direction and
vx ¼ P=ρc for a plane wave traveling in thex direction. However, it is common
to use the velocity in the direction of the traveling wave for both the reflected and
transmitted waves to define these reflection and transmission coefficients, so if we
let Vt be the velocity amplitude of the transmitted wave in the transmitted wave
120 6 Reflection and Transmission of Bulk Waves
direction and Vr be the velocity amplitude of the reflected wave in the reflected
wave direction, we have V t ¼ Pt =ρt ct , V r ¼ Pr =ρr cr for both these waves in terms
of their pressure amplitudes and acoustic impedances with no difference in sign.
Thus, with this convention we have the pressure and velocity amplitudes of all the
plane waves (incident, reflected, and transmitted) given by the same relationship,
P ¼ ρcV, as found in many texts on wave motion. In this book we will also use this
convention and define the velocity-based reflection and transmission coefficients
(Rv, Tv) as
Pr =ρ1 c1 ρ c2 ρ1 c1
Rv ¼ ¼ Rp ¼ 2
Pi =ρ1 c1 ρ1 c1 þ ρ2 c2
ð6:25Þ
Pt =ρ2 c2 ρ1 c1 2ρ1 c1
Tv ¼ ¼ Tp ¼ ;
Pi =ρ1 c1 ρ2 c2 ρ1 c 1 þ ρ2 c 2
which are also valid when the corresponding displacement ratios are used instead.
If a plane harmonic pressure wave strikes an interface between two fluid media at an
angle θi as shown in Fig. 6.6, the incident, reflected, and transmitted waves are
given by
y
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 121
where θr and θt are the reflected and transmitted angles, respectively. From the
continuity of pressure on the interface y ¼ 0, we find
For Eq. (6.27) to be satisfied at all x, the phase terms must all match, giving
and
sin θi sin θt
¼ : ð6:29Þ
c1 c2
Equation (6.28) is merely a statement that the angle of incidence equals the angle of
reflection while Eq. (6.29) is called Snell’s Law. From Eq. (6.27) the amplitudes
then satisfy
Pi þ Pr ¼ Pt : ð6:30Þ
Solving Eq. (6.32) for the reflection and transmission coefficients, we find
ρ2 c2 cos θi ρ1 c1 cos θt
Rp ¼
ρ2 c2 cos θi þ ρ1 c1 cos θt
ð6:33Þ
2ρ2 c2 cos θi
Tp ¼ :
ρ2 c2 cos θi þ ρ1 c1 cos θt
1=2
cos θt ¼ ð1 sin 2 θt Þ
1=2
¼ 1 c22 sin 2 θi =c21
122 6 Reflection and Transmission of Bulk Waves
both the reflection and transmission coefficients can be written in terms of θi only as
1=2
ρ2 c2 cos θi ρ1 c1 1 c22 sin 2 θi =c21
Rp ¼ 1=2
ρ2 c2 cos θi þ ρ1 c1 1 c22 sin 2 θi =c21
ð6:34aÞ
2ρ2 c2 cos θi
Tp ¼ 1=2
ρ2 c2 cos θi þ ρ1 c1 1 c22 sin 2 θi =c21
where tv, and rv are the transmission and reflection coefficients, ang is the incident
angle in medium one (in degrees), and (d1, c1) and (d2,c2) are the density and wave
speed pairs for the first and second media, respectively (units are arbitrary since
only their ratios are present in the reflection and transmission coefficients).
Although the problem being considered here involves the plane interface between
two fluids, let the properties of the first medium be water with density and wave
speeds d1 ¼ 1.0, c1 ¼ 1480 m/s and the second be aluminum (Al) with density and
(compressional) wave speeds d2 ¼ 2.7, c2 ¼ 6420 m/s, to produce the same change
of properties found in immersion testing setups. Writing these reflection and
transmission coefficients as complex numbers in the forms Rv ¼ jRv jexpðiϕR Þ, T v
¼ jT v jexpðiϕT Þ the magnitudes (|Rv|, |Tv|) and phases (ϕT, ϕR) of these reflection
and transmission coefficients generated by the MATLAB fluid_fluid function are
plotted versus the incident angle in Figs. 6.7 and 6.8. Note, however, all these
results are based on a fluid-fluid model and, as mentioned previously, except at
normal incidence we must use fluid-solid models to correctly represent the results
for fluid/solid interfaces.
For all incident angles that satisfy the condition sin θi c1 =c2 , Eq. (6.34) shows
that the reflection and transmission coefficients are both real. For a water-Al this
corresponds to an incident angle of about 13.3 and, as seen in Figs. 6.7 and 6.8, the
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 123
coefficients indeed are purely real (phases ¼ 0). This means that in this case the
reflected and transmitted waves are just plane waves whose amplitudes are propor-
tional to the incident plane wave. Note that the very large impedance mismatch
between the water and the aluminum causes the reflection coefficient to be close to
1.0 and the transmission coefficient to be small (approx. 0.16) so we would expect
that most of the incident wave is reflected back into the first medium.
For sin θi > c1 =c2 the square root terms in Eq. (6.34) are no longer real,
corresponding to the appearance of non-zero phases in both the reflected and
transmitted waves of Figs. 6.7 and 6.8, and we must interpret what these solutions
mean physically. The angle at which this change of behavior first occurs is called
the critical angle, θcr where
As long as the wave speed of the second medium is faster than that of the first
medium, Eq. (6.35) shows that a critical angle can indeed exist. Beyond this critical
angle, the cos θt term in Eq. (6.33) becomes imaginary so we need to examine the
meaning of this behavior. Note that this term came from the pressure term of the
transmitted wave given by
so that in order to guarantee that the transmitted pressure does not become infinitely
large for all frequencies (both positive and negative) as y ! 1, we must take, when
θi > sin 1 ðc1 =c2 Þ
1=2
cos θt ¼ isgnω c22 sin 2 θi =c21 1 ð6:37Þ
where
(
1 for ω > 0
sgnω ¼ : ð6:38Þ
1 for ω < 0
In this case Eq. (6.36) shows that the transmitted wave is no longer a plane wave
propagating into the second medium, but instead is an inhomogeneous wave which
travels along the interface and decays exponentially with depth from the interface
(Fig. 6.9) From Eq. (6.36) and Snell’s law, it follows that the pressure in this
inhomogeneous wave can be written as
1 ia
Rp ¼ ; ð6:40Þ
1 þ ia
y
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 125
Similarly, since
T p ¼ Rp þ 1
¼ expð2iχ Þ þ 1 ð6:43Þ
¼ 2 cos χexpðiχ Þ
When θi θcr , the incident plane wave generates both plane reflected and trans-
mitted waves. If we define a set of cylinders along the incident, reflected and
transmitted directions, as shown in Fig. 6.10, where the cylinders all intersect a
common area, ΔA, on the interface, then since
we find
cos θt
Ir þ It ¼ Ii ; ð6:47Þ
cos θi
126 6 Reflection and Transmission of Bulk Waves
where Ii, It and Ir are the incident, transmitted and reflected intensities. In terms of
the pressure-based reflection and transmission coefficients, Eq. (6.47) gives
ρ1 c1 cos θt 2
R2p þ T ¼ 1; ð6:48Þ
ρ2 c2 cos θi p
which can also be verified by direct substitution of the Rp and Tp of Eq. (6.33) into
Eq. (6.48).
Once the critical angle is exceeded, the transmitted wave in Fig. 6.10 no longer
exists and we must reconsider the energy terms in this situation. Since the incident
plane wave is unchanged, the incident intensity in the incident wave direction is
given by
In the reflected wave direction, the reflected wave carries the energy flux
P2i
Πr ¼ cos 2 ½ωðx sin θi =c1 y cos θi =c1 tÞ 2χ ð6:50Þ
ρ1 c 1
so that I r ¼ I i . Because the intensity of the reflected wave is identical to that of the
incident wave when the critical angle is exceeded, this situation is often referred to
as total reflection.
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 127
For the inhomogeneous wave in medium two, consider the velocity components
1 ∂ptrans 1 ∂ptrans
vx ¼ vy ¼ : ð6:52Þ
iωρ2 ∂x iωρ2 ∂y
Calculating these components for the inhomogeneous wave from Eq. (6.44), the
power flux across the interface y ¼ 0 (in the y-direction), Πy, is given by
Πy ¼ Re½ptrans Re vy
2 cos 2 χ P2i α sgn ω ð6:53Þ
¼ sin ½2ðk1 sin θi x ωt χ Þ;
ρ2 c2
which averages to zero over one complete cycle so I y ¼ 0 for the inhomogeneous
wave. Similarly, the power flux in the x-direction, Πx, is given by
Πx ¼ Re½ptrans Re½vx
4 cos 2 χ P2i sin θi ð6:54Þ
¼ exp½2jωjα y=c2 cos 2 ½k1 sin θi x ωt χ
ρ2 c1
which, when averaged over one cycle, gives the intensity, Ix, for the inhomogeneous
wave as:
2 cos 2 χ P2i sin θi
Ix ¼ exp½2jωjαy=c2 : ð6:55Þ
ρ2 c 1
When the incident angle exceeds the critical angle, the appearance of the frequency
dependent cos θt term in both the reflected wave and inhomogeneous wave expres-
sions has a profound effect on the shape of the reflected and transmitted pulses
which correspond to these frequency domain terms. To see this, first consider the
reflected wave (Eq. (6.42)) which can be rewritten as
where the phase term ur ¼ ðt x sin θi =c1 þ y cos θi =c1 Þ and we can extract the
frequency dependent sgn ω term explicitly by using the definition of χ in both
cos 2χ and sin 2χ:
and
sin 2χ ¼ sin ½2sgnω tan 1 ðz1 α=z2 cos θi Þ
¼ sgnω sin ½2 tan 1 ðz1 α=z2 cos θi Þ ð6:57bÞ
¼ sgnω sin 2χ 0
Following a similar procedure for the inhomogeneous wave expression (Eq. (6.44)),
we find
with ut ¼ t x sin θi =c1 . Now, let Pi ¼ FðωÞ=2π, where F(ω) is the Fourier
transform of a time domain function, f(t), and integrate over all frequencies:
ð
þ1
cos 2χ 0
preflt ðtÞ ¼ FðωÞexpðiωur Þdω
2π
1
ð6:60Þ
ð
þ1
i sin 2χ o
FðωÞsgnωexpðiωur Þdω:
2π
1
ð
þ1
1
FðωÞexpðiωur Þdω ¼ f ður Þ; ð6:61Þ
2π
1
so now consider the second integral. Using the definition of the Fourier transform
and interchanging orders of integration we have
8 þ1 9
ð
þ1 ð
þ1
<ð =
FðωÞsgnωexpðiωur Þdω ¼ f ðtÞ sgnωexp½iωðτ ur Þdω dτ:
: ;
1 1 1
ð6:62Þ
The inner integral does not formally converge but consider it as the limit of the
following integral which does converge:
ð
þ1
I ¼ lim sgn ω exp jωjε exp½iωðτ ur Þdω: ð6:63Þ
ε!0
1
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 129
Since the sgn function is an odd function we can rewrite Eq. (6.63) as
ð
þ1
I ¼ lim 2i exp jωj ε sin ½ωðτ ur Þdω; ð6:64Þ
ε!0
0
2i
I¼ : ð6:65Þ
τ ur
Placing this result into Eq. (6.62) and using Eq. (6.61), Eq. (6.60) then becomes
where
ð
þ1
1 f ðτÞdτ
H½f ; ur ¼ ð6:67Þ
π τ ur
1
is the Hilbert transform of f and the integral in Eq. (6.67) is interpreted in the
principal value sense (note that the bold symbol used here for the Hilbert transform
does not mean it is a vector quantity). Appendix D lists a number of functions and
their Hilbert transforms. Because of the existence of the Hilbert transform term in
Eq. (6.66) the shape of the reflected pulse is no longer of the same form as the
incident pulse. This phenomena is known as pulse distortion and arises from the
fact that the reflection coefficient was frequency dependent. Below the critical
angle, such distortion, of course, disappears. Later, we will give a physical expla-
nation for the existence of this distortion.
In the same fashion we can consider the inhomogeneous wave where we have
ð
þ1
2 cos 2 χ 0
ptrans ðtÞ ¼ FðωÞexpðjωjαy=c2 Þexpðiωut Þdω
2π
1
ð6:68Þ
ð
þ1
2i cos χ 0 sin χ 0
FðωÞsgnωexpðjωjαy=c2 Þexpðiωut Þdω:
2π
1
Again using the relationship between F and f and interchanging orders of integra-
tion gives for the first term in Eq. (6.68)
81 9
ð
þ1
<ð =
2 cos 2 χ 0
f ðτ Þ cos ½ωðτ ut Þexpðωαy=c2 Þdω dτ; ð6:69Þ
2π : ;
1 0
130 6 Reflection and Transmission of Bulk Waves
ð
1
αy=c2
cos ½ωðτ ut Þexpðωαy=c2 Þdω ¼ ; ð6:70Þ
ðτ ut Þ2 þ α2 y2 =c22
0
ð
þ1
2 cos 2 χ 0 f ðτÞαy=c2 dτ
: ð6:71Þ
π ðτ ut Þ2 þ α2 y2 =c22
1
Following an entirely similar procedure for the second term of Eq. (6.68), we obtain
81 9
ð
þ1
<ð =
2 cos χ 0 sin χ 0
f ðτ Þ sin ½ωðτ ut Þexpðjωjαy=c2 Þdω dτ; ð6:72Þ
π : ;
1 0
ð
1
ð τ ut Þ
sin ½ωðτ ut Þexpðjωjαy=c2 Þdω ¼ ð6:73Þ
ðτ ut Þ2 þ α2 y2 =c22
0
ð
þ1
2 cos χ 0 sin χ 0 ðτ ut Þf ðτÞdτ
: ð6:74Þ
π ðτ ut Þ2 þ α2 y2 =c22
1
Placing Eqs. (6.74) and (6.71) back into Eq. (6.68) for the inhomogeneous wave and
writing the coefficients in terms of the angle 2χ 0 we have
ð
þ1
ð cos 2χ 0 þ 1Þ f ðτÞαy=c2 dτ
ptrans ðtÞ ¼
π ðτ ut Þ2 þ α2 y2 =c22
1
ð6:75Þ
ð
þ1
sin 2χ 0 ðτ ut Þf ðτÞ dτ
þ :
π ðτ ut Þ2 þ α2 y2 =c22
1
It is interesting to note that the integrand of the first term in Eq. (6.75) represents a
delta function in the limit as y ! 0, i.e.,
1 αy=c2
lim ¼ δ ð τ ut Þ ð6:76Þ
y!0 π ðτ ut Þ2 þ α2 y2 =c22
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 131
and the integrand of the second term becomes just the Hilbert transform of f, so that
on y ¼ 0
Combining this result with Eq. (6.66) for the reflected wave, on y ¼ 0 we find that
indeed the pressure is continuous at the interface since
The behavior of the reflected and inhomogeneous waves when we are past the
critical angle raises some interesting questions. First, from our results on energy
flux (Eqs. (6.51) and (6.55)) we found that the intensity of the reflected wave was
equal to the intensity of the incident wave (total reflection) yet there was also a
non-zero energy flux parallel to the surface in the inhomogeneous wave. Where did
this additional energy in the inhomogeneous wave come from? Second, the exis-
tence of pulse distortion means that the reflected wave often has a “precursor”
behavior which begins at t ¼ 1 even when the incident pulse is identically zero
until some finite time. For example, if the incident pulse is a delta function, δ(t),
(see Appendix D) the Hilbert transform is given by 1=π t which is plotted in
Fig. 6.11, showing the precursor explicitly. How can such a disturbance exist in the
reflected wave before the incident wave arrives?
To answer these questions, it is useful to consider a problem where a point
source acts near an interface as shown in Fig. 6.12. At early times when the incident
(curved) wave front is incident on the interface at an angle below the critical angle
(Fig. 6.12a) the incident, reflected, and transmitted wave fronts are all attached at
the interface and model approximately our plane wave reflection problem locally at
the interface. However, at later times (Fig. 6.12b), if the wave speed of the second
material is faster than the first, the transmitted wave outruns the incident wave and
generates a head wave which precedes the reflected wave as shown. In fact, at much
later times this head wave can be infinitely distant from the reflected wave in the
first material. Thus, the precursor found previously is just a residual head wave
contribution to the disturbance in material one when the head wave is at infinity.
b
reflected wave incident wave
ρ1, c1
head wave
ρ2, c2
Figure 6.12b also shows schematically that beyond the critical angle the energy flux
from the source into the inhomogeneous wave in the second medium, follows a path
quite different from that of the energy of the reflected wave. Therefore, the energy
of the inhomogeneous wave comes from the source by a separate route, not from the
incident wave front directly, and there is no violation of energy balances when the
incident wave is totally reflected.
In ultrasonic pulse-echo immersion testing, the waves scattered from a flaw return
to the transducer along the same path taken during transmission (Fig. 6.13).
Assume, for the present, that the waves traveling in both directions can be consid-
ered to be plane waves and let the two materials involved be modeled as fluids.
Then, if an interface is crossed from material one to material two as shown, pressure
reflection and transmission coefficients, R12 and T12, respectively, will be involved
in going from the transducer to the flaw. Similarly, pressure reflection and trans-
mission coefficients, R21 and T21, respectively, will be involved in going from the
flaw back to the transducer, i.e., from material two to material one, on a completely
reversed path. Although we can independently solve for each of these “reverse”
reflection and transmission coefficients, this is not necessary since they are directly
related to the original coefficients through the so-called Stokes’ relations. To obtain
Stokes’ relations, consider first the waves generated in going from material one to
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 133
v v
c d
2 T12T21
R12 R12
θi θr θi
x x
material two (Fig. 6.14a). We note that the solution for this problem is completely
unchanged if we simply change the direction of every wave in Fig. 6.14a. Thus, the
waves shown in Fig. 6.14b also satisfy all the same equations and boundary
conditions as the original problem. Now, consider the two auxiliary problems
shown in Fig. 6.14c, d, which involve incident waves in medium one and two,
respectively, traveling along the same reversed paths as shown in Fig. 6.14b. If we
add the solutions for the waves in these two auxiliary problems and compare the
result to Fig. 6.14b, we see that we must have
and
R212 þ T 21 T 12 ¼ 1: ð6:79bÞ
which is the first Stokes relation. Also, from the conservation of energy (Eq. (6.48))
we found that
ρ1 c1 cos θt 2
R212 þ T ¼1 ð6:81Þ
ρ2 c2 cos θi 12
ρ1 c1 cos θt
T 21 ¼ T 12 ; ð6:82aÞ
ρ2 c2 cos θi
which is the second Stokes’ relation. Note that here the transmission coefficients are
based on pressure ratios. If we use velocity ratios instead we can write the
transmission coefficients with a v superscript and Stokes’ second relation is
ρ2 c2 cos θt v
v
T 21 ¼ T 12 : ð6:82bÞ
ρ1 c1 cos θi
Since the incident and reflected waves are traveling in the same medium, the first
Stokes’ relation is unaffected by a change to velocity ratios instead of pressure
ratios for the reflection coefficients. All of these relations, of course, can be verified
by direct solution of the problem where an incident wave in material two strikes the
interface at an angle θt. Obtaining such reversed solutions is very easy for the fluid-
fluid interface case so that the economy of using Stokes’ relations is not very great
here. For more complicated fluid-solid and solid-solid interface problems, however,
Stokes’ relations are a very convenient way to obtain reversed transmission coef-
ficients when they are needed.
In treating the reflection and refraction at an interface between two fluid media, we
considered a local (x, y, z) coordinate system (recall Fig. 6.6) where all the waves
were assumed to lie in the xy plane (see Fig. 6.15). In this section we will
reconsider the solution of that interface problem in a general three-dimensional
setup where all measurements are with respect to an arbitrary fixed coordinate
system (X, Y, Z) instead (Fig. 6.15). This generalization is important since all NDE
wave problems are inherently three-dimensional and we will use plane waves as
building blocks for more general problems such as the sound beams generated by
ultrasonic transducers. In this (X, Y, Z) system we can locate an arbitrary point P in
either medium by the position vector, X, (Fig. 6.16) and write the pressure in the
incident, reflected, and transmitted waves as
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 135
ptrans
n et
y
ρ 2 , c2
x ρ2 , c2
n
P
n ρ2 , c2
where ei, er, and et are unit vectors in the incident, reflected, and transmitted
directions, respectively. The planar interface is defined as those points Xp where
Xp n ¼ D and where D is the perpendicular distance from O0 to O in the n direction
(Fig. 6.17). From the boundary conditions
136 6 Reflection and Transmission of Bulk Waves
ti ¼ tr ¼ tt ð6:88Þ
and
Since the unit vector ti lies in the plane of incidence formed by ei and n, Eq. (6.88)
says that the reflected and transmitted waves also lie in this plane (Fig. 6.18), a fact
that we implicitly assumed before (see Fig. 6.6). Equation (6.89) is then the
statement in 3-D that the angle of incidence equals angle of reflection for the
reflected wave and a statement of Snell’s law for the transmitted wave. Now
consider the unit vectors er and ei. Since their magnitudes are both unity, we have
ti
et
n interface
plane of incidence
Using Eq. (6.89) we can eliminate the tangential components and solve for er n as
e r n ¼ ð e i nÞ ð6:91Þ
Since the reflected wave must be traveling away from the interface (in a sense
opposite to the incident wave), we must choose the minus sign. Then, since the
tangential components of ei and er are equal, we have
e i ðe i n Þ n ¼ e r ðer n Þ n ð6:92Þ
ð e i nÞ 2 þ ð e i t i Þ 2 ¼ ð e t nÞ 2 þ ð e t t t Þ 2 ¼ 1 ð6:94Þ
In this case the choice of sign depends on the direction of n. For n as shown in
Fig. 6.18, we must choose the positive square root value since the transmitted wave
must be traveling away from the interface in the same sense as n. Once the
appropriate sign is chosen in Eq. (6.95), et can be written explicitly in terms of ei
and n as
Now, going back to the boundary conditions, after phase matching we find
and
2ρ2 c2 ðei nÞ
Pt ¼ Pi exp½ik1 Dðei nÞ ik2 Dðet nÞ
ρ2 c2 ðei nÞ þ ρ1 c1 ðet nÞ ð6:99Þ
¼ T p Pi exp½ik1 Dðei nÞ ik2 Dðet nÞ:
where, from Eqs. (6.98–99) and Eq. (6.95), Rp and Tp can be written as functions
only of ðei nÞ. Thus, the reflected and transmitted waves are explicitly (omitting
the expðiωtÞ time dependency)
The general phase terms found in the 3-D fluid-fluid interface problem of the
previous section will be particularly useful when modeling a variety of later
problems since the form of such phase expressions are the same for both fluid
and solid media cases. Here, we wish to consider the interface problem shown in
0 0 0
Fig. 6.19, where a plane wave travels from point y ¼ x ; y ; z to point x ¼ ðx; y; zÞ
through an interface, which for convenience has been located with its normal
n along the z-axis. If we let k2 et ¼ p2 and k1 ei ¼ p1 then since jp2 j ¼ k2 and jp1 j
¼ k1 we have in the (x, y, z) coordinate system of Fig. 6.19
p1 ¼ px ex þ py ey þ p1z ez
ð6:101Þ
p2 ¼ px ex þ py ey þ p2z ez ;
where
6.2 Reflection and Refraction at a Fluid-Fluid Interface (Oblique Incidence) 139
1=2
p1z ¼ k21 p2x p2y
1=2 ð6:102Þ
p2z ¼ k22 p2x p2y
and px, py are the same for both p1 and p2 by Snell’s law. Now, consider the phase
term of a wave transmitted through the interface (Eq. (6.100)):
Using the definitions of p1 and p2, ϕ can then be written in terms of px and py only as
ϕ px ; py ¼ p2 X þ p1z D p2z D; ð6:104Þ
If we fix points x and y and let px and py vary then we will now show that of all
the possible values of the transmitted phase, the value of ϕ when it is stationary, i.e.,
∂ϕ=∂px ¼ ∂ϕ=∂py ¼ 0, is when ϕ corresponds to the one path from y to x through
the interface that satisfies Snell’s law (Fig. 6.20). To prove this result, we let
D2
x (x,y,z)
D
ρ1, c1 ρ2, c2
interface
140 6 Reflection and Transmission of Bulk Waves
0 0
∂ϕ=∂px ¼ x x px D=p1z px z z D =p2z ¼ 0
0 0 ð6:106Þ
∂ϕ=∂py ¼ y y py D=p1z py z z D =p2z ¼ 0:
which is nothing more than a statement of Snell’s law for a wave traveling from y to
xI and xI to x, where xI ¼ ðxI ; yI ; zI Þ is on the interface (Fig. 6.20). To see that
Eq. (6.107) is indeed a solution to the equations of Eq. (6.106), we first note that at
the stationary phase point, we have
" 0 2 #
0 2 1=2
xI x yI y k1 D
p1z ¼ k1 1 ¼
D21 D21 D1
" #1=2 ð6:108Þ
0
ðx xI Þ2 ðy yI Þ2 k2 z z D
p2z ¼ k2 1 ¼
D22 D22 D2
so that, for example, placing these results and Eq. (6.107) back into the first
expression of Eq. (6.106), we find it is satisfied identically, i.e.,
0 0
0
k1 xI x D=D1 k2 ðx xI Þ z z D =D2
xx
k1 D=D k2 ðz z0 DÞ=D2
0 0
1
¼ x x xI x ðx xI Þ ¼ 0:
which is also satisfied identically. To obtain the value of the phase ϕ at the
stationary phase point, we first rewrite Eq. (6.104) as
However, since Snell’s law is satisfied at the stationary point, we may express the
first term in Eq. (6.109) as
6.3 Reflection and Refraction at a Fluid-Solid Interface (Oblique Incidence) 141
Recognizing that the solutions of Eq. (6.107) imply that at the stationary phase
point
ϕ ¼ k1 D1 þ k2 D2 ; ð6:113Þ
where recall D1 and D2 are distances from y to the interface and from the interface
to x along a “ray” that satisfies Snell’s law. This intimate connection that we have
shown to exist between stationary phase conditions and Snell’s law is at this point
only a formal result. However, in later Chapters, we will see that this relationship is
of significant practical importance in the modeling of the propagation of bounded
beams of ultrasound through a planar interface. Also, we should point out that our
result here is equivalent to Fermat’s principle that states that a ray wave path
between two fixed points is a path of stationary time between those points, a
principle that is frequently used in the field of seismology.
θs2 Ψtrans
y
pinc ¼ Pi exp ikp1 x sin θp1 þ y cos θp1 iωt
preflt ¼ Pr exp ikp1 x sin θp1 y cos θp1 iωt
ð6:114Þ
ϕtrans ¼ At exp ikp2 x sin θp2 þ y cos θp2 iωt
ψ trans ¼ Bt exp½iks2 ðx sin θs2 þ y cos θs2 Þ iωt;
where we have anticipated the results of phase matching on the interface, as seen in
the fluid-fluid case, and set the angle of the reflected pressure wave with respect to
the y-axis equal to the corresponding angle of the incident wave. Recall, for the
solid the stresses τyy, τxy and the vertical displacement uy are given in terms of the
potentials by
h
i
2 2
τyy ¼ ρc2s2 ω2 ϕ=c2s2 2 ∂ ϕ=∂x2 þ ∂ ψ=∂x∂y
h i
2 2 2
τxy ¼ ρc2s2 2 ∂ ϕ=∂x∂y þ ∂ ψ=∂y2 ∂ ψ=∂x2 ð6:115Þ
uy ¼ ∂ϕ=∂y ∂ψ=∂x
1 ∂p
uy ¼ : ð6:116Þ
ρω2 ∂y
As in the fluid-fluid problem, phase matching the incident and transmitted waves at
the interface will lead to a generalized Snell’s law given by:
6.3 Reflection and Refraction at a Fluid-Solid Interface (Oblique Incidence) 143
The interface conditions, then reduce to a set of linear equations for the unknown
wave amplitudes which can be put in matrix form as
2 32 Pr 3 2 cos θp1 3
cos θp1 cos θp2 sin θs2 P
6 ρ1 cp1 ω2 76 7 6 ρ1 cp1 ω2 i 7
6 cp2 cs2 6
76 7 6 7
6 76 7 6 7
6 76 7 6 7
6 ρ2 ω2 cos 2θs2 ρ2 ω2 sin 2θs2 76 7 6 Pi 7
76 At 7 6 7:
1
6
6 76 7¼6 7
6 76 7 6 7
6 76 7 6 7
4 sin 2θp2 cos 2θs2 54 7 6 7
0 5 4 5
c2p2 c2s2 0
Bt
ð6:119Þ
Pr Δ2 Δ1
¼
Pi Δ2 þ Δ1
At 2cp2 cos θp1 cos 2θs2
¼ ð6:121Þ
Pi ρ1 cp1 ω2 Δ
Bt 2c2s2 cos θp1 sin 2θp2
¼ :
Pi ρ1 cp1 cp2 ω2 Δ
The above amplitude ratios in the solid involve the potential amplitudes. Trans-
mission coefficients based on stress/pressure ratios would be instead
where Tnn and Tns are the normal and shear stresses in the P and S waves respec-
tively (see Eqs. (4.28) and (4.32)) and Tα;β
mn denotes the transmission coefficient
going from medium m to medium n for a transmitted wave of type α due to an
144 6 Reflection and Transmission of Bulk Waves
incident wave of type β. These transmission coefficients are similar to the pressure
ratio coefficients used in the fluid-fluid case although note that there is a change in
sign because of the use of (tensile) stress/pressure ratios here instead of the
pressure/pressure ratios used in the fluid-fluid case. If we use velocity amplitude
ratios instead to define these transmission coefficients we have
ðn est Þ
e⊥ ¼ : ð6:125Þ
jðn est Þj
If we include the phase terms of the transmitted wave in a coordinate invariant form
as in Eq. (6.100), the complete displacement vectors uα for a transmitted wave of
type α can be written (omitting the expðiωtÞ time dependency) as
x x
n
s
d
θp2 θs2
ept = dp est
y y
transmitted P-wave transmitted SV-wave
Fig. 6.22 Unit vectors eαt ðα ¼ P, SÞ in the direction of propagation for the transmitted P- and SV-
waves and the corresponding polarization unit vectors dα
6.3 Reflection and Refraction at a Fluid-Solid Interface (Oblique Incidence) 145
uα ¼ Uα exp ikα2 X eαt þ ikp1 D epi n ikα2 Dðeαt nÞ
α;P
T 12 Pi α
¼ d exp ikα2 X eαt þ ikp1 D epi n ikα2 Dðeαt nÞ
iωρ2 cα2
v α;P
T 12 Pi
¼ dα exp ikα2 X eαt þ ikp1 D epi n ikα2 Dðeαt nÞ ;
iωρ1 cp1
ð6:126Þ
where kαm ¼ ω=cαm (m ¼ 1,2). In a similar fashion we can write the pressure in the
reflected wave in coordinate invariant form as
preflt ¼ RP;P
12 Pi exp ik p1 X epr þ 2ik p1 D epi n ð6:127Þ
α;β
where Rmn is the reflection coefficient (based on a pressure ratio) for a reflected
wave of type α due to a wave of type β when reflecting off an interface between
medium m (containing the incident wave) and medium n.
Consider now the behavior of the reflected and transmitted waves. From gener-
alized Snell’s law it follows that there are two critical angles present in the solid
medium. Beyond the first critical angle, ðθcr Þ1 ¼ sin 1 cp1 =cp2 , the transmitted P-
wave turns into an inhomogeneous wave (Fig. 6.23) and the terms
h 2 i1=2
cos θp2 ¼ isgn ω sin 2 θp1 cp2 =cp1 1
ð6:128Þ
2i sgn ω cp2 sin θp1 h 2 i1=2
sin 2θp2 ¼ sin 2 θp1 cp2 =cp1 1
cp1
Pt St
(inhomogeneous wave)
146 6 Reflection and Transmission of Bulk Waves
Pt, St
(inhomogeneous waves)
Table 6.1 Critical angles for Materials (θcr)1 (deg) (θcr)2 (deg)
(a) water-steel and (b) water-
Water-steel 14.53 27.55
aluminum
Water-aluminum 13.33 29.13
h 2 i1=2
cos θs2 ¼ isgn ω sin 2 θp1 cs2 =cp1 1
ð6:129Þ
2i sgn ω cs2 sin θp1 h 2 i1=2
sin 2θs2 ¼ sin 2 θp1 cs2 =cp1 1
cp1
also become imaginary and functions of frequency. Thus, pulse distortion is present
whenever any of these critical angles are exceeded. To illustrate the type of critical
angles present for interfaces between water and typical structural materials,
Table 6.1 gives the critical angles for water-steel and water-aluminum cases.
The MATLAB® function fluid_solid implements the velocity-based transmis-
sion coefficients for a fluid-solid interface, which are the coefficients of particular
interest in immersion testing setups. The calling sequence for this function is
P;P v S;P
where (tpp, tps) are the transmission coefficients (vT12 , T12 ), respectively, (d1, d2)
are densities of the fluid and solid, cp1 is the compressional wave speed of the fluid,
and (cp2, cs2) are the compressional and shear wave speeds of the solid. Figure 6.25
shows the behavior of the magnitude of these two transmission coefficients for a
water-Al interface, where we can see the change in their behavior at the two critical
angles listed in Table 6.1. Figure 6.26 shows an expanded view of the magnitude of
these coefficients for angles below the second critical angle. It can be seen that the
tpp coefficients has a magnitude that is approximately constant for most angles
below the first critical angle, a behavior we also saw in the fluid-fluid case
(Fig. 6.7). In both the fluid-fluid and fluid solid cases, therefore, the magnitudes
of these velocity-based transmission coefficients for the transmitted P-wave are
nearly the same as the normal incidence (θi ¼ 0) case where these coefficients are in
fact identical. The magnitude of the transmitted shear wave (Fig. 6.26) is zero at
normal incidence (no mode conversion) and increases approximately linearly for
angles below the first critical angle. The phases of the transmitted waves are shown
6.3 Reflection and Refraction at a Fluid-Solid Interface (Oblique Incidence) 147
In an elastic solid, the instantaneous power/ unit area delivered by a wave across a
surface S with unit normal, n, is given by
Π ¼ tðnÞ v; ð6:130Þ
where t(n) is the traction vector and v is the velocity. Alternately, in terms of the
stress and velocity components we have
Π ¼ τkl nk vl ð6:131Þ
v ¼ V e e f ðt x e=c1 Þ
ð6:132Þ
tk ¼ T kl il f ðt x e=c1 Þ;
where Ve is the velocity amplitude and Tkl are the stress amplitude components with
respect to the coordinate planes normal to the unit vectors il. Then, in terms of
components with respect to these coordinates
and Eq. (6.132), one finds that the stress and velocity amplitudes are related through
which gives
T lk ek ¼ ρc1 V e el : ð6:135Þ
which is of an identical form as for the case of a fluid (see Eq. (6.12)). Thus, for a
harmonic wave f ¼ cos ½ωðx e=c1 tÞ and n ¼ e, the intensity I is given by
ρc1 V 2n T2
I¼ ¼ nn ; ð6:137Þ
2 2ρc1
where T nn ¼ ρc1 V n is the normal stress of the wave with respect to its own wave
front.
Now, consider instead a plane S-wave traveling in the e direction with polariza-
tion in the t direction (Fig. 6.28). Then
v ¼ V t t f ðt x e=c2 Þ
ð6:138Þ
tk ¼ T kl il f ðt x e=c2 Þ:
Following the same steps followed in deriving Eq. (6.135), in this case we find
T kl ek ¼ ρc2 V t tl : ð6:139Þ
so that again if f ¼ cos ½ωðx e=c2 tÞ and e ¼ n, the intensity of the wave is given
by
ρc2 V 2t T2
I¼ ¼ nt ; ð6:144Þ
2 2ρc2
P2i P2r T2 T2
ΔAp1 ¼ ΔAp1 þ nn ΔAp2 þ nt ΔAs2 : ð6:145Þ
2ρ1 cp1 2ρ1 cp1 2ρ2 cp2 2ρ2 cs2
In this case the areas of the various cylindrical “beams” are related to the common
area, ΔA, on the interface through
Δ As 2 Tnn
Tnt
6.3 Reflection and Refraction at a Fluid-Solid Interface (Oblique Incidence) 151
P2i cos θp1 P2r cos θp1 T 2nn cos θp2 T 2nt cos θs2
¼ þ þ ð6:147Þ
2ρ1 cp1 2ρ1 cp1 2ρ2 cp2 2ρ2 cs2
For the fluid-fluid interface, we found that the transmission coefficient in going
from material one to material two could be related directly to the corresponding
transmission coefficient when going from material two to material one on a
completely reversed path, a situation, as we remarked previously, could be found
when conducting a pulse-echo immersion inspection (Fig. 6.13). In this case we
consider Stokes’ relations for transmission through a fluid-solid interface. The
original waves and their pressure (or stress) amplitudes (for an incident wave of
unit pressure amplitude) are shown in Fig. 6.30. As in the fluid-fluid case, the
solution for this problem is unchanged in form if we reverse the directions of all the
waves, resulting in the solution shown in Fig. 6.31a. This same reversed solution
can be obtained as a superposition of the three problems shown in Fig. 6.31b–d,
where the wave amplitudes are given in terms of the appropriate reflection and
transmission coefficients. Thus, considering the reversed incident wave, we obtain
P;P
θp 2 T 12 P;P P;P θp 2
T 12 R 12
θs 2
S;P
T 12 S;P P;P
θs 2
T 12 R 12
c d
P;P P;P P;S S;P
T 21 T 12 T 21 T 12
θp 1 θp 1
2
RP;P
12 þ T P;P P;P P;S S;P
21 T 12 þ T 21 T 12 ¼ 1: ð6:149Þ
Note that although this result is for reflection and transmission coefficients based on
P;P P;P S;P
pressure ratios (for R12 ) and stress/pressure or pressure/stress ratios (for (T12 , T12 )
P;P P;S
and (T21 , T21 ), respectively) expressing all these coefficients in velocity or dis-
placement ratios does not change Eq. (6.149) since the terms that represent those
changes occur in canceling pairs.
When combined with the energy balance equation (Eq. (6.148)), we obtain the
set of Stokes’ relations
But these Stokes, relation do depend on the choice of the coefficients, as found in
the fluid-fluid case. In Eq. (6.150a) the coefficients involve pressure/stress or stress/
pressure ratios. For coefficients based on velocity (or displacement) ratios we have
instead
6.4 Reflection and Refraction at a Solid-Solid Interface (Smooth Contact) 153
Angle beam transducers used in NDE testing consist of a contact P-wave transducer
placed at an angle on a solid wedge as shown in Fig. 6.32. The wedge itself is kept
in contact with a solid to be tested through a thin layer of fluid couplant. We will
model a similar setup here where we consider the response of a planar interface
between two solids to an incident plane P-wave, and where the thickness of the fluid
couplant will be neglected. The influence of the fluid on the transmission process
will instead be included by considering the two solids to be in smooth contact, i.e.,
the shear stress will be assumed to vanish at the interface. In this case, the incident,
reflected, and shear waves at the interface are shown in Fig. 6.33. The waves in this
case (in terms of the potentials ϕ and ψ) are
Pt
St
x
thin fluid layer of
thickness t ≅ 0. θp 2 φtrans
ρ2, cp 2, cs 2 θs 2 Ψtrans
y
154 6 Reflection and Transmission of Bulk Waves
ϕinc ¼ Ai exp ikp1 x sin θp1 þ y cos θp1 iωt
ϕreflt ¼ Ar exp ikp1 x sin θp1 y cos θp1 iωt
ψ reflt ¼ Br exp½iks1 ðx sin θs1 y cos θs1 Þ iωt ð6:151Þ
ϕtrans ¼ At exp ikp2 x sin θp2 þ y cos θp2 iωt
ψ trans ¼ Bt exp½iks2 ðx sin θs2 þ y cos θs2 Þ iωt
Applying these boundary conditions and phase matching all the terms on the
boundary gives again generalized Snell’s law, which in this case can be written as
Defining
" #
cp1 cos θp2 c2s1 sin 2θs1 sin 2θp1
Δ1 ¼ cos 2θs1 þ
2
cp2 cos θp1 c2p1
" #
ρ2 c2s2 sin 2θs2 sin 2θp2
Δ2 ¼ cos 2θs2 þ
2
ρ1 c2p2 ð6:155Þ
" #
cp1 cos θp2 c2s1 sin 2θs1 sin 2θp1
Δ3 ¼ cos 2θs1
2
cp2 cos θp1 c2p1
Δ ¼ Δ1 þ Δ2 ;
6.4 Reflection and Refraction at a Solid-Solid Interface (Smooth Contact) 155
When the incident angle is below the first critical angle, both transmitted compres-
sion and shear waves will be present and such an angle beam transducer (Fig. 6.32)
would be operating in a mixed mode configuration. However, in most uses of such a
transducer, the incident angle is chosen so that only the first critical angle ðθcr Þ1
¼ sin 1 cp1 =cp2 is exceeded. In this case the transmitted P-wave becomes an
inhomogeneous wave and there is only a transmitted shear wave in the second solid
(Fig. 6.34). A transducer operating in this configuration is called an angle beam
shear wave transducer. One common use of angle beam shear waves is in the testing
of welds in plates and pipes. Note that we have not discussed what happens to the
reflected P- and S-waves in the first material (the wedge). If those waves were to
Pt
St
156 6 Reflection and Transmission of Bulk Waves
Pt , St
pass into the part to be tested, they would also produce a response from whatever
reflectors or flaws are present in the second material, and could make interpretation
of the received response more difficult. Thus, various means are used in practice to
try to minimize or eliminate these reflections, including making the geometry of the
wedge such that the path to escape from the wedge is long (Fig. 6.35a), grooving the
wedge surface to scatter the reflected waves in many directions (Fig. 6.35b), or
adding absorbing material to eliminate any further reflections (Fig. 6.35c).
If the incident angle is chosen such that the second critical angle ðθcr Þ2 ¼ sin 1
cp1 =cs2 is also exceeded (Fig. 6.36) then both inhomogeneous P- and S-waves are
produced traveling on the surface of the second material. In the next chapter we will
see how this configuration is also of practical importance for generating Rayleigh
(surface) waves.
The MATLAB function solid_f_solid implements the velocity-based transmis-
sion coefficients of Eq. (6.156b). The calling sequence for this function is
>> [tpp, tps] ¼ solid_f_solid(iangd, d1, d2, cp1, cs1, cp2, cs2);
6.5 Reflection and Refraction at a Solid-Solid Interface (Welded Contact) 157
where iangd is the incident angle in degrees, (d1, d2) are the densities of the first and
second media, (cp1, cs1) are the P- wave speed and S-wave speed for the first
medium and (cp2, cs2) are the P-wave speed and S-wave speed for the second
medium. The coefficients (tpp, tps) are the velocity-based transmission coefficients
P;P v S;P
(vT12 , T12 ). Figure 6.37 plots these coefficients below the second critical angle for
a Lucite-Al interface, a case which simulates the type of properties typically found
in angle beam shear wave setups. As seen in Fig. 6.37 the behavior of these
transmission coefficients closely follow the patterns seen in the fluid-solid case
(Fig. 6.26) but with differing scales. Note that if we set cs1 ¼ 0 in the call to this
function, then the function simply returns the velocity-based fluid-solid interface
transmission coefficients, a fact that can be verified directly by making this replace-
ment in Eq. (6.156b) and comparing to Eq. (6.122b).
In this section we will develop the plane wave reflection and transmission coeffi-
cients for two elastic solids that are in perfect (welded) contact along a planar
interface. As for the interface problem with smooth contact just considered, the
incident waves will be taken to lie in a “vertical” xy plane (see Fig. 6.38).
However, in the welded contact case we will treat both incident P- and S-waves
in the first medium. For an incident S-wave we can identify separately two cases:
(1) where the displacement vector of the S-waves is in the vertical plane (vertically
polarized shear or SV-waves) and (2) where the displacement is along the z-axis,
158 6 Reflection and Transmission of Bulk Waves
θp 2 φtrans (At)
ρ2 , cp 2 , cs2
θs 2
y Ψtrans (Bt)
i.e., in a horizontal plane (horizontally polarized shear or SH-waves). Since the SH-
waves do not couple to either P- or SV-waves for this interface problem, as we will
see below, we will solve each of these two cases separately.
Figure 6.38 shows the geometry and coordinate system chosen for P- or SV-waves
incident on a welded solid-solid interface. We will assume that these incident waves
generate both reflected and transmitted P- and SV-waves. Using potentials to
represent these waves, we have
ϕinc ¼ Ai exp ikp1 x sin θp1 þ y cos θp1 iωt
ψ inc ¼ Bi exp½iks1 ðx sin θs1 þ y cos θs1 Þ iωt
ϕreflt ¼ Ar exp ikp1 x sin θp1 y cos θp1 iωt
ð6:157Þ
ψ reflt ¼ Br exp½iks1 ðx sin θs1 y cos θs1 Þ iωt
ϕtrans ¼ At exp ikp2 x sin θp2 þ y cos θp2 iωt
ψ trans ¼ Bt exp½iks2 ðx sin θs2 þ y cos θs2 Þ iωt:
ðux Þ1 ¼ ðux Þ2
uy 1 ¼ u y 2
ð6:159Þ
τyy 1 ¼ τyy 2
τxy 1 ¼ τxy 2 :
Placing the waves of Eq. (6.157) into these boundary conditions, we again must
satisfy generalized Snell’s law (see Eq. (6.153)) and we obtain a set of four
equations in four unknowns
Rather than solve this entire set simultaneously, we will follow the approach of
Ewing, Jardetsky, and Press [2] and note that Eqs. (6.160a) and (6.160c) are in
terms of ðAi þ Ar Þ and ðBi Br Þ while Eqs. (6.160b) and (6.160d) are in terms of
ðAi Ar Þ and ðBi þ Br Þ. Thus, we can solve these two sets of two equations
separately to obtain
l1 m1
Ai þ A r ¼ At þ Bt ð6:161aÞ
Δ1 Δ1
l2 m2
Bi B r ¼ At þ Bt ð6:161bÞ
Δ1 Δ1
l3 m3
Ai A r ¼ At þ Bt ð6:161cÞ
Δ2 Δ2
l4 m4
Bi þ B r ¼ At þ Bt ; ð6:161dÞ
Δ2 Δ2
160 6 Reflection and Transmission of Bulk Waves
where
cs1
Δ1 ¼ sin 2θs1 sin θp1 þ cos 2θs1 cos θs1
cp1
cs1 ð6:162Þ
Δ2 ¼ sin 2θp1 sin θs1 þ cos 2θs1 cos θp1
cp1
and
cs1 ρ
l1 ¼ sin 2θs1 sin θp2 þ 2 cos 2θs2 cos θs1
cp2 ρ1
cs1 ρ
m1 ¼ sin 2θs1 cos θs2 2 sin 2θs2 cos θs1
cs2 ρ1
cs1 ρ2
l2 ¼ cos 2θs1 sin θp2 cos 2θs2 sin θs1
cp2 ρ1
cs1 ρ
m2 ¼ cos 2θs1 cos θs2 þ 2 sin 2θs2 sin θs1
cs2 ρ1
cp1 ρ c2 ð6:163Þ
l3 ¼ cos 2θs1 cos θp2 þ 2 2s2 sin 2θp2 sin θp1
cp2 ρ1 cp2
cp1 ρ
m3 ¼ cos 2θs1 sin θs2 þ 2 cos 2θs2 sin θp1
cs2 ρ1
c2s1 ρ c2
l4 ¼ sin 2θp1 cos θp2 þ 2 2s2 sin 2θp2 cos θp1
cp1 cp2 ρ1 cp2
c2s1 ρ
m4 ¼ sin 2θp1 sin θs2 þ 2 cos 2θs2 cos θp1 :
cp1 cs2 ρ1
By simply adding both Eqs. (6.161a) and (6.161c) and Eqs. (6.161b) and (6.161d),
Ar and Br can be eliminated and we obtain two equations for the unknowns At and
Bt. Solving this pair of equations and substituting the results back into
Eqs. (6.161a)–(6.161d), yields Ar and Br. For an incident P-wave only (Bi ¼ 0),
we obtain the transmission and reflection amplitude ratios
m2 m4
2 þ
At Δ1 Δ2
¼
Ai Δ
l2 l4
2 þ
Bt Δ1 Δ2
¼
Ai Δ
ð6:164Þ
l2 l4 m1 m3 l1 l3 m2 m4
þ þ
Ar Δ1 Δ2 Δ1 Δ2 Δ1 Δ2 Δ1 Δ2
¼
Ai Δ
l2 m4 m2 l4
2 2
Br Δ1 Δ2 Δ1 Δ2
¼ :
Ai Δ
6.5 Reflection and Refraction at a Solid-Solid Interface (Welded Contact) 161
where
l2 l4 m1 m3 l1 l3 m2 m4
Δ¼ þ þ þ þ ð6:166Þ
Δ1 Δ2 Δ1 Δ2 Δ1 Δ2 Δ1 Δ2
All of the above results are in terms of potential amplitudes. However, from
Eqs. (4.26) and (4.30), the displacement amplitudes are given in terms of such
potential amplitudes through
U n ¼ ikp Φ ¼ iωΦ=cp
ð6:167Þ
U s ¼ iks Ψ ¼ iωΨ =cs
where again we have indicated that these transmission and reflection coefficients
are based on velocity ratios by using a v superscript, and have explicitly indicated
that the shear polarization is one of vertical (SV) polarization (in the xy plane of
Fig. 6.37).
Needless to say, the reflection and transmission coefficients for this problem are
algebraically rather complex. Ewing, Jardetsky, and Press [2] have plotted the
behavior of similar coefficients (based on energy ratios) for different material
162 6 Reflection and Transmission of Bulk Waves
>> [tp, ts, rp, rs] ¼ solid_solid(iangd, d1, d2, cp1, cs1, cp2, cs2,
type);
where (tp, ts, rp, rs) are the P-wave and SV-wave transmission coefficients and P-
wave and SV-wave reflection coefficients, respectively, for a plane wave incident
on a solid-solid interface at the angle(s), iangd, (in degrees). The quantities (d1, d2)
are the densities of the first and second media, and (cp1, cs1) and (cp2, cs2) are the
compressional and shear wave speeds for the first and second media. The input
variable type is a string (‘P’ or ‘S’) that indicates the type of incident wave. If
cs1 ¼ 0 and type ¼ ‘P’ solid_solid returns the transmitted and reflected waves at a
fluid-solid interface. Welded solid-solid interfaces are not as commonly encoun-
tered in NDE problems as fluid-solid and two solids in smooth contact. We have
evaluated this function for a P-wave incident on a Lucite- Al interface and plotted
in Fig. 6.39 the transmitted P-wave and SV-wave coefficients to illustrate the
difference between two solids that are in welded contact and the smooth contact
case considered in Fig. 6.37. Below the first critical angle the transmitted P-wave
curves are nearly identical because the values vary little from the normal incidence
case, a case which gives the same transmission coefficients for fluid-fluid, fluid-
solid, solid-solid (smooth contact) and the solid-solid problem if the same wave
speeds are used in all of these models. The transmitted S-wave again shows a linear
behavior below the first critical angle but with some amplitude differences. Above
the first critical angle, the behavior of the welded contact and smooth contact curves
in Figs. 6.37 and 6.39 have similar behavior but with some changes in shape and
amplitude.
If a plane shear wave of horizontal (SH) polarization traveling in the xy plane is
incident on a welded solid-solid interface (Fig. 6.40) the motion generated is one of
anti-plane strain. As shown in Chap. 3, in this case it is unnecessary to use potentials
to describe the waves since there is only one displacement component
w ¼ wðx; y; tÞ and it satisfies the wave equation (or Helmholtz’s equation for
harmonic waves) itself. Thus, in this case the incident, reflected, and transmitted
waves can be written as
ðwÞ1 ¼ ðwÞ2
ð6:170Þ
ρ1 c2s1 ð∂w=∂yÞ1 ¼ ρ2 c2s2 ð∂w=∂yÞ2 :
Using these boundary conditions and phase matching all the terms on the interface
as before, one obtains a set of equations similar in form to Eq. (6.33) for a fluid-fluid
interface, namely
" #" # " #
1 1 Wt Wi
¼ : ð6:171Þ
ρ2 cs2 cos θs2 ρ1 cs1 cos θs1 Wr ρ1 cs1 cos θs1 W i
The solution of Eq. (6.171) gives the transmission and reflection coefficients in
terms of velocity (or displacement) ratios as
Equations (6.164)–(6.165) and (6.172) show that the P- and SV-wave problem is
indeed entirely decoupled from the SH-wave problem. Thus, an incident P- or SV-
wave produces no reflected or transmitted SH-waves, and likewise an incident SH-
wave generates no reflected or transmitted P- and SV-waves. Consequently, there
are no “mixed” transmission and reflection coefficients, i.e.,
v α;SH α;SH
T 12 ¼ v R12 ¼0 ðα ¼ P, SV Þ
ð6:173Þ
v SH;β v SH;β
T 12 ¼ R12 ¼0 ðβ ¼ P, SV Þ:
Note that this result was obtained for plane waves incident on a planar interface.
When a 3-D wave front is incident on a curved interface, waves of all types are
coupled. However, if the frequency of the incident waves is sufficiently high
(so that the interface appears locally as planar) and if the incident wave front can
also be treated locally as planar, these results may still be useable in those more
general cases.
A final boundary value problem that we will consider in this Chapter is the
reflection of an incident P- or SV-wave from a stress-free plane surface as shown
in Fig. 6.41. This problem can be solved as a special case of the general interface
problem considered in the previous section by simply neglecting the transmitted
wave terms. Thus, the incident and reflected waves
ϕinc ¼ Ai exp ikp1 x sin θp1 þ y cos θp1 iωt
ψ inc ¼ Bi exp½iks1 ðx sin θs1 þ y cos θs1 Þ iωt
ð6:174Þ
ϕreflt ¼ Ar exp ikp1 x sin θp1 y cos θp1 iωt
ψ reflt ¼ Br exp½iks1 ðx sin θs1 y cos θs1 Þ iωt
y
6.6 Reflection at a Stress-Free Surface 165
result in two equations that are identical to Eqs. (6.160c), (6.160d) with
At ¼ Bt ¼ 0, i.e.,
Many authors, including Graff [3], Achenbach [4], Eringen and Suhubi [5], and
Harker [1] have given plots of the behavior of the reflection coefficients for this
problem. The MATLAB® function stress_freeP returns the velocity-based reflec-
tion coefficients of Eq. (6.177b) for an incident P-wave. The calling sequence for
this function is
where rp is the reflected P-wave coefficient and rs the reflected SV-wave coeffi-
cient. The input variable ang is the angle of the incident P-wave (in degrees) and
(cp, cs) are the compressional and shear wave speeds of the solid. This function was
used to generate Fig. 6.42 which shows the reflection coefficients for a P-wave
incident on a stress-free aluminum surface. Note that at normal incidence v RP;P
¼ 1 and the polarization of the reflected P-wave is opposite to the incident wave
so the total velocity is twice that of the incident wave at the free surface.
The previous sections have concentrated on examining the behavior of plane waves
at a plane interface. This focus may appear to be of limited applicability to more
realistic problems where the waves and interfaces may not be planar. However, as
6.7 Reflection, Transmission, and the Kirchhoff Approximation 167
we will see in this section, many of the concepts introduced can be used in more
general settings. To keep the discussion as simple as possible, we will examine the
case of pressure waves interacting with a curved interface between two fluids, but
the more general elastic wave case and mixed fluid/elastic problems lead to
identical results. For more details in those generalized cases see [6].
Consider a harmonic pressure wave incident on a curved interface between two
fluids (Fig. 6.43). We will take the incident, reflected and transmitted waves of the
form
pinc ¼ Pinc ðxi Þexp iω t T inc ðxi Þ
preflt ¼ Preflt ðxi Þexp iω t T reflt ðxi Þ ð6:180Þ
p trans
¼P trans
ðxi Þexp½iωðt T trans
ðxi ÞÞ:
These expressions include the plane waves and spherical waves we have already
discussed since in those cases we use the general forms:
Plane wave traveling in the e-direction:
P ð x i Þ ¼ P0 , T ðxi Þ ¼ t0 þ ei xi =c ð6:181aÞ
Spherical wave:
pffiffiffiffiffiffiffi
Pðxi Þ ¼ P0 =r, T ðxi Þ ¼ r=c r ¼ xi xi ð6:181bÞ
where in Eq. (6.181c) the implied summation on α is over the (1, 2) subscripts only.
A Gaussian beam is similar to Eq. (6.181c) but it does not fit the forms seen in
Eq. (6.180), where T(xi) is assumed to be real, because of the complex parameters
contained in the Gaussian. Nevertheless, in the next section we will see there is a
168 6 Reflection and Transmission of Bulk Waves
where ni are components of the unit normal to the interface. Now, let us assume that
the phases of these waves are matched at the interface, i.e., set T inc ¼ T reflt ¼ T trans .
Then the equations for the amplitudes become
for every point on the curved interface. But Eq. (6.185) is nothing more than the
boundary conditions for a plane wave incident on a plane interface (whose normal
6.7 Reflection, Transmission, and the Kirchhoff Approximation 169
is in the n-direction) written here in terms of the slowness vectors (see Eq. (6.85)).
Thus, at high frequencies, we can relate the incident, reflected and transmitted wave
amplitudes at a general point on that interface as if they were plane wave ampli-
tudes incident on a plane interface at that point whose normal coincides with that of
the actual interface. This is the foundation of the Kirchhoff approximation, which
we will use in modeling wave propagation from transducers through interfaces in
Chap. 8, when considering wave scattering from flaws in Chap. 10, and in devel-
oping quantitative flaw sizing algorithms in Chap. 15.
We now need to return to the phase matching conditions T inc ¼ T reflt ¼ T trans .
We will satisfy these conditions at a point x ¼ ðz1 ; z2 ; z3 Þ on the interface in a
neighborhood of a fixed point, x0, on the interface (Fig. 6.43). Expanding any these
functions to second order at that point we have
∂T ðx0 Þ 2
1 ∂ T ð x0 Þ
T ðxÞ ¼ T x0 þ zi þ zi zj ; ð6:186Þ
∂zi 2 ∂zi ∂zj
where we have assumed the (z1, z2) coordinates are in the tangent plane of the
interface at x0 and z3-coordinate is normal to the interface (in the n-direction). If the
interface is curved, we will express that curved surface in the form z3 ¼ f ðz1 ; z2 Þ
which in the neighborhood of x0 we can write to second order as
2
1 ∂ f
z3 ¼ zα zβ ; ð6:187Þ
2 ∂zα ∂zβ
where the (α, β) indices only range over the (1, 2) values only. Then Eq. (6.186)
becomes
∂T ðx0 Þ 2
1 ∂ T ð x0 Þ
2
1 ∂T ðx0 Þ ∂ f
T ð xÞ ¼ T x 0 þ zα þ zi zj þ zα zβ : ð6:188Þ
∂zα 2 ∂zi ∂zj 2 ∂z3 ∂zα ∂zβ
To match the constant and O(|zα|) phases of all the waves we must have
The first equation of Eq. (6.189) simply says that the waves must have a common
constant delay term at the interface, while the second equation says that the
components of the slowness vectors in the tangent plane of the interface must be
identical, which is just another form of Snell’s law. We can see this easily since
∇T ¼ s ¼ e=c so if we let (u1, u2) be unit vectors in the tangent plane of the
interface in the (z1, z2) directions then the second equation in Eq. (6.189) can be
written as
170 6 Reflection and Transmission of Bulk Waves
At high frequencies we can satisfy this equation for all frequencies by setting the
coefficients of the (ω2, ω) terms equal to zero and ignoring the last remaining term
[7] to yield
∇T ∇T ¼ 1=c2 ð6:193Þ
and
Equations (6.193) and (6.194) are two basic equations of high frequency ray theory.
Equation (6.193) is called the eikonal equation and Eq. (6.194) is called the
transport equation. The eikonal equation shows that the slowness vector s ¼ ∇T
indeed has a magnitude equal to the reciprocal of the wave speed while the transport
equation shows that the behavior of the amplitude, P(xi), during propagation is
controlled by the curvature (quadratic derivatives) of the wave front, T(xi). In this
book we will not treat ray theory in depth but we will see in Chap. 8 that in the
development of the use of the stationary phase approximation for the integrals
involved in describing the transmission and reflection of ultrasonic transducer wave
fields at interfaces gives results that also can derived with the use of ray theory. For
in-depth discussions of ray theory, the reader is referred to [7–9].
6.8 Reflection and Transmission of a Gaussian Beam at a Curved Interface 171
In this section we will consider a Gaussian beam in the form considered in Chap. 4
that is incident obliquely on an interface between two fluids and we will solve
for the transmitted and reflected Gaussian beams, as shown in Fig. 6.44. We will
assume that the incident Gaussian beam is launched at point xs and has traveled
a distance D to the interface and we will write the incident Gaussian beam in
(x1, x2, x3) coordinates where the beam is propagating along the x3-direction
and (x1, x2) are normal to that propagation direction. We will take origin of the
(x1, x2, x3) coordinates, like that of the (z1, z2, z3) coordinates, to be at the point
O where the incident beam strikes the interface so that the two coordinates systems
differ by simply a rotation. [Important note: the origin of the (x1, x2, x3) coordinates
is shown displaced in Fig. 6.44 merely to allow us to display those coordinates
separately from the (z1, z2, z3) coordinates. The same is true for the other coordi-
nates as well]. We will also assume that the (z1, z3) plane is the plane of incidence so
that the x2 and z2 axes are parallel to each other. As seen in Chap. 4 the incident
Gaussian beam can be written in these (x1, x2, x3) coordinates as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i i
q10 q20
p inc
¼ P0
q ðD þ x3 Þq2i ðD þ x3 Þ
i
1 ð6:195Þ
ik1 x21 x22
exp ik1 D þ ik1 x3 þ þ iωt ;
2 q1i ðD þ x3 Þ q2i ðD þ x3 Þ
where D is the distance that the Gaussian has traveled to the interface, P0 is the
starting amplitude of the Gaussian, and
q1i ðD þ x3 Þ ¼ D þ x3 þ q10
i
ð6:196Þ
q2i ðD þ x3 Þ ¼ D þ x3 þ q20
i
:
As seen in the last section, the interface boundary conditions are best described in
the (z1, z2, z3) coordinates. Thus, we need to express the incident Gaussian in those
coordinates. This can be done by rewriting Eq. (6.195) in the form
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i i
q10 q20 ik1 inc
p ¼ P0
inc
exp ik1 D þ ik1 x3 þ L xm xn iωt
q1i ðD þ x3 Þq2i ðD þ x3 Þ 2 mn
ik1 inc
¼ P ðx3 Þexp ik1 D þ ik1 x3 þ
inc
L xm xn iωt ;
2 mn
ð6:197Þ
This form is particularly convenient since to place this beam in the (z1, z2, z3)
coordinates we simply rotate the (x1, x2, x3) axes by an acute angle θinc about the
x2 axis, where
xm ¼ Ginc
nm zn ð6:199Þ
ð6:201Þ
In a similar fashion, we can express the transmitted wave in (y1, y2, y3) coordinates,
where the origin of these coordinates is again at O, the y3-axis is along the
transmitted beam direction, and (y1, y3) plane is in the plane of incidence, giving
6.8 Reflection and Transmission of a Gaussian Beam at a Curved Interface 173
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qtr10 qtr20 ik2 tr
p trans
¼ Pt exp iωttr þ ik2 y3 þ L y y iωt
qtr1 ðy3 Þqtr2 ðy3 Þ 2 mn m n
ð6:202Þ
ik2 tr
¼ P ðy3 Þexp iωttr þ ik2 y3 þ
tr
L y y iωt
2 mn m n
with
qtr1 ¼ y3 þ qtr10
ð6:203Þ
qtr2 ¼ y3 þ qtr20
and
2 3
1=qtr1 ðy3 Þ 0 0
6 7
Ltr ¼ 4 0 1=qtr2 ðy3 Þ 0 5: ð6:204Þ
0 0 0
If the transmitted wave makes an acute angle θtr with respect to the positive z3-axis
then the (y1, y2, y3) coordinates can be expressed in terms of the (z1, z2, z3) coordi-
nates by the rotation matrix
ym ¼ Gtrnm zn ; ð6:205Þ
where
2 3
cos θtr 0 sin θtr
6 7
Gtr ¼ 4 0 1 0 5; ð6:206Þ
sin θtr 0 cos θtr
giving
ik2 tr tr tr
ptrans ¼ Ptr ðzm Þexp iωttr þ ik2 ðz1 sin θtr þ z3 cos θtr Þ þ Lmn Gim Gjn zi zj iωt
2
¼ Ptr ðzm Þexp½iωϕtr ðzm Þ iωt:
ð6:207Þ
For the reflected Gaussian beam, we will use the (r1, r2, r3) axes, where again the
origin is at O, the r3 axis is along the direction of propagation, and the (r1, r3) plane
is in the plane of incidence, giving
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qrf10 qrf20 ik1 rf
p reflt
¼ Pr exp iωtrf þ ik1 r 3 þ L r m r n iωt
qrf1 ðr 3
Þqrf2 ðr 3 Þ 2 mn ð6:208Þ
ik1 rf
¼ P ðr 3 Þexp iωtrf þ ik1 r 3 þ
rf
L r m r n iωt ;
2 mn
174 6 Reflection and Transmission of Bulk Waves
with
qrf1 ¼ r 3 þ qrf10
ð6:209Þ
qrf2 ¼ r 3 þ qrf20
and
2 3
1=qrf1 ðr 3 Þ 0 0
6 7
Lrf ¼ 6
4 0 1=qrf2 ðy3 Þ 0 7
5: ð6:210Þ
0 0 0
In this case, if we let θrf be the acute angle that the r3-axis makes with the negative
z3-direction, we can transform from the (r1, r2, r3) coordinates to the (z1, z2, z3)
coordinates with
r m ¼ Grfnm zn ; ð6:211Þ
where
2 3
cos θrf 0 sin θrf
Grf ¼ 4 0 1 0 5; ð6:212Þ
sin θrf 0 cos θrf
The boundary conditions at the interface are just Eq. (6.182), which give
Pinc exp iωϕinc þ Prf exp iωϕrf ¼ Ptr expðiωϕtr Þ
iω ∂ϕinc inc iω ∂ϕrf rf
P exp iωϕinc þ P exp iωϕrf
ρ1 ∂z3 ρ1 ∂z3
iω ∂ϕtr tr ð6:214Þ
¼ P expðiωϕ Þ tr
ρ2 ∂z3inc
1 ∂P 1 ∂Prf 1 ∂Ptr
exp iωϕ inc
þ exp iωϕ
rf
expðiωϕ Þ :
tr
ρ1 ∂z3 ρ1 ∂z3 ρ2 ∂z3
6.8 Reflection and Transmission of a Gaussian Beam at a Curved Interface 175
If, as done in the last section we assume at high frequencies the last term on the right
side of the second boundary condition in Eq. (6.214) is much smaller than the
remaining terms, these equations reduce to
Pinc exp iωϕinc þ Prf exp iωϕrf ¼ Ptr expðiωϕtr Þ
1 ∂ϕinc inc 1 ∂ϕrf rf
P exp iωϕinc þ P exp iωϕrf
ρ1 ∂z3 ρ1 ∂z3 ð6:215Þ
1 ∂ϕtr tr
¼ P expðiωϕtr Þ:
ρ2 ∂z3
Assuming for the moment all the phase terms are matched, i.e.,
and placing the explicit expressions we have for those phases into Eq. (6.215), if we
satisfy those boundary conditions at O we find
which, as expected from the last section, are identical to the plane wave relations at
a plane interface so that
pffiffiffiffiffiffiffiffiffiffiffiffi
i i
q10 q20
P t ¼ T p P ð 0Þ ¼ T p P0
inc
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q1i ðDÞq2i ðDÞ
pffiffiffiffiffiffiffiffiffiffiffiffi ; ð6:218Þ
i i
q10 q20
Pr ¼ Rp P ð0Þ ¼ Rp P0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
inc
q1i ðDÞq2i ðDÞ
where (Tp, Rp) are the plane wave transmission coefficients at a plane interface
whose normal n coincides with that of the curved surface at point O (Fig. 6.44).
Now consider the phase matching conditions of Eq. (6.216), which are:
1
x3 ¼ hαβ zα zβ ; ð6:220Þ
2
where the (α, β) summations both only run over the values (1,2). For a general
curved surface, hαβ is the symmetrical 2 2 curvature matrix. However, if we try to
match phases in this general case we will find that we need to consider more general
Gaussian beams with L matrices that are not diagonal, as assumed in Eqs. (6.198),
(6.204), (6.210) for all our elliptical cross-section Gaussian beams. Such general-
izations are certainly possible, but here we wish to keep the simplicity of the
Gaussian beam expressions we have been considering and instead assume that the
principal axes of the curved surface are aligned with the z1 and z2 axes so that the
surface curvature matrix is diagonal:
1=R1 0
h¼ ; ð6:221Þ
0 1=R2
which also implies that principal curvature directions of the interface lie normal to
the plane of incidence and in that plane of incidence. In the special case where the
incident Gaussian beam is normal to the interface, this does not imply any real
restriction on our solutions since in that case we can always choose the plane of
incidence to be along the principle curvature directions. For oblique incidence,
restrictions do exist. For a beam incident on a cylindrical interface, for example, the
plane of incidence of the beam must either be parallel to the axis of the cylinder or
perpendicular to it (Fig. 6.45).
Placing Eq. (6.220) into Eq. (6.219), with the h matrix given by Eq. (6.221) we
have
But, since L13 ¼ L31 ¼ L23 ¼ L33 ¼ 0 for all the L matrices involved, these
conditions reduce to
D z1 sin θinc 1
inc inc inc
þ þ Lμη Gαμ Gβη þ cos θinc hαβ zα zβ
c1 c1 2c1
z1 sin θtr 1
tr tr tr
¼ ttr þ þ Lμη Gαμ Gβη þ cos θtr hαβ zα zβ ð6:223Þ
c2 2c2
z1 sin θrf 1
rf rf rf
¼ trf þ þ Lμη Gαμ Gβη cos θrf hαβ zα zβ ;
c1 2c1
where all the subscripts involved range over (1,2) only and now matrices such as
Lμη, Gαμ are 2 2 sub-matrices of the original L and G matrices whose subscripts
again only range over (1,2).
If we phase match all the constant terms in Eq. (6.223) we have
so all of the waves have the same common time delay at point O. Matching the
coefficients of the linear terms gives
which is just Snell’s law, and matching the quadratic terms yields
Linc G inc inc
μη αμ βη G þ cos θ inc αβ =c1
h
¼ Ltrμη Gtrαμ Gtrβη þ cos θtr hαβ =c2 ð6:226Þ
¼ Lrfμη Grfαμ Grfβη cos θrf hαβ =c1
178 6 Reflection and Transmission of Bulk Waves
or, equivalently,
c2 inc tr 1 h tr i1 inc inc tr 1 h tr i1 c2
Ltrδσ ¼ Lμη Gδα Gσβ Gαμ Gβη þ Gδα Gσβ cos θinc cos θtr hαβ
c1 c
h i1 h i1 h i1 h i1 1
Lrfδσ ¼ Linc
μη Gδα
rf
Grfσβ Ginc rf
αμ Gβη þ Gδα
inc
Grfσβ cos θinc þ cos θrf hαβ :
ð6:227Þ
and
2 3 2 3 2 3
1 1 2
0 0 0
6 qrf ð0Þ 7 6 q i ðDÞ 7 6 R cos θ 7
6 1 7 6 1 7 6 1 inc 7;
6 7¼6 7þ4 ð6:229Þ
4 1 5 4 1 5 2 cos θinc 5
0 0 0
qrf2 ð0Þ q2 ðDÞ
i R2
where we have used the fact that θrf ¼ θinc in Eq. (6.229). These matrix expressions
can also be written in terms of their scalar components as
Recall, from Eq. (4.121) we can express these relations in terms of beam width and
wave front curvatures since the incident, reflected, and transmitted waves all can be
placed in the form
6.8 Reflection and Transmission of a Gaussian Beam at a Curved Interface 179
1 1 λ
¼ þi 2
q1m ðzÞ R1m ðzÞ π w1 ðzÞ
m
1 1 λ ðm ¼ i, tr, rf Þ; ð6:231Þ
¼ þi 2
q2m ðzÞ R2m ðzÞ π w ðzÞ
m
2
1 1 2 cos θinc
¼ þ
Rrf2 ð0Þ R2i ðDÞ R2
and
cos θtr i
wtr1 ð0Þ ¼ w ðDÞ
cos θinc 1
wtr2 ð0Þ ¼ w2i ðDÞ ð6:233Þ
wrf1 ð0Þ ¼ w1i ðDÞ
wrf2 ð0Þ ¼ w2i ðDÞ:
Equation (6.232) shows the effect the interface curvature has on the focusing or
defocusing of the Gaussian beam. The easiest way to illustrate this is to let the
incident Gaussian beam
wave front radii be infinite at the interface
R1i ðDÞ ¼ R2i ðDÞ ¼ 1 so that we have planar wave fronts impinging on the curved
interface. Then
It is easy to show, using Snell’s law, that the factor ðc2 cos θinc =c1 cos θtr Þ is
positive when c2 =c1 > 1 and negative when c2 =c1 < 1. Then, for example, if c2 =
c1 > 1 and R1 > 0 it follows that Rtr1 ð0Þ > 0 so that the transmitted wave front in the
plane of incidence is spreading (or being defocused) in the second medium.
Similarly,
tr if c2 =c1 > 1 and R1 < 0 the transmitted wave front is being focused
R1 ð0Þ < 0 . These are the two cases (a) and (b) shown in Fig. 6.46. In case (a) we
say that the curved interface is a defocusing interface for the transmitted waves
while in case (b) we say it is a focusing interface. The behavior of the radius of
curvature, R2, affects the transmitted wave front radius of curvature Rtr2 (0) in
exactly the same manner. Note that if c2 =c1 < 1 then by the same argument a
curved interface where R1, 2 > 0 is a focusing interface while R1, 2 < 0 is a
defocusing interface for the transmitted waves, as shown in Fig. 6.47. Equation
(6.234) also shows that for the reflected wave the wave front radii of curvature have
the same signs as those of the radii of curvature of the interface.
Equation (6.233) says that all the beam widths of the Gaussians are identical,
except for the transmitted beam width in the plane of incidence. This is easy to
show if we draw a set of parallel lines of width winc in the plane of incidence,
representing an incident beam, as shown in Fig. 6.48, and follow those lines as they
are transmitted or reflected from the interface according to Snell’s law. If we
assume that in the plane of incidence all the waves have a common width, w0,
along the interface, we see that the width of the reflected wave, wref, is the same as
the incident wave in the plane of incidence, while the width of the transmitted wave,
cos θtr
wtr ¼ cos θinc winc , in agreement with Eq. (6.233). Out of the plane of incidence the
same argument shows that all the beam widths must be identical.
We can collect all our results for the fluid-fluid problem and write the incident,
transmitted and reflected Gaussians (see Eq. (6.195), Eq. (6.202) and (6.208)) as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi " ! #
i
q10 q20i
0 ik1 x21 x22
p ¼ P0
inc 0 0 exp ik1 x3 þ 0 þ i 0 iωt ð6:235Þ
q1i x3 q2i x3 2 q1i x3 q2 x 3
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
qtr10 qtr20 ik2 y1 y22
ptrans
¼ Pt exp ik 1 D þ ik 2 y þ þ iωt
qtr1 ðy3 Þqtr2 ðy3 Þ 3
2 qtr1 ðy3 Þ qtr2 ðy3 Þ
ð6:236Þ
182 6 Reflection and Transmission of Bulk Waves
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi " ! #
qrf10 qrf20 ik1 r 21 r 22
p reflt
¼ Pr exp ik 1 D þ ik 1 r 3 þ þ iωt ;
qrf1 ðr 3 Þqrf2 ðr 3 Þ 2 qrf1 ðr 3 Þ qrf2 ðr 3 Þ
ð6:237Þ
0
where we have written the incident wave in terms of the distance x3 ¼ D þ x3
0
whose origin x3 ¼ 0 will be taken as the starting point of the Gaussian in the first
medium and we have used Eq. (6.224), i.e., ttr ¼ trf ¼ D=c1 . The q-parameters for
these waves are
0 0
qmi x3 ¼ x3 þ qm0 i
for m ¼ 1, 2. All of these expressions can be written in terms of the unknowns (P0,
qi10 , qi20 ) since we found for the amplitudes (Eq. (6.218))
pffiffiffiffiffiffiffiffiffiffiffiffi
i i
q10 q20
Pt ¼ T p P ð0Þ ¼ T p P0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
inc
q1 ðDÞq2i ðDÞ
i
pffiffiffiffiffiffiffiffiffiffiffiffi ; ð6:239Þ
i i
q10 q20
Pr ¼ Rp Pinc ð0Þ ¼ Rp P0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q1i ðDÞq2i ðDÞ
and for the transmitted and reflected q-parameters (Eq. (6.230) with qtrm0 qtrm ð0Þ,
qrfm0 qrfm ð0Þ, m ¼ 1, 2):
1 1 2 cos θinc
¼ i þ :
qrf20 D þ q10 R2
If we place the amplitude terms (Eq. (6.239)) into these Gaussians and drop the
common expðiωtÞ term, then we have, finally
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi " !#
q10i
q20i
0 ik1 x21 x22
p inc
¼ P0 0 0 exp ik1 x3 þ 0 þ i 0 ð6:241Þ
q1i x3 q2i x3 2 q1i x3 q2 x 3
6.8 Reflection and Transmission of a Gaussian Beam at a Curved Interface 183
pffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q10i i
q20 qtr10 qtr20
ptrans ¼ P0 T p pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q1i ðDÞq2i ðDÞ q1 ðy3 Þq2 ðy3 Þ
tr tr
2 ð6:242Þ
ik2 y1 y22
exp ik1 D þ ik2 y3 þ þ
2 qtr1 ðy3 Þ qtr2 ðy3 Þ
pffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q10i
q20i
qrf10 qrf20
preflt ¼ P0 Rp pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rf rf
" q1 ðDÞq2 ðDÞ q1 ðr 3 Þq2 ðr 3 Þ
i i
!# ð6:243Þ
ik1 r 21 r 22
exp ik1 D þ ik1 r 3 þ þ :
2 qrf1 ðr 3 Þ qrf2 ðr 3 Þ
The results of the last section can be directly used for a Gaussian beam incident on a
curved fluid-solid interface or solid-solid interface since the velocity
(or displacement) in the solid, like the pressure in the fluid, satisfy the paraxial
wave equation and the satisfaction of the boundary conditions at the fluid-solid
interface proceed in an identical fashion to the fluid-fluid interface case. As
indicated in Chap. 4, the differences we need to account for the fluid-solid case
are the different wave modes (P and SV) possible in the solid and the corresponding
polarizations of the waves (see Fig. 6.49). Thus, the waves for the fluid-solid
interface can be written as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qi;β
10 q20
i;β
vinc ¼ v0 diβ 0
0
qi;β x qi;β x
" 1 3 2 3 !# ð6:244Þ
β 0 ik1β x21 x22
exp ik1 x3 þ 0 þ i;β 0
2 qi;β 1 x3 q2 x 3
and
qffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
γ;β γ
qi;β
10 q20
i;β
qrf10;γ qrf20;γ
vreflt ¼ v0 R12 drf qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qi;β i;β qrf1 ;γ ðr 3 Þqrf2 ;γ ðr 3 Þ
1 ðDÞq2 ðDÞ ð6:246Þ
" !#
β γ ik1γ r 21 r 22
exp ik1 D þ ik1 r 3 þ þ
2 q1rf ;γ ðr 3 Þ qrf2 ;γ ðr 3 Þ
for the transmitted and reflected Gaussian where (dβi , dαtr , dγrf ) are the polarizations
for the incident, transmitted, and reflected waves of type (β, α, γ), respectively. The
plane wave transmission and reflection coefficients here are based on velocity
(or displacement) ratios. For the fluid-solid interface problem obviously there are
only P-waves in the fluid so β ¼ γ ¼ P while α ¼ ðP; SV Þ for transmitted P- and
SV-waves. For two solids in smooth contact, as found in angle beam shear wave
transducer problems, again we would have β ¼ P but γ ¼ ðP; SV Þ and α ¼ ðP; SV Þ.
Note that also the wave numbers are given by k1β ¼ ω=c1β , k2α ¼ ω=c2α , k1γ ¼ ω=c1γ in
terms of the wave speeds (cβ1 , cα2 , cγ1 ) and we have
0 0 i;β
m x3 ¼ x3 þ qm0
qi;β
tr;α
m ðy3 Þ ¼ y3 þ qm0
qtr;α ð6:247Þ
qrfm ;γ ðr 3 Þ ¼ r 3 þ qrfm0;γ :
α β β α
1 β
c2α cos 2 θinc 1 c 2 cos θ =c
inc 1 cos θ tr 1
¼ β
þ 2θ α
qtr;α
10 c1 cos 2 θtrα i;β
D þ q10 cos tr R 1
1 c2α 1
1
β
¼ β
þ c2α cos θinc =c1β cos θtrα
qtr;α c1 D þ q10 i;β R2
20 ð6:248Þ
1 1 2 1
rf ;γ
¼ i;β
þ β
q10 D þ q10 cos θinc R1
β
1 1 2 cos θinc
rf ;γ
¼ i;β
þ :
q20 Dþ q10 R2
Except for the reflected wave relations in Eq. (6.248) these solutions can also be
used for a Gaussian beam incident on an interface between two solids as long as the
polarization of the incident wave lies in the plane of incidence, a condition that only
is of importance for incident shear waves since an incident P-wave automatically
lies in the plane of incidence (which by definition is the plane containing the normal
to the interface at the point where an incident ray intersects the interface and the
incident wave direction, which is also the polarization of a P-wave). A shear wave,
however, polarized perpendicular to the plane of incidence has a different trans-
mission coefficient, as seen earlier in this chapter, and so this component of the
transmitted shear wave has different properties from a transmitted shear wave
whose polarization lies in the plane of incidence. For a consistent way to handle
shear waves of arbitrary polarization at an interface see [6] or [7]. [Note that in this
book we will also refer to shear waves polarized in the plane of incidence as SV-
waves and shear waves polarized normal to the plane of incidence as SH-waves,
even when the plane of incidence is not a vertical plane as assumed in many of the
derivations in this chapter]. The reflected wave relations of Eq. (6.248) need to be
modified for a solid-solid interface since those relations assume γ ¼ β ¼ P,
c1γ ¼ c1β , and θrfγ ¼ θinc
β
. It is easy to obtain the proper reflected wave relations by
noting that we can view a reflected wave simply as a transmitted wave traveling
with a wave speed ctrα ¼ crfγ in the θtrα ¼ π θrfγ direction, where θγrf is the acute
angle of the reflected wave with respect to the negative z3-axis. Thus, if we make
the replacements ctrα ! crfγ , θtrα ! π θrfγ in the transmitted wave relations of
Eq. (6.248), the reflected wave relations for a solid-solid interface become
γ β β γ
1 c1γ cos 2 θinc
β
1 c 1 cos θ =c
inc 1 þ cos θ rf 1
rf ;γ
¼ þ
q10 c1β cos 2 θrfγ i;β
D þ q10 cos θinc
2 β R 1
ð6:249Þ
1 c1γ 1
1
rf ;γ
¼ β i;β
þ c1γ cos θinc
β
=c1β þ cos θrfγ :
q20 c1 D þ q10 R2
transmitted and reflected through general curved interfaces using analytical expres-
sions like the ones developed in this Chapter, a multi-Gaussian transducer beam
model is a particularly effective model, one that does away with numerical ray
tracing and the need for thousands of elements and it is a model devoid of singular
behavior at focal points and caustics. These same characteristics are retained for
Gaussian beams in homogeneous, anisotropic elastic media [7]. For inhomoge-
neous elastic media Gaussian beam models can also be used but they do require
numerical ray tracing to determine the path of the propagating beam [7].
It should be obvious from the various problems discussed in this Chapter that
Snell’s Law plays a fundamental role in the transmission of ultrasonic waves.
Thus, it is useful to have a versatile tool that makes it easy to apply this law in a
variety of ways. A MATLAB function, snells_law has been written to meet that
need. That function has the calling sequence:
where the function returns the variables (ang_in, ang_out), which are the incident
and refracted angles, respectively, (in degrees) at a plane interface between two
materials with (c1, c2) being the wave speeds in the first and second media,
respectively. This function can handle two types of problems that we designate as
forward and reverse problems. In a forward problem the input parameter type is
the string ‘f’ and the input parameter, ang, (in degrees) is the incident angle in
the first medium (which travels with wave speed c1) and the calculated refracted
angle, ang_out is returned (for a wave traveling in the second medium with wave
speed c2) along with the incident angle (ang_in ¼ ang). In a reverse problem where
type ¼ ‘ r ’, the input parameter, ang, is the value of the refracted angle, which is
returned (ang ¼ ang_out) along with the calculated incident angle, ang_in. To
illustrate various uses of this function, let us consider a water-steel interface where
cp1 ¼ 1480 m/s, cp2 ¼ 5900 m/s, cs2 ¼ 3200 m/s. A forward problem would be, for
example, where we specify the incident angle, say, ang ¼ 5 and calculate the
refracted P-wave angle. We find from MATLAB:
so the refracted P-wave angle is about 20.3 . A reverse problem would be where we
want, for example, a refracted shear wave to travel at 45 in the steel and need to
calculate the incident angle required in the water:
6.9 Snell’s Law: A Discussion and Numerical Examples 187
and we see that incident angle needed is approximately 19.1 . We can also use this
function to determine the critical angles for this water-steel interface by setting the
refracted angle equal to 90 for both P- and S-waves:
Two final cases we will consider is when we go from a faster medium to a slower
medium. For going from steel to water, for example, for a specified angle for the P-
waves in the steel (forward problem) we find:
and, as expected, the refracted wave now travels a smaller angle than the incident
wave. However, this means that we may not always be able to find an incident angle
that produces a specified refracted angle (reverse problem) in such cases and the
MATLAB function flags this fact:
188 6 Reflection and Transmission of Bulk Waves
The amplitude changes of a plane wave of type β (β ¼ P, SV, SH) that passes
from a medium m through a plane interface to medium n, then emerges as a
plane wave of type α (α ¼ P, SV, SH) is determined by a plane wave
transmission coefficient
T 1 ðωÞ ¼ T αmn, β
T 2 ðωÞ ¼ T αmn, β
These transmission coefficients are also present in models for the propagation
of sound beams across curved interfaces. Note that the coefficients are
different, depending on the ratios of quantities we choose to use in their
definitions, and through Stokes’ relations we can write these coefficients in
terms of those along reversed paths so the specific coefficients that appear in
an LTI model depend on such choices. Also note that below critical angles
these transmission coefficients are real quantities and independent of
frequency.
Most texts on wave propagation in fluids and solids treat some reflection and
transmission problems at interfaces. There are a number of books, however,
where interface problems (and multiple interfaces) play a central role, including
Ewing, Jardetsky, and Press [2], Brekhovskikh [10], White [3], Officer [4], and
Kennett [11].
The Stokes’ relations discussed in this Chapter do not appear to be widely
appreciated. Towne [12] discusses the basic relations for electromagnetic wave
problems and Qu, Achenbach, and Roberts [13] have given a very general deriva-
tion valid for fluid-solid interfaces where the solid can be anisotropic. The discus-
sion of the connection between Snell’s law and stationary phase follows the
treatment used by Stamnes [14] in evaluating angular plane spectrum integrals by
6.10 About the Literature 189
the method of stationary phase, a technique we will use frequently in later chapters.
Candel and Crance [5] give a nice discussion of how stationary phase results can be
used to evaluate the transmitted wave fields at an interface between two fluids (for
two-dimensional waves) and Ben-Menahem and Singh [15] describe the connection
between Fermat’s theorem and Snell’s law. For more details on the use of Gaussian
beams for NDE problems, see Schmerr and Song [6]. For extensions of Gaussian
beam modeling to anisotropic and inhomogeneous elastic wave problems see the
book by Cerveny [7].
ρ1 , c1 ρ2 , c2
ρ1, c1 ρ2, c2
Fig. P6.1 Reflection and transmission of a compressional wave at normal incidence to a rough
surface
x
R
h
x=0 x=h
A Ar At
1/4”
Fig. P6.3 Transmitted and reflected waves for an aluminum plate in water
6.11 Problems
6.1. In problem 3.1, the equation of motion for a bar of variable cross section A(x)
was derived and the wave speed, cb, determined for a bar of constant area. Let
AðxÞ ¼ A0 expð2αxÞ, where A0 and α are constants. For this exponential
“horn”, assume traveling harmonic waves of the form
Show that f(x) must satisfy a “damped oscillator” equation of the form
00 0 0
f þ 2Af þ B2 f ¼ 0 whereð Þ ¼ dð Þ=dx
τxx ðx ¼ 0 , tÞ ¼ τxx ðx ¼ 0þ , tÞ ¼ 0
6.11 Problems 191
Thus, it is reasonable to assume for the whole interface that Eq. (P6.1) is
satisfied. For the other boundary condition note that
uðx ¼ 0 , tÞ 6¼ uðx ¼ 0þ , tÞ
in general since there can be some relative motion of the two sides of the
interface that are not in contact. It is, therefore, also reasonable to expect that
the amount of this relative motion is related to the stress experienced at the
interface so we assume
Fig. P6.4 Transducer crystal (a) radiating directly into water, (b) with a facing layer, (c) with a
quarter wave plate layer
192 6 Reflection and Transmission of Bulk Waves
(b) What is the asymptotic behavior of the transmission coefficient when the
frequency, f, is large? Using this asymptotic expression, determine how
the shape of the transmitted pulse across an imperfect interface is related
to the shape of an incident pulse.
6.4. Consider an elastic layer as shown in Fig. P6.2 in which we have 1-D waves
traveling in both the plus and minus x-directions whose displacements are
given by
where T is a 2 2 matrix. Since τxx and vx are continuous at both faces of the
layer, if we want to stack up multiple layers we can do so by just multiplying
their transfer matrices together.
(a) Determine the transfer matrix T for a single layer
(b) If an elastic layer, of thickness h, density ρ2 and wave speed, c2, is
imbedded in another elastic medium, with density ρ1 and wave speed
c1, and a longitudinal wave
ρp 1 , cp 1 ρp 2 , cp 2 ρp 3 , cp 3
A0 Ar A1 A2 At
x
ui ur u1 u2 ut
h
x=0 x=h
(crystal) (layer) (water)
Fig. P6.5 Layer geometry for modeling a transducer crystal with a quarter wave plate layer
6.11 Problems 193
P -wave transducer
α
Lucite
steel
St
θ = 45°, 60°
R21 T 12 T 21 expð2ik2 hÞ
R ¼ R12 þ ;
1 R221 expð2ik2 hÞ
where T12, T21 and R12, R21 are the transmission and reflection coeffi-
cients for the individual interfaces. Then, by expanding the denominator
of this expression, express this coefficient as a series of multiple waves.
P
D
d1
h1
θ1 ρ1 , c1
x d2 h2
θ2
ρ2 , c2 Q
h ¼ nλ2 =4 n ¼ 1, 3, 5, ::::
If the specific acoustic impedance, z2, of this quarter wave layer is chosen
appropriately in terms of the impedance, z1, of the crystal, and z3, the imped-
ance of the water, then the wave reflected back into the crystal will vanish and
we will have the solution shown in Fig. P6.4c.
Determine the z2 (in terms of z1 and z3) that will cause the reflection to
vanish by solving the one-dimensional layer problem shown in Fig. P6.5.
Assuming harmonic wave disturbances, apply the continuity conditions at
x ¼ 0, h, giving four equations for the displacement amplitudes Ao, A1, A2, At
to obtain two equations for Ao and At only.
Solve these two equations for R ¼ Ar =A0 . In this R expression, set h ¼ nλ2
=4 and find z2 so that R ¼ 0. Alternately, use the transfer matrix approach of
problem 3 (which eliminates the A1 and A2 terms automatically) and follow
similar steps.
6.7. A Lucite wedge is in contact with a steel part as shown in Fig. P6.6. If we place
a compressional wave transducer on this wedge, what angle, α, must we have
for the wedge to be at the first critical angle? At the second critical angle?
What angle, α, produces a 45 shear wave in the solid? A 60 shear wave in the
solid? (where the angles mentioned are measured from the normal to the
surface).
6.8. Consider the oblique reflection of a plane P-wave at a stress free surface. Show
that if we neglect mode conversion, i.e., assume only a reflected P-wave is
present that no solution is possible other than the trivial solution.
References 195
References
1. I.S. Gradshteyn, I.M. Ryzhik, Table of integrals, series, and products (Academic Press,
New York, 1980)
2. W. Ewing, W. Jardetsky, F. Press, Elastic Waves in Layered Media (McGraw-Hill, New York,
1957)
3. J.E. White, Seismic Radiation, Transmission, Attenuation (McGraw-Hill, New York, 1965)
4. C.B. Officer, Introduction to the Theory of Sound Transmission (McGraw-Hill, New York,
1958)
5. S.M. Candel, C. Crance, Direct Fourier synthesis of waves in layered media and the method of
stationary phase. J. Sound. Vib. 74, 477–498 (1981)
6. L.W. Schmerr, S.-J. Song, Ultrasonic nondestructive evaluation systems—models and mea-
surements (Springer, New York, 2007)
7. V. Cerveny, Seismic Ray Theory (Cambridge University Press, Cambridge, 2001)
8. M.A. Slawinski, Waves and Rays in Elastic Continua, 3rd edn. (World Scientific, Singapore,
2015)
9. J. Pujol, Elastic Wave Propagation and Generation in Seismology (Cambridge University
Press, Cambridge, 2003)
10. L.M. Brekhovskikh, Waves in Layered Media (Academic Press, New York, 1960)
11. B.L.N. Kennett, Seismic Wave Propagation in Stratified Media (Cambridge University Press,
New York, 1983)
12. D.H. Towne, Wave Phenomena (Dover, New York, 1967)
13. J. Qu, J.D. Achenbach, R.A. Roberts, Reciprocal relations for transmission coefficients: theory
and application. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 36, 280–286 (1989)
14. J.J. Stamnes, Waves in Focal Regions (Adam Hilger, Boston, 1986)
15. A. Ben-Menahem, S.J. Singh, Seismic Waves and Sources (Springer, New York, 1981)
Chapter 7
Propagation of Surface and Plate Waves
Although bulk P- and S-waves are the types of waves most commonly used in NDE
testing, there are other types of waves that can exist in elastic solids that are useful
as well. In this Chapter, we will consider briefly the properties of surface (Rayleigh)
waves and plate (SH and Lamb) waves that make them attractive for special NDE
inspection applications.
Consider the plane stress-free surface of an elastic solid (Fig. 7.1). Lord Rayleigh,
in 1887, sought to find solutions of the equation of motion that represented traveling
wave solutions which were confined primarily to a region near the surface.
Following in the footsteps of Rayleigh, we will consider the case of
two-dimensional plane strain disturbances which are mixtures of both longitudinal
and shear wave motions that satisfy the governing equations for the displacement
potentials given by (for harmonic waves of time dependency of expðiωtÞ):
2 2
∂ ϕ=∂x2 þ ∂ ϕ=∂y2 þ ω2 ϕ=c2p ¼ 0
2 2 ð7:1Þ
∂ ψ=∂x2 þ ∂ ψ=∂y2 þ ω2 ψ=c2s ¼ 0:
Since we are looking for disturbances that decay in amplitude away from the
surface, we will assume disturbances of a form similar to what we found for
inhomogeneous waves of Chap. 6, namely
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_7) contains supplementary material, which is available to authorized users.
ϕ ¼ AexpðαyÞexp½ikðx ctÞ
ð7:2Þ
ψ ¼ BexpðβyÞexp½ikðx ctÞ;
where c is the wave speed of the propagating wave, k ¼ ω=c and α and β are
unknown exponential decay coefficients. Placing these expressions into the equa-
tions of motion Eq. (7.1), gives
k2 þ α2 þ ω2 =c2p AexpðαyÞexp½ikðx ctÞ ¼ 0
2 ð7:3Þ
k þ β2 þ ω2 =c2s BexpðβyÞexp½ikðx ctÞ ¼ 0:
Note that if the solutions of the type given in Eq. (7.2) exist and are confined to the
surface (so α and β are both real) then, as Eq. (7.4) shows, we must have
c < cs < cp , i.e., the surface wave must travel slower than a bulk shear wave in
the same material.
Placing Eq. (7.2) into the expressions for the displacements, ux and uy, and the
stresses τyy and τxy, gives
To determine an explicit value for the wave speed, c, the boundary conditions at
the surface y ¼ 0 must be satisfied, i.e.,
2
β þ k2 A þ 2ikβ B ¼ 0
ð7:6Þ
2ikαA þ β2 þ k2 B ¼ 0;
(where, with ks ¼ ω=cs , we have used the relationship 2k2 k2s ¼ β2 þ k2 ) which
produces two homogeneous equations for the amplitudes A and B. For a non-trivial
7.1 Rayleigh Surface Waves 199
solution to exist for this set of equations, the determinant of the matrix of coefficients
must vanish, giving
2
β2 þ k2 4k2 αβ ¼ 0 ð7:7Þ
where (cp, cs) are the compressional and shear wave speeds, respectively, and cr is
the Rayleigh wave speed in the same units that cp and cs are given. For example, for
mild steel where cp ¼ 5.9 mm/μs and cs ¼ 3.2 mm/μs, cr ¼ 2.96 mm/μs. If you do
not have MATLAB® available, a simple equation that gives a good fit to the roots of
Eq. (7.8) is
0:862 þ 1:14ν
cR ¼ cs ; ð7:9Þ
1þν
where ν is Poisson’s ratio. Table 7.1 shows some of the typical values for the
Rayleigh wave speed obtained from this fitting equation over the range of possible
Poisson ratio values. As can be seen, the Rayleigh wave speed is roughly 90 % of
the shear wave speed over this entire range.
Rayleigh waves can be efficiently generated by an angle beam transducer if the
incident angle in the wedge is chosen correctly (Fig. 7.2). To see this, consider the
transmitted waves at the solid-solid interface (smooth contact) considered in
Chap. 6:
Rayleigh Wave
ρ2, cp 2, cs 2, cR 2
ϕtrans ¼ At exp ikp2 x sin θp2 þ y cos θp2 iωt
ð7:10Þ
ψ trans ¼ Bt exp½iks2 ðx sin θs2 þ y cos θs2 Þ iωt:
Beyond the second critical angle, these transmitted waves are both inhomogeneous
waves of the form
" pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
jωjysin 2 θp2 1
ϕtrans ¼ At exp exp ikp2 x sin θp2 cp2 t
cp2
" pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi# ð7:11Þ
jωjy sin 2 θs2 1
ψ trans ¼ Bt exp exp½iks2 ðx sin θs2 cs2 tÞ:
cs2
kx ¼ kp2 sin θp2 ¼ ks2 sin θs2 ¼ kp1 sin θp1 ; ð7:12Þ
where kx is the common wave number along the boundary and kx ¼ ω=cx , where cx
is the common apparent wave speed along the boundary. Then Eq. (7.11) can be
rewritten in terms of kx and cx as
h qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
ϕtrans ¼ At exp yjkx j 1 c2x =c2p2 exp½ikx ðx cx tÞ
h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii ð7:13Þ
ψ trans ¼ Bt exp yjkx j 1 c2x =c2s2 exp½ikx ðx cx tÞ:
Comparing Eqs. (7.13) with (7.2), we see that these equations are identical to those
for a Rayleigh wave traveling in material two if we set cx ¼ cR2 .
Then since kx ¼ ω=cR2 ¼ ω sin θp1 =cp1 , we have
θp1 ¼ ðθcr ÞR ¼ sin 1 cp1 =cR2 ð7:14Þ
satisfies Eq. (7.14), it is necessary to have cp1 < cR2 , i.e., the longitudinal wave
speed of the wedge must be less than the Rayleigh wave speed of the material being
examined. A common material that will satisfy this condition for most structural
materials is Lucite (Plexiglas), which has the longitudinal and shear wave speeds
given approximately by (see Table 3.1)
then the incident angle, θp1, needed to produce a set of inhomogeneous waves under
the wedge that have exactly the same properties as the Rayleigh wave is given by
ðθcr ÞR ¼ sin 1 cp1 =cR2 ffi 71:9 :
When these inhomogeneous waves reach the end of the wedge, they already satisfy
the free stress boundary conditions so they can continue to propagate outside the
wedge as a Rayleigh surface wave.
Rayleigh waves are of great value in NDE applications where one needs to
inspect the surface or near-surface region of a part since these waves are quite
sensitive to the presence of surface defects. Because the depth of penetration of
Rayleigh waves is frequency dependent (see problem 7.4), one can use this depen-
dency to control the extent of the near-surface region being inspected. Rayleigh
waves also exhibit less amplitude loss due to beam spreading than bulk waves since
they are confined to the surface of the part. This property allows Rayleigh waves to
travel long distances on smooth surfaces to regions of the part where one may not
have direct access. Surface roughness, of course, can limit the travel capability of
Rayleigh waves and also generate significant competing noise in the received
Rayleigh wave signals.
In a plate geometry, as shown in Fig. 7.3, new types of waves called plate wave
disturbances are possible. First, we will examine plate waves for a plate that has
stress-free faces and is assumed to be subjected to horizontal shearing (SH) motions
202 7 Propagation of Surface and Plate Waves
A solution for each value of m is called a mode. Consider first the mode m ¼ 0. In
this case k0 ¼ ω=cs , c ¼ cs and the displacement is given by
This is just an ordinary horizontally polarized bulk shear wave traveling along the
axis of the plate, with uniform displacement across the thickness (Fig. 7.4). This
mode is also called a SH0 mode.
Now, consider the other modes for m ¼ 2, 4, 6, etc. From Eq. (7.23) the
displacement for these modes is given by (see Fig. 7.5)
y
2h
m=4
x
204 7 Propagation of Surface and Plate Waves
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
km ¼ ω2 =c2s ðmπ=2hÞ2 ð7:26Þ
and
cs ω
cm ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð7:27Þ
ω2 ðmπ=2hÞ2 c2s
Unlike bulk waves and the Rayleigh wave, where the wave speeds were strictly a
function of material properties, Eq. (7.27) shows that, except for the SH0 mode, the
wave speeds of all other plate wave modes are also functions of frequency. Because
of this behavior the modes are said to be dispersive and relations such as Eq. (7.23)
(or, equivalently Eqs. (7.26) and (7.27)) are called dispersion relations. In disper-
sive wave propagation problems, because all the frequency components which
make up a pulse do not travel in “lock step” with a common wave speed, some
frequency components will travel slower or faster than others, leading to a change
in the overall pulse shape with time. We can illustrate this phenomenon with a
simple example. If A( f ) is the Fourier transform of a(t) then we have for a
propagating non-dispersive plane wave traveling in the +x direction with a constant
phase velocity, c0, the inverse Fourier transform relationship
ð
þ1
which shows that the wave always propagates with the same profile, a(t), that it had
at x ¼ 0. However, if the phase velocity is a function of frequency, then as the wave
propagates its profile is instead
ð
þ1
To illustrate the difference between ad(t, x) and aðt x=c0 Þ, let the initial
pulse profile a(t) at x ¼ 0 be the box function shown in Fig. 7.6a (whose
Fourier transform, A( f ), is known—see Eq. (2.11)) and suppose the wave speed
is given as cðf Þ ¼ 6ð1 þ f =50Þ mm/μs, where f is measured in MHz. Since A( f ) has
frequency content at all frequencies, it was multiplied by a low pass filter that
eliminated the frequencies above 50 MHz before performing the inverse Fourier
transform at a sampling frequency of 100 MHz in order to prevent aliasing. Our
7.2 Plate Waves: Horizontal Shearing Motions 205
Fig. 7.6 Propagation of a box-like wave form in a dispersive medium, starting at (a), x ¼ 0, and
propagating to (b), x ¼ 10 mm, (c), x ¼ 30 mm, and (d), x ¼ 50 mm
choice for the wave speed here is rather artificial but it is used simply to illustrate
the effects of dispersion. Performing the inverse Fourier transform numerically with
the Fast Fourier transform, then when the undispersed wave is at x ¼ 10 mm, the
dispersed profile is as shown in Fig. 7.6b, with the subsequent profiles at x ¼ 30 mm
and x ¼ 50 mm shown in Fig. 7.6c, d. The undispersed traveling box wave form is
also shown as dashed lines in all these cases. It can be seen that the wave profile
changes significantly from the box function and these changes vary with propaga-
tion distance. There is a distinct high frequency precursor that develops in the early
time profiles for the dispersive waves. This occurs because the wave speed in the
dispersive case considered here increases with increasing frequency so that these
higher frequency components “out run” the lower frequency components in the
wave. This simple example shows that dispersion can have a profound effect on the
pulse shape of a wave as it propagates. Such changes can make it difficult to
discern other changes, such as those induced when the propagating wave interacts
with flaws.
206 7 Propagation of Surface and Plate Waves
Below this frequency, Eq. (7.26) shows that the wave number, km, of the mode
becomes imaginary and the mode is said to be cut off. Thus, for each mode there is
only a certain minimum frequency below which the mode can propagate, and this
minimum frequency increases as the mode number increases, as shown in Fig. 7.7,
where the normalized wave speed, cm/cs, is plotted versus the non-dimensional
frequency, ωh/cs, for the m ¼ 0 to m ¼ 8 symmetrical modes as well as the anti-
symmetrical modes that we will now discuss.
For the anti-symmetrical modes we have
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
w ¼ B sin ω2 =c2s k2 y exp½ikðx ctÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð7:31Þ
∂w=∂y ¼ B cos ω2 =c2s k2 y ω2 =c2s k2 exp½ikðx ctÞ:
In this case the satisfaction of the boundary conditions leads to the requirement
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos ω2 =c2s k2 h ¼ 0; ð7:32Þ
7.2 Plate Waves: Horizontal Shearing Motions 207
which is satisfied by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ω2 =c2s k2m h ¼ mπ=2 m ¼ 1, 3, 5, . . . ð7:33Þ
i.e., the same equation as for the symmetrical modes except now the anti-
symmetrical modes correspond to odd values for m. Thus, all the expressions
obtained previously for km and cm are valid for this case as well. In Fig. 7.7 the
plot of cm/cs versus ωh/cs also shows the m ¼ 1 to m ¼ 7 anti-symmetrical modes.
The fact that cm ! 1 at the cutoff frequencies may seem to imply some
unphysical behavior at those frequencies. However, in dispersive propagation
problems we must distinguish between the frequency dependent velocity, c(ω),
which is called the phase velocity, and another frequency dependent velocity, cg(ω),
called the group velocity [2]. While the phase velocity cðωÞ ¼ ω=k, the group
velocity satisfies cg ðωÞ ¼ dω=dk, or equivalently
c ð ωÞ
c g ð ωÞ ¼ : ð7:34Þ
ω dcðωÞ
1
cðωÞ dω
For non-dispersive problems where c is a constant, the group velocity is just the
same as the ordinary phase velocity. The significance of the group velocity is that it,
not the phase velocity, represents the speed at which energy propagates in a
dispersive wave. Thus, unlike the phase velocity, the group velocity is always
well behaved and does not go to infinity. For this case of SH-mode plate waves
the group velocities, cgm, for the modes can be found analytically as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
mπ 2
c 2ffi
s
cgm ¼ cs 1 ðm ¼ 0, 1, 2, 3, . . .Þ: ð7:35Þ
2h ω
These group velocities, normalized by the shear wave speed, cs, are plotted versus
ωh/cs in Fig. 7.8. We see that except for the non-dispersive m ¼ 0 mode, all the
group velocities increase from a value of zero at the cutoff frequencies to a value
equal to the wave speed, cs, as the frequency goes to infinity. Our primary emphasis
in this book will be to describe the behavior of non-dispersive bulk waves, so we
will not give an in-depth discussion of dispersion in this chapter, but since such
behavior is inherent to plate waves, the reader should consult works that describe
those waves in more detail such as Rose [3].
In addition to the SH plate waves discussed in the last section there are also plate
waves that can exist that are combinations of P- and SV-waves. These plate waves
are called Lamb waves. Again, we consider the plate geometry of Fig. 7.3 where,
assuming plane strain conditions the displacement potentials are taken of the form
ϕ ¼ f ðyÞexp½ikðx ctÞ
ð7:36Þ
ψ ¼ gðyÞexp½ikðx ctÞ:
Placing these relations into the wave equations for the potentials then gives, as
before
d2 f =dy2 þ k2p k2 f ¼ 0
ð7:37Þ
d2 g=dy2 þ k2s k2 g ¼ 0
f ¼ AcoshðαyÞ þ CsinhðαyÞ
ð7:38Þ
g ¼ DcoshðβyÞ þ BsinhðβyÞ;
where
α2 ¼ k2 k2p
ð7:39Þ
β2 ¼ k2 k2s :
ϕ ¼ AcoshðαyÞexp½ikðx ctÞ
ð7:40Þ
ψ ¼ BsinhðβyÞexp½ikðx ctÞ;
Equation Eq. (7.41) shows that the x-component of the displacement is even in
y and the y-component is odd, resulting in the deformation of the plate for these
waves as shown in Fig. 7.9. The corresponding stresses are
τyy ¼ ρc2s β2 þ k2 AcoshðαyÞ 2ikBcoshðβyÞ exp½ikðx ctÞ
ð7:42Þ
τxy ¼ ρc2s 2ikαAsinhðαyÞ þ β2 þ k2 BsinhðβyÞ exp½ikðx ctÞ:
tanhðβhÞ 4k2 αβ
2 ¼ 0; ð7:44Þ
tanhðαhÞ k2 þ β2
which must be solved for the wave number, k, or phase velocity c. Figure 7.10
shows the dispersion curves for a number of these extensional modes for an
aluminum plate, where the phase velocity, c, (in mm/μs) is plotted versus fh (the
frequency, f, times half plate width h (in MHz-mm)). As can be seen there are cutoff
frequencies for all the modes except one, which is called the fundamental exten-
sional mode. We can extract some analytical results from Eq. (7.44) for this mode.
For example, at low frequencies the hyperbolic tangent terms in Eq. (7.44) can be
approximated as
tanhðαhÞ ffi αh
tanhðβhÞ ffi βh
Expanding out the terms in Eq. (7.45) and using the relationships between the wave
speeds cp, cs and the elastic constants:
Eð1 νÞ E
c2p ¼ , c2s ¼
ρð1 þ νÞð1 2νÞ 2ρð1 þ νÞ
The wave speed, c0, found in this manner for the lowest order (fundamental)
extensional mode, where cs < c0 < cp , agrees with the extensional wave speed
found in plates from elementary plate theory [4]. In contrast, at high frequencies,
the tanh functions in Eq. (7.44) can be approximated as
tanhðαhÞ ffi 1
tanhðβhÞ ffi 1
7.3 Lamb Waves 211
giving
2
k2 þ β2 ¼ 4k2 αβ ð7:47Þ
which is just the equation for Rayleigh surface waves, whose solution is c ¼ cR . Thus,
in the fundamental mode the phase velocity varies between these two limits. The other
higher order modes all have cutoff frequencies below which they do not exist and it
can be shown that they asymptotically go to the shear wave speed, cs, a high
frequencies. We will not discuss these details further here but much more information
can be found in sources such as Mindlin [5] and Meeker and Meitzler [6].
ϕ ¼ CsinhðαyÞexp½ikðx ctÞ
ð7:48Þ
ψ ¼ DcoshðβyÞexp½ikðx ctÞ
which are even in y for uy and odd in y for ux, resulting in the flexural (bending)
deformations of the plate as shown in Fig. 7.11. The corresponding stresses in this
case are
τyy ¼ ρc2s k2 þ β2 CsinhðαyÞ 2ikβDsinhðβyÞ exp½ikðx ctÞ
ð7:50Þ
τxy ¼ ρc2s 2ikαCcoshðαyÞ þ k2 þ β2 DcoshðβyÞ exp½ikðx ctÞ
and setting the determinant of the coefficients of this equation equal to zero now
yields
2 2
tanhðβhÞ k þ β2
¼ 0: ð7:52Þ
tanhðαhÞ 4k2 αβ
Figure 7.12 shows the plot of the dispersion curves where again the phase velocity,
c, in mm/μs is plotted versus fh in MHz-mm for an aluminum plate. Again, at low
and high frequencies we can obtain some analytical results from Eq. (7.52) for the
one mode that does not have a cutoff frequency, which is called the fundamental
flexural mode. At low frequencies, an expansion similar to that for the extensional
case can be done, although the algebra is rather lengthy and care must be taken to
keep sufficient lowest order terms in the expansion. For explicit details, we refer the
reader to Bland [7], and merely quote the result for the phase velocity, cb, of the
fundamental flexural mode which is
1=4
Dp
cb ¼ ω1=2 ; ð7:53Þ
2ρh
where
8μðλ þ μÞh3
Dp ¼ ð7:54aÞ
3ðλ þ 2μÞ
is the flexural rigidity of the plate, which can also be written in the alternate forms
EI 8 c2s
2
Dp ¼ ¼ c cs ρh ;
2 3
ð7:54bÞ
1 ν2 3 c2p p
7.3 Lamb Waves 213
where c is the mode phase velocity, (cp, cs) are the compressional and shear wave
speeds, fh is the frequency times the half-thickness of the plate and type is a string
given either as ‘s’ for the symmetrical extensional modes or ‘a’ for the anti-
214 7 Propagation of Surface and Plate Waves
symmetrical bending modes. This approach encounters difficulties when the phase
velocity and fh values are both very small, as can occur only for the fundamental
flexural mode. This is because the actual value of the function in that case is a result
of the cancellation of terms, as mentioned earlier, a process that is difficult to
capture numerically. Although it is useful to have plots of the dispersion curves for
the modes, to use that data one needs to have the actual values of the phase velocity
curves separately for each particular mode. These values can also be easily obtained
by extracting the individual curves (contours) from the internal data in the
MATLAB® plotting function contour. This extraction is done with another
MATLAB® script, dispersion_curves. This script again evaluates the contours of
the dispersion functions as done in dispersion_plots and then extracts and reorders
the individual dispersion curves according to the lowest frequency at which they
exist. The script asks for user inputs for the type of mode (‘s’ for symmetrical
(extensional) modes and ‘a’ for anti-symmetrical (flexural) modes) and the number
of the mode, n, which the user wants plotted, where n ¼ 1 for the fundamental mode
and n ¼ 2, 3, . . . are the higher modes, ordered starting with the one with the lowest
cutoff frequency and proceeding to the curve with the next lowest cutoff, etc. For
the fundamental flexural mode, the analytical values of Eq. (7.51) are used at small
fh values (below fh ¼ 0.02) instead of the unreliable values found from contour.
Figure 7.13a, b show the first (fundamental) and second extensional modes for an
aluminum plate obtained from dispersion_curves while Fig. 7.14a, b show the first
and second flexural modes.
Fig. 7.13 Plots of wave speed (in mm/μs) versus fh for (a) the fundamental extensional mode, and
(b) the next higher extensional mode
7.6 Problems 215
Fig. 7.14 Plots of wave speed (in mm/μs) versus fh for (a) the fundamental flexural mode, and (b)
the next higher flexural mode.
Rayleigh waves and plate waves are not the only types of waves possible in elastic
media with boundaries. Waves in cylindrical and spherical geometries, Stoneley
(surface) waves at fluid-solid and solid-solid interfaces, and Love waves in a layer
over a half-space are all examples of other waves that can exist and be potentially
used in NDE applications. Further details on the properties of these waves can be
found in a variety of wave propagation and seismology references such as [8, 10].
Our treatment of surface and plate waves in this Chapter has been brief. More
detailed discussions of surface waves and waves in plates and rods can be found in
the monographs by Viktorov [11], Redwood [12], and Nayfeh [13] and in the books
of Rose [3, 14], Achenbach [1], Miklowitz [15], Bland [7], and Brekhovskikh
[16]. Mindlin [5], and Meeker and Meitzler [6] also summarize many of the main
results found in the literature.
7.6 Problems
7.1. Let c=cs ¼ 1 þ Δ in our equation for the Rayleigh wave speed Eq. (7.8). Since
we know that the Rayleigh wave speed is close to the shear wave speed for
most structural materials, Δ is typically small. By expanding Eq. (7.8) for
216 7 Propagation of Surface and Plate Waves
c 0:875 þ 1:125ν
¼ ;
cs 1þν
References
1. J.D. Achenbach, Wave Propagation in Elastic Solids (American Elsevier, New York, 1973)
2. J. Lighthill, Waves in Fluids, 2nd edn. (Cambridge University Press, Cambridge, 2001)
3. J.L. Rose, Ultrasonic Waves in Solid Media (Cambridge University Press, Cambridge, 1999)
4. J.A. Hudson, The Excitation and Propagation of Elastic Waves (Cambridge University Press,
New York, 1980)
5. R.D. Mindlin, Waves and vibrations in isotropic elastic plates, in Structural Mechanics, ed. by
J.N. Goodier, N.J. Hoff (Pergamon Press, New York, 1960)
6. T.R. Meeker, A.H. Meitzler, Guided wave propagation in elongated cylinders and plates, in
Physical Acoustics, ed. by W.P. Mason (Academic Press, New York, 1964)
7. D.R. Bland, Wave Theory and Applications (Oxford University Press, New York, 1988)
8. J. Pujol, Elastic Wave Propagation and Generation in Seismology (Cambridge University
Press, Cambridge, 2003)
9. F. Honarvar, E. Enjilela, A.N. Sinclair, An alternative method for plotting dispersion curves.
Ultrasonics 49, 15–18 (2009)
10. K. Aki, P.G. Richards, Quantitative Seismology (W.H. Freeman, San Francisco, 1980)
11. I.A. Viktorov, Rayleigh and Lamb Waves (Plenum Press, New York, 1967)
12. M. Redwood, Mechanical Waveguides (Pergamon Press, New York, 1960)
13. A.H. Nayfeh, Wave Propagation in Layered Anisotropic Media (Elsevier, New York, 1995)
14. J.L. Rose, Ultrasonic Guided Waves in Solid Media (Cambridge University Press, Cambridge,
2014)
15. J. Miklowitz, The Theory of Elastic Waves and Waveguides (North Holland, Amsterdam,
1978)
16. L.M. Brekhovskikh, Waves in Layered Media (Academic, New York, 1960)
Chapter 8
Ultrasonic Transducer Radiation
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_8) contains supplementary material, which is available to authorized users.
First, we go back to our integral representation theorem for the pressure in a fluid
(Eq. (5.11)) with f ¼ 0 (no body forces). Applying that theorem to the half-space V
ðz
0Þ in Fig. 8.1 and assuming the radiated pressure field satisfies the Sommerfeld
radiation conditions, then the pressure at a point x in V can be written as:
ð
pðx; ωÞ ¼ ½Gðx; y; ωÞ∂pðy; ωÞ=∂nðyÞ pðy; ωÞ∂Gðx; y; ωÞ=∂nðyÞdSðyÞ:
Sp
ð8:1Þ
This representation theorem is not directly useful since in order to calculate the
pressure in the half-space V ðz
0Þ, we would have to know both the pressure,
p, and its normal derivative, on the x–y plane. However, recall that to obtain this
representation theorem we used a fundamental solution, G, given by
expðikr Þ
G¼ ; ð8:2Þ
4πr
If we now consider a new fundamental solution, G*, due to a point source in V and
an image source (Fig. 8.2) outside V, where
expðikr Þ exp ikr *
G ¼
*
þ ð8:4Þ
4πr 4πr*
8.1 Planar Piston Transducer in a Fluid 221
y Sp
V (z ≥ 0)
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ jx yj ¼ ðx1 y1 Þ2 þ ðx2 y2 Þ2 þ ðx3 y3 Þ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:5Þ
r * ¼ x* y ¼ ðx1 y1 Þ2 þ ðx2 y2 Þ2 þ ðx3 þ y3 Þ2 ;
Thus, if we replace G by G* in the reciprocal theorem, and follow the same steps
taken to obtain Eq. (5.11), we would find instead, for any point x in V
ð
1 ∂pðy; ωÞ expðikr Þ
pðx; ωÞ ¼ dSðyÞ: ð8:7Þ
2π ∂nðyÞ r
Sp
But
(the negative sign occurs because the outward unit normal to V is in the negative z-
direction on Sp) and we can write, finally
ð
iωρ expðikr Þ
pðx; ωÞ ¼ vz ðy; ωÞ dSðyÞ: ð8:8Þ
2π r
Sp
examples of both integral types).] In this form, this integral model is valid for any
assumed velocity distribution on the planar surface. For the special case where the
velocity vz ¼ v0 ðωÞ is spatially constant on S and vz ¼ 0 elsewhere, we find
ð
iωρv0 ðωÞ expðikr Þ
pðx; ωÞ ¼ dS: ð8:9Þ
2π r
S
Thus, the motion of the surface of the transducer in this model acts in unison as a
“piston”, and this piston is surrounded by an infinite planar “baffle” that is motion-
less. A real ultrasonic transducer, of course, has a finite active crystal area that
moves and is attached to the relatively rigid frame of a finite housing. However, if
the crystal is excited uniformly over its face and if the sound beam is primarily
generated directly ahead of the active area into the fluid, the piston transducer
model of Eq. (8.9) will likely give a good representation of the sound field
generated. Physically, we see that Eq. (8.9) says that the pressure wave field of
the transducer can be considered to arise from a superposition of elementary point
sources (spherical waves) over the face of the transducer so this model can also be
considered to be a point source model.
Along the central axis of a circular piston transducer of radius a the integral in
Eq. (8.9) can be readily evaluated to obtain an analytical expression for the pressure
wave field. To see this, note that because of the symmetry of this case (Fig. 8.3) and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
since r ¼ ρ2 þ z2 , the area differential dS ¼ 2πρ dρ ¼ 2πr dr, so that Eq. (8.9)
reduces to the one-dimensional integral
pffiffiffiffiffiffiffiffiffi
r¼ ð
z2 þa2
edge wave
ρcvo δ(t – z 2 + a 2 / c)
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffii
pðz; ωÞ ¼ ρcv0 ðωÞ expðikzÞ exp ik z2 þ a2 : ð8:11Þ
Equation (8.11) shows that the on-axis pressure expression contains two propaga-
tion terms, one that appears to be a wave that has directly traveled in the z-direction
from the face of the transducer to the evaluation point, and a second wave that
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
appears to have gone through a distance z2 þ a2 , which is the distance from the
transducer edge to the evaluation point (Fig. 8.4). These two waves are called the
direct wave, and edge wave, respectively. If we assume v0 ðωÞ ¼ v0 where v0 is
independent of frequency (so that in the time domain v0 ðtÞ ¼ v0 δðtÞ, i.e. the piston
source has an impulsive-like motion), then multiplying Eq. (8.11) by expðiωtÞ=2π
and inverting the resulting expression into the time domain gives
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
pðz; tÞ ¼ ρcv0 δðt z=cÞ δ t z2 þ a2 =c ; ð8:12Þ
where the on-axis direct and edge wave pulses appear explicitly (Fig. 8.5) as
isolated wave arrivals. Although the on-axis direct and edge waves both appear
as plane-wave terms, an evaluation of the entire pressure wave field (for an
impulsive input) would show that only the direct wave is indeed a plane wave
generated directly from the face of the transducer but contained to a cylindrical
region extending normally from the transducer face (Fig. 8.6), a region which we
will call the “main beam” of the transducer. However, the edge wave is not a plane
wave at all but has a curved wave front which propagates from the entire transducer
edge in an overlapping doughnut-like fashion in 3-D (obtained by rotating the edge
waves shown in the vertical plane of Fig. (8.6) about the z-axis).
224 8 Ultrasonic Transducer Radiation
edge wave
To examine the explicit behavior of the on-axis pressure, first note that this pressure
can be rewritten as
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
pðz; ωÞ ¼ 2iρcv0 ðωÞexp ik z2 þ a2 þ z =2 sin k z2 þ a2 z =2 :
ð8:13Þ
This expression shows that there will be on axis nulls (locations of zero pressure),
where
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
sin k z2 þ a2 z =2 ¼ 0: ð8:14Þ
Solving for the square root term in Eq. (8.15) and then squaring both sides of the
resulting equation, the quadratic term in z cancels, leaving an equation that can be
solved for the null locations, zn, as
ðkaÞ2 ð2nπ Þ2
z ¼ zn ¼ : ð8:16Þ
4nπk
ðkaÞ2 a2
zn ¼ ¼ ; ð8:17Þ
4πnk 2nλ
where λ is the wavelength and where the distance N ¼ a2 =λ is called the near field
distance. In terms of this distance, the null locations then are
N
zn ¼ n ¼ 1, 2, 3, . . . ; ð8:18Þ
2n
which shows that the null farthest from the transducer (n ¼ 1) occurs at approxi-
mately one-half a near field distance.
Since there are on-axis pressure nulls in the wave field of a circular piston
transducer, we expect there will also be corresponding locations of local maximum
pressure. To find the on-axis maxima, we first solve for the square of the magnitude
of the on-axis pressure from Eq. (8.13), which gives
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i2
jpðz; ωÞj2 ¼ 4 ρcv0 ðωÞ sin k z2 þ a2 z =2 : ð8:19Þ
Rejecting the null locations where the sine term in Eq. (8.20) vanishes, it follows
that we must have
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
cos k z2 þ a2 z =2 ¼ 0; ð8:21Þ
Since the form of Eq. (8.22) is identical to that found for the nulls (Eq. (8.15)) if we
make the replacement n ! ð2m þ 1Þ=2, it follows that the locations of the on-axis
maxima, zm, are given exactly by
226 8 Ultrasonic Transducer Radiation
For ka large, as in the case of the nulls, we can write the maxima locations in terms
of the near field distance N by neglecting the second term in the numerator of
Eq. (8.23). The result is, approximately,
N
zm ¼ m ¼ 0, 1, 2 . . . ; ð8:24Þ
2m þ 1
which shows that the last on-axis maximum occurs at one near field distance.
The pressure in the far field region of the transducer, therefore, is in the form of a
spherical spreading wave. This is reasonable since as the distance from the trans-
ducer increases, the transducer eventually acts like a point source.
If we assume v0 is a constant in frequency, multiply Eq. (8.25) by expðiωtÞ=2π,
and sum over all frequencies we can obtain the far field on-axis time domain
response as
ð
þ1
ρv0 a2
pðz; tÞ ¼ iωexp½iωðt z=cÞdω
4πz
8
1 9
ð
þ1
ρv0 a d <
2 = ð8:26Þ
¼ 1=2π exp½iωðt z=cÞdω
2z dt: ;
1
ρv0 a2 dδðt z=cÞ
¼ ;
2z dt
where dδ/dt is the derivative of the delta function, which behaves like a “doublet”
(see Fig. 8.11).
Figure 8.7 shows an on-axis plot of the magnitude of the pressure obtained from
Eq. (8.11) or Eq. (8.13) versus normalized distance, z/N (varying from 0.18 to 4.0)
for a transducer with ka ¼ 100. At distances less than one near field distance we can
see the multiple nulls and maxima. This strong structure disappears beyond one
near field distance and gradually approaches the far field or spherical spreading
8.1 Planar Piston Transducer in a Fluid 227
Fig. 8.7 Magnitude of the normalized on-axis pressure for a plane piston transducer (ka ¼ 100)
showing several on-axis nulls and maxima and the spherical decay of the wave field in the far field
region
region where Eq. (8.25) is valid. As a rule of thumb, the far field region where
Eq. (8.25) can be used, occurs at distances greater than about three near field
distances, as can be verified from Fig. 8.7.
From our previous discussions it is obvious that a transducer does not generate
purely a plane wave type of disturbance. There are in general both direct (plane)
wave contributions and edge waves, and these waves arrive separately at different
times. However, a real limited bandwidth transducer has a motion of its surface of
finite time duration. Thus, if we are not too close to the transducer face, the direct
and edge waves will merge together in what appears to be a single response. In
addition, since most NDE transducers operate at high frequencies ðka
1Þ this
response travels approximately normal to the transducer face in the form of a well
collimated beam. In this case, it makes sense to consider the transducer output as a
quasi-plane wave, i.e. a plane wave that is modified by a frequency dependent
diffraction correction.
To obtain the on-axis diffraction correction for a circular piston transducer, we
go back to the exact on-axis response (Eq. (8.11)) and make the paraxial approx-
imation (see Chap. 6), i.e. we assume that all the waves arriving at point z on the
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
axis travel nearly parallel to the z-axis so that we have z2 þ a2 ffi z þ a2 =2z. In this
case the on-axis pressure reduces to
228 8 Ultrasonic Transducer Radiation
pðz; ωÞ ¼ ρcv0 ðωÞexpðikzÞ 1 exp ika2 =2z ; ð8:27Þ
which is indeed in the form of a plane wave modified by a diffraction correction, C1,
where
C1 ða; ω; zÞ ¼ 1 exp ika2 =2z : ð8:28Þ
For points not on the axis of a transducer, it is in general not possible to obtain
completely analytical results for the pressure in the frequency domain, even when
the transducer is circular (although later in this chapter we will obtain some exact
analytical time domain expressions for an impulsive input to a circular transducer).
In the off-axis far field limit, however, we can obtain some explicit results in the
frequency domain. To see this, consider a circular transducer of radius, a, and a
general off-axis point x as shown in Fig. 8.8. From the geometry we have
r 2 ¼ d2 þ z2 ¼ d2 þ R2 cos 2 θ
ð8:30Þ
d 2 ¼ ρ2 þ ðR sin θÞ2 2ρR sin θ cos ϕ;
so that
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ R2 þ ρ2 2Rρ sin θ cos ϕ: ð8:31Þ
ð ða
2π
iωρv0 ðωÞ expðikRÞ
pðx; ωÞ ¼ exp½ikρ sin θ cos ϕρ dρ dϕ: ð8:32Þ
2π R
0 0
2ðπ
ða
pðx; ωÞ ¼ iωρv0 ðωÞ J 0 ðkρ sin θÞρ dρ: ð8:34Þ
0
ða
aJ 1 ðαaÞ
J 0 ðαρÞρ dρ ¼ ; ð8:35Þ
α
0
Equation (8.36) shows that the far field off-axis behavior consists of three terms.
The first term is merely a frequency dependent coefficient, while the second term
represents a spherical spreading wave. The third term describes the angular depen-
dence of the amplitude of this spherical wave. Figure 8.9 plots the angular term
explicitly versus θ, showing the existence of a definite “lobe” structure in the far
field. The first null location, θnull, of the J1(ka sin θ)/ka sin θ function occurs when
ka sin θ ffi 3:83. This defines the main central lobe of the angular diffraction pattern
of the transducer. Since θnull is often small and θnull ffi xnull =z where xnull is the
perpendicular distance (radius) of the evaluation point x from the axis (Fig. 8.10),
this null location can be approximately written as
Fig. 8.9 Angular diffraction pattern in the far field of a circular plane piston transducer
We can also obtain the far field off-axis pressure versus time from Eq. (8.36) when
v0 is a constant by again multiplying Eq. (8.36) by expðiωtÞ=2π and summing over
all frequencies to obtain
ð
þ1
iρacv0
pðx; tÞ ¼ J 1 ðωa sin θ=cÞexp½iωðt R=cÞdω: ð8:39Þ
2πR sin θ
1
But, since the J1 function is an odd function in frequency, only the product with the
odd (sine) part of the complex exponential is non-zero, and the integral can be
rewritten as
ð
1
ρacv0
pðx; tÞ ¼ J 1 ðωa sin θ=cÞ sin ½ωðt R=cÞdω; ð8:40Þ
πR sin θ
0
where we have used the fact that the integrand is even to halve the integration range.
The ω integration can then be evaluated since [1]
8.1 Planar Piston Transducer in a Fluid 231
8
ð
1 β=α
< pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jβ j < α
J 1 ðαxÞ sin ðβxÞdx ¼ : ð8:41Þ
: α β
2 2
0 0 jβ j > α
This yields
8
< ρacv0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
> ðt R=cÞ=ða sin θ=cÞ
ffi jt R=cj < a sin θ=c
pðx; tÞ ¼ πR sin θ a2 sin 2 θ=c2 ðt R=cÞ2 : ð8:42Þ
>
:
0 otherwise
The behavior of Eq. (8.42) is plotted in Fig. 8.11. As can be seen from that figure,
this impulse response of the transducer in the far field consists of a finite duration,
anti-symmetrical pulse which has singularities showing up at the earliest and latest
times in the pulse. These singularities are called “flash points” and correspond to the
edge wave contributions coming from the nearest and farthest points on the edge,
respectively, to the evaluation point x. As the angle θ decreases, these flash points
come closer together, finally merging into the on-axis doublet found previously
(Eq. (8.26)).
Now, we return to the Rayleigh-Sommerfeld equation and note that we can drop a
line parallel to the z-axis from a general off-axis point x to a point y0 in the plane of
the transducer and set up a set of cylindrical coordinates (ρ, ϕ) in the transducer
plane about this point (Fig. 8.12). In terms of these cylindrical coordinates we can
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
write r ¼ z2 þ ρ2 and dS ¼ ρ dρ dϕ, as in the on-axis case. In the paraxial
approximation we have r ffi z þ ρ2 =2z so that Eq. (8.9), becomes
ðð
iωρv0 ðωÞexpðikzÞ
pðx; ωÞ ¼ exp ikρ2 =2z ρ dρ dϕ; ð8:43Þ
2πz
232 8 Ultrasonic Transducer Radiation
b y
ρ φ y
y0
x
where we have kept only the lowest order approximation for r (r ¼ z) in the
denominator of Eq. (8.9) but both terms in the approximation for the phase of the
complex exponential exp(ikr). To further evaluate Eq. (8.43), we need to consider
separately three cases: (1) point x in the main beam of the transducer, (2) point
x outside this main beam, and (3) point x on the edge of the main beam.
Case 1: Point x in the main beam
In this case the point y0 lies within the active area of the transducer face as shown in
Fig. 8.12 and ρ is a single-valued function of ϕ which varies from 0 to ρ ¼ ρe ðϕÞ (its
value on the edge) so that Eq. (8.43) becomes
ð ρðe
2π
iωρv0 ðωÞexpðikzÞ
pðx; ωÞ ¼ exp ikρ2 =2z ρ dρ dϕ: ð8:44Þ
2πz
0 0
ð
2π
ρcv0 ðωÞexpðikzÞ 2
pðx; ωÞ ¼ exp ikρe =2z 1 dϕ: ð8:45Þ
2π
0
The ϕ integration for the constant integrand can also be done, to yield
8 2ðπ
9
< 1 2 =
pðx; ωÞ ¼ ρcv0 ðωÞexpðikzÞ 1 exp ikρe =2z dϕ : ð8:46Þ
: 2π ;
0
8.1 Planar Piston Transducer in a Fluid 233
b
φ
φ = φ–
y0
ρe–
C+ ρe+
C–
φ = φ– + γ
ϕðþγ ρðþ
e
iωρv0 ðωÞexpðikzÞ
pðx; ωÞ ¼ exp ikρ2 =2z ρ dρ dϕ ð8:47Þ
2πz
ϕ ρ
e
ϕðþγ
ρcv0 ðωÞexpðikzÞ n h i h io
exp ik ρþ =2z exp ik ρ
2 2
pðx; ωÞ ¼ e e =2z dϕ:
2π
ϕ
ð8:48Þ
φ ≅ φ– + π
8 9
< ð =
ρcvo ðωÞexpðikzÞ
pðx; ωÞ ¼ exp ikρ2e =2z dϕ ; ð8:49Þ
2π : ;
C
where the edge integral is interpreted in the principal value sense, i.e.
8.1 Planar Piston Transducer in a Fluid 235
ð ð
p:v: exp ikρ2e =2z dϕ ¼ lim exp ikρ2e =2z dϕ: ð8:53Þ
Cε !0
C CCε
Comparing Eqs. (8.53), (8.49), and (8.50) for all three cases, we see that for any
location x we can write the off-axis pressure in the single form
8 9
< ð
1 2 =
pðx; ωÞ ¼ ρcv0 ðωÞexpðikzÞ Θ exp ikρe =2z dϕ ; ð8:54Þ
: 2π ;
C
where
8
<1 x inside the main beam
Θ ¼ 1=2 x on the beam edge : ð8:55Þ
:
0 x outside the main beam
and, again we note that the integral is interpreted in the principal value sense when
x is on the edge of the main beam. From Eq. (8.55) it follows that the diffraction
correction, C1, for an arbitrarily shaped planar piston probe and any point, x, in that
transducer’s wave field can be written as
8 9
< ð
1 2 =
C1 ðx; ωÞ ¼ Θ exp ikρe =2z dϕ ; ð8:56Þ
: 2π ;
C
which reduces to the explicit expression found earlier (Eq. (8.28)) for a circular
transducer when x is on the transducer axis.
Within the paraxial approximation, Eq. (8.54) gives the pressure anywhere in the
wave field of a planar piston transducer. However, that result can be easily
generalized to obtain an exact evaluation of the pressure wave field everywhere.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
To see this, recall Fig. 8.12 and note that we have r ¼ ρ2 þ z2 so that we can
rewrite the area integration in the Rayleigh-Sommerfeld equation in terms of r as
dS ¼ r dr dϕ Thus, Eq. (8.9) reduces to
ð
iωρv0 ðωÞ
pðx; ωÞ ¼ expðikr Þdr dϕ; ð8:57Þ
2π
S
which is similar in form to Eq. (8.43). Therefore, following exactly the same steps,
as done before to evaluate Eq. (8.43), we find in this case
236 8 Ultrasonic Transducer Radiation
8 9
< ð =
1
pðx; ωÞ ¼ ρcv0 ðωÞ Θ expðikzÞ expðikr e Þdϕ ; ð8:58Þ
: 2π ;
C
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where re ¼ ρ2e þ z2 . Again, inside the main beam r is a single-valued function of
ϕ so that Eq. (8.58) is equivalent to
8 2ðπ
9
< 1 =
pðx; ωÞ ¼ ρcv0 ðωÞ expðikzÞ expðikr e Þdϕ : ð8:59Þ
: 2π ;
0
Outside the main beam Eq. (8.58) reduces, as before, to two integrals:
ϕðþγ
ρcv0 ðωÞ
pðx; ωÞ ¼ exp ikr þ
e exp ikr e dϕ; ð8:60Þ
2π
ϕ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi
where r e ¼ ρe þ z2 and r þ
e ¼ ρþ e þ z2 . On the boundary of the main
beam, the integral in Eq. (8.58) again is interpreted in the principal value sense.
Although the integrals in these exact expressions for the pressure must in general be
done numerically, this evaluation is not difficult and so Eq. (8.58) serves as a
convenient form to evaluate the piston transducer wave field exactly. Note that the
original Rayleigh-Sommerfeld integral (Eq. (8.9)) can also be evaluated numeri-
cally, but this would involve double integrals, so that the use of Eq. (8.58) is more
efficient and also shows explicitly the existence of the separate direct and edge
wave terms.
Although the coordinates (ρ, ϕ) are convenient for describing the pressure wave
field in simple terms, the numerical evaluation of the edge wave integral may be
more conveniently accomplished for any shape of a planar piston transducer by
rewriting Eq. (8.58) in a completely coordinate-invariant form. First consider the
case when x is inside the main beam as shown in Fig. 8.15. If we let er and ea be unit
vectors as shown, and define et to be a unit tangent vector on the transducer edge as
shown (where the edge is transversed in a counterclockwise direction), then from
the geometry it follows that
and
ds ¼ ρe dϕ= cos α;
φ re er
y0 θ x
ea
z
et α er × ea
y0 ρe
ðe r e a Þ et 2 2
¼ ρe =r e dϕ
re
and we have
ρ2e
¼ sin 2 θ ¼ 1 cos 2 θ ¼ 1 ðer ea Þ2 ;
r 2e
so that, finally
1 ðn er Þ et ds
dϕ ¼ h i ð8:61Þ
r e 1 ðer nÞ2
in terms of the unit normal to the transducer n ¼ ea . When x is outside the main
beam (Fig. 8.16a, b), an entirely similar analysis shows that
1 ðn er Þ et ds
dϕ ¼ h i ð8:62Þ
r
e 1 ð er n Þ2
b
y0 ea
x
er
φ ρe+ re+
C+
et
where r e ¼ r þ
e on Cþ and r e ¼ r e on C as before and the integral is interpreted in
the principal value sense when x is on the edge of the main beam.
where ln is the length of the segment. The distance, ρe, to the edge can be
approximated as the distance to each centroid of a segment and the distance from
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the segment to the point in the fluid, r e ¼ ρ2e þ z2 can likewise be taken to that
centroid, so that
8.1 Planar Piston Transducer in a Fluid 239
Fig. 8.17 Geometry parameters for approximating a boundary diffraction wave transducer model
for (a) when in the main beam of the transducer, and (b) when outside that main beam
xnþ1 þ xn
xcn ¼
2
y þ yn
ycn ¼ nþ1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
ρen ¼ ðxcn ρ0 Þ2 þ y2cn
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r en ¼ ρ2en þ z2 :
All of these approximations can be used for evaluation of the wave field when in the
main beam (Fig. 8.17a) or outside that main beam (Fig. 8.17b). When ρ0 a
(Fig. 8.17a) so that we are in the main beam of the transducer the normalized
pressure wave field, p/ρcv0, can be evaluated by assuming the segments are
sufficiently small so that the integrand of the edge integral, exp(ikre), can be
approximated by its (constant) centroidal value, exp(ikren) over each segment,
and we find
pðρ0 ; z; ωÞ 1 XN
¼ expðikzÞ expðik r en ÞΔϕn : ð8:64aÞ
ρcv0 ðωÞ 2π n¼1
[Note: In implementing Eq. (8.64) it is convenient to define the segments so that the
centroid of the first (n ¼ 1) segment lies along the x-axis, i.e yc1 ¼ 0. When ρ0 < a,
then Δϕ1 ! π for this first segment as ρ0 ! xc1 ffi a and expðik re1 Þ ! expðik zÞ so
that the first element produces a value of exp(ikz)/2, which when combined with the
first term in Eq. (8.64a), gives the proper value seen in Eq. (8.58) when on the
boundary of the main beam. The principal value of the edge integral is then
automatically found from the sum of all the other segments].
When ρ0 > a, we need to distinguish the Cþ and C parts of the boundary. This
can be done by noting that when the angle ψ n to the centroid of a segment satisfies
ψ n < ψ or ψ n > 2π ψ we are on C while if ψ < ψ n < 2π ψ we are on Cþ
240 8 Ultrasonic Transducer Radiation
(Fig. 8.17b), where ψ ¼ cos 1 ða=ρ0 Þ. Assuming that the segments are set up so
that yc1 ¼ 0, as mentioned above, then ψ 1 ¼ 0 and all the ψ n values are given from
h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
Δψ ¼ cos 1 xc2 = x2c2 þ y2c2
ψ n ¼ ðn 1ÞΔψ ðn ¼ 1, 2, . . . , N Þ:
Thus, if we identify the values of n for those segments that satisfy these constraints
and call them ðnþ ; n Þ, respectively, for Cþ and C , respectively, for outside the
main beam we have
pðρ0 ; z; ωÞ 1 X 1 X
¼ expðik r en ÞΔϕn expðik r en ÞΔϕn : ð8:64bÞ
ρcv0 ðωÞ 2π n 2π nþ
Again, near the boundary of the main beam this gives the proper limit since when
ρ0 ffi a only the first element is in C and Δϕ1 ! π so the first sum in Eq. (8.64b)
becomes simply exp(ikz)/2 and the second sum over all other elements generates
the principal value integral.
Equations (8.64a, 8.64b) have been implemented in MATLAB® for a circular
planar piston transducer using the functions bdw_model and bdw_fluid. The calling
sequence for the bdw_model function is
where (rho, z) are the radial and axial locations (in mm) where the normalized
pressure, p ¼ pðρ0 ; z; ωÞ=ρcv0 ðωÞ, is to be evaluated at a frequency, f, (in MHz) for
a transducer of radius, a, (in mm) radiating into a fluid with a wave speed,
c, (in m/s). The last parameter, N, specifies the number of segments to use to
approximate the edge integral. This function evaluates the pressure in the (ρ0, z)
cylindrical coordinates and requires all of its input values to be scalars, so that one
must call the function multiple times for different combinations of parameters. To
use this function in a more versatile manner a second MATLAB® function
bdw_fluid has been written whose calling sequence is
This function uses Cartesian coordinates (x, y, z) in the fluid and allows those
coordinates to be scalars, vectors, or two-dimensional matrices. The bdw_fluid
function converts the inputs to cylindrical coordinates and makes the necessary
calls to the function bdw_model to evaluate the normalized pressure. The bdw_fluid
model also has a seventh optional input parameter, Nopt, which specifies the
number of segments used to approximate the edge integral. If Nopt is not given
then the number of segments is automatically chosen to make segment length equal
to a tenth of a wavelength, which is generally a very conservative value that allows
one to consider the edge integral integrand, exp(ikre), to be a constant over each
8.1 Planar Piston Transducer in a Fluid 241
Fig. 8.18 Magnitude of the on-axis pressure for a 6.35 mm radius circular piston transducer
radiating into water at 5 MHz as calculated with the MATLAB® function bdw_fluid
segment, as assumed in Eqs. (8.64a, 8.64b). However, Nopt can be given explicitly
in the call to bdw_fluid to override the default and specify the number of segments
to use. Generally a value of N corresponding to a segment length of one wavelength
or less should be sufficient for most NDE transducers.
The frequency input parameter, f, must be a scalar so that if one wants to use
bdw_fluid to generate a pulse, one must make multiple calls to this function at the
necessary frequencies and then perform an inverse FFT (see Appendix A).
The function bdw_fluid can serve as a benchmark model to test other more
complex and more approximate beam models. The function also lets one gain a
better understanding of the complex wave field of the transducer in the near field.
For a small sampling of such capabilities, consider a 6.35 mm radius transducer
radiating into water at 5 MHz. Figure 8.18 shows the magnitude of the on-axis
pressure in the near field generated by bdw_fluid for this transducer when Nopt is
not specified. As can be seen from Fig. 8.18 the last on-axis maximum occurs at a
z-distance of about 140 mm. Figure 8.19 shows the cross-axis pressure profile at
this location, with the expected on-axis maximum and visible side lobes. If instead
we evaluate the cross-axis wave field at z ¼ 70 mm, which is near the last on-axis
null, we see that null appear in the profile shown in Fig. 8.20 together with much
larger off-axis values. Figure 8.21 shows a full 2-D cross-sectional view of the wave
field of the transducer in the near field, where the on-axis nulls and maxima are
clearly visible as is the complex structure in the beam cross section. For cases like
this one, when there are many calls made to bdw_model it may be useful to choose
an N explicitly that is smaller than the default value to reduce the calculation time.
With the default value for N for this example (N ¼ 1350), the calculation time was
29.2 s to calculate the 100,000 points in the image of Fig. 8.21, but when N was
specified instead as N ¼ 100, the calculation time was reduced to 4.9 s.
242 8 Ultrasonic Transducer Radiation
Fig. 8.19 Magnitude of the cross-axis pressure at z ¼ 140 mm for a 6.35 mm radius circular piston
transducer radiating into water at 5 MHz as calculated with the MATLAB® function bdw_fluid
Fig. 8.20 Magnitude of the cross-axis pressure at z ¼ 70 mm for a 6.35 mm radius circular piston
transducer radiating into water at 5 MHz as calculated with the MATLAB® function bdw_fluid
Another wave to write the pressure wave field for a circular transducer of
radius a is to use Eq. (8.63) and express the edge wave term in terms of polar
coordinates (ρ0, ψ) with respect to the center of the transducer (Fig. 8.22). In this
case we have
8.1 Planar Piston Transducer in a Fluid 243
Fig. 8.21 Cross sectional image of the magnitude of the pressure wave field of a 6.25 mm radius
circular piston transducer radiating into water at 5 MHz as calculated with the MATLAB®
function bdw_fluid. (Note the unequal scales on z and ρ0)
1 ðn er Þ et ds cos α ds a cos α dψ
h i¼ ¼
r e 1 ðer nÞ2 ρe ρe
aρe cos α dψ aða ρ0 cos ψ Þdψ
¼ ¼ 2
ρ2e a þ ρ20 2aρ0 cos ψ
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r e ðψ Þ ¼ a2 þ ρ20 þ z2 2aρ0 cos ψ ;
244 8 Ultrasonic Transducer Radiation
Although Eq. (8.58) (or, equivalently, Eq. (8.63)) is an exact result for the pressure
wave field in the frequency domain, the edge wave integral must in general be
evaluated numerically, as shown in the last sub-section. However, an exact analyt-
ical evaluation of the pressure is possible for the impulse response of a circular
transducer in the time-domain. To see this, multiply the Rayleigh-Sommerfeld
equation, Eq. (8.9), by expðiωtÞ=2π and sum over all frequencies. This leads to
an expression that can be formally inverted into the time domain as:
ð ð
þ1
iωρ v0 ðωÞexp½ikðr ctÞ
pðx; tÞ ¼ 2
dr dω
4π r
81 S 9 ð8:66Þ
< ð
ρ d v0 ðt r=cÞ =
¼ dS
2π dt: r ;
S
for a general input velocity vz ¼ v0 ðtÞ where v0(t) is the function whose Fourier
transform is v0(ω). For an impulsive input velocity vz ¼ v0 δðtÞ, then we would have
8 9
ð
ρv0 d < δðt r=cÞ =
pðx; tÞ ¼ dS : ð8:67Þ
2π dt : r ;
S
ρ0
a
ð
δðt r=cÞ
I¼ dS; ð8:68Þ
r
S
which, using the properties of the delta function and dS ¼ r dr dϕ, can be rewritten
formally as
8 9
ð >
rmax Φðr Þ
< ð >
=
I¼c dϕ δðr ctÞdr: ð8:69Þ
>
: >
;
r min 0
Performing the ϕ integration and using the sampling properties of the delta func-
tion, we find
where Φ(ct) can be interpreted as the angular extent of the “wave front” r ¼ ct on
the active face of the transducer. The unit step functions are present to guarantee
that the contribution to the pressure is zero outside the smallest and largest possible
radius values rmin and rmax, respectively, on the transducer active face. The explicit
determination of the terms, rmin, rmax, Φ(r), in Eq. (8.70) is most conveniently done
by considering separately the following three cases:
Case (1): Point x in the main beam and 0 < ρ < ða ρ0 Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In this case we would have r min ¼ z, r max ¼ ða ρ0 Þ2 þ z2 , and Φðr Þ ¼ 2π
(Fig. 8.23) so that I becomes
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
I ¼ 2πcH ðct zÞH ða ρ0 Þ þ z ct : 2 ð8:71Þ
Physically, this corresponds to the response for times after the direct plane wave has
arrived at x but before the earliest edge wave arrival.
Case (2): Point x in the main beam and a ρ0 < ρ < a þ ρ0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Now, we have r min ¼ ða ρo Þ2 þ z2 , r max ¼ ða þ ρ0 Þ2 þ z2 and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Φðr Þ ¼ Φ ρ2 þ z2 as shown in Fig. 8.24a so that
246 8 Ultrasonic Transducer Radiation
a ρ
Φ/2 π – Φ/2
ψ
O ρ0 y0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I ¼ cΦðctÞH ct ða ρ0 Þ2 þ z2 H ða þ ρ0 Þ2 þ z2 ct : ð8:72Þ
and
so that
Placing this result into Eq. (8.72) we then obtain the edge wave contribution to the
pressure in the main beam.
8.1 Planar Piston Transducer in a Fluid 247
b
a ρ
ψ Φ/2
O y0
ρ0
Case (3): Point x outside the main beam and ðρ0 aÞ < ρ < ðρ0 þ aÞ
In this case, where there is only an edge wave contribution since we are outside the
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
main beam, we have r min ¼ ða ρ0 Þ2 þ z2 , r max ¼ ða þ ρ0 Þ2 þ z2 again and Φ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðr Þ ¼ Φ ρ2 þ z2 as shown in Fig. 8.25a so that, as in case (2), we have
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
I ¼ cΦðctÞH ct ða ρ0 Þ þ z H 2 ða þ ρ0 Þ þ z ct : 2 ð8:75Þ
and, again,
so that Φ(ct) in this case is in fact identical to the expression found for case (2),
Eq. (8.74).
Collecting these results for the integral I, we have, finally, the pressure wave
field due to an impulsive input velocity on a circular transducer in the form
ρcv0 d
pðx; tÞ ¼ fΦðctÞH ðct r min ÞH ðr max ctÞg; ð8:76Þ
2π dt
where all the terms are known explicitly and summarized in Table 8.1.
As shown in the previous sections, the pressure wave field of a planar piston
transducer can be expressed as a superposition of sources (spherical waves) over
248 8 Ultrasonic Transducer Radiation
x
y0 Cε
z
for x inside the main beam, on the main beam edge (assuming the edge is smooth),
and outside the main beam, respectively, so that we can rewrite Eq. (8.63) as
ð
ρcv0 ðωÞ expðikr e Þ ðn er Þ et ds
pðx; ωÞ ¼ h i; ð8:78Þ
2π re 1 ðer nÞ2
Cr
8.1 Planar Piston Transducer in a Fluid 249
where
8 9
ð <ð ð ð=
¼ lim þ þ ; ð8:79Þ
Cε !0 : ;
Cr C Γ1 þΓ2 Cε
and the integrals on Γ1 and Γ2 can be omitted since they cancel as shown in
Fig. 8.21. We will refer to Eq. (8.78) as an expression of the pressure solely in
terms of boundary diffraction waves. Young, in 1802, first expressed the qualitative
idea that the diffraction of light through an aperture could be expressed in terms of a
direct wave obtainable from geometrical optics, and a wave originating at the
boundary of the aperture. Subsequently, Maggi (1888), and Rubinowicz (1917)
obtained explicit mathematical expressions verifying Young’s ideas [2]. Much
later, Miyamoto and Wolf (1962) [3, 4], and Marchand and Wolf (1962) [5]
developed a general theory of boundary diffraction waves, including its application
to the Rayleigh-Sommerfeld equation, as shown above. The value of such boundary
diffraction wave expressions is that, as mentioned previously, they show explicitly
the physics of the beam generated by the transducer while reducing the evaluation
of the general pressure wave field to single integrals only.
It is difficult to directly extend either the case of the Rayleigh-Sommerfeld
theory (Eq. (8.8)) or the boundary diffraction wave theory (Eq. (8.78)), to predict
the wave field generated by a transducer on the other side of a planar interface.
However, recall from Chap. 4 that the spherical waves present in the Rayleigh-
Sommerfeld equation can be represented in terms of an angular spectrum of plane
waves (Eq. (4.53)) as
ð þ1
þ1 ð
expðikr Þ i exp½ip ðx yÞdpx dpy
¼ ; ð8:80Þ
r 2π pz
1 1
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ jx yj ¼ ðx x0 Þ2 þ ðy y0 Þ2 þ ðz z0 Þ2 ð8:81Þ
p ¼ p x e x þ p y e y þ pz e z
0
and, for z z > 0
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< i p2 þ p2 k2 if p2 þ p2 > k2
x y x y
pz ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð8:82Þ
: k2 p2 p2 if k2 > p2 þ p2
x y x y
0
with pz ! pz if z z < 0. Placing Eq. (8.80), which is called Weyl’s integral, into
the Rayleigh-Sommerfeld integral, then gives
250 8 Ultrasonic Transducer Radiation
8 9
ð < þ1
ð þ1
ð
ωρv0 ðωÞ exp½i p ðx yÞdpx dpy=
pðx; ωÞ ¼ dSðyÞ: ð8:83Þ
4π 2 : pz ;
S 1 1
(where ∇y indicates that the gradient operator operates on the coordinates of point
y here). If we let n be the unit vector normal to the transducer surface S, place this
into Eq. (8.83) where S ¼ Sr is taken to be inside the extended contour Cr, and apply
Stokes’ theorem in the form
ð ð
n ∇y f dSðyÞ ¼ f et dsðyÞ ð8:85Þ
Sr Cr
we find
8 9
ð < þ1 ð
ð þ1
iρcv0 n ep et exp½ip ðx yÞdpx dpy=
pðx; ωÞ ¼ h 2 i dsðyÞ; ð8:86Þ
4π 2 : p 1 ne ;
Cr 1 1 z p
ð
ð þ1
þ1
expðikr e Þ ðn er Þ et i n ep et exp½ip ðx yÞdpx dpy
h i¼ h 2 i ;
re 1 ðn er Þ2 2π p 1 n ep
1 1 z
ð8:87Þ
A planar transducer produces a broad beam that is particularly useful when trying to
detect flaws. Once a flaw is found, however, it is often desirable to have the beam
focused into a small region near the flaw. Such focusing can be accomplished by
using a curved transducer crystal or by placing an acoustic lens in front of a planar
crystal to produce a non-planar wave front. In this section we will consider models
of the wave field of a spherically focused transducer radiating into a fluid.
where Sf is now the spherical surface of Fig. 8.27. Recent comparisons of more
exact numerical models [7] have shown that the O’Neil theory is indeed a good
252 8 Ultrasonic Transducer Radiation
v0
O
approximation under all but the most extreme conditions. However, there are
models of spherically focused probes other than the O’Neil model available
[8, 9]. One of those other models is based directly on the Rayleigh-Sommerfeld
formulation where the integration is on a planar surface S, but where the normal
velocity on the transducer face is taken to be that due to a spherically converging
wave. Thus, since vz ¼ ð1=iωρÞ∂p=∂z we can write the Rayleigh-Sommerfeld
integral in the form
ð
1 ∂pinc expðikr Þ
pðx; ωÞ ¼ dS; ð8:89Þ
2π ∂z r
S
where pinc is the pressure for a spherically converging wave, with origin (focus)
at O, over the planar aperture (Fig. 8.28). If we let pinc be given by the spherical
wave (Fig. 8.29)
p0 ðωÞR0 h
0
i
pinc ¼ 0 exp ik R0 r ; ð8:90Þ
r
planar aperture
ð
p0 ðωÞR0 z1 expðikR0 Þ 1 ∂ expðikr 1 Þ expðikr Þ
pðx; ωÞ ¼ dS: ð8:92Þ
2π ρ ∂ρ r1 r
S
where we have defined the pressure amplitude in terms of a velocity v0(ω) via
p0 ðωÞ ¼ ρ1 cv0 ðωÞ. Note that here we have used the symbol ρ1 to represent the
density of the fluid so as to distinguish it from the radius, ρ. In the following
sections, we will primarily use O’Neil’s model (Eq. (8.88)) to evaluate the beam of
ultrasound radiated from a spherically focused transducer.
254 8 Ultrasonic Transducer Radiation
R0 – q 0R 0
Although other models (Eq. (8.92) or Eq. (8.94)) are based on a different set of
basic assumptions, in most cases there is little difference in the radiated fields that
they predict (see the comparisons of Stamnes [8], for example). Thus, the choice of
the model in many cases is more a matter of convenience than of accuracy.
First, we will consider the O’Neil model and the case where point x is on the
axis of a spherically focused transducer of radius, a, and focal length
R0 (Fig. 8.30). The distance, h, from the back to the front of the spherical surface
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
is given by h ¼ R0 R20 a2 and we let q0 ¼ 1 z=R0 be the normalized
distance from the geometric focus. On a spherical surface of radius
R0 (Fig. 8.31a) we can write the area element dS ¼ R20 sin α dα dϕ, which can be
used for the geometry shown in Fig. 8.31b where the angle ϕ (not shown) is an
angle in the plane perpendicular to the z-axis. From the law of cosines for triangle
ABC in Fig. 8.31b we have R21 ¼ 2R20 2R20 cos α so that differentiating both sides
of this expression we have R1 dR1 ¼ R20 sin α dα. Placing this result into the
expression for dS, we see dS ¼ R1 dR1 dϕ. But, also from triangle ABC we have,
from the law of cosines, R1 ¼ 2R0 cos θ, and from triangle ABD
dS ¼ r dr dϕ=q0 ; ð8:95Þ
which is the generalization of the same result found previously for the planar
transducer. In fact, as R0 ! 1, q0 ! 1 and we simply recover our previous result.
Placing Eq. (8.95) into the O’Neil expression, Eq. (8.88), then we find
8.2 Spherically Focused Piston Transducer in a Fluid 255
b z
2
dS = R 0 sin αdαdφ
α
C R0
y
φ
x
2ðπ ð
re
iωρv0 ðωÞ
pðz; ωÞ ¼ expðikr Þdr dϕ; ð8:96Þ
2πq0
0 z
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where r e ¼ ðz hÞ2 þ a2 is the distance from x to the transducer edge.
Performing the integrations then gives finally the on-axis pressure explicitly as
ρcv0 ðωÞ
pðz; ωÞ ¼ ½expðikzÞ expðikr e Þ: ð8:97Þ
q0
The structure of Eq. (8.97) is very similar to that for the planar probe. Letting
v0 ðωÞ ¼ v0 , a constant, and inverting Eq. (8.97) into the time domain we find
ρcv0
pðz; tÞ ¼ ½δðt z=cÞ δðt r e =cÞ; ð8:98Þ
q0
which shows that the on-axis response consists of a direct spherical wave (with
focus at z ¼ R0 ) and an edge wave term.
Although the q0 term in Eqs. (8.97) and (8.98) goes to zero when z ¼ R0 , these
expressions are well-behaved near the geometric focus. To see this, consider
Eq. (8.97) which we rewrite as
256 8 Ultrasonic Transducer Radiation
ρcv0 ðωÞexpðikzÞ
pðz; ωÞ ¼ f1 exp½ikðr e zÞg: ð8:99Þ
q0
ρcv0 ðωÞexpðikR0 Þ
pðR0 ; ωÞ ¼ lim f1 ½1 þ ikq0 h þ g
q0 !0 q0 ð8:102Þ
¼ ρcv0 ðωÞðikhÞexpðikR0 Þ:
Note that the maximum pressure for the focused probe does not occur at the
geometric focus except at infinitely high frequencies. However, Eq. (8.102)
shows that the effect of tight focusing and high frequencies (kh large) increases
the amplitude at the geometric focus from that of the planar probe, which recall had
a maximum pressure amplitude of 2ρcv0(ω) (Fig. 8.7). Thus, defining the amplifi-
cation factor, Af, as the ratio of the pressure amplitude at the geometric focus of the
spherically focused probe to the maximum pressure in the planar probe, we find
Af ¼ kh=2: ð8:103Þ
As an example, for a 0.50 in. diameter, 3.0 in. focal length transducer radiating into
water at 10 MHz, we obtain Af ¼ 5:63 approximately.
8.2 Spherically Focused Piston Transducer in a Fluid 257
Because the structure of Eq. (8.99) for the on-axis pressure is very similar in form to
that of the planar probe, we expect that the on-axis behavior of peaks and nulls
might also be similar. To some extent that is true, although, as we will see, the effect
of focusing produces some distinct differences as well. As in the planar case, we can
rewrite Eq. (8.99) in the form
which shows that on-axis pressure nulls occur in the response when
so that
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k ðz hÞ2 þ a2 z =2 ¼ nπ n ¼ 1, 2, . . . : ð8:106Þ
Equation (8.106) looks very similar to the planar case, except now we see that both
positive and negative multiples of π are retained on the right side of Eq. (8.106).
Following the same procedures as in the planar case, we can solve for the location,
zn, of the on-axis nulls which in this case gives
a 2 þ h 2 n2 λ 2
zn ¼ n ¼ 1, 2, . . . : ð8:107Þ
2h 2nλ
In the planar probe case h ¼ 0 and Eq. (8.107) reduces to that found for the planar
case if the positive sign is taken in the denominator of Eq. (8.107) (the negative sign
must be rejected for planar probes since it would lead to negative zn values). For the
focused probe, the minus sign is acceptable as long as it does not lead to negative
values of zn. This places the restriction that h
nλ, or
n Int½h=λ; ð8:108Þ
where Int[h/λ] is the largest positive integer h=λ. To give a physical meaning to
the choice of the sign in Eq. (8.107), note that for most cases a2 þ h2
n2 λ2 so
that
a2 þ h2 h
zn ffi ¼ R0 ; ð8:109Þ
2h 2nλ h nλ
258 8 Ultrasonic Transducer Radiation
where we have used the relationship a2 þ h2 ¼ 2R0 h. From Eq. (8.109) it follows
that for the plus sign zn < R0 , while for the negative sign zn > R0 , so that the choice
of sign simply corresponds to whether the null exists on one side of the geometric
focus or the other. However, the nulls beyond the geometric focus ðzn > R0 Þ must
satisfy the rather restrictive inequality of Eq. (8.108) if they are to exist at all.
To find the location of the on-axis pressure maxima, we again follow the planar case
approach and consider the square of the magnitude of the on-axis pressure, which in
the focused case is given by
2ρcv0 ðωÞ sin ðkðr e zÞ=2Þ 2
jpðz; ωÞj2 ¼ : ð8:110Þ
q0
Setting the derivative of Eq. (8.110) equal to zero, and dividing by the factor
4ðρcv0 Þ2 sin ½kðr e zÞ=2, which is non-zero since we are not at a null, we find,
after some algebra
Figure 8.32 shows the behavior of the magnitude of the normalized on-axis pressure
versus the non-dimensional distance, z/R0, for a 1/2 in. diameter, 10 MHz, 3 in.
focal length transducer radiating into water. As can be seen in Fig. 8.32, nulls exist
on both sides of the geometric focus and the maximum pressure occurs at a location
(the “true” focus) which is closer to the transducer than the geometric focus. As the
frequency of the transducer increases, this true focus moves closer to the geometric
focus, and the amplification due to focusing increases, as expected from
Eq. (8.103). These results were generated with the MATLAB® function onaxis_foc,
whose calling sequence is
>> p ¼ onaxis_foc(z,f,a,R,c);
8.2 Spherically Focused Piston Transducer in a Fluid 259
which returns the normalized pressure, p/ρcv0, at locations z (in mm) on the axis of
a spherically focused transducer of radius, a, (in mm) and focal length, R, (in mm)
radiating into a fluid with wave speed, c, (in m/s). The function uses the O-Neil
model. If R ¼ inf the function returns the on-axis pressure of a planar transducer of
radius a.
Since most ultrasonic transducers used in NDE applications are not very tightly
focused (h a) and since we are primarily interested in the fields in the neighbor-
hood of the focus, the paraxial approximation (z
a) can be assumed in many
applications. We have exactly
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
re z ¼z2 2zh þ h2 þ a2 z
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ z2 2hðz R0 Þ z
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ z2 þ 2hq0 R0 z;
ρcv0 ðωÞ
pðz; ωÞ ¼ expðikzÞ 1 exp ika2 q0 =2z : ð8:113Þ
q0
From Eq. (8.113) it follows that the on-axis diffraction correction, for the spheri-
cally focused transducer, C1, is given by
1
C1 ðz; ωÞ ¼ 1 exp ika2 q0 =2z ; ð8:114Þ
q0
It is interesting to compare these results of O’Neil’s model with the planar aperture
model (Eq. (8.92)) presented earlier, since exact on-axis results can be obtained
from that model as well. Writing the planar area element in Eq. (8.92) as
dS ¼ 2πρ dρ, that equation becomes
ða
∂ expðikr 1 Þ expðikr Þ
pðz; ωÞ ¼ p0 ðωÞR0 z1 expðikR0 Þ dρ: ð8:115Þ
∂ρ r1 r
0
However,
∂ expðikr 1 Þ expðikr Þ ∂ exp½ikðr r 1 Þ
¼ ; ð8:116Þ
∂ρ r1 r ∂ρ r 1 ðr r 1 Þ
which can be verified by direct differentiation, so that the integral can be done
exactly as ρ varies from 0 to a, r1 varies from z1 to R0 and r varies from z to re
(Fig. 8.29)) to yield
exp½ikðr e R0 Þ exp½ikðz R0 Þ
pðz; ωÞ ¼ p0 ðωÞR0 z1 expðikR0 Þ
R 0 ð r e R0 Þ z 1 ð z R0 Þ
ð8:117Þ
p z1 q0 expðikr e Þ
¼ 0 expðikzÞ :
q0 ð R0 r e Þ
Comparing this result with that of the O’Neil model (Eq. (8.97)), we see that both
“exact” results are different. However, in the paraxial approximation
z1 q0 =ðR0 r e Þ ! 1 and r e ! z þ a2 q0 =2z again so if we let p0 ðωÞ ¼ ρcv0 ðωÞ,
where now ρ is the density, both models predict exactly the same diffraction
correction.
8.2 Spherically Focused Piston Transducer in a Fluid 261
D
focus x
Sf
When point x is not on the axis of the focused transducer then an exact evaluation of
the pressure is still possible for the O’Neil model in terms of a decomposition into
direct and edge wave terms, as in the planar transducer case. In the focused case, we
consider the geometry formed by dropping a line from x through the geometric
focus, to a point y0 on the spherical surface as shown in Fig. 8.33. This figure is
identical to that of the on-axis case (Fig. 8.31b) considered previously if we let
z ! D. Thus, following the same arguments which led to Eq. (8.95), we find
r dr dβ
dS ¼ R1 dR1 dβ ¼ ; ð8:118Þ
q0
As in the planar case, the explicit evaluation of Eq. (8.119) is most conveniently
done by considering separately the following three cases:
Case 1: Point x inside the main beam
For the focused transducer, the “main beam” consists of the region inside the cone
formed by extending the normals to the spherical surface Sf into the surrounding
medium (Fig. 8.34). In this case point y0 lies on Sf and we have
ð ðre
2π
iωρv0 ðωÞ
pðx; ωÞ ¼ expðikr Þdr dβ ð8:120Þ
2πq0
0 D
C
x
focus
main beam region
C re–
x
re+
2 2ðπ
3
ρcv0 ðωÞ 4 1
pðx; ωÞ ¼ expðikDÞ expðikr e Þdβ5: ð8:121Þ
q0 2π
0
which, as in the planar case, can be put into the compact form
2 3
ð
ρcv0 ðωÞ 4
pðx; ωÞ ¼ expðikr e Þdβ5; ð8:123Þ
2πq0
C
Recall, in the on-axis case we found r 2 ¼ R21 q0 þ R20 ð1 q0 Þ2 . In this case, then we
2
have, for r ¼ re and R1 ¼ R1e (i.e. on the transducer edge), r 2e ¼ R1e q0 þ D2 . Thus,
in the paraxial approximation, we have
264 8 Ultrasonic Transducer Radiation
planar D
x
aperture
" 2 #
R1e q0
re ffi D 1 þ ð8:127Þ
2D2
Equation (8.128) is the direct generalization of the planar probe result, (Eq. (8.54))
to the focused case. This expression can be made substantially easier to evaluate by
noting that within the paraxial approximation we expect R1e ffi ρe and dβ ffi dϕ,
where ρe is the distance to the edge as shown in Fig. 8.37 and ϕ is measured in the
plane perpendicular to the transducer axis so that we find
8 9
ð
ρcv0 ðωÞexpðikDÞ < 1 2 =
pðx; ωÞ ¼ Θ exp ikρe q0 =2D dϕ ð8:129Þ
q0 : 2π ;
C
For the planar probe case, the diffraction correction term analogous to Eq. (8.130)
could be applied essentially anywhere in the wave field of the transducer (except
close to the transducer where the paraxial approximation would fail). However,
there is another implicit restriction inherent in Eq. (8.130). For approximations such
as Eq. (8.127) to be valid, the line extending from point x to the transducer surface
8.2 Spherically Focused Piston Transducer in a Fluid 265
must intersect that surface either within the main beam region or at least lie very
close to it. Thus, Eq. (8.127) cannot be used, for example, when x is in the plane of
the geometric focus ðz ¼ R0 Þ. In such cases, to obtain the pressure wave field values
one would instead have to use the exact expression, Eq. (8.126), or use the explicit
expression that is derived in Sect. 8.2.3.4. This same type of restriction on the
region of validity of diffraction correction terms for focused transducers will also be
present in later sections for more general focused probe problems. In those prob-
lems one does not typically have more exact solutions to fall back on as for this
single medium problem. However, the multi-Gaussian beam models discussed later
in this Chapter provide diffraction correction terms for focused circular transducers,
even in complex geometries (such as radiation through curved interfaces) that are
well-behaved, so that a multi-Gaussian beam model is the most effective choice to
calculate diffraction corrections.
As in the planar transducer case, it is possible to put the general off-axis expression,
Eq. (8.126), in a coordinate-invariant form. To see this, consider the geometry of
Fig. 8.38 where ea and er are unit vectors pointing from x to y0 and x to y,
respectively. As point y moves along the edge C through the vector displacement
ds ¼ et ds, in the plane perpendicular to the line from x to y0, we have on Γ
(Fig. 8.39)
where Γ is the image of C in this plane and ds⊥ is the component of ds in this
plane. In this plane, we can use the result from the geometry of the planar case,
Eq. (8.61), as
1 ðer ea Þ ds⊥
dβ ¼ h i: ð8:132Þ
r e 1 ðer e a Þ2
since er ea is perpendicular to ea. Placing Eq. (8.133) into Eq. (8.126) then we
have
8 9
ð
ρcv0 ðωÞ < 1 expðikr e Þ ðer ea Þ ds =
pðx; ωÞ ¼ Θ expðikzÞ h i : ð8:134Þ
q0 : 2π re 1 ðe e Þ2 ; r a
C
One of the nice applications of the coordinate invariant form of Eq. (8.134) is to use
it to express the integral in Eq. (8.134) in terms of coordinates centered in a plane
perpendicular to the central transducer axis (Fig. 8.40), as was done for the planar
case. Letting ds ¼ a sin ϕ dϕ ex þ a cos ϕ dϕ ey and
(
y ¼ jyj D > R0
x ¼ ðz hÞez þ y ey
y ¼ jyj D < R0 ð8:135Þ
y ¼ a cos ϕex þ a sin ϕ ey
h
z
8.2 Spherically Focused Piston Transducer in a Fluid 267
and performing all the vector operations in Eq. (8.134), after considerable algebra
we obtain
(
ρcv0 ðωÞ
pðy; z; ωÞ ¼ Θ exp½ikðR0 þ d Þ
ðd=R0 Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 9
a y2 þ ðz R0 Þ2 ð expðikre Þ½jyj sin ϕðh R0 Þ ajz R0 jdϕ =
2π
þ ;
2π ½jyj sin ϕðh R0 Þ ajz R0 j2 þ y2 R20 cos 2 ϕ;
0
ð8:136Þ
where
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
< þ ðz R0 Þ2 þ y2 D > R0
d¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
: ðz R Þ2 þ y 2 D < R
0 0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:137Þ
r e ¼ a2 þ y2 2 ay sin ϕ þ ðz hÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h ¼ R0 R20 a2
d=R0 ! 1
R0 þ d ! z ð8:138Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r e ! a2 þ y2 2ay sin ϕ þ z2
ð
2π
a expðikr e Þ½a þ jyj sin ϕdϕ
I¼ : ð8:139Þ
2π a2 þ y2 þ 2ajyj sin ϕ
0
However, the angle ϕ here is not the same as that in the planar case which is measured
0
from the radius in the plane through y0 (Fig. 8.41). Thus if we let ϕ ¼ ϕ þ 3π=2 we
0
have sin ϕ ¼ cos ϕ so that finally, Eq. (8.136) reduces to
8 9
ð 0
< 1
2π
expðikre Þa a jyj cos ϕ dϕ =
0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0
where r e ¼ a2 þ y2 2ajyj cos ϕ þ z2 . Note that the change of limits on ϕ0 from
ð3π=2, π=2Þ to (0, 2π) is permissible since
268 8 Ultrasonic Transducer Radiation
φ x
R0
y
z
φ′ focus
ð
2πþα
0 2ðπ
0
0 0
f cos ϕ dϕ ¼ f cos ϕ dϕ
α 0
for any function f and angle α. Equation (8.140) is identical with our previous planar
0
transducer result (Eq. (8.65)) if we simply let ϕ ¼ ψ, jyj ¼ ρ0 .
An important result that follows directly from Eq. (8.136) is an expression for the
off-axis pressure on a plane passing through the geometric focus and perpendicular
to the transducer central axis. In this case Θ ¼ 0 except at the single point on the
axis and z ¼ R0 so that Eq. (8.136) reduces to
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
ð
2π
sin ϕ R20 a2 exp ik R20 þ y2 2ay sin ϕ dϕ
ρcv0 ðωÞaR0
pðR0 ; y; ωÞ ¼ :
2πy R20 a2 sin 2 ϕ
0
ð8:141Þ
2ðπ
ρcv0 ðωÞaexp ikR0
pðR0 ; y; ωÞ ffi sin ϕ exp ikay sin ϕ=R0 dϕ ð8:142Þ
2πy
0
focus
where θ is shown in Fig. 8.42. It is interesting to note that Eq. (8.143) is identical to
the result we found previously for the far field diffraction pattern of a planar piston
transducer. This expression can also be used to estimate the focal spot size of our
spherically focused transducer. If we let Wf be twice the distance from the on-axis
maximum to an off-axis node (Fig. 8.43), we have approximately (with R0 ffi R0 )
R0
W f ¼ 7:66 : ð8:144aÞ
ka
However, in practice other criteria (such as 6 dB drop off points) are often used
instead to define a focal point size. For the 6 dB beam width, W f6dB , we use the
distance where the amplitude has dropped to one half of its maximum value, which
occurs when kay=R0 ¼ 2:2244 so that approximately
R0
W f6db ¼ 4:45 : ð8:144bÞ
ka
To obtain focusing action at the location of the geometric focus we certainly must
have the beam width to be less than the beam width of an equivalent planar
transducer, which is approximately equal to the diameter of the transducer. Thus,
if we require, for example, W f6dB < 2a this implies
270 8 Ultrasonic Transducer Radiation
a2
R0 < 2:83 :
λ
Since the effective far field occurs when z > πa2 =λ, where DR ¼ πa2 =λ is called the
Rayleigh distance, we see that focusing is only possible if the geometrical focal
length is in the near field of the equivalent planar transducer.
Neither the O’Neil model or the plane aperture model contains an acoustic lens as
found in the actual construction of many focused probes. Instead, these models
produce focusing by either placing the sources on a curved surface (O’Neil) or by
taking the sources to be due to an incident curved wave front (planar aperture
model). Here, we will present a simple model that does include a lens directly and
compare the results of this model with our previous results.
Specifically, we will consider a spherical lens in front of a circular planar
aperture S of radius a (Fig. 8.44), where the lens is treated as an equivalent fluid
medium of wave speed c2 and density ρ2 and the wave speed and density of the
actual fluid are taken as c1 and ρ1, respectively. Following Schlengermann [9] we
will evaluate the pressure at a point x on the axis of the transducer by using the
Rayleigh-Sommerfeld formulation:
ð
iωρ1 v0 ðωÞ expðik1 r Þ
pðx; ωÞ ¼ dS ð8:145Þ
2π r
S
and take the lens into account by writing the radius r as the distance
8.2 Spherically Focused Piston Transducer in a Fluid 271
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ ðz dÞ2 þ ρ2 þ nd; ð8:146Þ
where n ¼ c1 =c2 accounts for the different wave speed in the lens and the total
distance is approximately the path followed by a ray from the planar aperture S,
through the lens, to a point x on the axis (Fig. 8.44). In this case the distance d is
given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ρ2
d ¼ R0 R20 ρ2 ffi ; ð8:147Þ
2R0
where R0 is the focal length of the lens. From Eq. (8.146) and (8.147) it follows that
for z
d
h i qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r2 ¼ ðz dÞ2 þ ρ2 þ 2nd ð z d Þ 2 þ ρ 2 þ n2 d 2
ð8:148Þ
ffi ½z2 þ ρ2 ð1 z=R0 Þ þ ðnρ2 =R0 Þ½z
where
0 0
q0 ¼ 1 z=R0
0 ð8:150Þ
R0 ¼ R0 =ð1 nÞ:
Thus, since the area dS ¼ 2πρ dρ in Eq. (8.145), the on-axis pressure can be
evaluated explicitly as
ða
r¼r
iωρ1 v0 ðωÞ
pðx; ωÞ ¼ 0 expðik1 r Þ dr,
q0
r¼z ð8:151Þ
ρ c1 v0 ðω Þ
¼ 1 0 fexpðikzÞ expðikr a Þg
q0
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ra ¼ ðz hÞ2 þ a2 þ nh
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:152Þ
h ¼ R0 R20 a2 :
272 8 Ultrasonic Transducer Radiation
Equation (8.151) is in the same form as the on-axis pressure expression obtained
from the O’Neil model. In fact, if we let c2 ! 1 so that the sources are instantly
transmitted from the plane aperture to the spherical surface of the lens, the on-axis
pressure results of the two models are identical. Even for a finite value of c2, in the
0
paraxial approximation we have ra ffi z þ a2 q0 =2z so that
ρ1 c1 v0 ðωÞexpðikzÞ h
0
i
pðx; ωÞ ¼ 0 1 exp ik1 a2 q0 =2z ; ð8:153Þ
q0
which is identical to the O’Neil model (Eq. (8.113)) if we simply make the
0 0
replacement R0 ! R0 . Since an “effective” value of R0 (or R0 ) is obtained exper-
imentally anyway (see Chap. 14), this difference in the models is likely not
significant, and we expect that all the models considered here predict essentially
the same behavior (at least in the paraxial approximation).
The previous sections of this chapter have given explicit models for the beam of
sound generated by a piezoelectric transducer radiating into a fluid. In NDE
immersion testing, this beam of sound must cross a fluid-solid boundary before it
strikes a scatterer so that it is necessary to include in our models the influence of the
interface on the sound beam.
Consider first the case shown in Fig. 8.45, where a planar piston probe in a fluid
medium (medium one) is oriented normal to the planar interface with a second
medium, which in this case will also be modeled as a fluid. From the Rayleigh-
Sommerfeld equation and the representation of a spherical wave in terms of an
angular spectrum of plane waves, the incident pressure in medium one generated by
the transducer (before interaction with the interface) is given by (see Eq. (8.83)):
8 9
ð þ1 ð þ1ð
ωρ1 v0 ðωÞ < exp½ip1 ðx yÞ =
pðx; ωÞ ¼ dp dp dSðyÞ; ð8:154Þ
4π 2 : p1z x y
;
S 1 1
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< k 2 p2 p2 k21 > p2x þ p2y
1 x y
p1z ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð8:155Þ
: i p 2 þ p2 k 2 p 2 þ p2 > k 2
x y 1 x y 1
For an incident plane wave exp½ip1 ðx yÞ, the plane wave transmitted through
the interface
into the second medium is (see Eq. (6.100))
T p ep1 n exp½ip2 ðx yÞ þ p1z D p2z D, where
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< k22 p2x p2y k22 > p2x þ p2y
p2z ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:156Þ
: i p 2 þ p2 k 2 p 2 þ p2 > k 2
x y 2 x y 2
ð8:157Þ
which is identical to the phase term considered previously in Chap. 6 (Eq. (6.104)).
If we evaluate the angular spectrum of plane waves integral in Eq. (8.157) by the
method of stationary phase then, as shown previously (Sect. 6.2.8), at the stationary
phase point we have
ϕ ¼ k1 D1 þ k2 D2
p1z ¼ k1 z1 =D1 ð8:159Þ
p2z ¼ k2 z2 =D2 ;
where z1, z2, D1, D2 are shown in Fig. (8.46). The first order partial derivatives of the
phase term ϕ are given by (see Eq. (6.106)):
274 8 Ultrasonic Transducer Radiation
z1 z2
0
∂ϕ=∂px ¼ x x px z1 =p1z px z2 =p2z
0 ð8:160Þ
∂ϕ=∂py ¼ y y py z1 =p1z py z2 =p2z
2
2 2 2
so that the Hessian H ¼ ∂ ϕ=∂p2x ∂ ϕ=∂p2y ∂ ϕ=∂px ∂py is
!
z1 z2 p2x z1 p2x z2 z1 z2 p2y z1 p2y z2
H¼ þ þ 3 þ 3 þ þ 3 þ 3
p1z p2z p1z p2z p1z p2z p1z p2z
2 ð8:162Þ
px py z1 px py z2
þ 3 :
p31z p2z
where
z1 z2 z1 z2
A¼ þ , B¼ 3
þ 3 ; ð8:164Þ
p1z p2z p1z p2z
z1 k21 z2 k22
AþB p2x þ p2y ¼ þ 3 ð8:166Þ
p31z p2z
z1 D1 z2 D2
¼ , ¼
p1z k1 p2z k2 ð8:168Þ
p21z ¼ k21 cos 2 θ1 , p22z ¼ k22 cos 2 θ2
so that, finally
D1 D2 D1 D2
H¼ þ þ
k1 k2 k1 cos 2 θ1 k
2 cos θ 2
2
ð8:169Þ
1 c2 c2 cos 2 θ1
¼ 2 D 1 þ D 2 D 1 þ D 2 :
k1 cos 2 θ1 c1 c1 cos 2 θ2
2 2
From Eqs. (8.161) and (8.169) it follows that ∂ ϕ=∂p2x < 0, ∂ ϕ=∂p2y < 0, and
H > 0 so that when applying the method of stationary phase to Eq. (8.57) (see
Eqs. (E.21) and (E.22)) the pressure in medium two reduces to:
ð
iωρ1 v0 ðωÞ T p exp½iðk1 D1 þ k2 D2 Þ
pðx; ωÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dS;
2π ðD1 þ c2 D2 =c1 Þ ðD1 þ c2 cos 2 θ1 D2 =c1 cos 2 θ2 Þ
S
ð8:170Þ
From Eq. (8.170) we can obtain a convenient expression for the diffraction correc-
tion of the transmitted beam in the paraxial approximation by following steps
similar to those for the single medium case considered previously. As in that
previous case we drop a line parallel to the z-axis from a general off-axis point
x to a point y0 in the plane of the transducer and set up a set of cylindrical
coordinates (ρ, ϕ) in the transducer plane about y0 (Fig. 8.46). In the paraxial
approximation we have
ε21 ε22
D 1 ffi z1 þ , D 2 ffi z2 þ ð8:171Þ
2z1 2z2
and
ε 1 ffi z 1 θ 1 , ε2 ffi z 2 θ 2
θ1 θ2 ð8:172Þ
ffi ðSnell’s LawÞ:
c1 c2
Keeping only the lowest order approximation for D1 and D2 in the amplitude part of
Eq. (8.170), but the first two orders in the phase ϕ ¼ k1 D1 þ k2 D2 , we find, using
Eq. (8.172)
k1 θ21 c2
ϕ ffi k 1 z1 þ k 2 z2 þ z1 þ z 2 : ð8:173Þ
2 c1
k 1 ρ2
ϕ ffi k 1 z1 þ k 2 z2 þ
c2
2z ð8:175Þ
z ¼ z1 þ z2
c1
2ðπ ð
ρe
iωρ1 v0 ðωÞT p ð0∘ Þexp½iðk1 z1 þ k2 z2 Þ ik1 ρ2
pðx; ωÞ ¼ exp ρ dρ dϕ;
2πz 2z
0 0
ð8:176Þ
8.3 Beam Propagation Through A Planar Interface: Planar Probe 277
where Tp(0 ) is now the transmission coefficient at normal incidence. The integral
in Eq. (8.176) is identical with that found in the single medium case if we simply
make the replacement z ! z. Entirely similar results then follow directly when x is
outside or on the edge of the main beam, so that we do not show all the details here
but merely quote the end result for the pressure at an arbitrary point x which, when
the integration on ρ is carried out, is
8 9
< ð =
1
pðx; ωÞ ¼ ρ1 c1 v0 ðωÞT p ð0 Þexp½iðk1 z1 þ k2 z2 Þ Θ exp ik1 ρ2e =2z dϕ
: 2π ;
C
ð8:177Þ
(see Eq. (8.54)). From Eq. (8.177) it follows that the diffraction correction, C1, is
given by
8 9
< ð =
1
C1 ðx; ωÞ ¼ Θ exp ik1 ρ2e =2z dϕ ; ð8:178Þ
: 2π ;
C
which is identical to the single medium case (Eq. (8.56)) if we again make the
replacement z ! z.
Although Eq. (8.170) is an explicit expression for the beam of sound transmitted
through the interface, because of the surface integration involved it is computa-
tionally rather expensive to evaluate. Using boundary diffraction wave theory,
however, it is possible to reduce this Rayleigh-Sommerfeld form to a combination
of direct-edge waves, as in Eq. (8.177), but without making the paraxial assump-
tion. To see this, we return to Eq. (8.154) and write it in the boundary diffraction
wave form (Eq. (8.86)) as
8 9
ð < þ1 ð
ð þ1
iρ c1 v0 ðωÞ n ep1 et exp½ip1 ðx yÞdpx dpy=
pðx; ωÞ ¼ 1 2 h 2 i dsðyÞ:
4π : p 1 ne ;
Cr 1 1 1z p1
ð8:179Þ
Transmitting the incident plane waves in Eq. (8.179) through the interface, as
before, we obtain the pressure in medium two in the form
278 8 Ultrasonic Transducer Radiation
8 þ1
ð < þ1
ð ð
iρ c1 v0 ðωÞ
pðx; ωÞ ¼ 1 2 T p n ep1 et
4π :
Cr 1 1 9 ð8:180Þ
exp½iðp2 ðx yÞ þ p1z D p2z DÞdpx dpy =
h 2 i dsðyÞ;
p1z 1 n ep1 ;
where T p ¼ T p ep1 n Evaluation of the angular spectrum of plane waves in
Eq. (8.180) by the method of stationary phase follows the same steps as described
previously. Since at the stationary phase point ep1 ¼ er1 where er1 is a unit vector
along the stationary phase ray path in medium one (see Fig. 8.46), we find
ð
ρ1 c1 v0 ðωÞ ðn er1 Þ et T p exp½iðk1 D1 þ k2 D2 Þ
pðx; ωÞ ¼ h ipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffids;
2π 1 ðn er1 Þ 2
D1 þ c2 D2 =c1 D1 þ c2 cos 2 θ1 D2 =c1 cos 2 θ2
Cr
ð8:181Þ
et
r
D1
ε θ1
θ2 D2 θ1
y0 x
Cε
8.3 Beam Propagation Through A Planar Interface: Planar Probe 279
r ¼ D1 þ ε2 = sin θ1
¼ D1 þ D2 sin θ2 = sin θ1 ð8:183Þ
¼ D1 þ c2 D2 =c1 :
where α ¼ ð2π, π, 0Þ depending on whether x is (inside, on the edge of, outside) the
main beam, respectively, so that finally
(
pðx; ωÞ ¼ ρ1 c1 v0 ðωÞ Θ T p ð0 Þexp½iðk1 z1 þ k2 z2 Þ
9
ð =
1 ðn er1 Þ et T p exp½iðk1 D1 þ k2 D2 Þ
h ipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ds ;
2π 1 ðn er1 Þ2
C
D1 þ c2 D2 =c1 D1 þ c2 cos 2 θ1 D2 =c1 cos 2 θ2 ;
ð8:185Þ
where Θ was defined in Eq. (8.55) and, as before, the integral is a principal value
integral when x is on the edge of the main beam. Comparing Eq. (8.185) with our
previous exact result (Eq. (8.63)) for the single medium case, we see that the present
case does reduce to the single medium results when the two fluid media are
the same.
When a planar piston transducer in an immersion testing setup radiates through the
interface with a solid, mode converted waves are present that are not accounted for
in the fluid-fluid interface model. However, many of the evaluation procedures are
identical to those already considered. For example, using the incident waves in the
fluid, given again by Eq. (8.154), and the plane wave transmission coefficients for a
fluid-solid interface, the displacement field in the solid is given by (see Eq. (6.126)):
280 8 Ultrasonic Transducer Radiation
X ωρ v0 ðωÞ
uðx; ωÞ ¼ 1
α¼P, S
4π 2 iωρ2 cα2
8 þ1 9
ð ð α;P α α
ð < þ1
α
T 12 d exp i p2 ðx yÞ þ p1z D p2z D dpx dpy =
dSðyÞ;
: p1z ;
S 1 1
ð8:186Þ
α:P
where cp1 is the wave speed of the fluid, T 12 ðα ¼ P, SÞ are the transmission
coefficients (based on stress/pressure ratios) for P-waves and S-waves,
cα2 , dα ðα ¼ P, SÞ are wave speeds and polarization vectors in the elastic solid for
P- and S-waves, and ρm ðm ¼ 1, 2Þ are the densities of the fluid and solid,
respectively. The wave number vectors p2α , ðα ¼ P, SÞ for P- and S-waves are
given by
p2α ¼ px ex þ py ey þ p2z α
ez
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
< k2α2 p2x p2y k2α2 > p2x þ p2y ð8:187Þ
α
p2z ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
>
: i p 2 þ p2 k 2 p2 þ p2 > k 2
x y α2 x y α2
where kα2 ðα ¼ P, SÞ is the wave number in the solid for P- and S-waves and the
coordinate system is the same one as defined for the fluid-fluid interface (Fig. 8.48).
Since the application of the method of stationary phase to Eq. (8.186) follows
exactly the same steps as illustrated previously, we present only the final result here
as
X ρ1 v0 ðωÞ
uðx; ωÞ ¼
ρ2 cα2
8, S
α¼P
9
<1 ð α;P α α α =
T 12 d exp i kp1 D1 þ kα2 D2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dSðyÞ;
:2π D1α þ cα2 D2α =cp1 D1α þ cα2 cos 2 θ1α D2α =cp1 cos 2 θ2α ;
S
ð8:188Þ
α;P α;P
where T 12 ¼ T 12 cos θ1α and Dmα ðm ¼ 1, 2Þ ðα ¼ P, SÞ are the distances
traveled through the interface along stationary phase paths that satisfy Snell’s law
for P- and S-waves, respectively, and θmα ððm ¼ 1, 2Þ, ðα ¼ P, SÞÞ are the associated
angles (see Fig. 8.48).
Applying the paraxial approximation to Eq. (8.188) also follows directly the steps
of the fluid-fluid case, so we find, formally (for point x in the main beam)
8.3 Beam Propagation Through A Planar Interface: Planar Probe 281
which shows that the diffraction correction (last term in the brackets in
Eq. (8.190a)) is in fact identical in form to that in the fluid-fluid interface case
(Eq. (8.178)). Note that the transmission coefficient in Eq. (8.190a) is, as mentioned
previously, based on a stress/pressure ratio, so if we rewrite it instead as a velocity
ratio then the velocity wave field v ¼ iω u is given by
P;P
vðx; ωÞ ¼ v0 ðωÞv T 12 ð0 Þez exp i kp1 z1 þ kp2 z2
8 9
< ð = ð8:190bÞ
1
Θ exp ikp1 ρ2e =2zp dϕ :
: 2π ;
C
282 8 Ultrasonic Transducer Radiation
ð8:192Þ
α;P α;P
where ¼
T 12 T 12 cos θ1α and er1
α
ðα
¼ P, SÞ is the unit vector along the part of the
stationary phase ray path in medium one for the path containing a P- or S-wave in
the solid, respectively. Using the same steps as carried out for the paraxial approx-
imation in the fluid-fluid problem to obtain the direct wave contribution from the
integral on Cε, we find finally
(
X ρ1 cp1 v0 ðωÞ α;P
uðx; ωÞ ¼ ΘT 12 0 ez exp i kp1 z1 þ kp2 z2
α¼P, S
iωρ2 cα2
9
ð α;P α >
=
1 T 12 d n er1 α
et exp i kp1 D1α þ kα2 D2α
h i q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffids ;
2π 1 n e α 2 α
D1 þ cα2 D2α =cp1
α
D1 þ cα2 cos 2 θ1α D2α =cp1 cos 2 θ2α >
;
C r1
ð8:193Þ
where now all the quantities in the integral in Eq. (8.193) are evaluated on the real
edge, C, of the transducer.
When point x is on the transducer axis, all distances and angles in the integral term
of Eq. (8.193) are constants and can be taken outside the integration, leaving
8.3 Beam Propagation Through A Planar Interface: Planar Probe 283
(
X ρ cp1 v0 ðωÞ α;P
uðz1 ; z2 ; ωÞ ¼ 1
T 12 0 ez exp i kp1 z1 þ kp2 z2
α¼P, S
iωρ2 cα2
9
α;P α ð α >
=
T 12 θ1 exp i kp1 D1α þ kα2 D2α 1 d n e α
e t
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
α ffi q
αffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h
r1
i ds :
D1 þ cα2 D2α =cp1 D1 þ cα2 cos 2 θ1α D2α =cp1 cos 2 θ2α 2π C 1 n er1 α 2 >
;
ð8:194Þ
However, we have
dα n er1
α
et dα α
h i ds ¼ α a dϕ ¼ d D1 þ cα2 D2 =cp1 dϕ ð8:195Þ
α 2
1 n er1 sin θ 1
so that on the axis we obtain the direct and edge wave terms given by
(
ρ1 cp1 v0 ðωÞ
uðz1 ; z2 ; ωÞ ¼ ez T P;P 12 0 exp i k p1 z1 þ kp2 z2
iωρ2 cp2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
D1 þ cp2 D2P =cp1 cos θ2P P
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P T P;P
12 θ 1 exp i k p1 D1 þ k p2 D2
P P
ð8:197Þ
When sufficiently far from the transducer θ1α , θ2α ! 0 and we can use the paraxial
approximation for the phase term, i.e.
b
S
dS θ2
ez
S
θ2
284 8 Ultrasonic Transducer Radiation
kp1 a2
kp1 D1α þ kα2 D2α ¼ kp1 z1 þ kα2 z2 þ ð8:198Þ
2 z1 þ cα2 z2 =cp1
(see Eq. (8.175)). Then Eq. (8.197) becomes, to first order in the far field
ρ1 cp1 v0
ikp1 a2
uðz1 ; z2 ; ωÞ ¼ ez T P;P
12 ð0 Þexp i kp1 z1 þ k p2 z2
:
iωρ2 cp2 2 z1 þ cp2 z2 =cp1
ð8:199aÞ
Or, in terms of a velocity ratio transmission coefficient and the velocity wave field,
v ¼ iω u, we find
ikp1 a2
vðz1 ; z2 ; ωÞ ¼ v0 ðωÞez v T P;P
12 ð0 Þexp i kp1 z1 þ kp2 z2
;
2 z1 þ cp2 z2 =cp1
ð8:199bÞ
which is very similar in form to our previous results for a single fluid (see Eq. (8.25)).
Now we will consider the case where a plane piston transducer of radius a is
oriented at oblique incidence to a plane fluid-fluid interface (Fig. 8.50). As in the
normal incidence case, we can represent the spherical wave term in the Rayleigh-
Sommerfeld integral by an angular spectrum of plane waves (see Eq. (8.154)),
propagate those plane waves through the interface (see Eq. (8.157)), and evaluate
the resulting angular spectrum by the method of stationary phase. Thus, Eq. (8.170)
is also valid for the oblique incidence case, i.e.
ð
iωρ1 v0 ðωÞ T p exp½iðk1 D1 þ k2 D2 Þ
pðx; ωÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dS;
2π D1 þ c2 D2 =c1 D1 þ c2 cos 2 θ1 D2 =c1 cos 2 θ2
S
ð8:200Þ
where T p ¼ T p ð cos θ1 Þ and the distances D1 and D2 are the distances along a ray
path from the face of the inclined transducer to the point x in the second medium
(Fig. 8.50) according to Snell’s law.
In the paraxial approximation we can obtain the diffraction correction for the
transmitted beam by following a similar approach to the normal incidence case.
8.3 Beam Propagation Through A Planar Interface: Planar Probe 285
In that case, we considered a general point, x, in the second medium and dropped a
line along the path of the direct wave ray through the interface and normal to the
transducer. In the oblique incidence case, however, the direct ray path from a
general point, x, that passes through the interface and ends up normal to the
transducer surface at some point y0 (Fig. 8.50) is an obliquely incident ray. In the
paraxial approximation, we want to expand the integrand of Eq. (8.200) about this
ray, for an arbitrary point y on the surface. Since, in this approximation, all rays are
nearly identical in orientation to the direct ray, we have θ1 ffi θ10 , θ2 ffi θ20 and the
amplitude part of Eq. (8.200) can be approximated as that along the direct ray,
giving
where e1 and e2 are unit vectors along the ray path in the first and second medium,
respectively, and xI is the point of intersection of the ray path with the interface. In
the paraxial approximation, we expect that the x0 and y0 components of these unit
vectors (in a (x0 , y0 , z0 ) coordinate system as shown in Fig. 8.51), are small. To obtain
an approximate expression for the phase ϕ under these conditions, we first rewrite
the phase as
286 8 Ultrasonic Transducer Radiation
x⬘ D1 xI
x
θ20
θ10
φ n
y⬘ D10
θ10
ρ
y0 z
y, y ⬘
0
ϕ ¼ k 2 e 2 x k 1 e 1 y þ ð k 1 e 1 k 2 e 2 Þ xI : ð8:203Þ
By Snell’s law, the last term in Eq. (8.203) can be written in terms of components
normal to the interface only, to obtain
0
ϕ ¼ k2 e2 x k1 e1 y þ ðk1 e1z k2 e2z ÞD0
0 0 ð8:204Þ
¼ k2 e2x x þ k2 e2z z k1 e1x0 x k1 e1y0 y þ ðk1 e1z k2 e2z ÞD0 ;
where D0 is the perpendicular distance from y0 to the interface (Fig. 8.51). For
convenience, we will write the phase in terms of the components of the wave
number vectors pm ¼ km em ðm ¼ 1, 2Þ as
0 0
ϕ ¼ p2x x þ p2z z p1x0 x p1y0 y þ ðp1z p2z ÞD0 ; ð8:205Þ
p2x ¼ p1x px
p2y ¼ p1y py ð8:206Þ
p2y0 ¼ p1y0 py0 :
For the choice of our coordinate systems we also have py ¼ py0 . The z-components
of p1 and p2 can then be written as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p1z ¼ k21 p2x p2y0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:207Þ
p2z ¼ k22 p2x p2y0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
px ¼ p1x ¼ px0 cos θ10 þ k21 p2x0 p2y0 sin θ10 ð8:208Þ
p2
p2x0 D10 D20 cos 2 θ10 y0 D10 D20
ϕ px ; py ¼ k1 D10 þ k2 D20
0 0 þ þ
2 k1 k2 cos 2 θ20 2 k1 k2
0 0
px0 x py0 y :
ð8:210Þ
Since the phase should be stationary on the path from y0 to x, we set ∂ϕ=∂px0 ¼
∂ϕ=∂py0 ¼ 0 which gives
0
k1 x
px ¼
0
Δx0
0 ð8:211Þ
k1 y
p y0 ¼ ;
Δy0
where
c2 cos 2 θ10
Δx0 ¼ D10 þ D20
c1 cos 2 θ20
ð8:212Þ
c2
Δy0 ¼ D10 þ D20 :
c1
Thus, we find that the stationary phase for the ray from y0 to x is given approxi-
mately as
0 2 0 2
k1 x k1 y
ϕ ¼ k1 D10 þ k2 D20 þ þ : ð8:213Þ
2Δxo 2Δy0
Note that Eq. (8.213) is a general paraxial result that makes no assumption on the
size of the incident angle θ10. When the transducer is normal to the interface, θ10 ,
288 8 Ultrasonic Transducer Radiation
θ20 ffi 0 and D10 ffi z1 , D20 ffi z2 so that Δx0 ¼ Δy0 ¼ z and Eq. (8.213) reduces to
the normal incidence case (Eq. (8.175)).
0 0
Letting x ¼ ρ cos ϕ, y ¼ ρ sin ϕ and dS ¼ ρ dρ dϕ, placing Eq. (8.213) into
Eq. (8.201) then gives (Fig. 8.51)
Equation (8.214) can be integrated on ρ for the cases when point x is inside, outside,
or on the edge of the main beam, as before, to give
pðx; ωÞ ¼ ρ1 c1 v0 ðωÞT p ðθ10 Þexp½iðk1 D10 þ k2 D20 ÞC1 ðx; ωÞ; ð8:215Þ
where Θ is as defined before (Eq. (8.55)) and ρe ¼ ρe ðϕÞ is the distance from point
y0 to the edge. For point x inside the main beam Θ ¼ 1 and the first integral can be
done [1] exactly as
2ðπ
cos 2 ϕ sin 2 ϕ pffiffiffiffiffiffiffipffiffiffiffiffiffiffi
dϕ= þ ¼ 2π Δx0 Δy0 ð8:217Þ
Δx0 Δy0
0
to yield
8 9
>
2ðπ >
ik1 ρ2e cos 2 ϕ sin 2 ϕ >>
>exp
< þ >
=
1 2 Δx0 Δy0
C1 ðx; ωÞ ¼ 1 pffiffiffiffiffiffiffipffiffiffiffiffiffiffi dϕ; ð8:218Þ
2π Δx0 Δy0 >
> cos 2 ϕ sin 2 ϕ >
>
0 >
: þ >
;
Δx0 Δy0
which shows explicitly the direct and edge wave terms. When point x is on the
central axis of the transducer, ρe ¼ a and the diffraction coefficient has the same
form as that of an elliptical shaped transducer radiating into a fluid. To see this,
consider the Rayleigh-Sommerfeld integral again, written out explicitly for the
on-axis response in the paraxial approximation (see Eq. (8.43)) as
8.3 Beam Propagation Through A Planar Interface: Planar Probe 289
ðð
iωρv0 ðωÞexpðikzÞ
pðz; ωÞ ¼ exp ik x2 þ y2 =2z dx dy: ð8:219Þ
2πz
For an elliptical transducer whose semi-major axes are ax, ay, we introduce the
transformations x ¼ ax u, y ¼ ay v and u ¼ σ cos α, v ¼ σ sin α so that Eq. (8.219)
becomes
iωρv0 ðωÞexpðikzÞax ay
pðz; ωÞ ¼
2πz
ð σ¼1
α¼2π ð h
i ð8:220Þ
exp ikσ 2 a2x cos 2 α þ a2y sin 2 α =2z σ dσ dα;
α¼0 σ¼0
where now the σ integration can be done explicitly (as well as one of the α
integrations—see Eq. (8.217)), to yield the on-axis pressure
2ðπ
8 h
i9
ax ay <exp ik a2x cos 2 α þ a2y sin 2 α =2z =
C1 ðz; ωÞ ¼ 1 dα; ð8:222Þ
2π : a2x cos 2 α þ a2y sin 2 α ;
0
which is identical with Eq. (8.218) if we let a2x =z ¼ a2 =Δx0 , a2y =z ¼ a2 =Δy0 .
As in fluid-fluid case, our previous model for the plane piston probe at normal
incidence on a plane fluid-solid interface (Eq. (8.188)) is also applicable when the
probe is at oblique incidence (Fig. 8.47):
X ρ v0 ðωÞ
uðx; ωÞ ¼ 1
ρ 2 cα2
8, S
α¼P
9
<1ð α;P α
T 12 d exp i kp1 D1α þ kα2 D2α =
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dSðyÞ;
:2π D1α þ cα2 D2α =cp1 D1α þ cα2 cos 2 θ1α D2α =cp1 cos 2 θ2α ;
S
ð8:223Þ
α;P α;P
where T 12 ¼ T 12 cos θ1α and recall, Dmα ðm ¼ 1, 2Þ ðα ¼ P, SÞ are the dis-
tances traveled through the interface along stationary phase paths that satisfy
290 8 Ultrasonic Transducer Radiation
Snell’s law for P- and S-waves, respectively, in the solid and θmα ðm ¼ 1, 2Þ
ðα ¼ P, SÞ are the associated angles (see Fig. 8.47).
Unlike, the normal incidence case, in the paraxial approximation both P- and
S-waves remain in the response. To see this, we first approximate, as before, the
amplitude terms in Eq. (8.223) along the direct ray from point x to the transducer
surface to obtain the P- and S-wave rays, i.e.
α;P α α ð
X ρ1 v0 ðωÞT 12 θ d
uðx; ωÞ ¼ pffiffiffiffiffiffiffi q10ffiffiffiffiffiffiffi0 exp i kp1 D1α þ kα2 D2α dSðyÞ; ð8:224Þ
α α
α¼P, S 2πρ2 cα2 Δx0 Δy0 S
where
α α α α α
Δx0 ¼ D10 þ cα2 cos 2 θ10 D20 =cp1 cos 2 θ20
α α α ð8:225Þ
Δy0 ¼ D10 þ cα2 D20 =cp1
and dα0 are the polarization vectors of the transmitted waves along the direct ray. We can
also approximate the phase terms in Eq. (8.224) using our previous paraxial results for
0 2 0 2
kp1 x kp1 y
kp1 D1α þ kα2 D2α ¼ α
kp1 D10 þ α
kα2 D20 þ α þ α ; ð8:226Þ
2Δx0 2Δy0
0 0
so that letting x ¼ ρ cos ϕ, y ¼ ρ sin ϕ and dS ¼ ρ dρ dϕ again, we can do the ρ
integration and obtain the displacement in terms of diffraction coefficients,
Cα1 (x, ω), as
X ρ cp1 v0 ðωÞ α;P α
α
uðx; ωÞ ¼ 1
T 12 θ10 d0α exp i kp1 D10
α α
þ kα2 D20 C1 ðx; ωÞ:
ρ c
α¼P, S 2 α2
iω
ð8:227aÞ
Or, again, we can use velocity ratio transmission coefficient and express the
velocity wave field v ¼ iω u as
X α;P α α α
α α
vðx; ωÞ ¼ v0 ðωÞv T 12 θ10 d0 exp i kp1 D10 þ kα2 D20 C1 ðx; ωÞ; ð8:227bÞ
α¼P, S
Here
8 " ! #9
>
> α ik ρ2 cos 2 ϕ sin 2 ϕ >
>
>Θ exp 1 e
2ðπ > þ >
>
1 < 2 α
Δx0 α
Δy0 =
C1α ðx; ωÞ ¼ pffiffiffiffiffiffiffiqffiffiffiffiffiffiffi ! dϕ
2π Δx0 α
Δy0 α >
> cos 2 ϕ sin 2 ϕ >
>
>
> þ >
>
0 : α α ;
Δx0 Δy0
ð8:228Þ
and
8
>
> 1 x in the main ðcylindricalÞ beam
<
α of refracted α waves ðα ¼ P, SÞ
Θ ¼ ð8:229Þ
>
> 1=2 x on the beam edge
:
0 otherwise
Now, we want to consider the case where a spherically focused transducer of radius
a and geometrical focal length R0 is at oblique incidence to a plane fluid-fluid
interface (Fig. 8.53). If we use the O’Neil model for the focused probe, then the
incident pressure field in the first medium is again given by Eq. (8.88). As in the
planar case, we can represent the spherical waves in the Rayleigh-Sommerfeld
integral by an angular spectrum of plane waves (Eq. (8.154)), propagate those plane
waves through the interface (Eq. (8.157)) and evaluate the resulting angular
spectrum by the method of stationary phase to obtain
ð
iωρ1 v0 ðωÞ T p ðθ1 Þexp½iðk1 D1 þ k2 D2 Þ
pðx; ωÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dS;
2π D1 þ c2 D2 =c1 D1 þ c2 cos 2 θ1 D2 =c1 cos 2 θ2
Sf
ð8:230Þ
which is identical with Eq. (8.170) for the piston probe except the integration is now
over the spherical surface, Sf and the distances D1 and D2 are along a ray from a
point y on the spherical surface to a point x in the second fluid (Fig. 8.53).
292 8 Ultrasonic Transducer Radiation
0 2 0 2
0 0 ρ2 x y
εz 0 x ; y ¼ ¼ þ ð8:233Þ
2R0 2R0 2R0
8.4 Beam Propagation Through a Planar Interface: Focused Probe 293
so that
0 2 0 2
k1 x ð1 Δx0 =R0 Þ k1 y 1 Δy0 =R0
k1 D1 þ k2 D2 ¼ k1 D10 þ k2 D20 þ þ :
2Δx0 2Δy0
ð8:234Þ
k1 D1 þ k2 D2 ¼ k1 D10 þ k2 D20
k1 ρ2 cos 2 ϕð1 Δx0 =R0 Þ sin 2 ϕ 1 Δy0 =R0
þ þ ; ð8:235Þ
2 Δx0 Δy0
which is very similar to the planar probe at oblique incidence case. The focusing
effects are due to the terms qx0 ¼ 1 Δx0 =R0 and qy0 ¼ 1 Δy0 =R0 which play the
same role here as the q0 term did for the single medium focused probe case.
Placing Eq. (8.234) into Eq. (8.231), we find
Comparing this result with the oblique incidence case for a planar probe
(Eq. (8.214)) we see that the forms are identical and if we let R0 ! 1 we do
recover the planar case exactly. However, because the focusing terms (qx0, qy0) can
be either positive or negative, when doing the ρ integration we must consider all
possibilities when evaluating the direct wave term. Since [1]
2ðπ
qx0 cos 2 ϕ qy0 sin 2 ϕ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
dϕ= þ ¼ 2πε Δx0 =jqx0 j Δy0 =qy0 ; ð8:237Þ
Δx0 Δy0
0
where
8
<1 if both qx0 , qy0 > 0
ε ¼ i if qx0 qy0 < 0 ; ð8:238Þ
: 1 if both qx0 , qy0 < 0
pðx; ωÞ ¼ ρ1 c1 v0 ðωÞT p ðθ10 Þexp½iðk1 D10 þ k2 D20 ÞC1 ðx; ωÞ; ð8:239Þ
294 8 Ultrasonic Transducer Radiation
with
and
8
<1 y0 in Sf
Θ ¼ 1=2 y0 on edge of Sf : ð8:242Þ
:
0 y0 outside Sf
When the probe is oriented normal to the planar interface, in the paraxial approx-
imation θ10 , θ20 ffi 0 and so Δx0 ffi Δy0 . In this case we have
pðx; ωÞ ¼ ρ1 c1 v0 ðωÞT p ð0 Þexp½iðk1 D10 þ k2 D20 ÞC1 ω; a; R0 ; Δy0 ð8:243Þ
which is in exactly the same form as the diffraction coefficient for the spherically
focused probe in a single medium (Eq. (8.130)) if we replace the distance D in that
case by Δy0.
For the special case when point x is on the axis of the transducer, ρe ¼ a,
Δy0 ¼ z1 þ c2 z2 =c1 , Θ ¼ 1 and the integral in Eq. (8.244) can be evaluated
explicitly to give
1
C1 ω; a; R0 ; Δy0 ¼ 1 exp ik1 a2 1 Δy0 =R0 =2Δy0 ; ð8:245Þ
1 Δy0 =R0
ð8:246Þ
y' P
D1
y⬘ y0
y
εz ′ plane P
y′
y0
ρ
where the plane surface is replaced now by the spherical surface Sf and the distances
are now measured along rays from point x to that surface. Thus, in the paraxial
approximation Eq. (8.224) also holds for this case by again letting S ! Sf :
α αð
X ρ1 v0 ðωÞT α;P
12 θ 10 d0
uðx; ωÞ ¼ pffiffiffiffiffiffiffiqffiffiffiffiffiffiffi exp i kp1 D1α þ kα2 D2α dSðyÞ: ð8:247Þ
α α
α¼P, S 2πρ2 cα2 Δx0 Δy0
S f
Using the same procedures followed for the focused probe on the fluid-fluid
interface, it follows that the phase terms of the P- and S-waves can be approximated
in the paraxial approximation as
" #
α
α
kp1 ρ2 qx0 cos 2 ϕ qy0 sin 2 ϕ
kp1 D1α þ kα2 D2α ¼ α
kp1 D10 þ α
kα2 D20 þ α þ α ; ð8:248Þ
2 Δx0 Δy0
8.4 Beam Propagation Through a Planar Interface: Focused Probe 297
with
α α
qx0 ¼ 1 Δx0 =R0
α α
ð8:249Þ
qy0 ¼ 1 Δy0 =R0
and all the other quantities as defined before. With this expression for the phase, the
ρ integration then yields, as before,
X ρ cp1 v0 ðωÞ α;P α
α
uðx; ωÞ ¼ 1
T 12 θ10 d0α exp i kp1 D10
α α
þ kα2 D20 C1 ðx; ωÞ:
ρ c
α¼P, S 2 α2
iω
ð8:250aÞ
Or, in terms of a transmission coefficient in terms of velocity ratios and the velocity
field, v ¼ iω u, we have
X α α α
vðx; ωÞ ¼ v0 ðωÞv T α;P α α
12 θ10 d0 exp i kp1 D10 þ k α2 D20 C1 ðx; ωÞ; ð8:250bÞ
α¼P, S
where now
8 " !#9
α
>
> α ik1 ρ2e qx0α
cos 2 ϕ qy0 sin 2 ϕ >>
2ðπ >
> Θ exp þ >
>
1 < 2 Δ α
x0 Δ α
y0
=
α !
C1 ðx; ωÞ ¼ pffiffiffiffiffiffiffiqffiffiffiffiffiffiffi dϕ
2π Δx0α
Δy0 α >
> qx0α α
cos 2 ϕ qy0 sin ϕ
2 >
>
0 >> þ >
>
: α α ;
Δx0 Δy0
ð8:251Þ
with
8
>
> 1 x in the main ðconicalÞ beam
<
of refracted α waves ðα ¼ P, SÞ
Θα ¼ : ð8:252Þ
>
> 1=2 x on the beam edge
:
0 otherwise
ð8:253Þ
298 8 Ultrasonic Transducer Radiation
with
8 α α
<1 if both qx0 , qy0 >0
α α
εα ¼ i if qx0 qy0 < 0 ð8:254Þ
: 1 if both qx0α α
, qy0 <0
and
α
α
qx0 cos 2 ϕ qy0 sin 2 ϕ
f α ðϕÞ ¼ α þ α : ð8:255Þ
Δx0 Δy0
where the diffraction coefficient is identical to that for the fluid-fluid problem (see
Eq. (8.244)).
First, we will consider the case where a fluid medium, with density ρ1 and wave
speed c1, is separated by a general smooth, curved interface, S, from a second fluid
medium, with density ρ2 and wave speed c2 and where a general transducer (planar
or focused) is located in the first medium (Fig. 8.56). In Chap. 5, we showed that if
the Sommerfeld radiation condition for medium two is satisfied, then the pressure at
an arbitrary point, x2, in medium two due to the waves generated by the transducer
can be represented as an integral over the curved interface S as
ð
pðx2 ; ωÞ ¼ ½G2 ðxs ; x2 ; ωÞ∂pþ ðxs ; ωÞ=∂nðxs Þ
ð8:257Þ
S
pþ ðxs ; ωÞ∂G2 ðxs ; x2 ; ωÞ=∂nðxs ÞdSðxs Þ;
where the plus superscript is used to denote that the field quantities in the integrand
are calculated in medium two at the interface. The function G2 is just the funda-
mental solution for medium two due to a point source of pressure acting in that
medium. At high frequencies, G2 and its normal derivative are given by
We will use the Rayleigh-Sommerfeld integral for a piston source (either planar or
focused) to represent the waves incident on the curved interface (Eq. (8.9)). Since
the Rayleigh-Sommerfeld integral is a superposition of point sources over the
transducer face, we will first examine the case where we have an incident pressure
field on the interface due to an isolated point source acting in medium one given by
er 2
r2
x2
where the minus superscript indicates that this field is evaluated on the interface on
the medium one side. The particular amplitude term chosen in Eq. (8.259) coincides
with coefficient appearing in the Rayleigh-Sommerfeld equation (Eq. (8.9)). Con-
sider now a small patch of the curved interface (Fig. 8.57)) where the point xs is
close to a fixed point, xI, in the patch and we let x ¼ xs xI . Then we have
approximately
r1 ffi r þ e x ð8:260Þ
and
where Tp is the plane wave transmission coefficient (based on pressure ratio) for a
planar interface and etr is a unit vector of a transmitted plane wave that satisfies
Snell’s law, i.e.
Using Eqs. (8.260) and (8.263) then the wave field on the medium two side of the
interface can be rewritten as
Since the entire interface can be broken up into multiple small patches similar to the
one we have just considered, we can take Eq. (8.264) as an approximation valid for
the entire curved interface.
To reiterate, the basic assumption used here was that the interaction of the
incident waves with the curved interface could locally be considered to be the
same as the interaction of a plane wave with an equivalent planar interface, where
the orientation (unit normal) of the equivalent planar interface coincides with that
of the unit normal to the actual curved interface. Under this assumption the wave
field at the interface in medium two can be obtained by simply solving a plane
wave/plane interface transmission problem.
This approximation is similar in spirit to the one used by Kirchhoff in treating
the transmission of waves through an aperture [10], so we will call it here the
Kirchhoff approximation (see Chap. 6). It has been used extensively in the optics,
electromagnetics and acoustics literature. We expect that the Kirchhoff approxi-
mation will be valid for smooth curved interfaces when the wavelength of the
ultrasound is much smaller than the curvature of the interface and the interface
geometry is fairly regular and slowly varying over the “footprint” of the incident
beam on the interface, a condition that is very likely to be satisfied in practice in
most NDE problems.
Placing Eqs. (8.258) and (8.264) into Eq. (8.257), the pressure in medium two
due to our point source,pδ(x2, ω), can be given explicitly as
ð
ω 2 ρ 1 v 0 ð ωÞ expði k1 r 1 Þ expði k2 r 2 Þ
p δ ð x 2 ; ωÞ ¼ 2
T p ½ðetr nÞ þ ðer2 nÞ dSðxs Þ:
8π c2 r1 r2
S
ð8:265Þ
Consequently, the pressure in medium two due to a transducer, pT(x2, ω), can be
obtained by integrating these point source solutions over the face of the transducer, i.e.
8
ð ð
ω 2 ρ 1 v 0 ð ωÞ < expði k1 r 1 Þ
p T ð x 2 ; ωÞ ¼ T p ½ðetr nÞ þ ðer2 nÞ
8π c22 : r1
ST S ð8:266Þ
expði k2 r 2 Þ
dSðxs Þ dSðx1 Þ:
r2
302 8 Ultrasonic Transducer Radiation
where
At high frequencies the phase in Eq. (8.267) is rapidly varying and we expect that
the major contributions to the integral will come from points where the phase term
ϕ is stationary (see Appendix E). Suppose that the point x0 is such a stationary
phase point on the interface. Then the phase term can be written as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ϕ ¼ k1 pðxsffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ r10 Þ ðxs þ r10 Þ
ð8:269Þ
þ k2 ðr20 xs Þ ðr20 xs Þ;
where
r10 ¼ x0 x1
ð8:270Þ
r20 ¼ x2 x0
and where xs here is the position of a point on the interface S as measured from x0.
Since r10 and r20 are two fixed vectors during the integration over the curved
interface, if we let xs ¼ xs ðs1 ; s2 Þ where sα ðα ¼ 1, 2Þ are orthogonal surface arc
length coordinates of xs as measured from x0, then the phase ϕ will be a function
only of those coordinates, i.e. ϕ ¼ ϕðs1 ; s2 Þ. Near the stationary point then we can
expand ϕ as
∂ϕ 1 ∂ ϕ
2
ϕðxs ; ωÞ ¼ ϕðx0 ; ωÞ þ sα þ sα sβ ; ð8:271Þ
∂sα x0 2 ∂sα ∂sβ
x0
where the terms on the right hand side of Eq. (8.271) can be obtained directly by
differentiating Eq. (8.269) to yield
8.5 Beam Propagation Through a Curved Interface 303
ϕðx0 ; ωÞ ¼ k1 r 10 þ k2 r 20
∂ϕ ∂xs ∂xs
¼ k 1 e r1 k 2 e r2
∂sα x0 ∂sα ∂sα
x0
∂ ϕ
2 2
∂ xs ∂xs ∂xs k1 k2
¼ ðk1 er1 k2 er2 Þ þ þ
∂sα ∂sβ ∂sα ∂sβ ∂sα ∂sβ r 10 r 20
x0
k1 ∂xs ∂xs k2 ∂xs ∂xs
er1 er1 er2 er2 :
r 10 ∂sα ∂sβ r 20 ∂sα ∂sβ x0
ð8:272Þ
which is just a statement of Snell’s law and, therefore, point x0 must lie on a ray
path from x1 to x2 that satisfies Snell’s law. This also means that at the stationary
phase point we have er2 ¼ etr and the amplitude part of the integrand in Eq. (8.268)
at that point becomes
2ðetr nÞT p
f ¼ : ð8:274Þ
r 10 r 20 x0
Taking this amplitude term as a constant near the stationary point then Eq. (8.267)
can be written approximately near that point as
ω2 ρ1 v0 ðωÞðetr nÞT p
pδ ðx2 ; ωÞ ¼ expði k1 r 10 þ i k2 r 20 Þ
4π 2 c2 r 10 r 202 3
ð 1 s2 ¼þε
s1 ¼þε ð 2
2 ð8:275Þ
i ∂ ϕ
exp4 sα sβ 5ds1 ds2 ;
2 ∂sα ∂sβ
s1 ¼ε1 s2 ¼ε2 x0
where it is implicitly understood that all the quantities appearing outside the
integral in Eq. (8.275) are evaluated at the stationary phase point. Now, consider
further the phase term remaining in the integral of Eq. (8.275). From differential
geometry (Appendix E) we have
2
∂ xs γ ∂xs
¼ Γαβ þ hαβ n; ð8:276Þ
∂sα ∂sβ ∂sγ
where Γγαβ is the Christoffel symbol of the second kind and hαβ is the curvature
tensor. Placing Eq. (8.276) into the expression for the second order derivatives of ϕ
appearing in Eq. (8.272) we find
304 8 Ultrasonic Transducer Radiation
∂ ϕ
2
γ ∂xs ∂xs
¼ Γαβ k1 er1 k2 er2
∂sα ∂sβ ∂sγ ∂sγ
x0
∂xs ∂xs k1 k2 ð8:277Þ
þ hαβ ½k1 ðer1 nÞ k2 ðer2 nÞ þ þ
∂s
α ∂sβ r10 r 20
k1 ∂xs ∂xs k2 ∂xs ∂xs
er1 er1 er2 er2 :
r 10 ∂sα ∂sβ r 20 ∂sα ∂sβ x0
But the first bracketed term on the right side of Eq. (8.277) vanishes again by Snell’s
law so that at the stationary phase point this quadratic phase term, which we will
denote by ϕ, αβ, is given by
∂xs ∂xs k1 k2
ϕ, αβ ¼ hαβ ½k1 ðer1 nÞ k2 ðer2 nÞ þ þ
∂sα ∂sβ r 10 r 20 ð8:278Þ
k1 ∂xs ∂xs k2 ∂xs ∂xs
er1 er1 er2 er2 :
r 10 ∂sα ∂sβ r 20 ∂sα ∂sβ x0
If the s1 surface coordinate is taken to lie in the plane of incidence formed by er1 and
n and the s2 coordinate is normal to that plane then it follows thatt1 ¼ ∂xs =∂s1 is a unit
vector tangent to the surface in the plane of incidence and t2 ¼ ∂xs =∂s2 is a unit vector
tangent to the surface in a plane normal to the plane of incidence (Fig. 8.58). Letting the
angles θ1 and θ2 be the acute angles between er1 and n and er2 and n, respectively, then
er1 n ¼ cos θ1 , er2 n ¼ cos θ2 and the components of ϕ, αβ are
k1 cos 2 θ1 k2 cos 2 θ2
ϕ, 11 ¼ þ h11 ðk1 cos θ1 k2 cos θ2 Þ
r 10 r 20
ϕ, 12 ¼ ϕ, 21 ¼ h12 ðk1 cos θ1 k2 cos θ2 Þ ð8:279Þ
k1 k2
ϕ, 22 ¼ þ h22 ðk1 cos θ1 k2 cos θ2 Þ:
r 10 r 20
To obtain a more explicit expression for the curvature terms in Eq. (8.279), let
the principal radii of curvature of the surface at x0 be denoted by R1 and R2 and
their associated directions (in the plane of the interface) be the unit vectors p1
and p2, respectively. Let the angle from the p1 axis to the t1 axis be φ (Fig. 8.59).
0 0 0
Thus, if we let h11 ¼ 1=R1 , h22 ¼ 1=R2 , h12 ¼ 0 we have
cos 2 φ sin 2 φ 1
h11 ¼ þ
R1 R2 R
I
1 1 1
h12 ¼ h21 ¼ sin φ cos φ ð8:281Þ
R 1 R2 RI0
sin 2 φ cos 2 φ 1
h22 ¼ þ ;
R1 R2 R0
which is a result from differential geometry also called Euler’s theorem [11]. Equa-
tion (8.279) then becomes
In the special case when the plane of incidence and the principal plane with radius
of curvature R1 coincide then φ ¼ 0 and 1=RI ! 1=R1 , 1=R0 ! 1=R2 , 1=RI0 ! 0.
To proceed with the evaluation of Eq. (8.275) by the method of stationary phase,
we first note that since the rapidly varying phase term in Eq. (8.275) causes the
306 8 Ultrasonic Transducer Radiation
integral to be negligible away from the stationary phase point we can extend the
limits of integration to 1, giving
where we have also used the relationship ðetr nÞ ¼ cos θ2 . In order to facilitate
our later comparisons with ray theory, we will introduce a change of variables
defined by
t1
s1 ¼ pffiffiffiffiffi
cos θ2 k2
t2
s2 ¼ pffiffiffiffiffi:
k2
Then Eq. (8.283) becomes
ωρ1 v0 ðωÞT p
pδ ð x 2 ; ω Þ ¼ expði k1 r 10 þ i k2 r 20 Þ
4π 2 r 10 r 20
ð t2 ¼þ1
t1 ¼þ1 ð h 0 i ð8:284Þ
exp iϕ, αβ tα tβ =2 dt1 dt2
t1 ¼1 t2 ¼1
with
If we now rotate the (t1, t2) coordinates to a new set of (u1, u2) coordinates where the
0
ϕ, αβ form in the phase term is diagonalized, i.e.
0 0 0
ϕ, 11 t21 þ 2ϕ, 12 t1 t2 þ ϕ, 22 t22 ! ϕp1 u21 þ ϕp2 u22 ; ð8:286Þ
8.5 Beam Propagation Through a Curved Interface 307
where
0 0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ϕ, 11 þ ϕ, 22 1 0 0
2 0 2
ϕp1 ¼ þ ϕ, 11 ϕ, 22 þ 4 ϕ, 12
0
2 0 2qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:287Þ
ϕ, 11 þ ϕ, 22 1 0 0
2 0 2
ϕp2 ¼ ϕ, 11 ϕ, 22 þ 4 ϕ, 12 ;
2 2
then we have
ω ρ1 v0 ðωÞT p
p δ ð x 2 ; ωÞ ¼ expði k1 r 10 þ i k2 r 20 Þ
4π 2 r 10 r 20
ð
u1 ¼þ1 u2 ð
¼þ1 ð8:288Þ
iϕp1 u21 iϕp2 u22
exp þ du1 du2 :
2 2
u1 ¼1 u2 ¼1
Performing both integrations directly (see Appendix E), we find the point source
response as
ω ρ1 v0 ðωÞT p exp iπ=4 sgn ϕp1 þ sgn ϕp2
p δ ð x 2 ; ωÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
expði k1 r 10 þ i k2 r 20 Þ
2π r 10 r 20 ϕp1 ϕp2
ð8:289Þ
where
π
σ¼ sgn ϕp1 þ sgn ϕp2 : ð8:291Þ
4
Equation (8.290) is the direct counterpart of Eq. (8.170) previously obtained for the
problem of beam transmission through a plane fluid-fluid interface. In fact, when
the principal radii of curvature of the curved interface are assumed to be infinitely
large, we find from the present model
0 1 c2 cos 2 θ1 1
ϕp1 ! ϕ, 11 ¼ þ
r 20 c1 cos 2 θ2 r 10
0 1 c2 1 ð8:292Þ
ϕp2 ! ϕ, 22 ¼ þ
r 20 c1 r 10
π
σ! ;
2
and
The distances (ρv1, ρv2) can be interpreted geometrically as the principal radii of
curvature of the transmitted wave front in medium two at the interface [12]. Thus,
Eq. (8.290) can be written in terms of these curvatures as
ð
i ω ρ1 v0 expði k1 r 10 Þ
pT ðx2 ; ωÞ ¼ Tp
2π r 10
ST
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:296Þ
jρv1 j j ρv2 j
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi expði k2 r 20 þ iσ iπ=2ÞdS:
j r20 þ ρv1 j j r 20 þ ρv2 j
The first term in the integrand of Eq. (8.296) is the incident pressure on the interface
from our original point source. The second term is the plane wave transmission
coefficient which, when multiplied by the incident pressure, gives the transmitted
pressure at the interface in medium two. The third term is a geometrical spreading
factor that modifies the transmitted pressure at the interface, according to the
changing area of a propagating ray tube, to obtain the corresponding pressure
amplitude at point x2. Finally, the fourth term in the integrand contains the phase
contribution due to propagation to point x2 from the interface and additional phase
terms that reproduce the well-known change in phase phenomena as one passes
through one (or more) singularities of the spreading factor [13]. Thus, our stationary
phase evaluation of the interface integral is completely equivalent to an approach
8.5 Beam Propagation Through a Curved Interface 309
ρ1,c1
ρ2,c2 > c1
based directly on geometrical ray theory. However, a nice feature of the stationary
phase approach is that explicit expressions for both the amplitude and phase
changes arise from the stationary phase evaluations, even in the complex case
considered here of an arbitrarily curved interface. Equivalent results to these were
quoted by Cerveny and Ravindra [12] based on the previous work of
Gel’chinskiy [14].
When the two distances ρv1, ρv2 appearing in the integration over the transducer
face (Eq. (8.290)) are both positive, the behavior of this integral is similar to that for
the planar interface (Eq. (8.170)), i.e. it is non-singular and can be used as the basis
for numerical evaluations. Such a situation occurs, for example, when both princi-
pal radii are negative ðR1 < 0, R2 < 0Þ, so that the interface is convex (see
Fig. 8.60a) and the beam is transmitted into a faster second medium ðc2 > c1 Þ. In
this case, the rays coming from a point source in medium one diverge in medium
two, as shown in Fig. 8.60a. When one or both the distances ρv1, ρv2 are negative,
however, then the amplitude of the transmitted wave calculated by geometric ray
theory can become infinite at locations in medium two where r 20 þ ρv1 ¼ 0 and/or
r 20 þ ρv2 ¼ 0 and the integrand appearing in Eq. (8.290) becomes singular. This can
happen, for example, for radiation through a concave interface ðR1 > 0, R2 > 0Þ
into a faster second medium ðc2 > c1 Þ where the ray paths cross (Fig. 8.60b). At
such points, which are called caustics, one can either perform a more detailed
asymptotic analysis of the wave field or resort to an entirely different procedure
(from that of the method of stationary phase) for evaluating the curved interface
integral. One such alternative approach, using edge elements, will be discussed in
Sect. 8.6.
In the planar interface problems considered previously in this chapter we were able
to obtain direct /edge wave expressions for the diffraction correction through the
use of the paraxial approximation. Here, we will show that this is also possible for
the general curved interface case. Invoking the paraxial approximation in this
curved interface case will also explicitly show the manner in which the integrand
becomes singular at caustics.
310 8 Ultrasonic Transducer Radiation
s1
r2
x2
xs D20
r1 θ20
n
y1 x0
s2
θ10
x1 y D10
y2
y0
curved interface
Fig. 8.61 Propagation of waves through a general curved interface in the neighborhood of a
stationary phase ray path
First, we return to Eq. (8.267) for the point source response through a curved
interface, i.e.
ð
ω 2 ρ 1 v 0 ð ωÞ expði k1 r 1 Þ expði k2 r 2 Þ
pδ ðx2 ; ωÞ ¼ T p ½ðetr nÞ þ ðer2 nÞ dSðxs Þ:
8π 2 c2 r1 r2
S
ð8:297Þ
To obtain a form valid in the paraxial approximation, we assume that the amplitude
terms in Eq. (8.297) can be approximated along a fixed ray direction from point x2
to a point y0 on a planar transducer surface, where the ray satisfies Snell’s law and
the ray is normal to the transducer surface at y0 (Fig. 8.61). Thus, as shown
previously in the stationary phase evaluation of Eq. (8.267) we obtain
ð
ω2 ρ1 v0 ðωÞðetr nÞT p
pδ ð x 2 ; ω Þ ¼ expði k1 r 1 þ i k2 r 2 ÞdS; ð8:298Þ
4π 2 c2 D10 D20
S
∂ϕ ∂ϕ
ϕðx1 ; xs ; ωÞ ffi ϕðy0 ; x0 ; ωÞ þ yα þ sα
∂y ∂sα y0 , x0
α y0 , x0
1
2
∂ ϕ ∂ ϕ
2
1 ∂ ϕ
2
þ yα yβ þ y α sβ þ sα sβ :
2 ∂yα ∂yβ ∂yα ∂sβ 2 ∂sα ∂sβ
y0 , x0 y0 , x0 y0 , x0
ð8:299Þ
where er1 ¼ r1 =r 1 , er2 ¼ r2 =r 2 are unit vectors along the ray paths in medium one
and two, respectively. However, since the fixed ray is chosen such that the ray is
perpendicular to the transducer surface and satisfies Snell’s law, both of these first
derivative terms vanish and one needs to calculate only the second derivative terms,
which after some algebra (and again using Snell’s law and the normality of the fixed
ray to the transducer surface) can be shown to be
∂ ϕ k1
2
¼ δαβ
∂yα ∂yβ r1 y0 , x0
y0 , x0
2
∂ ϕ k1 ∂y ∂xs
¼
∂yα ∂sβ r 1 ∂yα ∂sβ y0 , x0
y0 , x0
∂ ϕ
2
k1 ∂xs ∂xs
¼ hαβ ½k1 ðer1 nÞ k2 ðer2 nÞ þ δαβ er1 er1
∂sα ∂sβ r1 ∂sα ∂sβ
y0 , x0
k2 ∂xs ∂xs
þ δαβ er2 er2 :
r ∂s ∂s
2 α β y0 , x0
ð8:302Þ
Here
k1 ∂xs ∂xs k2 ∂xs ∂xs
K αβ ¼ δαβ er1 er1 þ δαβ er2 er2
D10 ∂sα ∂sβ D20 ∂sα ∂sβ
þ hαβ fk1 ðer1 nÞ k2 ðer2 nÞg
ð8:304Þ
and all the terms appearing in Eqs. (8.303) and (8.304) are implicitly understood to
be evaluated along the fixed ray. If we take the y1 and s1 axes to be in the plane of
incidence and the y2 and s2 axes to be normal to the plane of incidence, then in terms
of the geometry of Fig. 8.61
∂y ∂xs
¼ cos θ10
∂y1 ∂s1
∂y ∂xs
¼1 ð8:305Þ
∂y2 ∂s2
∂y ∂xs ∂y ∂xs
¼ ¼0
∂y1 ∂s2 ∂y2 ∂s1
and
k1 k2
K 11 ¼ cos 2 θ10 þ cos 2 θ20 h11 fk1 cos θ10 k2 cos θ20 g
D10 D20
K 12 ¼ K 21 ¼ h12 fk1 cos θ10 k2 cos θ20 g ð8:306Þ
k1 k2
K 22 ¼ þ h22 fk1 cos θ10 k2 cos θ20 g:
D10 D20
Now, if we choose a rotated set of coordinates (u1, u2) in the tangent plane to the
interface at point x0 defined by
uγ ¼ Qγα sα
or ð8:307Þ
sα ¼ Qαγ
T
uγ
such that the rotation matrix Qαβ diagonalizes the Kαβ tensor, then the phase term
simplifies to
1 k1 k1 ∂y ∂xs
ϕðx1 ; xs ; ωÞ ¼ k1 D10 þ k2 D20 þ y y T
Qβγ y α uγ
2 D10 α α D10 ∂yα ∂sβ ð8:308Þ
1 1
þ K p1 u21 þ K p2 u22 ;
2 2
8.5 Beam Propagation Through a Curved Interface 313
where Kp1 and Kp2 are the principal values of the Kαβ matrix:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
K 11 þ K 22 1
K p1, p2 ¼ ðK 11 K 22 Þ2 þ 4K 212 : ð8:309Þ
2 2
0
Note that all these values are simply related to the ϕ, αβ defined previously (see
Eq. (8.285)) through
0
K 11 ¼ k2 cos 2 θ20 ϕ, 11
0
K 22 ¼ k2 ϕ, 22 ð8:310Þ
0
K 12 ¼ k2 cos θ20 ϕ, 12
Placing the phase expression of Eq. (8.308) into Eq. (8.298) and assuming that the
integration of the quadratic approximation for the phase term over the interface can
be replaced by a integration over the entire tangent plane at x0, we find
where
k1 ∂y ∂xs
dγ ¼ T
Qβγ yα ð8:313Þ
D10 ∂yα ∂sβ
is a coefficient that is linear in terms of the (y1, y2) coordinates. As long as both Kp1
and Kp2 are non-zero, the integrations on u1 and u2 can be performed explicitly since
ð
þ1 rffiffiffiffiffiffi
π
exp ibx þ iax dx ¼ 2
exp ib2 =4a þ π sgn a=4 ð8:314Þ
j aj
1
with
π π
σ¼ sgn K p1 þ sgn K p2 sgn ϕp1 þ sgn ϕp2 : ð8:316Þ
4 4
Note that we would also obtain exactly this same result by evaluating the interface
integral in Eq. (8.312) directly by the method of stationary phase. Thus, Eq. (8.315)
is the generalization of the result obtained previously by the method of stationary
phase (Eq. (8.289)) that is now valid, in the paraxial approximation, for a bundle of
rays about the stationary phase path from x1 to x2 and reduces to our former result
identically when y ¼ 0. If this result is integrated over the face of the transducer,
using Eq. (8.311), then
pT ðx2 ; ωÞ ¼ ρ1 c1 v0 ðωÞT p expðik1 D10 þ ik2 D20 ÞC1 ðST ; ωÞ; ð8:318Þ
where
Consider first the special case when the plane of incidence and a principal plane
of the surface are aligned. For this case the Kαβ is already diagonal and so no
rotation is needed, i.e. Qαβ ¼ δαβ and Eq. (8.313) gives
k1 cos θ10 y1
d1 ¼
D10 ð8:320Þ
k 1 y2
d2 ¼ :
D10
If we place this result into Eq. (8.319) and express the (y1, y2) coordinates in terms
of polar coordinates (ρ, ϕ), the ρ integration can be done explicitly and we obtain a
form similar to the previous planar interface problems, namely
8.5 Beam Propagation Through a Curved Interface 315
8 9
ð
ik2 cos θ20 expðiσ Þ < Θ exp ik1 ρ2e gðϕÞ=2 =
C 1 ð S T ; ωÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi dϕ ; ð8:321Þ
2πD10 D20 K p1 K p2 : C gð ϕÞ ;
where
1 k1 cos 2 θ10 1 k1
gð ϕÞ ¼ 1 cos ϕ þ
2
1 sin 2 ϕ ð8:322Þ
D10 D10 K p1 D10 D10 K p2
and
k1 k2 1
K p1 ¼ cos 2 θ10 þ cos 2 θ20 ðk1 cos θ10 k2 cos θ20 Þ
D10 D20 R1
ð8:323Þ
k1 k2 1
K p2 ¼ þ ðk1 cos θ10 k2 cos θ20 Þ
D10 D20 R2
with R1 being the principal curvature of the interface in the plane of incidence and
R2 the principal curvature in a plane normal to the plane of incidence.
For the more general case when a principal curvature plane is not aligned to the
plane of incidence, the rotation matrix Qαβ is needed. Then
k1
d1 ¼ ðQ cos θ10 y1 þ Q12 y2 Þ
D10 11
ð8:324Þ
k1
d2 ¼ ðQ cos θ10 y1 þ Q22 y2 Þ:
D10 21
Placing this result into Eq. (8.319) and introducing polar coordinates, we again
obtain a diffraction correction expression in the form of Eq. (8.321) but where now
2
1 k1 Q Q2
gð ϕÞ ¼ 1 cos 2 θ10 11 þ 21 cos 2 ϕ
D10 D10 K p1 K p2
2 k1 Q11 Q12 Q21 Q22
cos θ10 þ cos ϕ sin ϕ ð8:325Þ
D10 D10 K p1 K p2
2 2
1 k1 Q12 Q22
þ 1 þ sin 2 ϕ:
D10 D10 K p1 K p2
This form can be placed in a form very similar to Eq. (8.322) if we define the quantities
1 Q2 Q2
¼ 11 þ 21
K I K p1 K p2
1 Q Q Q Q
¼ 11 12 þ 21 22 ð8:326Þ
K I0 K p1 K p2
1 Q212 Q222
¼ þ
K 0 K p1 K p2
316 8 Ultrasonic Transducer Radiation
in much the same manner as in Eq. (8.281) we previously defined the “normal
section” curvature terms (1/RI, 1/RI0, 1/R0) (see [12]). In this general case, the
values of Kp1 and Kp2 are given by Eq. (8.309), or equivalently, by
Having developed the paraxial approximation for the radiation of a planar trans-
ducer through a curved interface, it is possible to directly extend that result to the
case where the transducer is spherically focused by modifying the phase term
appropriately, as done previously for the plane interface problems. If we let Rs be
the geometrical focal length of the transducer the total phase on the face of the
transducer, in the paraxial approximation, is then
ϕ ¼ k1 r 1 þ k 2 r 2
ð8:328Þ
ffi i k1 yα yα =2D10 id 21 =2K p1 id22 =2K p2 ik1 yα yα =2Rs
so that in the general case we again obtain Eq. (8.321), but where now
2
1 k1 Q Q2 D10
gð ϕÞ ¼ 1 cos 2 θ10 11 þ 21 cos 2 ϕ:
D10 D10 K p1 K p2 Rs
2 k1 Q11 Q12 Q21 Q22
cos θ10 þ cos ϕ sin ϕ ð8:329Þ
D10 D10 K p1 K p2
1 k1 Q212 Q222 D10
þ 1 þ sin 2 ϕ:
D10 D10 K p1 K p2 Rs
8.5.1.3 Caustics
When either Kp1 or Kp2 are zero (i.e. at a caustic) Eq. (8.315) predicts an infinite
pressure. Going back to Eq. (8.312), the nature of this singularity can be made more
explicit by writing out separately both the u1 and u2 integrations as
1 1
ð8:330Þ
8.5 Beam Propagation Through a Curved Interface 317
Equation (8.330) shows that if either Kp1 or Kp2 vanish, then the corresponding
integral becomes a delta function. For example, if only K p1 ¼ 0 then the integral on
u1 becomes
ð
þ1
When this result is integrated over the face of the transducer, the resulting response
is finite since although the delta function is singular, it is integrable. Similar results
hold either when only Kp2 vanishes or when both Kp1 and Kp2 are zero. Thus, the
diffraction correction does remain finite even at caustics. However, since
the predicted fields themselves are singular at caustics, which is clearly incorrect,
the finite diffraction correction values predicted at or near these locations cannot be
expected to be accurate and one is likely forced to go to other beam models of the
type discussed in Sect. 8.6.
The procedures for handling a fluid-solid interface follow closely those just
outlined for the fluid-fluid case. For example, we start with an integral representa-
tion for the components of the displacement vector, uδ, for the waves generated in
the solid by a point source in the fluid as (see Eq. (5.45)):
ð
uδ n ðx2 ; ωÞ ¼ Cklij nk ðxs Þ Gln ðxs ; x2 ; ωÞ∂uδ i ðxs Þ=∂xj
ð8:333Þ
S
uδ l ðxs Þ∂Gin ðxs ; x2 ; ωÞ=∂xj dSðxs Þ;
where it is assumed that the Sommerfeld radiation conditions are again satisfied and
the quantities appearing in the integrand are being evaluated in medium two
(Fig. 8.62). If we again view the wave incident on the interface in medium one as
approximately that of a plane wave (Eq. (8.261)) and use the Kirchhoff approxi-
mation to find the displacements on the interface in medium two, we obtain on a
small patch P (Fig. 8.63)
318 8 Ultrasonic Transducer Radiation
X T α;P ρ v0 ðωÞ exp ikp1 r
α
uδ i ðx; ωÞ ¼ 12 1
di expðikα2 eα xÞ
α¼P, S
2πρ 2 c α2 r 1
ð8:334Þ
X T α;P ρ v0 ðωÞ exp ikp1 r 1
α
¼ 12 1
di ;
α¼P, S
2πρ2 cα2 r1
where Tα;P
12 is the transmission coefficient for waves of type α due to incident
P-waves (based on a stress/pressure ratio), cα2 and kα2 are the wave speed and
wave number, respectively, of waves of type α in medium two, dαi are the compo-
nents of the polarization vector, and eα defines a unit vector in the transmitted wave
direction that satisfies Snell’s law (analogous to etr in the fluid-fluid problem). It
follows then that the derivatives of this displacement are given by
α;P
∂uδ i ðx; ωÞ X ikα2 ejα T 12 ρ1 v0 ðωÞ α exp ikp1 r 1
¼ di ð8:335Þ
∂xj α¼P, S
2πρ2 cα2 r1
However, at high frequencies, the far field forms of the fundamental solution and its
derivative for a wave of type γ ðγ ¼ P, SÞ in medium two are
f lnγ exp ikγ2 r 2
Glnγ ¼
4πρ2 c2γ2 r2
ð8:337Þ
∂Ginγ
¼ ikγ2 νjγ Ginγ ;
∂xj
where
The superscripts used here on να ðα ¼ P, SÞ are just to indicate which part of the
fundamental solution these terms are associated with and do not at this point denote
two different directions (see Figs. 8.62 and 8.63). However, when performing a
stationary phase analysis later on, these terms, να ðα ¼ P, SÞ, will be evaluated in
two different ray directions and so we are also anticipating that fact here.
8.5 Beam Propagation Through a Curved Interface 319
x2
where
1 γ h i
γ
I αn γ ¼ f Cklij d α
ik α2 n k e α
þ ik γ2 n j ν : ð8:340Þ
ρ2 c2γ2 ln i j k
If, as in the fluid-fluid case, we take this point source solution and integrate it over
the face of the transducer then we obtain the components of the displacement
vector, uγT , of type γ ðγ ¼ P, SÞ generated in the solid by the transducer as
8 9
X ð <ð α;P =
γ ρ v 0 ð ω Þ T 12 α γ exp ik γ2 r 2 exp ik p1 r 1
uTn ð x2 ; ω Þ ¼ 1
I dS ð x s Þ dSðx1 Þ
8π 2 ρ2 α¼P, S : cα2 n
r2 r1 ;
ST S
ðγ ¼ P, SÞ:
ð8:341Þ
Again, we can consider eliminating the interface integral in Eq. (8.341). To see this,
consider Eq. (8.339) which is very similar in form to that of the fluid-fluid interface
problem (Eq. (8.266)) so that the evaluation by the method of stationary phase
proceeds as shown before. At the stationary phase point(s), therefore, we have
X ρ v0 ðωÞT α;P
uδγ n ðx2 ; ωÞ ¼ 1
γ γ
12
α¼P, S
4π ρ 2 cα2 r 10 r 20
h
i
I αn γ exp iπ=4 sgn ϕp1 γ
þ sgn ϕp2 γ ð8:342Þ
γ γ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
exp ik r
p1 10 þ ik r
γ2 20
γ γ
kγ2 cos θ2γ ϕp1 ϕp2
and all the quantities are understood to be evaluated at the appropriate stationary
phase point(s), with
γ γ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
γ ϕ, 11 þ ϕ, 22 1 γ γ 2 γ 2
ϕp1 ¼ þ ϕ, 11 ϕ, 22 þ 4 ϕ, 12
2 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:343Þ
γ ϕ, γ þ ϕ, 22
γ
1 γ γ 2 γ 2
ϕp2 ¼ 11 ϕ, 11 ϕ, 22 þ 4 ϕ, 12
2 2
and
γ 1 cγ2 cos 2 θ1P 1 cγ2 cos θ1P =cp1 cos θ2γ
ϕ, 11 ¼ γ þ γ
r 20 cp1 cos 2 θ2γ r 10 RIγ cos 2 θ2γ
γ cγ2 cos θ1P =cp1 cos θ2γ
ϕ, 12 ¼ γ ð8:344Þ
RI0 cos θ2γ
γ 1 cγ2 1 cγ2 cos θ1P =cp1 cos θ2γ
ϕ, 22 ¼ γ þ γ ;
r 20 cp1 r 10 R0γ
8.5 Beam Propagation Through a Curved Interface 321
where the superscript γ ðγ ¼ P, SÞ is used on both the distances, angles, and the
radii of curvature terms in Eq. (8.344) to denote the dependency of those variables
on the mode (in the solid) of the particular ray being considered.
Equation (8.342) can be put into an even closer correspondence with the fluid-
fluid case result by examining the evaluation of the Iαγ
n coefficient at the stationary
points. If we expand out the elastic constant tensor this coefficient is explicitly
1 γ α h γ
i
I αn γ ¼ f d λδ kl δ ij þ μδ ki δlj þ μδ kj δli ik α2 n k e α
þ ik γ2 n j ν k : ð8:345Þ
ρ2 c2γ2 ln i j
Consider first the case where α ¼ γ ¼ P. Then d iP ¼ eiP , νiP ¼ eiP , f lnP ¼ elP enP and
IPP
n reduces considerably to just
2ikp2
I Pn P ¼ ðλ þ 2μÞ nl elP enP
ρ2 c2p2 ð8:346Þ
¼ 2ikp2 cos θ2γ dnP :
If instead we let α ¼ P, γ ¼ S then d iP ¼ eiP , νiS ¼ eiS , f lnS ¼ δln elS enS and
" P
!#
iω δln elS enS nl ni d iP elP eiS d iP nl nj ejP dl nj ejS d lP
I Pn S ¼ λ þμ þ þ þ ;
ρ2 c2s2 cp2 cp2 cs2 cp2 cs2
ð8:347Þ
for any constant A. Although Eq. (8.347) appears rather complex, by using Snell’s
law in the form
eiS emS nm ni eiP emP nm ni
¼ ð8:349Þ
cs2 cp2
together with Eq. (8.348), after considerable algebra one finds in fact
I Pn S ¼ 0: ð8:350Þ
ð8:353Þ
ð8:354Þ
where
π
γ γ
σγ ¼ sgn ϕp1 þ sgn ϕp2 : ð8:355Þ
4
Equation (8.354) can be used directly for curved interfaces where caustics do not
occur in the solid. For focusing interfaces where caustics are present, again one must
resort to different evaluation strategies, as mentioned previously for the fluid-fluid case.
Except for the leading coefficient and polarization term, Eq. (8.354) is identical in
form to the fluid-fluid case result (Eq. (8.290)) so that in the paraxial approximation
the displacement in the solid can be written in terms of propagation and transmis-
sion terms along a fixed ray and diffraction coefficients, C1γ ðγ ¼ P, SÞ, as
where Cγ1 can be obtained directly from the fluid-fluid diffraction correction expres-
sion by explicitly identifying the mode associated with the particular fixed ray
under consideration through replacements such as
8.6 The Numerical Evaluation of Beam Models 323
k1 ! kp1
k2 ! kγ2
γ γ
D10 , D20 ! D10 , D20 ð8:357Þ
γ γ
θ10 , θ20 ! θ10 , θ20
R1 , R2 ! R1γ , R2γ ;
with similar generalization of terms contained in Eq. (8.358). For example, for a
spherically focused probe and a general curved interface one has
( " # )
2 γ
γ 1 kp1 2 γ Q11 Q221 D10
g ðϕÞ ¼ γ 1 γ cos θ10 γ þ γ cos 2 ϕ
D10 D10 K p1 K p2 Rs
( " #)
2 kp1 γ Q11 Q12 Q21 Q22
γ γ cos θ 10 γ þ γ cos ϕ sin ϕ ð8:359Þ
D10 D10 K p1 K p2
( " # )
γ
1 kp1 Q212 Q222 D10
þ γ 1 γ γ þ γ sin 2 ϕ;
D10 D10 K p1 K p2 Rs
where the rotation matrix, Qαβ, also depends on the mode type γ in the solid and
terms such as (Kγp1 , Kγp2 ) again follow directly from the fluid-fluid case with the
appropriate replacements given in Eq. (8.357).
Evaluating the fields generated by a transducer is one of the more complex tasks
encountered in modeling ultrasonic systems. In the previous sections of this chapter
we have analyzed a variety of transducer problems using approximations such as
the method of stationary phase, the paraxial approximation, the Kirchhoff approx-
imation, etc. Those problems and their governing equations are summarized in
Table 8.2. In general, we can split the cases shown in Table 8.2 into two catego-
ries—(1) where the formulation leads to expressions in terms of single or multiple
surface integrals, and (2) where the formulation leads to a combination of explicit
324 8 Ultrasonic Transducer Radiation
Table 8.2 Two surface, one surface, and line integral ultrasonic beam models
Two surface One surface Line integral bd. Line integral
integrals integral diff.wave paraxial
(1) Planar transducer
(1a) Single medium N.a. Eq. (8.9) Eqs. (8.63, Eq. (8.56)
8.64a, 8.64b)
(1b) Planar interface- Special case Eq. (8.170) Eq. (8.185) f-f Eq. (8.178) f-f
normal of (1d) f-f
Eq. (8.188) Eq. (8.193) f-s Eq. (8.190a, b)
f-s f-s
(1c) Planar interface- Special case Eq. (8.200) – Eq. (8.218) f-f
obliq. of (1d) f-f
Eq. (8.223) Eq. (8.228) f-s
f-s
(1d) Curved interface Eq. (8.266) Eq. (8.290)a – Eq. (8.318) f-f
f-f f-f
Eq. (8.305) Eq. (8.317)a Eq. (8.356) f-s
f-s f-s
(2) Focused transducer
(2a) Single medium N.a. Eq. (8.88) Eqs. (8.134, Eq. (8.130)c
8.140)b
(2b) Planar interface- Special case Eq. (8.230) – Eq. (8.244)c f-f
normal of (2d) f-f
Eq. (8.246) Eq. (8.256a, b)c
f-s f-s
(2c) Planar interface- Special case Eq. (8.230) – Eq. (8.240)c f-f
obliq. of (2d) f-f
Eq. (8.246) Eq. (8.253)c f-s
f-s
(2d) Curved interface Eq. (8.266) Eq. (8.290)a – Eq. (8.321)c,a
f-f f-f f-f
Eq. (8.305) Eq. (8.317)a Eq. (8.358)c,a
f-s f-s f-s
f-f ¼ fluid-fluid interface
f-s ¼ fluid-solid interface
a
Expressions are singular at caustics, focal points
b
Principal value integral evaluation requires care
c
Results valid only near the cone of the direct wave
terms (direct waves) and line integrals (edge waves). For either category we are still
left with integrals to evaluate numerically. Figure 8.64 uses an immersion type of
inspection setup to graphically illustrate four types of numerical modeling strate-
gies that will be discussed in this book.
First, consider the models involving single surface integrals (case (a) in
Fig. 8.64) These models can be described as point source models since they
typically involve the integration of the responses produced by spherical waves
whose source points are distributed over the face of the transducer, generating a
8.6 The Numerical Evaluation of Beam Models 325
Fig. 8.64 Types of beam models: (a) single surface integral (point source) model, (b) line integral
(boundary diffraction wave) model, (c) multiple surface integral model (see edge elements), (d)
multi-Gaussian beam model (Gaussians are shown displaced only for display purposes)
the interface, is inadequate and one must resort to more complex evaluations. This
is one of the main disadvantages of point source models. However, for plane
interfaces and non-focusing curved interfaces, point source models are well-
behaved and can be used effectively without such complications.
The sound beam generated by piston transducers physically consists of the
superposition of a direct wave and an edge wave within the “main beam” of the
transducer and purely edge waves outside that main beam. Boundary diffraction
wave theory allows us to transform the surface integrations in point source models
into direct wave and line integral edge wave forms. For a planar circular transducer
radiating into a fluid this leads, as shown in Sect. 8.1.3.2 to a relatively simple and
effective beam model for this canonical problem. However, as we see in Table 8.2,
except for planar probes, there are limitations that make these line integral forms
not as well suited for numerical evaluation as other models. For planar probes the
numerical evaluation of these line integrals can easily be accomplished by breaking
the line integral into small elements, using low order approximations for the
integrands, and summing the contributions from each element. However, one
normally needs an order of magnitude fewer line segments than in the surface
integral models, and for the line integral models that use the paraxial approximation
no numerical ray tracing is needed since the models only involve a single fixed ray
that goes from the point of evaluation to the transducer face and is normal to that
face. In general, it is known that the paraxial approximation can break down in the
near field and at high refracted angles for transducers radiating at oblique incidence
through a planar interface [16], and the curvature of the surface must also not vary
rapidly over the “footprint” of the ultrasonic beam on a curved interface in order to
approximate that curvature as a constant along a fixed ray direction, as done in the
paraxial approximation. Fortunately, many of these cases where the paraxial
approximation fails are rather extreme so that paraxial models can be used as a
very efficient method for modeling many practical ultrasonic problems. Line
integral models still involve superposition of point sources over the transducer
edge, so just like point source models they can lead to singular responses for curved
interface problems.
If one wants to calculate transducer wave fields across curved interfaces without
using the stationary phase approximation to eliminate the integrations over the
interface one can to go to models based on multiple surface integrals. This auto-
matically raises the amount of numerical computations to be performed by an order
of magnitude or more since, although one does not have to do numerical ray tracing,
one needs to sum integrals over all combinations of elements on both surfaces as
shown in Fig. 8.64c. Direct, 2-D numerical evaluation of the surface integral
expressions is certainly possible, as shown, for example, by Stamnes [8], but
these schemes are difficult to implement in general, particularly when cases like
curved transducer surfaces or interfaces are involved. Thus, it is useful to consider
other approaches that can effectively handle such complex cases. In the next section
we will present one such procedure called the edge element method which can be
used for both single and multiple surface models.
8.6 The Numerical Evaluation of Beam Models 327
The single surface integral beam models appearing in Table 8.2 all are in terms of
integrals of the form
ð
I ¼ f expðiϕÞdS: ð8:360Þ
S
In this section, we will first outline in general terms the edge element approach for
dealing with the evaluation of such integrals and then illustrate this general
approach with several examples.
In the edge element approach, we begin by breaking the surface S up into small
planar elements, ΔSm ðm ¼ 1, . . . MÞ, over each of which f can be considered a
constant, to obtain
X
M ð
I¼ f 0m expðiϕÞdS: ð8:361Þ
m¼1
ΔSm
Furthermore, the phase term, ϕ, is expanded in each element to first order in the
form ϕ ffi ϕ0m þ i p0m x, where ϕ0m and p0m are constants over each element, so
that Eq. (8.361) reduces to
328 8 Ultrasonic Transducer Radiation
X
M ð
I¼ f 0m expðiϕ0m Þ expðip0m xÞdSðxÞ: ð8:362Þ
m¼1
ΔSm
Recall, however, that the plane wave term appearing in Eq. (8.362) can be put into a
form (see Eq. (8.84)) such that the surface integral can be reduced, via Stokes’
theorem, to a line integral around the edge, ΔCm, of the element to give
X
M ð
I¼ f 0m g0m expðiϕ0m Þ expðip0m xÞdCðxÞ; ð8:363Þ
m¼1
ΔCm
where g0m are known functions. If the edges of each element are then taken as a
series of N straight line segments (N ¼ 3 for triangular elements, etc.) it will be
shown that the line integrals in Eq. (8.363) can then be reduced even further to
analytical forms, i.e.
ð X
N
I m g0m expðip0m xÞdCðxÞ ¼ I mn ; ð8:364Þ
n¼1
ΔCm
where Imn are known functions so that Eq. (8.363) becomes, finally
M X
X N
I¼ f 0m I mn expðiϕ0m Þ: ð8:365Þ
m¼1 n¼1
x
r
y
m
r0
y⬘ m
e0
ym
0
ΔSm
The remaining surface integrals in Eq. (8.367) can be reduced to line integrals
around the edges of each element by writing the integrand in the form
h
0 i
exp ike0m y ¼ nm ∇ g y ; ð8:368Þ
where
330 8 Ultrasonic Transducer Radiation
m
e
ΔCm
0 0
nm e0m exp ik e0m y
g y ¼h 2 i ð8:369Þ
1 nm e0m ik
and nm is the unit normal to the element. An application of Stokes’ theorem (see
Eqs. (8.84) and (8.85) and Appendix C, Eq. (C.12)) to Eq. (8.367) gives
ρcv0 ðωÞ X
M
exp ikr 0m
pðx; ωÞ ¼ Im ; ð8:370Þ
2π m¼1 r 0m
where
ð h
i
nm e0m 0
Im ¼ h 2 i exp ik e m
0 y ds: ð8:371Þ
1 nm e0m ΔC m
If we let θm be the
angle measured
between the element normal, nm, and the unit
vector, em m m m
0 , and let ejj , e⊥ be unit vectors in the plane of the element, where ejj is
along the direction of the projection of em m
0 onto the element plane and e⊥ is
m
perpendicular to ejj (see Fig. 8.66), then
where Dm m
n extends from y0 to the centroid of the nth line segment, s is the distance
in the direction of the unit vector etmn along this segment from its own centroid, and
Lmn is the length of the segment. By changing the variable of integration to η ¼ s=
Lnm and using Eq. (8.374) for each of the N segments defining the boundary of the
element, we find
(
X
N
1 m h
i
Im ¼ e⊥ Lnm exp ik sin θm ejjm Dnm
sin θ m
n¼1 9
ð
þ1=2
h
i > = ð8:375Þ
exp ik sin θm ejjm Lnm η dη :
>
;
1=2
where
h
i
h
i sin k sin θm ejjm Lnm =2
nm e0m Lnm
I mn ¼ exp ik sin θm ejjm Dnm h
i ;
sin 2 θm k sin θm ejjm Lnm =2
ð8:378Þ
Because of the presence of the sin θm terms in the denominator of Imn, it appears that
there is a singularity as θm ! 0. However, if we add all the segment contributions to
the closed contour around a given element, we find to first order
N m
X
e Lm
⊥
Im ffi ik
n
e⊥m Lnm þ O ðθ m Þ
ejjm Dnm
n¼1 (
θ m
) ( ) ð8:379Þ
e⊥m X N X N
¼ m Lnm ik e⊥m Lnm ejjm Dnm þ Oðθm Þ:
θ n¼1 n¼1
The first term in brackets in the second line of Eq. (8.379) vanishes identically,
removing this apparent singularity, and the second term in brackets can be shown to
simply be the area, ΔSm, of this planar element, so that there is always a finite limit
for Im given by
To demonstrate the use of the edge element method, it was applied to a one-half
inch diameter, four inch focal length spherically focused transducer radiating into
water at 5 MHz [16]. The transducer surface was broken up into 2048 planar area
elements (32 radial divisions by 64 angular divisions). The boundary diffraction
wave single integral expression of Eq. (8.136) was used as the “exact” solution to
test the edge element approach. For the boundary diffraction wave model, the edge
integral was performed by dividing the transducer rim into 128 elements and
assuming the integrand was a constant in each element. In both cases, increasing
the number of elements further did not measurably change the predicted results.
Figure 8.68 shows an on-axis magnitude of pressure profile as computed by both
methods and Fig. 8.69 shows a cross axis profile computed at an axial distance
z ¼ 4.1 cm, which is well in the near field of this transducer. As can be seen from
those figures, there is no discernible difference between the two on-axis profiles and
only a slight mismatch of the two cross-axis profiles at the on-axis minimum. Of
course, the edge element method, because of the much larger number of elements
needed to cover the entire surface adequately, took approximately 70 times as long
as the boundary diffraction wave model to compute the same profiles.
Note that there is nothing in the derivation of Eq. (8.377) that restricts its use to
only the spherically focused transducers. In fact, any curved radiating surface
(cylindrical, bi-cylindrical, etc.) could be approximated by planar elements and
Eq. (8.377) applied to obtain the radiated wave field.
8.6 The Numerical Evaluation of Beam Models 333
6.0
Exact
Edge Elements
4.0
Ipl
2.0
0.0
2.0 6.0 10.0 14.0 18.0
z, cm
Fig. 8.68 The on-axis incident pressure profile for a 0.5 in. diameter, 4 in. focal length, 5 MHz
spherically focused transducer
2.40
Exact
Edge
Elements
1.60
Ipl
0.80
0.00
–1.00 –0.50 0.00 0.50 1.00
x, cm
Fig. 8.69 The cross-axis incident pressure profile at z ¼ 4.1 cm for a 0.5 in. diameter, 4 in. focal
length, 5 MHz spherically focused transducer
An examination of Eq. (8.378) for a single radiating edge segment shows a strong
similarity to the well-known Fraunhofer expressions for a rectangular transducer
[17]. In fact, as will be shown here, if four such segment contributions are added
together to form a rectangular element, one obtains simply the classical Fraunhofer
334 8 Ultrasonic Transducer Radiation
L4
b
expression for the far field radiation pattern of a rectangular piston transducer.
Since we will be working in this section with only one element, for clarity we will
henceforth remove the m superscripts from all the pertinent expressions. In this
case, the total pressure for one rectangular element is given by
ρcv0 expðikr 0 Þ X4
ðn e0 Þ Ln
pðx; ωÞ ¼
2π r0 sin 2 θ
n¼1 ð8:381Þ
sin k sin θ ejj Ln =2
exp ik sin θ ejj Dn :
k sin θ ejj Ln =2
Figure 8.70 shows the geometry of this rectangular element, where we have let
a and b be the lengths of the two sides. If we write the unit vector e0 in terms of
spherical coordinates (θ, ϕ) then in terms of its components along the (x, y, z) axes
we have
and the dot and cross products appearing in Eq. (8.381) become
ðn e0 Þ L1 a cos ϕ ðn e0 Þ L3 a cos ϕ
¼ , ¼
sin 2 θ sin θ sin 2 θ sin θ
ejj L1 ¼ a sin ϕ, ejj L3 ¼ a sin ϕ ð8:383Þ
b b
ejj D1 ¼ cos ϕ, ejj D3 ¼ cos ϕ:
2 2
8.6 The Numerical Evaluation of Beam Models 335
where
A ¼ kða=2Þ sin θ sin ϕ
ð8:385Þ
B ¼ kðb=2Þ sin θ cos ϕ:
so that the total pressure of Eq. (8.381) from all four sides is given finally by
which is just the Fraunhofer expression for a rectangular transducer [17]. Thus, the
edge element method can be considered to be the decomposition of the transducer
wave field into contributions from a set of transducer elements, each of which is
small enough so that the Fraunhofer approximation is valid and where each radiating
element contribution is further decomposed into a series of radiating straight edges.
The key part in the edge element approach is the ability to approximate the phase in
terms of a linear function of the coordinates of an arbitrary point in an element.
Therefore, if we consider the case of a planar (or focused) transducer radiating at
oblique incidence to a plane fluid-solid interface (see Fig. 8.52 and Eq. (8.188)), we
must expand the phase terms
for a point y in the mth element on the transducer surface close to a fixed point, ym
0,
in that element (see Fig. 8.71). From Fig. 8.71 the distances appearing in this phase
term can be written as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D1α ¼ ð xI y Þ ð x I y Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8:389Þ
α
D2 ¼ ðx xI Þ ðx xI Þ;
336 8 Ultrasonic Transducer Radiation
α αm
D1 e20
αm αm
nm Θ10 xT
αm
y θ10
αm
D10
y⬘ αm
e10
y0m
ΔSm
Placing Eq. (8.390) into Eq. (8.389) and expanding the phase term of Eq. (8.388) to
at most linear terms in y0 and xαT m , one finds
0
ϕα ffi kp1 Dα10m þ kα2 Dα20m þ kp1 eα10m xαT m kα2 eα20m xαT m kp1 eα10m y
0
ð8:391Þ
¼ kp1 Dα10m þ kα2 Dα20m kp1 eα10m y ;
where use has been made of the fact that the term in the parenthesis in Eq. (8.391)
vanishes by Snell’s law for the ray path from ym 0 to x. Using this approximation for
the phase in Eq. (8.188) and approximating all the remaining parts of the integrand
in terms of their constant values (for a given element) along a ray from ym 0 to x, the
displacement field in the solid can be written as
8
X ρ1 v0 ðωÞ X M ><T α;P Θα m dα m expik Dα m þ k Dα m
p1 10 α2 20
uðx; ωÞ ¼ 12 10 0
pffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffi
α¼P, S
2πρ 2 c α2 m¼1
>
: αm
Δx0 Δy0 αm
9 ð8:392Þ
ð h
i
0 =
0
exp ikp1 eα10m y dS y
;
ΔSm
8.6 The Numerical Evaluation of Beam Models 337
with
cα2 cos 2 Θα10m α m
Δαx0m ¼ Dα10m þ D
cp1 cos 2 Θα20m 20 ð8:393Þ
cα2 α m
Δαy0m ¼ Dα10m þ D :
cp1 20
Since the integral term in Eq. (8.392) is identical in form to the one appearing in
Eq. (8.367), Eq. (8.392) reduces in a similar fashion to
X i ρ v0 ðωÞ
uðx; ωÞ ¼ 1
α¼P, S
2π k p1 ρ cα2
8 2 9
XM X N ><T α;P Θα m dα m expik Dα m þ k Dα m > = ð8:394aÞ
p1 10 α2 20 α
12 10 0
p ffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffi I mn
>
m¼1 n¼1 : Δαx0m Δαy0m >
;
or, in terms of the velocity transmission coefficient and the velocity, v ¼ iω u,
8 9
>
< α m α m > =
X v0 ðωÞ X X T 12 Θ10 d0 exp i kp1 D10 þ kα2 D20
M N v α;P α m α m
α
vðx; ωÞ ¼ pffiffiffiffiffiffiffiffiffiq ffiffiffiffiffiffiffiffiffi I mn ;
α¼P, S
2π m¼1 n¼1 : > Δαx0m Δαy0m >
;
ð8:394bÞ
where
h
i
m m h
i αm αm
αm
n e10 Ln sin k p1 sin θ 10 e jj L n =2
m
α
I mn ¼ αm exp ikp1 sin θα10m eαjj m Dnm h
i :
sin θ10
2
kp1 sin θα10m eαjj m Lnm =2
ð8:395Þ
In [16], this edge element approach was also tested using a set of parameters to
simulate a typical angle beam shear wave transducer. Although an angle beam shear
wave transducer involves a compressional wave transducer in contact with a solid
wedge that is, in turn, in contact (through a thin couplant layer) with the underlying
solid being interrogated, since the compressional wave transducer generates pri-
marily P-waves in the wedge, it seems appropriate to neglect the shear strength of
the wedge entirely and replace the wedge with an equivalent fluid medium. In fact,
in Sect. 8.8, where a full elastic model of an angle beam probe is developed, it will
be shown that the fluid-solid model does indeed represent the angle beam probe
fields accurately and such a replacement is well justified.
For the specific studies conducted in [16], the compressional wave speed and
density of Lucite were used to model the “fluid” wedge and the underlying solid
was taken to be steel. The transducer was modeled as a 5 MHz, 1/2 in. diameter
planar transducer oriented so as to produce a refracted angle ranging from 30 to
338 8 Ultrasonic Transducer Radiation
0.60
Surface Integral
Edge Element
0.40
lul
0.20
0.00
0.00 1.00 2.00 3.00 4.00
z, [depth normal to the interface in cm]
Fig. 8.73 Central axis displacement profiles at a refracted angle of 75 into steel for a 0.5 in.
diameter transducer radiating at 5 MHz on a Lucite wedge
75 in the steel. The ultrasonic travel distance in the Lucite was taken as approx-
imately 1.8 cm to correspond to a typical path length found in commercial trans-
ducers. The magnitude of the displacement wave field was plotted in the steel along
a central axis of the transducer (Fig. 8.72a) and for a cross axis scan parallel to the
interface (Fig. 8.72b). The edge element method was compared to a direct numer-
ical integration of the surface integral of Eq. (8.223), using the approach of Stamnes
[8] where the radial integration was done with a Gauss-Legendre technique and the
angular integration was done with the impulse-response method. In the edge
element model, 2048 area elements were used (32 radial divisions by 64 angular
divisions) whose contour contributions were summed as shown in Eq. (8.395). The
2-D numerical integration scheme used 1024 area elements (4 radial divisions by
256 angular divisions). Upon finer division, neither method predicted any signifi-
cant changes in the wave fields so that the number of elements was deemed
adequate. Figures 8.73 and 8.74 show a comparison of the two methods for (1) a
central axis profile (Fig. 8.73) and (2) a cross-axis profile taken at a depth of 2 cm.
into the steel (Fig. 8.74), both taken with a high refracted angle of 75 . In both
cases, one sees an excellent agreement between the two methods. This agreement
remained excellent at all the other angles tested as well.
8.6 The Numerical Evaluation of Beam Models 339
0.60
Surface Integral
Edge Element
lul 0.40
0.20
0.00
3.0 5.8 8.5 11.3 14.0
x, cm
Fig. 8.74 Cross-axis displacement profiles at a refracted angle of 75 into steel for a 0.5 in.
diameter transducer radiating at 5 MHz on a Lucite wedge
As seen in Sect. 8.5, the combined use of the Kirchhoff approximation and the
method of stationary phase leads to a single integral formulation for curved
interface problems where the integrand is singular for certain cases. One way to
avoid that singular behavior is to only use the Kirchhoff approximation, which
leaves the model in terms of two surface integrals. To evaluate both those integrals
numerically it is also possible to use the edge element approach. To see this,
consider first the fluid-fluid curved interface problem.
From Eq. (8.266), we can write the response from a piston transducer (planar or
focused) through a curved interface in the form
8 9
ð <ð =
i ωρ1 v0 ðωÞ expði k1 r 1 Þ expði k2 r 2 Þ
pT ðx2 ; ωÞ ¼ T F dS ð x Þ dSðx1 Þ;
8π 2 : p r1 r2
s
;
ST SI
ð8:396Þ
where
and where N is the unit vector normal to the interface (outward normal from
medium two). If we break up both the transducer and interface surfaces into planar
elements and approximate, as before, the amplitude parts of the integrands in
Eq. (8.396) as constants over each of those elements then we obtain
mq
i ωρ1 v0 ðωÞ XM X
T p Θmq
Q
10 F
pT ðx2 ; ωÞ ¼
8π 2 r mq q
10 r 20
ð ð m¼1 q¼1
0
0 ð8:398Þ
exp½iðk1 r 1 þ k2 r 2 ÞdST y dSI z :
ΔSTm ΔSIq
where now we have defined eP, mq emq tr as the unit vector of a transmitted plane
q
P-wave satisfying Snell’s law and νq e20 is a unit vector pointing from a fixed
point in the qth element on the interface to point x (Fig. 8.75). Note that now the
superscript on the unit vector, νq, is not being used to distinguish a particular mode
direction arising out of a stationary phase analysis as in Eq. (8.338), but instead
merely indicates the unit vector associated with the qth interface element as shown
in Fig. 8.75.
If we expand the phase term to first order, following the same steps as used
previously to obtain Eq. (8.391), we find
mq q 0 0
k1 r 1 þ k2 r 2 ffi k1 r mq q mq
10 þ k 2 r 20 þ k1 e10 k2 e20 z k1 e10 y
0 0
ð8:400Þ
¼ k1 r mq q mq mq
10 þ k2 r 20 p0 z k1 e10 y ;
where
pmq q mq
0 ¼ k2 e20 k 1 e10 : ð8:401Þ
Note that because r1 and r2 do not lie on a stationary path from y0 to x2 satisfying
Snell’s law, the term with brackets in Eq. (8.359) does not vanish in general and
Eq. (8.357) becomes
mq
i ωρ1 v0 ðωÞ XM X
T p Θmq q
Q
10 F
pT ðx2 ; ωÞ ¼ 2 mq q exp i k1 r mq
10 þ k2 r 20
8π r 10 r 20
ð
m¼1 q¼1
0 ð
0
0 0
exp ipmq 0 z dS I z exp ik 1 e mq
10 y dS T y :
ΔSIq ΔSTm
ð8:402Þ
8.6 The Numerical Evaluation of Beam Models 341
x
r2
q
r 20
nm Nq q
v q ≡ e20
r1
mq z⬘
mq
θ10 Θ10 e
P;mq
≡ etr
mq
y⬘
mq mq
mq
r10 ΘγP
e10
m
ΔST mq
Pγ0 q
ΔSI
m th planar element
on the transducer q th planar element
surface on the curved
interface
Fig. 8.75 Planar element geometries for the transducer and interface surfaces when using edge
elements for a curved interface (fluid-fluid) problem
Since both of the remaining integrals are decoupled and in the same form as
considered previously, application of the edge element approach can reduce these
integrals to the sum of explicit terms over the edges of the elements on both the
transducer face and the interface given by
mq
i ρ1 c1 v0 ðωÞ XM X Q X S X R
T p Θmq 10 F
mq q T
pT ðx2 ; ωÞ ¼ mq q mq exp i k1 r 10 þ k 2 r 20 I mqs I mqr
I
8π 2 m¼1 q¼1 s¼1 r¼1 r 10 r 20 p0
ð8:403Þ
with
T;m h
i
nm emq
10 Ls mq mq
T
I mqs ¼ mq exp ik 1 sin θ e jj D T;m
hsin θ10
2 10 s
i
sin k1 sin θmq mq
10 ejj Ls
T;m
=2
h
i
k1 sin θmq emq LsT;m =2
10 jj
ð8:404Þ
Nq emqp0 Lr
I;q h
i
I
I mqr ¼ mq exp i pmq sin θmq emq DI;q
sin 2 θ 0 p0 pjj r
h p0
i
sin pmq mq mq
0 sin θ p0 epjj Lr
I;q
=2
h
i ;
p sin θ
mq mq mq
p0 epjj Lr =2
I;q
0
where Dt;m
n is the vector from a fixed point in the mth element to the centroid of the
nth edge segment of that element, where the element is on the transducer surface
ðt ¼ T Þ or interface ðt ¼ I Þ, respectively, and similarly Lt;m
n ðt ¼ T, I Þ is a vector
defining the nth directed edge segment of the mth element. The vectors emq mq
jj and epjj
342 8 Ultrasonic Transducer Radiation
mq mq
are unit vectors in the direction of the projection of emq mq
10 and ep0 ¼ p0 = p0 onto
the plane of the transducer and interface, respectively.
The same edge element approach described for the fluid-fluid problem is also
directly applicable to the fluid-solid case as well. From Eq. (8.341) the displace-
ments in the solid are given by
8
ð ð
γ ρ1 v0 ðωÞ X < T α;P 12 α γ
uT n ð x 2 ; ωÞ ¼ I
8π 2 ρ2 α¼P, S : cα2 n
T ð8:405Þ
S S
exp ikγ2 r 2 exp ikp1 r 1
dSðxs Þ dSðx1 Þ ðγ ¼ P, SÞ:
r2 r1
If we approximate the transducer surfaces and interface again as planar elements we have
α;P mq αγ ; mq
ρ1 v0 ðωÞ X X M X
Θ10 I n
Q
γ T 12
uTn ð x 2 ; ωÞ ¼
8π ρ2 α¼P, S m¼1 q¼1
2 r mq q
10 r 20
ð ð
0
0 ð8:406Þ
exp i kp1 r 1 þ kγ2 r 2 dST y dSI z ;
ΔSTm ΔSIq
where
1 γ;q h i
α;mq q α;mq
I nα γ;m q ¼ f C klij d ik α2 N e þ ik γ2 N q q
ν ð8:407Þ
ρ2 c2γ2 ln i k j j k
and
f P;q νlq νnq
ln ¼ ð8:408Þ
f ln ¼ δln νlq νnq :
S;q
Here, Nq are components of a unit vector normal to the interface (pointing into
material one), νq are components of a unit vector pointing from a fixed point in the
qth element to point x2 and eα;mq ðα ¼ P, SÞ are unit vectors in the second medium,
as determined by Snell’s law, for P- and S-waves, respectively, which is similar to
the notation used before for the fluid-fluid case (see Fig. 8.75). As in Eq. (8.400), we
have approximately
0 0
kp1 r 1 þ kγ2 r 2 ffi kp1 r mq q mq mq
10 þ k γ2 r 20 pγ0 z k p1 e10 y ; ð8:409Þ
8.6 The Numerical Evaluation of Beam Models 343
pmq mq q
γ0 ¼ k p1 e10 kγ2 e20 : ð8:410Þ
α;P mq αγ ; mq
ρ1 v0 ðωÞ X X M X
Θ10 I n q
Q
γ T 12
uTn ð x 2 ; ωÞ ¼ mq q exp i kp1 r mq 10 þ kγ2 r 20
8π ρ2 α¼P, S m¼1 q¼1
2 r 10 r 20
ð
0 ð
0
0 0
exp ipmq
γ0 z dS I z exp ik p1 e mq
10 y dS T y ;
ΔSIq ΔSTm
ð8:411Þ
α;P mq α γ ; m q
ρ1 cp1 v0 ðωÞ X X M X Q XS X
γ
R
T 12 Θ10 I n
uTn ð x 2 ; ωÞ ¼
8π ρ2 ω α¼P, S m¼1 q¼1 s¼1 r¼1 r r q pmq
2 mq ð8:412Þ
10 20 γ0
q T
exp i kp1 r mq
10 þ kγ2 r 20 I mqs I mqr
I;γ
with
T;m h
i
nm emq
10 Ls mq mq
T
I mqs ¼ mq exp ik p1 sin θ e jj D T;m
h θ10
2 10 s
sin
i
sin kp1 sin θmq mq
10 ejj Ls
T;m
=2
h
i
kp1 sin θmq emq LsT;m =2
10 jj
h
i ð8:413Þ
Nq emq γp LrI;q mq mq
I I;γ ¼ exp i p γ0 sin θ mq
γp e γpjj D I;q
mqr
sin 2 θγp
mq r
h
i
sin pmq mq mq
γ0 sin θγp eγpjj Lr
I;q
=2
h
i :
mq mq
pγ0 sin θmqγp eγpjj Lr
I;q
=2
All the quantities appearing in Eq. (8.413) are identical to those in the fluid-fluid
mq mq
problem except now for each element emqγpjj is the projection of eγp ¼ pγ0 =pγ0 onto
mq
This edge element approach will be very computationally intensive for either the
fluid-fluid or fluid-solid problems because of the requirement to approximate the
fields on two separate surfaces. Note that in these models involving two surface
integrals no numerical ray tracing necessary but the computational requirements are
aggravated by the fact that, unlike the fields on the transducer, which are confined to
its surface, the interface surface fields have no distinct limits so that a surface mesh
must be large enough to capture the “footprint” of the transducer beam on the
interface where these field amplitudes are non-negligible. However, the edge
element approach is adaptable to interfaces of any shape so that the method should
be able to act as general beam model “standard” to test other more approximate
formulations. It should be noted that the edge element method is a very low order
numerical scheme since the amplitude of the integrands are taken to be a constant
(zeroth order approximation) and the phase is only expanded to at most a linear
(first order) term. Thus, higher order numerical schemes of a similar nature could
make such methods more computationally efficient.
All the models considered previously in this Chapter were for immersion trans-
ducers and led to equations which were analogous to the Rayleigh-Sommerfeld
equation for a single fluid medium. In this section, we will consider the case of a
compressional wave transducer in contact with the plane, stress-free surface of an
elastic solid. A model of the radiated compressional and shear bulk waves will be
developed following the angular spectrum of plane wave integral approach of
Vezzetti [18]. Evaluation of those angular spectrum integrals by the method of
stationary phase will be shown to lead to modified Rayleigh-Sommerfeld types of
integrals containing both spherical waves terms and angular directivity functions.
This contact transducer model will also be used as part of the angle beam shear
wave transducer model considered in the next section.
Consider a compressional wave transducer in contact with the plane surface of
an infinite elastic solid as shown in Fig. 8.76a. Since the driving crystal of the
transducer is in contact with the solid through a thin fluid couplant layer, it is
reasonable to model this transducer as that of a uniform pressure, p0(ω), acting over
some active area, ST, of the interface (Fig. 8.76b). For this problem, we will write
the displacement vector, u(x, ω), in the solid (assuming harmonic waves of exp
ðiωtÞ time dependency) as an angular spectrum of plane compressional and shear
waves in the form
8.7 Contact Transducer 345
b
p0
ex ′
ey ′ x′
ST ez ′
y′ z′
0 X 3 ð þ1
þ1 ð h i
ðiÞ 0
u x ;ω ¼ βðiÞ UðiÞ exp ip1 x dpx0 dpy0 ; ð8:414Þ
i¼1
1 1
where are β(i) unknown coefficients, U(i) are unit polarization vectors,
0 0 0 0
x ¼ x ; y ; z and
ðiÞ ðiÞ
p1 ¼ px0 ex0 þ py0 ey0 þ p1z0 ez0
with
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ k2s1 p2x0 p2y0 ; ð8:415Þ
ð1Þ ð2Þ
p1z0 ¼ p1z0
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð3Þ
p1z0 ¼ k2p1 p2x0 p2y0
and where (kp1, ks1) are the wave numbers for compressional and shear waves in the
ð1Þ ð2Þ
solid. Thus, (β(1), U(1), p1 ) and (β(2), U(2), p1 ) are associated with shear waves and
ð3Þ
(β(3), U(3), p1 ) are associated with compressional waves, respectively. In order for
the equations of motion for the solid to be satisfied, the polarization vectors must be
chosen so that
Since we are interested here in modeling only the bulk waves generated by the
transducer in the solid (and not other waves such as surface waves or head
waves, both of which can propagate along the free surface) a convenient
choice for representing the polarization vectors that satisfies the conditions of
Eq. (8.416) is
346 8 Ultrasonic Transducer Radiation
1h i
ð1Þ
e 0 p1
Uð1Þ ¼ x ¼ pð10Þ ey0 þ py0 ez0
ð1Þ 1z
ex0 p1 g
1 h 2 i
ð2Þ
p Uð1Þ ð1Þ
Uð2Þ ¼ 1 ¼
g e x 0 p 0p 0 e 0 p 0p 0 e 0
x y y x 1z z ð8:417Þ
ð2Þ
p1 Uð1Þ ks1 g
1h i
ð3Þ
p ð3Þ
Uð3Þ ¼ 1 ¼ px0 ex0 þ py0 ey0 þ p1z0 ez0 ;
ð3Þ
p1 kp1
where qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g¼ k2s1 p2x0 : ð8:418Þ
This representation for the shear wave unit vectors is degenerate ðg ¼ 0Þ only when
the shear wave propagation is in the ex0 direction (along the interface).
To obtain the coefficient terms, βðiÞ ði ¼ 1, 2, 3Þ, the boundary conditions on the
0
surface must be satisfied. These conditions are, on the plane z ¼ 0
p0 ðωÞ on ST
τz0 z0 ¼
0 otherwise ð8:419Þ
τ x0 z 0 ¼ τy0 z0 ¼ 0:
When the appropriate derivatives of Eq. (8.414) are taken and placed into
Eq. (8.419), the boundary conditions become
3 ð ð n
þ1 þ1
X
ðiÞ ð iÞ ðiÞ ðiÞ
iρ1 c2p1 2c2s1 βðiÞ U x0 px0 þ Uy0 py0 þ U z0 p1z0
i¼1
1 1
o h
i
ðiÞ ðiÞ 0 0 p0 ðωÞ on ST
þ 2c2s1 βðiÞ Uz0 p1z0 exp i px0 x þ py0 y dpx0 dpy0 ¼
0 otherwise
X
3 ð
þ1 þ1ð n
o h
i
ðiÞ ðiÞ ðiÞ 0 0
iρ1 c2s1 βðiÞ U z0 px0 þ U x0 p1z0 exp i px0 x þ py0 y dpx0 dpy0 ¼ 0
i¼1
1 1
3 ð ð n
þ1 þ1
X
o h
i
ðiÞ ðiÞ ðiÞ 0 0
iρ1 c2s1 βðiÞ U z0 py0 þ U y0 p1z0 exp i px0 x þ py0 y dpx0 dpy0 ¼ 0:
i¼1
1 1
ð8:420Þ
All of the terms on the left hand side of Eq. (8.420) are in the form of
two-dimensional spatial Fourier transforms. Thus, if we take the 2-D inverse spatial
Fourier transform of these equations we obtain
8.7 Contact Transducer 347
3 n
X
o
ðiÞ ðiÞ ðiÞ ðiÞ ðiÞ ðiÞ
iρ1 c2p1 2c2s1 βðiÞ U x0 px0 þ U y0 py0 þ U z0 p1z0 þ 2c2s1 βðiÞ U z0 p1z0
i¼1 ð h
i 00 00
p0 ðωÞ 00 00
¼ 2
exp i px0 x þ py0 y dx dy
ð2π Þ
ST
ð8:421Þ
X
3
ðiÞ ðiÞ ðiÞ
ðiÞ
β U z0 px0 þ U x0 p1z0 ¼0
i¼1
X3
ðiÞ ðiÞ ðiÞ
βðiÞ U z0 py0 þ U y0 p1z0 ¼ 0;
i¼1
which is a set of three linear equations to solve for the three unknowns (β(1), β(2), β(3)).
The solution of these equations is
ð3Þ
2py0 k2s1 p1z0 Φ
βð1Þ ¼ "
2 #
2
ð 1Þ ð 3 Þ ð 1Þ
ρ1 c2s1 g 4p1z0 p1z0 p2⊥0 þ p1z0 p2⊥0
ð3Þ ð1Þ
ð2Þ
2px0 ks1 p1z0 p1z0 Φ
β ¼ "
2 #
ð 1Þ ð 3 Þ
ρ1 c2s1 g 4p1z0 p1z0 p2⊥0 þ
ð 1Þ 2
p1z0 p2⊥0 ð8:422Þ
2
ð1Þ
kp1 p1z0 p⊥0 Φ 2
βð3Þ ¼ "
2 #
ð1Þ ð3Þ 2 ð1Þ 2
ρ1 cs1 4p1z0 p1z0 p⊥0 þ
2 p1z0 p⊥02
with ð
i h
00 00
i 00 00
Φ¼ 2
p0 ðωÞexp i px0 x þ py0 y dx dy ð8:423Þ
ð2π Þ
ST
and
p2⊥0 ¼ p2x0 þ p2y0 : ð8:424Þ
Placing this solution into Eq. (8.414) and interchanging orders of integration gives
8 9
0 X 3 ð < þ1
ð þ1
ð h
i =
ip0 ðωÞ ðiÞ
β^ UðiÞ exp ip1 x x dpx0 dpy0 dS
ð iÞ 0 00
u x ;ω ¼
i¼1
4π 2 : ;
ST 1 1
ð8:425Þ
00 00 00 00 00
with dS ¼ dx dy , x ¼ x ex0 þ y ey0 , and
348 8 Ultrasonic Transducer Radiation
ðiÞ
β^ ¼ βðiÞ =Φ: ð8:426Þ
The term in the brackets in Eq. (8.425) can be evaluated directly by the method of
stationary phase following the steps outlined in Appendix E. It
follows
from those
results that for a general 2-D integral with amplitude term, A px0 ; py0 , and phase
term, ϕ px0 ; py0 , the stationary phase contribution of the integral is
ð þ1
þ1 ð
h
i
A px0 ; py0 exp iϕ px0 ; py0 dpx0 dpy0
1 1
ð8:427Þ
2πA pxs0 ; pys0 h
i
¼ pffiffiffiffiffiffiffi exp iϕ pxs0 ; pys0 þ iπσ=4 ;
jH j
(see Eq. (E.22)) and where the s superscript indicates that the quantities are being
evaluated at the stationary phase point. For example, for the compressional wave
term of Eq. (8.425) we have
ð3Þ
A px0 ; py0 ¼ β^ Uð3Þ
0 00 0 00
ð8:429Þ
ð3Þ 0
ϕ px0 ; py0 ¼ px0 x x þ py0 y y þ p1z0 z
and the stationary phase point pxs0 ; pys0 is given by
0 00
pxs0 ¼ kp1 x x =D
0 00 ð8:430Þ
pys0 ¼ kp1 y y =D;
where D is the distance between points x00 and x0 and the angle θ0 is related to the
0 0
geometry through z ¼ D cos θ (see Fig. 8.77). At this stationary point we find simply
ϕ pxs0 ; pys0 ¼ kp1 D
D4 ð8:431Þ
H¼ 2
kp1 z0
σ ¼ 2:
8.7 Contact Transducer 349
ST
z′, z′′
x′′
y′, y′′ θ′ S
d1
P
d1
x′
The evaluation of the amplitude term, A pxs0 ; pys0 , is considerably more involved.
After
some algebra,
it can be shown that at the stationary phase point the combination
of A px0 ; py0 and the Hessian, H, can be written in terms of an angularly dependent
s s
where
0 0
0 cos θ κ 21 κ21 =2 sin 2 θ
Kp θ ¼ 0
2G sin θ ð8:433Þ
0 00 0 00 0
x x y y z
d1 ¼
P
ex0 þ ey0 þ ez0
D D D
and the amplitude term again yields a directivity function and unit polarization
vector, i.e.
ð1Þ ð2Þ 0
A pxs0 ; pys0 β^ Uð1Þ þ β^ Uð2Þ K s θ d1S
pffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffi ¼ ; ð8:436Þ
jH j jH j ρ1 c2s1 D
where now
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 0 0
0 κ31 cos θ sin θ 1 κ21 sin 2 θ
Ks θ ¼ 0
2G κ1 sin θ ð8:437Þ
0 00 0 0 00 0
x x cos θ y y cos θ 0
d1S ¼ 0 ex 0 þ 0 ey0 þ sin θ ez0 :
D sin θ D sin θ
Using all these results, the displacement field in the solid given by Eq. (8.425)
becomes
ð
0 p ðω Þ 0 expðiks1 DÞ
u x ; ω ¼ 0 2 K s θ d1S dS
2πρ1 cs1 D
ST
ð
ð8:438Þ
p0 ðωÞ 0
P exp ik p1 D
þ K p θ d 1 dS:
2πρ1 c2p1 D
ST
1.20
Kp 0.80
0.40
0.00
0.0 25.0 50.0 75.0 100.0
angle theta_prime, degrees
Fig. 8.78 The compressional wave directivity function, Kp(θ0 ), for a contact compressional wave
transducer on Lucite
0.60
0.40
Ks
0.20
0.00
0.0 10.0 20.0 30.0 40.0
angle theta_prime, degrees
Fig. 8.79 The shear wave directivity function, Ks(θ0 ), for a contact compressional wave trans-
ducer on Lucite
Except for the constant coefficient, Eq. (8.439) is exactly the same form as for a
piston transducer radiating into a fluid. This shows that in the paraxial approxima-
tion, the compressional waves in the solid behave the same as for those in an
equivalent fluid medium, as previously noted by Schmerr and Sedov [19].
Recall, Fig. 6.34 showed the general configuration of an angle beam shear wave
transducer where a contact compressional wave transducer is used to radiate waves
into a solid wedge which itself is in smooth contact with a second solid via a thin
layer of fluid couplant. Normally the angle of the wedge is set up so that the first
critical angle is exceeded for compressional waves traveling normal to the contact
transducer, thus generating primarily transmitted shear waves into the second solid.
Notice, however, that Fig. 6.34 is somewhat misleading since, as seen in the last
section, the contact transducer produces both P- and S-waves in the wedge and these
waves do not travel in paths only normal to the contact transducer surface. Thus, in
general waves of both P- and S-type will also be generated, through mode conver-
sion, in the second medium. In this section we will set up a general model of an angle
beam shear wave transducer that properly accounts for all the directly transmitted
body waves present in both the wedge and the second medium. The model will
ignore any contributions in the second medium from multiply reflected waves in the
wedge (or other conversions at various edges) since those contributions will be
assumed to be made small by the design configuration of the wedge as illustrated in
Fig. 6.35a–c. Similarly, head waves and surface wave contributions to the transducer
response will be neglected.
To set up a model for this problem, we can start with the solution for the contact
transducer, Eq. (8.425), which can be used to represent the displacement field in the
wedge (medium one) in terms of an angular spectrum of plane waves. To transmit
these plane wave contributions through the plane interface with the second medium,
we need only to modify Eq. (8.425) in three ways. First, the incident phase term
must be changed to represent a refracted plane wave traveling in the second
medium, second the appropriate incident amplitude in the wedge must be modified
by a plane wave transmission coefficient, and third, the polarization of the incident
waves in the wedge must be changed to that of the waves traveling in the second
medium. After carrying out all three of these modifications to Eq. (8.425), we will
obtain a proper representation for the transmitted displacement field in the second
medium, which can be written in the form
8.8 Angle Beam Shear Wave Transducer 353
y ', y "
ex ' x ', x "
ey '
x" = (x ", y ", 0) ey O ex
x
O ez ' ez
y
z ', z " z
D⊥ θ'1α;P
d1P θ0
α;P
θ1
α;P
D1 xI = (x 'I, y 'I, z 'I)
= (x I, y I, z I)
α;P SV;P
θ2 d2 (α = SV )
ns α;P
D2 P ;P
d2 (α = P )
x = (x, y, z)
Fig. 8.81 Transmission of a P-wave in the first medium and the associated geometry variables
00 00 00
and coordinate systems. The x , y , z ¼ 0 coordinates are used to describe a general point on the
face of the transducer. The (x0 , y0 , z0 ) coordinates, which are tangential (x0 , y0 ) and normal (z0 ) to the
transducer, and the (x, y, z) coordinates (see insert) which are parallel (x, y) and normal (z) to the
interface, are both used to describe points in the first or second media. The origin of all these
coordinates is O
8 2 þ1 þ1
X <ip ðωÞ ð ð ð α;SV hn ð1Þ o i
uðx; ωÞ ¼ 0 4 v
T ^ Uð1Þ þ β^ ð2Þ U2 dSV dα;SV
β
α¼P, SV
: 4π 2 12 1 2
1 1
h
S T i
ðαÞ ðαÞ ðSV Þ
exp i p2 X p2z D⊥ þ p1z D⊥ dpx0 dpy0
ð þ1
þ1 ð h i
v α;P ^ ð3Þ ð3Þ
þ T 12 β U d1P d2α;P
h
1 1 ii o
ðαÞ ðαÞ ðPÞ
exp i p2 X p2z D⊥ þ p1z D⊥ dpx0 dpy0 dS ;
ð8:440Þ
00
where X ¼ x x . In terms of an (x, y, z) coordinate system, which is aligned with
the x-y plane parallel to the plane of the interface and origin O at the transducer (see
the inserts in Figs. 8.81 and 8.82), we have
ðαÞ ðαÞ
p2 ¼ p2x ex þ p2y ey þ p2z ez
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðαÞ
p2z ¼ k2α2 p22x p22y ð8:441Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðαÞ
p1z ¼ k2α1 p21x p21y ðα ¼ P, SV Þ
354 8 Ultrasonic Transducer Radiation
and the perpendicular distance from point x00 to the plane interface is D⊥ (Fig. 8.81).
Note that (following the notation we have used previously in this chapter) the
superscript α; β ðα ¼ P, SV Þ, ðβ ¼ P, SV Þ is being used in terms in Eq. (8.440) and
will be used in later equations to indicate the dependency of the underlying
expressions on mode, α, of the particular transmitted ray being considered and
the corresponding mode, β, in the first medium.
To explain the origin of all the terms in Eq. (8.440), consider first the contribu-
tion from the incident P-wave in the wedge (Fig. 8.81). The amplitude of that
^ ð3Þ ð3Þ
is just the component of β U along the direction of
contribution in medium one
ð3Þ ð3Þ
propagation, d1P ¼ p1 =p1 . To obtain the displacement amplitude for a wave of
type α ðα ¼ P, SV Þ in medium two, the displacement amplitude in medium one
must be multiplied by the transmission coefficients (based on a velocity or dis-
α;P
placement ratio), vT12 , for two solid media in smooth contact, which can be
obtained from the results of Chap. 6. There is no SH-wave contribution in medium
two, since as also shown in Chap. 6 there is no coupling between plane P-waves and
SH-waves at a plane interface. The polarizations of the transmitted wave in medium
ð3Þ
two, d2α : P , lies in the plane of incidence formed by p1 and the unit normal to the
interface, ns, and again can be determined as part of the plane wave transmission
problem solution. The phase term appearing in Eq. (8.440) also was obtained in
Chap. 6 where we showed in going from medium one to medium two
h
i h
i
ð3Þ ðαÞ ðαÞ ðPÞ
exp i p1 X ! exp i p2 X p2z D⊥ þ p1z D⊥ ð8:442Þ
(see Eq. (6.104)). Combining all these contributions then gives the P wave expres-
sion shown in Eq. (8.440).
8.8 Angle Beam Shear Wave Transducer 355
The shear wave term in Eq. (8.440) can be obtained in a similar fashion (see
Fig. 8.82), however, it is somewhat more complex because the total shear wave
ð1Þ ð2Þ
vector amplitude in medium one, given by β^ Uð1Þ þ β^ Uð2Þ has components that
ð1Þ
lie both in and out of the plane of incidence formed by p1 and ns. Since the
component of this amplitude normal to the plane of incidence is an SH-component
that is not transmitted through the interface, we first must eliminate that component
and compute only the part of the incident amplitude in the plane of incidence. To
obtain this in-plane part, we first note that a unit polarization vector normal to the
plane of incidence, dSH1 , can be obtain through the cross product
ð1Þ
n p1
s
1 ¼
dSH ð8:443Þ
ð1Þ
ns p1
and, consequently, a unit polarization vector in the plane of incidence can be found
from
ð1Þ ð1Þ
dSV
1 ¼ d SH
1 p1 = p1 ; ð8:444Þ
as shown in Fig. 8.82. Thus, the component of the incident SV-wave in medium one
h ð1Þ ð2Þ
i
can be obtained from β^ Uð1Þ þ β^ Uð2Þ dSV , as given in Eq. (8.440). Multi-
1
plying this amplitude by the transmission coefficients for two solids in smooth
contact, vTα;SV
12 , gives the amplitude of the transmitted P- and SV-waves, and again
there are no transmitted SH-wave components. The polarization vectors in medium
two, dα;SV
2 , can again be found as part of solving the plane wave transmission
problem. Finally, the phase term for the transmitted P- and SV-waves arises from
the following correspondence when going from medium one to two
h
i h
i
ð1Þ ðαÞ ðαÞ ðSV Þ
exp i p1 X ! exp i p2 X p2z D⊥ þ p1z D⊥ : ð8:445Þ
where dP1 and dS1 are the same polarization vectors in medium one given by
Eqs. (8.433) and (8.437) for the contact transducer but these quantities now also
depend on the particular α and β modes being considered in medium one and two,
respectively, so this dependency will be indicated by the appropriate superscript
term α; β (this superscript is not shown explicitly in Figs. 8.81 and 8.82).
0 α:SV
" 0 00 α;SV 0 α;SV
#
α;SV cos 2θ 1 0 α;SV xI x cos θ 1 sin θ0
d1S dSV ¼ 0 α;SV
sin θ 1
2
cos θ0 þ
D1α;SV
1
sin θ1α;SV sin θ 1
ð8:447Þ
and
with the angles and distances are as shown in Figs. 8.81 and 8.82. Similarly, the
phase terms become
h
i
ðαÞ ðαÞ ðPÞ
exp i p2 X p2z D⊥ þ p1z D⊥ ¼ exp i kp1 D1α;P þ kα2 D2α;P
h
i h
i ð8:449Þ
ðαÞ ðαÞ ðSV Þ
exp i p2 X p2z D⊥ þ p1z D⊥ ¼ exp i ks1 D1α;SV þ kα2 D2α;SV
ðx xI ÞP;SV ðy yI ÞP;SV
dP;SV
2 ¼ ex þ ey þ cos θP;SV
2 ez
D2P;SV DP;SV
2
ðx xI ÞSV;SV cos θ2SV;SV ðy yI ÞSV;SV cos θ2SV;SV
d2SV;SV ¼ ex þ ey sin θ2SV;SV ez
D2SV;SV sin θ2SV;SV D2SV;SV sin θ2SV;SV
ðx xI ÞSV;P cos θ2SV;P ðy yI ÞSV;P cos θ2SV;P
d2SV;P ¼ ex þ ey sin θ2SV;P ez
D2SV;P sin θ2SV;P D2SV;P sin θ2SV;P
ðx xI ÞP;P ðy yI Þ P;P
dP;P
2 ¼ ex þ ey þ cos θP;P
2 ez :
D2P;P DP;P
2
ð8:450Þ
ð
p 0 ð ωÞ X 0 α;SV α;SV
uðx; ωÞ ¼ K s θ 1 d1S dSV
1 dS
2πρ1 c2s1 α¼P, SV
STh
i
d2α;SV v T 12
α;SV
exp i ks1 D1α;SV þ kα2 D2α;SV
qffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffi
Δxα;SV Δyα;SV
X ð K p θ0 α;P dα;P v T α;P expikp1 Dα;P þ kα2 Dα;P
p ð ωÞ 1 2 12 1 2
þ 0 2 qffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffi dS:
2πρ1 cp1 α¼P, SV Δα;P Δα;P
ST x y
ð8:451Þ
There are a total of four wave contributions present in Eq. (8.451). To examine the
relative contributions of each of these terms, the edge element method was used to
compute the wave fields in the second medium. Since the details follow closely
those for the fluid-solid interface discussed in Sect. 8.6.1 we will only outline
briefly the steps here. As in the fluid-solid case, the amplitude terms in the
integrands of Eq. (8.451) for a particular element can be taken to be a constant
and the phase terms can be approximated to first order as
m
kβ1 D1α;β þ kα2 D2α;β ffi kβ1 D10
α;β α;β
þ kα2 D20 α;β
kp1 e01 y; ð8:452Þ
where y is an arbitrary point in the element of interest and the Dα;β 10 and D20
α;β
m
distances are from a fixed point, y0 , in the element (usually taken as the centroid)
to the field point x along a stationary ray path satisfying Snell’s law for a wave of
type β in medium one and type α in medium two. The vector (eα;β m
01 ) is a unit vector
taken along this “fixed” ray path in medium one (see Fig. 8.83). Once these
approximations are made, the surface integration over each element can be reduced,
as before, to one dimensional line integrals over the element edges. These edge
integrals, in turn, can be evaluated explicitly when the edges are straight. The
resulting expression for the displacement in medium two is then
8 h
0 α;P i 9
>
> v α;P α;P α;P >
>
p0 X < X
M X N T 12 m
K p θ 1 d 2m exp iϕ m
=
α;P
uðx; ωÞ ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi r m
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I
2 >
2πρ1 cp1 α¼P, SV >m¼1 n¼1 α;P α;P
mn
>
>
: Δx0 Δy0 ;
m
8
h
i
m 9
>
> v α;SV 0 α;SV S SV α;SV α;SV α;SV >
>
p0 X < X
M X N T 12 K s θ 1 d 1 d 1 m d 2m exp iϕm
=
α;SV
þ m
r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m
r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I ;
2πρ1 cs1 α¼P, SV >
2
> α;SV α;SV
mn
>
>
:m¼1 n¼1 Δx0 Δy0 ;
m m
ð8:453Þ
358 8 Ultrasonic Transducer Radiation
nm ym
0 Lm
α;β n
(e01 )m mth element
α;β nth edge
α;β
D1
(D 10 )m element
interface
α;β
D2
α;β
(D 20 )m
where
m
m
α;β α;β α;β
ϕm ¼ kβ1 D10 þ kα2 D20 ð8:454Þ
and
m
α;β
nm e01 Lnm h
0 α;β m
m i
α;β
Im ¼
0 α;β m exp ikβ1 sin θ 1 ejjα;β Dnm
sin 2 θ 1
h
0 α;β m
m i ð8:455Þ
sin kβ1 sin θ 1 ejjα;β Lnm =2
h
α;β m
m i :
ejjα;β Lnm =2
0
kβ1 sin θ 1
For the mth element, nm is the normal to the transducer, (θ0 α;β m
1 ) is the angle between
m
α;β m
nm and (e01 ) , and eα;β
jj
α;β m
is a unit vector along the projection of (e01 ) on the
transducer surface, i.e.
m
0 α;β m
m
0 α;β m
α;β
e01 ¼ sin θ 1 ejjα;β þ cos θ 1 nm ; ð8:456Þ
0.000
0.0 4.0 8.0 12.0 16.0
z, cm
0.000
0.0 3.0 6.0 9.0 12.0
z, cm
the fluid-solid model of Sect. 8.6.1. As in Sect. 8.6.1 the wedge material was taken
as Lucite and the underlying material steel, and the same number of edge elements
were used for the present model as for the fluid-solid model of Sect. 8.6.1. Fig-
ure 8.84 shows all the various wave contributions to an on-axis plot when the wedge
angle is such that a compressional wave traveling normal to the contact transducer
will produce a refracted shear wave angle of 30 in the steel. As can be seen from
that figure the major response, as expected, comes from the P ! SV contribution.
All the other wave contributions are totally negligible except for the P ! P wave in
the very near field. For a refracted shear wave angle of 45 (see Fig. 8.85) or higher,
even the P ! P becomes negligible so that only the dominant P ! SV wave needs
to be considered. Since these results show that the shear waves in the wedge have
little effect on the wave fields of typical angle beam shear wave transducer
configurations, it seems likely that one can simply neglect the shear waves entirely
in the wedge and use a fluid-solid model instead, as discussed in Sect. 8.6.1.
Figures 8.86 and 8.87 show that this replacement is indeed valid. At a refracted
360 8 Ultrasonic Transducer Radiation
0.000
0.0 2.5 5.0 7.5 10.0
z, cm
0.0000
0.0 2.5 5.0 7.5 10.0
z, cm
where (x1, x2) are coordinates on the face of the transducer, which is located on the
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
plane x3 ¼ 0. The distance ρ ¼ x21 þ x22 . Wen and Breazeale [20] used a
non-linear optimization approach to fit Eq. (8.457) to the sum of a set of circularly
symmetric Gaussians of the form
X
n
vinc ðx1 ; x2 ; 0; ωÞ ¼ v0 ðωÞ Ar dp exp Br ρ2 =a2 ; ð8:458Þ
r¼1
where the polarization dp ¼ e3 ¼ ð0; 0; 1Þ, i.e. it is in the x3-direction, and (Ar, Br)
are complex-valued fitting coefficients. In [20] Wen and Breazeale obtained a very
good model with only n ¼ 10 coefficients, while in [21] they used a modified
procedure and n ¼ 15 coefficients. Tests with both sets of coefficients showed that
the transducer wave field they model closely agrees with that of a piston probe
except in a region near the face of the transducer, with the fifteen coefficients giving
the field properly somewhat closer to the transducer [22, 23]. We will not discuss
the details of the optimization techniques used since once a good set of coefficients
have been found, they can be placed in a look-up table and re-used for the any
specific case under consideration. The 15 coefficients obtained by Wen and
Brezeale, for example, have been placed in the MATLAB® function gauss_15,
which has the calling sequence
and simply returns the coefficients. These coefficients give us the starting values
for each of the Gaussian beams in the fluid. Recall from Chap. 6, Eq. (6.195) that
the pressure of a Gaussian beam in a fluid is of the form (with β ¼ P)
362 8 Ultrasonic Transducer Radiation
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i i
q10 q20
pinc ¼ P0
q i ðx3 Þq2i ðx3 Þ
1 2 ð8:459Þ
ik1 x1 x22
exp ik1 x3 þ þ ;
2 q1i ðx3 Þ q2i ðx3 Þ
where here we have shifted the x-coordinates so the origin is at the face of the
transducer and omitted the expðiωtÞ term which is common to all Gaussian beam
expressions, including those at the transducer face. Similarly, the velocity field is
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i i
P0 p q10 q20
vinc ¼ d
ρ1 cp1 q1i ðx3 Þq2i ðx3 Þ ð8:460Þ
2
ik1 x1 x22
exp ik1 x3 þ þ
2 q1i ðx3 Þ q2i ðx3 Þ
so that at the transducer face we must have for each Gaussian (r ¼ 1, 2, . . . , 15)
P0 =ρ1 cp1 r
¼ A r v0 ð8:461aÞ
and
1 1 iB
i
¼ i ¼ r ; ð8:461bÞ
q10 r
q 20 r DR
whereDR ¼ ka2 =2is called the Rayleigh distance. A multi-Gaussian beam model for
the pressure wave field of a circular piston transducer, therefore, can be written as
X15
Ar
pðx1 ; x2 ; x3 ; ωÞ ¼ ρ1 cp1 v0 ðωÞ
1 þ iBr x3 =DR
r¼1 ð8:462Þ
k1 ðBr =DR Þ 2
exp ik1 x3 x þ x2 :
2
2ð1 þ iBr x3 =DR Þ 1
The velocity wave field of this transducer for a beam of type α ðα ¼ P, SV Þ in the
solid, is therefore given by (see Eq. (6.245))
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tr;α tr;α ffi
X
15
Ar α;β q q
vtrans;α ¼ v 0 ð ωÞ v
T dα 10
r 20 r
iB D=DR þ 1 12 tr q1tr;α ðy3 Þ r q2tr;α ðy3 Þ r
r¼1 r
" !# ð8:463Þ
β α ik2α y21 y22
exp ik1 D þ ik2 y3 þ tr;α þ tr;α ;
2 q1 ðy3 Þ r q2 ð y 3 Þ r
where D is the distance from the transducer face to the fluid-solid interface along
the center line of the transducer and β ¼ P since the first medium is a fluid. Note that
8.9 Multi-Gaussian Beam Models 363
although we only needed Gaussian beams of circular cross section in the fluid to
represent the fluid wave field of the circular transducer, elliptical cross section
Gaussians of the form given in Eq. (4.63) are produced in the solid. For elliptical or
rectangular transducers, elliptical cross section Gaussians would also be needed in
the fluid [22]. Equation (8.463) can be put into a more compact form by letting
β
β
c2α cos θinc =c1β cos θtrα
α 1 c2α cos 2 θinc iBr =DR 1
m1 r ¼ ¼ þ
tr;α
q10 r c1β cos 2 θtrα iBr D=DR þ 1 cos 2 θtrα R1
α 1 c2α iBr =DR α β β α 1
m2 r ¼ ¼ þ c cos θ =c cos θ
c1β iBr D=DR þ 1
tr;α 2 inc 1 tr
q20 r
R2
ð8:464Þ
and
tr;α
q1 ðy3 Þ r ¼ m1α r y3 þ 1 = m1α r
tr;α ð8:465Þ
q2 ðy3 Þ r ¼ m2α r y3 þ 1 = m2α r
so that
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffisffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
15
Ar 1 1
v α;β α
v trans;α
¼ v 0 ð ωÞ T 12 dtr α α
r¼1
iB r D=D R þ 1 m 1 r 3 y þ 1 m 2 r y3 þ 1
" !# ð8:466Þ
β α ik2α m1α r y21 m2α r y22
exp ik1 D þ ik2 y3 þ α þ α :
2 m1 r y3 þ 1 m2 r y3 þ 1
By changing the transmission coefficient in Eq. (8.466) to be the coefficient for two
solids in smooth contact and by letting v0 ðωÞ ¼ p0 ðωÞ=ρ1 cp1 , where p0(ω) is the
constant pressure on the face of a contact circular P-wave transducer acting on a
wedge, Eq. (8.466) can also be used for modeling angle beam shear wave trans-
ducers by letting α ¼ SV and orienting the transducer so that the first critical angle is
exceeded.
For paraxial models like a multi-Gaussian beam model it is easy to modify the
planar transducer result to include the effects of focusing. For example, consider a
spherical wave propagating inwards to a point O, where O is at a radius R0 from
the plane of a transducer (see Fig. 8.88). This inward wave can be written as
exp½ikðr s R0 Þ where we have added the constant phase term exp(ikR0) so that
the spherical wave is at the coordinate origin, O, at time t ¼ 0. On the plane z ¼ 0,
this phase can be written in terms of ρ, the radius. In the paraxial approximation
ρ=R0 1, which assumes that the spherical wave is propagating essentially along
the z-axis, so that in this case the spherical wave becomes
364 8 Ultrasonic Transducer Radiation
h
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
exp½ikðr s R0 Þ ¼ exp ik ρ2 þ R20 R0
ð8:467Þ
ρ2
ffi exp ik :
2R0
Now, let us assume that a piston velocity field which includes this phase term acts
on the face of a planar circular piston transducer of radius, a. Then the Rayleigh-
Sommerfeld model gives
ð
iωρ1 v0 ðωÞ expðikr Þ
pðx; ωÞ ¼ exp ikρ2 =2R0 dS : ð8:468Þ
2π r
ST
ða h
iωρ1 v0 ðωÞexpðikzÞ q i
pðz; ωÞ ¼ exp ikρ2 0 ð2πρdρÞ
2πz 2z
0 ð8:469Þ
ρ cp1 v0 ðωÞexpðikzÞ
¼ 1 1 exp ika2 q0 =2z :
q0
X
15
vz ¼ v0 ðωÞ Ar exp Br ρ2 =a2 exp ikρ2 =2R0
r¼1
ð8:470Þ
X
15 0
¼ v0 ðωÞ Ar exp Br ρ2 =a2 ;
r¼1
where
0
Br ¼ Br þ ika2 =2R0 : ð8:471Þ
Thus, we merely need to redefine the Wen and Breazeale Br coefficients with
Eq. (8.471) to produce the coefficients needed to model a spherically focused piston
transducer of radius a.
A multi-Gaussian beam model has been implemented in MATLAB® using both
a script and a function. The script, called parameters2, contains all the input
parameters needed to define an immersion or angle beam shear wave inspection
and the locations at which the transmitted velocity wave field is to be calculated.
Those locations can be either a scalar (evaluation at a single point), a vector
(evaluation along a line) or a 2 2 matrix (evaluation in a plane, i.e. a cross-
sectional image). When the locations are in a 2 2 matrix, the frequency, f, must be
a scalar but for scalar or vector-valued locations the frequency itself can also be a
vector. When the wave field values are calculated at multiple frequencies, use of an
inverse FFT makes the simulation of pulses simple. The script parameters2 also
places all of the input parameters in a MATLAB structure named setup. This
structure is passed as an input argument to the MG_beam2 function, which returns
the normalized
trans;α amplitude
of the transmitted velocity wave field of the transducer,
α
V¼ v dtr =v0 ðωÞ, at the specified locations and/or frequencies. The calling
sequence for this function is simply
>> V ¼ MG_beam2(setup);
If one wishes to recall the parameters in the structure setup, one can do this with
help of the function display_struct whose calling sequence is
>> display_struct(setup)
There are a total of 19 parameters contained in the structure setup. These include
the frequency, f, in MHz, the types of waves to be considered in materials one and
two (specified by strings ‘p’ or ‘s’ for P-waves and SV-waves), the type of interface
(specified by strings ‘fs’ or ‘ss’ for fluid-solid or two solids in smooth contact), the
366 8 Ultrasonic Transducer Radiation
diameter, d, and focal length, fl, of the circular transducer (both in mm), where
fl ¼ inf for a planar transducer. The geometry parameters include the axial
distances, D, and y3, along a “central” ray from the transducer in the first and
second media to the point at which the beam velocity is to be evaluated, and the
distances (y1, y2) in the plane of incidence and normal to the plane of incidence
from the central ray to that evaluation point. All of these distances are in measured
in mm. Other geometry parameters are the incident angle of the transducer with
respect to the normal to the interface (in degrees), and the principal curvatures (R1,
R2) of the interface (in mm) in the plane of incidence and normal to the plane of
incidence. For a plane interface, R1 ¼ inf, R2 ¼ inf. Finally, there are six material
parameters specified starting with the densities of the two media (d1, d2), in
arbitrary units, and the P-wave and S-wave speeds in the two media (cp1, cs1,
cp2, cs2), all measured in m/s. No attenuation parameters for the media are
contained in setup but they can be easily added and included in the beam model.
When the script parameters2 is evaluated and the default structure setup generated
is displayed we find:
>> parameters2
>> display_struct(setup)
setup.f ¼ 10
setup.type1 ¼ ’p’
setup.type2 ¼ ’p’
setup.int ¼ ’fs’
setup.trans ¼
d: 12.7000
fl: Inf
setup.geom ¼
D: 10
y3: [1x500 double]
y1: 0
y2: 0
i_ang: 0
R1: Inf
R2: Inf
setup.matl ¼
d1: 1
d2: 7.9
cp1: 1480
cs1: 0
cp2: 5900
cs2: 3200
which shows that the default setup is for a 10 MHz, 12.7 mm (0.5 in.) diameter
planar transducer generating and radiating a P-wave through a plane fluid-solid
8.9 Multi-Gaussian Beam Models 367
Fig. 8.89 The magnitude of the on-axis normalized velocity calculated with a multi-Gaussian
beam model for a 10 MHz, 12.7 mm diameter transducer radiating through a water-steel interface
as a function of the depth in the steel
interface at normal incidence, where the beam field is to evaluated along the central
ray of the transducer. The distance, D, is shown explicitly. We can see the explicit
minimum and maximum values of the 500 values contained in the vector y3
through the commands:
>> min(setup.geom.y3)
ans ¼ 10
>> max(setup.geom.y3)
ans ¼ 200
The densities and wave speeds of both materials are specified to be that of water/
steel, so this default structure is for the evaluation of the velocity of a transducer
radiating across a water-steel interface along the axis from y3 ¼ 10 mm to 200 mm.
The magnitude of this velocity is shown in Fig. 8.89. It would be useful to have a
more exact result for this case to serve as a reference. This is possible, since we
found in Sect. 8.3.2.3 that we could obtain an explicit boundary diffraction wave
model for the on-axis wave field of a transducer at normal incidence to a fluid-solid
interface (Eq. (8.197)) which we have rewritten here for the velocity of the
transmitted P-waves in terms of the velocity-based transmission coefficients:
vðz1 ; z2 ; ωÞ ¼ v0 ðωÞez v T P;P 12 0 exp i k p1 z1 þ kp2 z2 9
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
D1 þ cp2 D2P =cp1 cos θ2P >
=
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
v P;P
T θ P
exp i k D
p1 1
P
þ k D
p2 2
P
:
D1 þ cp2 cos 2 θ1P D2P =cp1 cos 2 θ2P
12 1
>
;
ð8:472Þ
368 8 Ultrasonic Transducer Radiation
where (z1, z2) are the distances in the first and section media (in mm), f is the
frequency (in MHz), a is the radius (in mm), (d1, d2) are the densities of the two
media, (cp1, cp2) are the compressional wave speeds of the media (in m/s) and cs2
is the shear wave speed, also in m/s. The function returns the normalized on-axis
velocity in the solid, v, as well as the paraxial approximation to Eq. (8.472), vp,
which is
P;P ∘
vðz1 ; z2 ; ωÞ ¼ v0 ðωÞez v T 12 ð0 Þexp i kp1 z1 þ kp2 z2 1 exp ikp1 a2 =2~z ;
ð8:473Þ
The edge wave term in Eq. (8.472) can be obtained by considering a ray from the
transducer edge to the evaluation point in the second medium (see Fig. 8.48) and
writing Snell’s law for this ray in terms of (a, z1, z2) and the radial distance, x, from
the transducer central axis, at which the ray intersects the interface. The solution of
Snell’s law for this radial distance is found numerically, using the MATLAB
function fzero. As mentioned previously, Ferrari’s method [15] can also be used
in place of fzero for faster evaluations, but in this case the difference is not great.
Knowing this radial distance we can determine all of the edge wave geometry terms
in Eq. (8.472). Figure 8.90 shows the evaluation of onaxis_interface for the same
parameters as considered previously with MG_beam2. Comparing Fig. 8.89 with
Fig. 8.90, one can see that MG_beam2 and the paraxial model of Eq. (8.473) both
agree very well with the onaxis_interface results.
To see the effects that an interface curvature has on a sound beam, consider now
the case of a 5 MHz, 12.7 mm diameter transducer radiating at normal incidence
through a water-steel interface. First, we will look at the on-axis wave field for the
planar interface. The changed input parameters from the two cases just discussed
are:
Fig. 8.90 The magnitude of the on-axis normalized velocity of a 10 MHz, 12.7 mm diameter
transducer radiating through a water-steel interface as a function of the depth in the steel. Results
calculated with a boundary diffraction wave model (solid line) and with a boundary diffraction
wave model and the paraxial approximation (dashed line)
Figure 8.91a shows the magnitude of the on-axis velocity for this case. However,
if we change the interface to have a cylindrical curvature where R1 ¼ 76 mm,
(R2 ¼ inf) the on-axis velocity is as shown in Fig. 8.91b, which clearly shows the
defocusing nature of the curved interface. If R1 ¼ 76 mm instead, Fig. 8.91c
shows that this is now a focusing cylindrical interface. These effects can be seen
more clearly for the entire beam by first going back to the planar interface case and
setting up a matrix of evaluation points to form an image:
>> setup.geom.R1 ¼ inf;
>> y3t ¼ linspace(0, 100,200);
>> y1t ¼ linspace(-20, 20, 100);
>> [y3, y1] ¼ meshgrid(y3t, y1t);
>> setup.geom.y3 ¼ y3;
>> setup.geom.y1 ¼ y1;
>> v ¼ MG_beam2(setup);
>> imagesc(y3t, y1t, abs(v))
>> colormap(jet)
Figure 8.92 shows the well collimated beam produced in this planar interface
case. If now the interface curvature is changed to R1 ¼ 76 mm, we find instead the
image shown in Fig. 8.93, which clearly shows an increased spreading of the beam
370 8 Ultrasonic Transducer Radiation
Fig. 8.91 The magnitude of the on-axis normalized velocity of a 5 MHz, 12.7 mm diameter
transducer radiating through a water-steel interface as a function of the depth in the steel. Results
calculated for (a) a planar interface, (b) a cylindrical interface with a 76 mm radius of curvature,
(c) a cylindrical interface with a 76 mm radius of curvature
and reduced amplitudes. However, when R1 ¼ 76 mm, the beam shown in
Fig. 8.94 is indeed more focused in the solid.
As an example of a non-normal incidence case, consider a 45 angle beam shear
wave transducer, where a 10 MHz, 12.7 mm diameter circular transducer is placed
on a Lucite wedge in smooth contact with an aluminum part. A MATLAB script
angle_beam2 specifies the parameters needed in the structure setup. According to
Snell’s law, the incident angle in the wedge must be 38.91 , which is greater than
the first critical angle of 24.9 . The distance from the transducer to the interface
along the central ray is specified as 20 mm, and since we have two solids in smooth
contact the transmission coefficient chosen must be for this case. These and other
specifications can be seen in the angle_beam2 script. The function MG_beam2 is
called with the setup parameters and an image displayed, as shown in Fig. 8.95,
which clearly shows a well-collimated SV-wave beam traveling at a 45 angle in the
aluminum.
As a final example of the use of MG_beam2, we will consider the generation of a
time-domain pulse for a circular transducer radiating into water. Here we specify a
8.9 Multi-Gaussian Beam Models 371
Fig. 8.92 The magnitude of the normalized velocity in the x–z plane of a 5 MHz, 12.7 mm
diameter transducer radiating through a planar water-steel interface
spectrum for the velocity, v0(ω), with a MATLAB function, spectrum, whose
calling sequence is
pffiffiffiffiffiffiffi
where a ¼ ln2=ðπ bwÞ. The maximum of the Gaussian is at the center frequency
fc (in MHz) and the 6 dB bandwidth is bw (also in MHz). Since there are only
positive frequencies in this spectrum if one takes twice the real part of the inverse
Fourier transform of this function one will recover a time domain signal, v(t), whose
amplitude at t ¼ 0 is B and which we can determine analytically as [22]
vðtÞ ¼ B cos ð2πf c tÞexp t2 =4a2 :
372 8 Ultrasonic Transducer Radiation
Fig. 8.93 The magnitude of the normalized velocity in the x–z plane of a 5 MHz, 12.7 mm
diameter transducer radiating through a cylindrical water-steel interface with radius of
curvature ¼ 76 mm
The MATLAB script pulse2 uses this function to simulate the pulse generated by a
12.7 mm diameter transducer radiating into water whose center frequency is
fc ¼ 5 MHz and whose bandwidth is bw ¼ 4 MHz. The location in the transducer
wave field where the pulse is generated is on-axis at z ¼ 135 mm, which is close to
one near field distance. The script sets up the parameters in the structure setup and
then evaluates the on-axis velocity with MG_beam2 function for a subset of
frequencies of 0–20 MHz over a full spectrum of 0–100 MHz. This on-axis velocity
is multiplied by the function spectrum over the same subset of frequencies and then
zeros are added to extend the results to the full spectrum of 100 MHz. This full
spectrum is then inverted into the time domain and plotted, taking into account that
the response only contains positive frequencies [22]. The results are shown in
Fig. 8.96.
These examples show some of the versatility and efficiency of a multi-Gaussian
beam model. We leave it to the reader to examine other cases of interest.
Fig. 8.94 The magnitude of the normalized velocity in the x–z plane of a 5 MHz, 12.7 mm
diameter transducer radiating through a cylindrical water-steel interface with radius of
curvature ¼ 76 mm
Fig. 8.95 The magnitude of the normalized velocity in the plane of incidence for a 10 MHz,
12.7 mm diameter P-wave transducer on a Lucite wedge radiating a 45 SV-wave into aluminum
374 8 Ultrasonic Transducer Radiation
Fig. 8.96 The on-axis time domain wave form of a transducer radiating into water with a 5 MHz
center frequency and a 6 dB bandwidth of 4 MHz, as calculated near one near field distance
a
z
2V0
Fig. P8.1 Planar transducer, with a non-uniform velocity distribution on its surface, radiating into
a fluid
2.5 in.
1.5 in.
diffraction corrections will be discussed in the next chapter. However, the use of
diffraction corrections as an integral part of an ultrasonic measurement model of the
type being discussed in this book is due to Thompson and Gray [27]. We have also
given a sampling of some additional papers that discuss various transducer model-
ing issues discussed in this Chapter in more depth [28–48].
a b
1.5 in.
Fig. P8.3 (a–c) Geometries for calculating incident wave fields
8.11 Problems
8.1. The last on-axis near field maximum of a circular-shaped planar transducer is
located at a distance, z0, given exactly by Eq. (8.23). At high frequencies, this
last maximum occurs approximately at z0 ffi N ¼ a2 =λ (see Eq. (8.24)). For a
0.5 in. diameter circular transducer radiating into water, compare the value of
z0, using both the exact and approximate expressions, at frequencies of 1, 5,
10, and 20 MHz. What can you conclude about the use of the approximate
expression?
8.2. In the far field, the radiation pattern of a circular plane piston transducer is
controlled by the directivity function J1(ka sin θ)/ka sin θ (Eq. (8.36)) which
has a series of lobes. The angle subtended by the first (main) lobe was given
by Eq. (8.37).
(a) For a 0.5 in. diameter transducer radiating into water at 1, 5, 10, and
20 MHz, what is the angle subtended by the main lobe?
(b) Determine an expression for the angle subtended by the second lobe of a
circular planar transducer. What conditions must be met for such a
second lobe to exist?
8.11 Problems 377
Plot both of these diffraction coefficients versus z/N for a 1/2 in. diameter
probe operating at 10 MHz. Where does the paraxial approximation start to
break down?
8.5. Consider a 0.5 in. diameter, 5 MHz transducer in a water bath, oriented
normal to a 1.5 in. thick aluminum plate (Fig. P8.2). If we model the
transducer as a circular piston probe which puts out a normal velocity, v0,
on its surface then in the water bath (Fig. P8.3a) the on-axis incident pressure
is given (neglecting reflections), in the paraxial approximation, as
pðz; ωÞ ¼ ρ1 c1 v0 ðωÞexpðik1 zÞC1 k1 a2 =2z ; ðP8:3Þ
where the diffraction coefficient is given in Eq. (P8.2). In the aluminum plate,
the on-axis pressure due the wave transmitted directly into the plate (again,
neglecting reflections) (Fig. P8.3b) would be given, in the paraxial approx-
imation, by
p ¼ ρ1 c1 v0 ðωÞexp½iðk1 z1 þ k2 z2 ÞT 12 ð0 ÞC1 k1 a2 =2z ; ðP8:4Þ
where
z ¼ z1 þ c2 z2 =c1 :
378 8 Ultrasonic Transducer Radiation
Similarly, on the other side of the plate (Fig. P8.3c) the pressure from the wave
directly transmitted across the plate to the other side is, in the paraxial approxima-
tion, given by
p ¼ ρ1 c1 v0 ðωÞexp½iðk1 z1 þ k2 z2 þ k1 z3 ÞT 12 ð0 ÞT 21 ð0 ÞC1 k1 a2 =2~z ; ðP8:5Þ
where
~z ¼ z1 þ c2 z2 =c1 þ z3 :
In Eq. (8.144a), the beam diameter, Wf, was defined to be twice the distance
from the axis to the first off-axis null. Here, we define a beam diameter,
Wf (3 dB), to be twice the distance from the axis to the location where the
pressure amplitude is reduced by 3 decibels (dB), i.e. ΔpdB ¼ 3, where
pðR0 ; y; ωÞ
ΔpdB ¼ 20 log10 : ðP8:7Þ
pðR0 ; 0; ωÞ
8.11 Problems 379
y y
r
x R sinθ
ρφ φ0
R φ0
θ
z
a O
Fig. P8.4 Geometry for calculating the far field radiation pattern of a rectangular piston
transducer
Now, define the 3 dB depth of focus, Dz(3 dB), to be twice the distance along
the axis from the geometric focus, z R0 ¼ q0 R0 , at which the pressure
amplitude is reduced by 3 decibels (dB), i.e. ΔpdB ¼ 3, where
pðz; ωÞ
ΔpdB ¼ 20 log10 : ðP8:11Þ
pðR0 ; ωÞ
Show that Dz ð3 dBÞ ¼ 1:8R20 λ=a2 . How is Dz(3 dB) related to Wf (3 dB) ? (see
the previous problem).
380 8 Ultrasonic Transducer Radiation
8.10. (a) Show that in the far field the Rayleigh-Sommerfeld equation for a
rectangular piston transducer, with sides of lengths a and b, reduces to
ð
b=2 ð
a=2
h
0 i 0 0
iωρv0 ðωÞexpðikRÞ 0
pðx; ωÞ ¼ exp ik sin θ x cos ϕ0 þ y sin ϕ0 dx dy :
2πR
b=2 a=2
ðP8:12Þ
(see Fig. P8.4 and follow the same procedure used for calculating the far
field radiation pattern for a circular transducer).
(b) By evaluating the integrals in Eq. (P8.12) exactly, determine an explicit
expression for the far field off-axis pressure.
(c) Using the results of (b), obtain expressions similar to Eqs. (8.37)–(8.38)
for the lateral extent of the main lobe in the x and y-directions.
8.11. Consider a rectangular piston transducer, with sides of lengths a and b,
radiating into a fluid.
(a) Using the Rayleigh-Sommerfeld equation, show that the on-axis pres-
sure can be obtained exactly in the form
2 3
2 ð
X qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
6 7
pðz; ωÞ ¼ ρcv0 ðωÞ4expðikzÞ exp ik C2i þ D2i = cos 2 ϕ 5dϕ; ðP8:13Þ
i¼1
Si
where the edge wave integrals can be written over the two sides Si and
where Ci and Di are coefficients along each side.
(b) By integrating the edge wave terms in Eq. (P8.13) numerically, deter-
mine the normalized on-axis pressure, p/ρcv0, in the near and far field for
a 5 MHz, 0.5 0.5 in. square transducer in water. Compare your results
with those for a 5 MHz, 0.5 in. diameter circular transducer.
8.12. (a) For a rectangular piston transducer, with sides of length a and b, use the
paraxial approximation to obtain an expression for the diffraction cor-
rection in terms of a product of two complex Fresnel integrals of the
form
cð2
F¼ exp iπu2 =2 du: ðP8:14Þ
c1
References
1. I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products (Academic, New York,
1980)
2. A. Rubinowicz, The Miyamoto-Wolf Diffraction Wave, in Progress in Optics, ed. by E. Wolf,
vol. 4 (Wiley, New York, 1965), pp. 201–240
3. K. Miyamoto, E. Wolf, Generalization of the Maggi-Rubinowicz theory of the boundary
diffraction wave—part I. J. Opt. Soc. Am. 52, 615–625 (1962)
4. K. Miyamoto, E. Wolf, Generalization of the Maggi-Rubinowicz theory of the boundary
diffraction wave—part II. J. Opt. Soc. Am. 52, 626–637 (1962)
5. E.W. Marchand, E. Wolf, Boundary diffraction wave in the domain of the Rayleigh-Kirchhoff
diffraction theory. J. Opt. Soc. Am. 52, 761–767 (1962)
6. H.T. O’Neil, Theory of focusing radiators. J. Acoust. Soc. Am. 21, 516–526 (1949)
7. F. Coulouvrat, Continuous field radiated by a geometrically focused transducer: numerical
investigation and comparison with an approximate model. J. Acoust. Soc. Am. 94, 1663–1675
(1993)
8. J.J. Stamnes, Waves in Focal Regions (Adam Hilge, Boston, 1986)
9. U. Schlengermann, Schallfeldausbildung bei ebenen Ultraschallquellen mit fokussierenden
linsen. Acustica 30, 291–300 (1974)
10. B.B. Baker, E.T. Copson, The Mathematical Theory of Huygen’s Principle, 2nd edn. (Oxford
University Press, Fair Lawn, 1950)
11. D.A. McNamara, C.W.I. Pistorius, J.A.G. Malherbe, Introduction to the Uniform Geometrical
Theory of Diffraction (Artech House, Boston, 1990)
12. V. Cerveny, R. Ravindra, Theory of Seismic Head Waves (University of Toronto Press,
Toronto, 1971)
13. G.L. James, Geometrical Theory of Diffraction for Electromagnetic Waves. Revised, 3rd edn.
(Peter Peregrinus, London, 1986)
14. B.Y. Gel’chinskiy, An Expression for the Spreading Function, in Problems in the Dynamic
Theory of Propagation of Seismic Waves, ed. by G.I. Petrashen, vol. 5 (Leningrad University
Press, Leningrad, 1961), pp. 47–53 [In Russian]
15. L.W. Schmerr, Fundamentals of Ultrasonic Phased Arrays (Springer, New York, 2014)
16. T.P. Lerch, Ultrasonic transducer characterization and transducer beam modeling for applica-
tions in nondestructive evaluation, Ph.D. thesis, Iowa State University, 1996
17. P.M. Morse, K. Uno Ingard, Theoretical Acoustics (Princeton University Press, Princeton,
1968)
18. D.J. Vezzetti, Propagation of bounded ultrasonic beams in anisotropic media. J. Acoust. Soc.
Am. 78, 1103–1108 (1985)
19. L.W. Schmerr, A. Sedov, An elastodynamic model for compressional and shear wave trans-
ducers. J. Acoust. Soc. Am. 86, 1988–1999 (1989)
382 8 Ultrasonic Transducer Radiation
20. J.J. Wen, M.A. Breazeale, A diffraction beam field expressed as the superposition of Gaussian
beams. J. Acoust. Soc. Am. 83, 1752–1756 (1988)
21. J.J. Wen, M.A. Breazeale, Computer Optimization of the Gaussian Beam Description of an
Ultrasonic Field, in Computational Acoustics: Scattering, Gaussian beams, and
Aeroacoustics, vol. 2 (Elsevier Science, Amsterdam, 1990), p. 181–196
22. L.W. Schmerr, S.J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and Mea-
surements (Springer, New York, 2007)
23. H.J. Kim, L.W. Schmerr, A. Sedov, Generation of the basis sets for multi-Gaussian ultrasonic
beam models—an overview. J. Acoust. Soc. Am. 119, 1971–1978 (2006)
24. G.R. Harris, Review of transient field theory for a baffled planar piston. J. Acoust. Soc. Am. 70,
10–20 (1981)
25. D.A. Hutchins, G. Hayward, Radiated Fields of Ultrasonic Transducers, in Physical Acoustics:
Ultrasonic Measurement Methods, ed. by R.N. Thurston, A.D. Pierce, vol. 19 (Academic
Press, New York, 1990), pp. 1–80
26. K. Gniadek, J. Petkiewicz, Applications of Optical Methods in the Diffraction Theory of
Elastic Waves, in Progress in Optics, ed. by E. Wolf, vol. 9 (American Elsevier, New York,
1971), pp. 283–310
27. R.B. Thompson, T.A. Gray, A model relating ultrasonic scattering measurements through
liquid-solid interfaces to unbounded medium scattering amplitudes. J. Acoust. Soc. Am. 74,
1279–1290 (1983) (see also Erratum, J. Acoust. Soc. Am., 75, 1645–1646, (1984))
28. A.O. Williams, The piston source at high frequencies. J. Acoust. Soc. Am. 23, 1–6 (1951)
29. M. Greenspan, Piston radiator: some extensions of the theory. J. Acoust. Soc. Am. 65, 608–621
(1979)
30. H.D. Mair, L. Bresse, D.A. Hutchins, Diffraction effects of planar transducers using a
numerical expression for edge waves. J. Acoust. Soc. Am. 84, 1517–1525 (1988)
31. J.C. Lockwood, J.G. Willette, High speed method for computing the exact solution for the
pressure variations in the nearfield of a baffled piston. J. Acoust. Soc. Am. 53, 735–741 (1973)
32. J.N. Tjotta, S. Tjotta, Nearfield and farfield of pulsed acoustic radiators. J. Acoust. Soc.
Am. 71, 824–834 (1982)
33. D.A. Hutchins, H.D. Mair, P.A. Puhach, J. Osei, Continuous-wave pressure fields of ultrasonic
transducers. J. Acoust. Soc. Am. 80, 1–12 (1986)
34. M.F. Hamilton, Comparison of three transient solutions for the axial pressure in a focussed
sound beam. J. Acoust. Soc. Am. 92, 527–532 (1992)
35. L.W. Schmerr, T.P. Lerch, A. Sedov, A Focussed Transducer/Scatterer Model for Ultrasonic
Reference Standards, in Review of Progress in Quantitative Nondestructive Evaluation, ed. by
D.O. Thompson, D.E. Chimenti, vol. 12A (Plenum Press, New York, 1993), pp. 925–932
36. U. Schlengermann, The characterization of focussing ultrasonic transducers by means of single
frequency analysis. Mater. Eval. 38, 73–79 (1980)
37. W.G.R. Gibson, R.S.C. Cobbold, F.S. Foster, Ultrasonic fields of a convex semispherical
transducer. J. Acoust. Soc. Am. 94, 1923–1929 (1993)
38. R.B. Thompson, T.A. Gray, J.H. Rose, V.G. Kogan, E.F. Lopes, The radiation of elliptical and
bicylindrically focused piston transducers. J. Acoust. Soc. Am. 82, 1818–1828 (1987)
39. A. Ilan, J.P. Weight, The propagation of short pulses of ultrasound from a circular source
coupled to an isotropic solid. J. Acoust. Soc. Am. 88, 1142–1151 (1990)
40. L.F. Bresse, D.A. Hutchins, Transient generation of elastic waves in solids by a disk-shaped
normal force source. J. Acoust. Soc. Am. 86, 810–817 (1989)
41. J.A. Archer-Hall, D. Gee, A single integral computer method for axisymmetric transducers
with various boundary conditions. NDT Int. 13, 95–101 (1980)
42. R.B. Thompson, T.A. Gray, Range of Applicability of Inversion Algorithms, in Review of
Progress in Quantitative Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chiment,
vol. 1 (Plenum Press, New York, 1982), pp. 233–249
References 383
43. R.B. Thompson, T.A. Gray, Analytical Diffraction Corrections to Ultrasonic Scattering Mea-
surements, in Review of Progress in Quantitative Nondestructive Evaluation, ed. by
D.O. Thompson, D.E. Chimenti, vol. 2A (Plenum Press, New York, 1983), pp. 567–586
44. R.C. Chivers, L. Bosselaar, P.R. Filmore, Effective area to be used in diffraction corrections.
J. Acoust. Soc. Am. 68, 80–84 (1980)
45. F. Amin, T.A. Gray, F.J. Margetan, A New Method to Estimate the Effective Focal Length and
Radius of Ultrasonic Focussed Probes, in Review of Progress in Quantitative Nondestructive
Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 10A (Plenum Press, New York, 1991),
pp. 861–865
46. B.P. Newberry, R.B. Thompson, A paraxial theory for the propagation of beams in anisotropic
materials. J. Acoust. Soc. Am. 85, 2290–2300 (1989)
47. Q.C. Guo, J.D. Achenbach, Radiation of ultrasound into an anisotropic solid. Ultrasonics 33,
449–456 (1996)
48. D.O. Thompson, D.E. Chimenti (eds.), Review of Progress in Quantitative Nondestructive
Evaluation. (American Institute of Physics, Melville, published annually, 1981–present)
Chapter 9
Material Properties and System Function
Determination
The models of wave propagation described in the previous Chapters have all treated
the underlying fluids and/or solids as perfect, non-attenuating media. Real mate-
rials, however, do exhibit an attenuation that must be accounted for in any complete
description of an ultrasonic measurement system. Since the processes that generate
material attenuation are generally quite complex, we will not attempt to model
those processes in detail from a fundamental standpoint. Instead we will use a
simple phenomenological 1-D attenuation model that is coupled with detailed
experimental measurements in an explicitly modeled calibration setup.
In a similar fashion, the transduction processes on transmission and reception
(where the voltage pulses generated by the pulser/receiver are transformed into
mechanical motion of the transducer face and vice-versa) are complex electrome-
chanical processes. Although sophisticated equivalent circuit models of transducers
are available (see [1, 2], for example), the precise equivalent circuits and transfer
functions for all the transduction processes are dependent on the specific transducer
(s) being used, the external control settings of the pulser/receiver (which can be
varied by the user over a wide range of values), the cabling used, etc. In spite of this
complexity, complete models of all the electrical and electromechanical compo-
nents in an ultrasonic measurement system have been developed [3], along with
methods for explicitly determining those modeled components with purely electri-
cal measurements. However, we will not describe in detail the transmission and
reception processes. Instead, we will use a single lumped term to account for all of
those components called the system function (or a closely related term called the
system efficiency factor) and we will show how this term can be found by detailed
experimental measurements in an explicitly modeled calibration setup.
For many modeling studies we may use tabulated values of density and wave
speed, such as those found in Table 3.1, but for simulation of results with specific
samples, one must measure those parameters. In Sect. 9.3 we will also discuss
briefly wave speed measurements, particularly in complex materials that are aniso-
tropic or inhomogeneous.
A1
¼ exp½αð f Þ d; ð9:1Þ
A0
A0 A 1 (< A 0)
Fig. 9.2 (a) Propagation in an ideal material without attenuation where amplitude changes are due
to beam spreading (diffraction) only, and (b) propagation in a real material with grain scattering
showing “metal noise” and where amplitude changes are due to both attenuation and beam
spreading
λ=D ffi 1 Stochastic Df 2
frequency like f 2. In water, for example the attenuation at room temperature has
been measured as [4]:
where f is the frequency in Hz and the dimensions of α are in Nepers (Np) per meter,
where a Neper is a dimensionless quantity. Similar effects are also important in
viscoelastic solids but those materials are not ordinarily used in ultrasonic testing.
There are, of course, other attenuation losses in materials, such as the losses due
to the conduction of heat from higher temperature regions in a wave to lower
temperature regions, etc. See the book by Fitting and Adler [5], for example, for
more details on those sources of attenuation.
The attenuation term, α, represents the coefficient of the decaying exponential
that describes the material attenuation effects. In many attenuation studies, authors
prefer to measure attenuation in terms of amplitude ratios on a decibel (dB) scale so
that it is necessary to relate these dB values to α. To obtain this relationship, let p1
be the pressure at z ¼ z1 in a 1-D plane wave traveling in the positive z-direction and
388 9 Material Properties and System Function Determination
let p2 be the pressure at z ¼ z2 , where z2 > z1 . Then the ratio of the magnitude of
these two pressures is given by
p1
¼ exp½α Δz; ð9:3Þ
p
2
where κ is a damping term and cf is the wave speed (without damping) of the fluid.
If we assume that the pressure p is characterized by an exponentially damped
harmonic plane wave propagating in the positive z-direction
and place this expression into Eq. (9.7), we find that in order to satisfy the damped
wave equation we must have
α2 k2 ¼ ω2 =c2f
ð9:9Þ
2αk ¼ κω=c2f ;
9.1 Sources of Attenuation 389
In order to have solutions that represent real positive values of k, we must choose
both plus signs in Eq. (9.12), which we can rewrite as
ω h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii1=2
k ¼ pffiffiffi 1 þ 1 þ κ2 =ω2 : ð9:13Þ
2 cf
Since the wave number k and wave speed c are related through k ¼ ω=c, Eq. (9.13)
shows that the wave speed (phase velocity) c is frequency dependent (dispersive)
and given by
pffiffiffi
2 cf
c¼h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii1=2 : ð9:14Þ
1 þ 1 þ κ 2 =ω2
ω hpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i1=2
α ¼ pffiffiffi 1 þ κ 2 =ω2 1 : ð9:15Þ
2 cf
390 9 Material Properties and System Function Determination
However, when the damping term is small, as would typically be the case in many
NDE applications, Eqs. (9.14) and (9.15) reduce to simply
c ffi cf
κ ð9:16Þ
αffi ;
2cf
which shows that our ad-hoc approach, which left the phase velocity unchanged and
merely added an exponentially damped attenuation factor, is valid in such
circumstances.
The fundamental absorption and scattering mechanisms that cause attenuation have
been studied for some time [6]. For the polycrystalline solids used in NDE appli-
cations, grain scattering, as described previously, is normally the major source of
attenuation. One of the most comprehensive models that directly relates the mate-
rial microstructure to attenuation (and phase velocity) is due to Stanke and Kino [7].
Their theory is quite general, allowing for both textured materials and non-uniform
grain shapes, and uses a second order Keller approximation to account for multiple
scattering effects. This theory, though quite general, was initially developed for the
special case of equiaxed, untextured materials with cubic crystallites. However,
Ahmed and Thompson [8] have included arbitrary texture and elongated grains.
Beltzer and Brauner [9] have modified the Stanke-Kino theory to make it satisfy
causality and provide a method to determine the attenuation and phase velocity that
satisfies the Kramers-Kronig relations. These detailed models relating microstruc-
ture to attenuation are essential for understanding the relationship between a
particular material and its attenuation. For many applications of ultrasonic model-
ing, however, the 1-D model-based approaches described earlier are likely
adequate.
Figure 9.3 shows pulse-echo ultrasonic immersion and contact measurement setups.
We will use these configurations to discuss modeling ultrasonic systems as LTI
systems following the outline of the general discussion in Chap. 2 but with more
details. Consider first the immersion case. In this setup, the pulser produces an
output voltage pulse which travels down a cable to the transducer where it is
converted from an electrical pulse to an acoustic pulse and mechanical motion at
the face of the transducer. As done in Chap. 8 we will assume that this transducer
9.2 LTI Models 391
motion is a spatially uniform normal velocity, v0(t), over the face of the transducer
(piston transducer model) whose frequency components are v0(ω). There is also an
output compressive force, FT(t) at the face of the transducer generated by a pressure
distribution acting on its face, whose frequency components are FT(ω). The force
FT(ω) and velocity v0(ω) are related through the acoustic radiation impedance of
the transmitting transducer T, ZT;a
r (ω) i.e.
FT ðωÞ ¼ Z T;a
r ðωÞv0 ðωÞ; ð9:17Þ
where the a superscript is used to emphasize that this is an acoustic impedance, not
an electrical impedance. This acoustic radiation impedance depends on the shape of
the transducer as well as the frequency in general. For example, for a circular
transducer of radius a Greenspan [10] has developed (from the Rayleigh-
Sommerfeld model for a piston transducer radiating into a fluid) an explicit
expression for this impedance:
where (ρ1, cp1) are the density and wave speed, respectively, of the fluid, ST ¼ πa2 is
the area of the moving face of the transducer, k is the wave number, and (J1, S1) are
Bessel and Struve functions, respectively. However, at the high frequencies that
ultrasonic NDE transducers typically operate, ka
1. For example, for a one half
inch diameter radiating into water at 5 MHz, ka ¼ 135 approximately. At such high
values, the radiation impedance of Eq. (9.18) is just the leading term in Eq. (9.18)
given by
a result that can also be shown to be true for any shaped piston transducer [3]. The
acoustic radiation impedance in this case is just the plane wave specific impedance
ρ1cp1 discussed in Chap. 6 times the active area, ST, of the transducer face. Note that
although Eq. (9.19) contains the plane wave specific impedance, at these high
392 9 Material Properties and System Function Determination
as the quantity we will use to characterize the output of the sending piston
transducer. This quantity has the dimensions of a force and is just proportional to
the piston output velocity. In general this is not the actual output force, FT(ω),
which is given by Eq. (9.17) but at the high ka values valid for most ultrasonic
transducers, there is no significant difference between Ft and FT so in those cases
there is little danger in viewing Ft as the output force. Nevertheless, we have placed
different subscripts on these forces to remind the reader these are not the same
forces, where a lowercase subscript “t” indicates this is the force being used in
defining the acoustic/elastic transfer function while the uppercase subscript “T”
indicates this is the actual force on the transmitting transducer T. The motion and
force on the face of the transmitting transducer is generated by the pulser, which can
be represented as an active circuit with a driving voltage source, Vi(ω) [3], and the
entire sound generation process will be characterized as an LTI system whose input
is Vi(ω) and whose output is Ft(ω). We will let Tg(ω) be the transfer function for the
sound generation process so we have
Now, consider the face of the receiving transducer when waves are incident on it,
as shown in Fig. 9.4. Let FR(ω) and vR(ω) be the force and normal velocity (normal
taken positive outward from the face of the transducer, as in the sending case)
generated by the interaction of the incident waves with this receiving transducer,
which again is assumed to act as a piston so that vR is a function of frequency only.
In this case we can decompose the reception problem into two parts as shown in
Fig 9.4: part (a) where the face of the transducer is held rigidly fixed (no motion)
while the incident waves strike it, and part (b) where the incident waves are absent
Fig. 9.4 Reception of waves incident on the receiving immersion transducer decomposed into
part (a) where the incident waves are present and the face of the transducer is held rigidly fixed and
part (b) where the incident waves are absent and the transducer radiates with a velocity that is the
same as in the original problem
9.2 LTI Models 393
but the transducer face moves with an outward velocity vR(ω) of our original
problem. In part (a), the force generated on the face of the transducer, FB(ω), is
called the blocked force. In part (b), the incident waves are absent so the force on
the face of the receiving transducer is just due to the motion of the face itself,
i.e. part (b) is just the same as when the transducer acts as a sender and the force in
part (b) is therefore just the acoustic radiation impedance of the receiving trans-
ducer R, ZrR;a (ω), times the velocity vR(ω). The total force, FR(ω), therefore, is just
the sum of the forces in parts (a) and (b), i.e.
In a detailed model of the entire reception process, in [3] it is shown that the blocked
force acts as natural driving (acoustical) source term for the reception process just
as Vi(ω) acts as the driving (electrical) source during sound generation. Also, in
Chap. 12 we will show that modeling the acoustic/elastic processes with reciprocity
principles leads directly to the blocked force. Thus, it is natural to relate the
measured output voltage frequency components, V0(ω), of the time-dependent
output voltage, V0(t), to this blocked force through a sound reception transfer
function, Tr(ω), i.e.
Since the force Ft is being used to characterize the output of the waves generated by
the sending transducer and the blocked force is being used to characterize the waves
driving the receiving transducer, it is appropriate to define an acoustic/elastic
transfer function, Ta(ω) as
F B ð ωÞ
T a ð ωÞ ¼ ð9:24Þ
F t ð ωÞ
In [3] it is shown how one can determine Tr(ω), Tg(ω) and Vi(ω) from a set of purely
electrical measurements. However, an alternative procedure is to lump all these
components into a single factor called the system function, sI(ω), for our immersion
system, where
and write our LTI model for the entire measurement system simply as
This simple LTI model, however, is only useful in this form if we have a way to
determine the system function directly. This is possible if we can perform a
calibration experiment in a configuration where we can model the acoustic/elastic
transfer function explicitly and measure the voltage output V0(ω). Thus, we need to
examine more carefully what is needed to determine that transfer function. To
determine the blocked force in general we would have to solve a boundary value
problem where the incident waves interact with a transducer whose face is held
rigidly fixed. However, from Chap. 6 we know that when an incident plane pressure
wave interacts with a rigid, immobile plane surface, the reflection coefficient is just
equal to one, so the total pressure is just twice that of the incident wave, a result
which is also true at high frequencies for non-planar waves (and non-planar
surfaces) as shown in Chap. 6. Thus, it is reasonable to take the blocked force as
twice the force of the incident waves acting on the receiving transducer as if that
transducer was not present. With this assumption, the acoustic/elastic transfer
function is
ð
2 pinc ðx; ωÞ
T a ðωÞ ¼ dSðxÞ
ST ρ1 cp1 v0 ðωÞ
SR
SR
¼2 p ðωÞ; ð9:28Þ
ST ave
where
ð
1 pinc ðx; ωÞ
pave ðωÞ ¼ dSðxÞ ð9:29Þ
SR ρ1 cp1 v0 ðωÞ
SR
is the normalized incident pressure, pinc, averaged over the face of the receiving
transducer (where pinc is calculated with the transducer absent). In the pulse-echo
case where SR ¼ ST the acoustic/elastic transfer function is twice this average
pressure. Note that since pinc(x, ω) is generated by the sending transducer, this
pressure is just proportional to ρ1cp1v0(ω) so that if we can model this normalized
pressure directly, we do not need to know the actual driving velocity on the face of
the sending transducer in order to model Ta(ω). We will see that we can indeed
model this normalized pressure and compute this average pressure in a number of
calibration setups.
Note that the LTI models used in this section are based on our particular
definitions of the transfer functions and those definitions are not the only ones
that could be used. For example, we could have taken T a ðωÞ ¼ pave ðωÞ and written
where βI(ω) has been called the system efficiency factor (for an immersion mea-
surement). In the first edition of this book the efficiency factor was used exclusively
9.2 LTI Models 395
SR
β I ðω Þ ¼ 2 sI ðωÞ: ð9:31Þ
ST
Thus, apart from their amplitudes, both have identical behavior and can be deter-
mined in precisely the same fashion.
When we model pave for a particular problem, we normally treat the materials
involved as completely ideal and ignore attenuation effects. However, as seen in the
last section we can use those ideal models and simply add in attenuation through a
series of exponentials with frequency dependent attenuation coefficients, αi(ω), for
the materials involved and the appropriate propagation lengths, di, of the waves in
those materials, giving a more complete LTI of a measurement system as
!
SR X n
V 0 ðωÞ ¼ 2 sI ðωÞpave ðωÞ exp αi ðωÞdi ð9:32Þ
ST i¼1
in terms of the system efficiency factor. Note that the dimensions of both sI(ω) and
βI(ω) are the same as V0(ω) (generally measured in volts-μsec). In the next section,
we will show a particular calibration setup where pave ðωÞ can be modeled explicitly.
This will allow us, as shown in later sections, to obtain sI(ω) or βI(ω) and the
attenuation terms αi(ω) experimentally from the general models of Eqs. (9.32)
and (9.33).
In the case of a contact P-wave transducer acting on the otherwise stress-free
surface, as shown in Fig. 9.3b, there are some physical differences with the
immersion case that must be considered in our LTI models. First, as discussed in
Chap. 8 it is not possible to assume the face of a sending transducer acts as a piston
source since the solid the transducer is in contact with is often more rigid than the
crystal of the transducer. Instead, it is more reasonable to assume that the P-wave
transducer, which is in contact with the solid through a thin fluid coupling layer,
produces a uniform pressure, p0(ω), on the surface of the solid. The compressive
force, F(ω), on the transducer face is then
Fig. 9.5 Reception of waves incident on the receiving contact transducer decomposed into part
(a) where the incident waves are present and the force on the transducer is zero (contact surface is
completely stress-free) and part (b) where the incident waves are absent and the force on the
contact transducer is the same as in the original problem
where vT(x, ω) is the velocity distribution over the transmitting transducer’s face.
The reception process for a contact transducer is shown in Fig. 9.5 where
incident waves strike that transducer and the otherwise stress-free surface on
which it sits and produces a force, FR(ω), and an average velocity, vR ðωÞ, on the
face of the transducer where
ð
1
v R ð ωÞ ¼ vR ðx; ωÞ dSðxÞ ð9:36Þ
SR
SR
in terms of the velocity distribution, vR(x, ω). Note that in the contact case we will
take vR and vR as positive when directed inwards towards the face of the transducer
as shown in Fig. 9.5. Again, we can decompose this problem into two parts. In part
(a), where the incident waves are present, the force acting on the transducer is zero,
and the average velocity, vfs ðωÞ, is
ð
1
vfs ðωÞ ¼ vfs ðx; ωÞ dSðxÞ; ð9:37Þ
SR
SR
where vfs(x, ω) is the total velocity field (incident and reflected waves) at a
completely stress-free surface (i.e. when the receiving transducer is absent). In
part (b) the incident waves are absent and the average velocity is called vs ðωÞ. The
force acting on the transducer in part (b) must be the total received force, FR(ω),
since there is no force present in part (a). Also note that in part (b) the transducer is
9.2 LTI Models 397
simply radiating waves into the solid (i.e. acting as a “sending” transducer) so we
can define the acoustic radiation impedance, ZrR;a (ω), as
where the minus sign is present because vs is positive acting in the direction opposite
to the external normal of the transducer face (the same as vR and vfs—see Fig. 9.5).
The total average velocity vR ðωÞ ¼ vfs ðωÞ þ vs ðωÞ is then
FR ð ω Þ
v R ð ωÞ ¼ v f s ð ωÞ : ð9:39Þ
r ð ωÞ
Z R;a
In a detailed model of the entire reception process, in [3] it is shown that the free
surface velocity, vfs ðωÞ, acts as natural driving source term for the contact case just
as the blocked force does for the immersion case. Also, in Chap. 12 we will see that
modeling the acoustic/elastic transfer function with reciprocity relations directly
leads to this free surface velocity. As in the immersion case, we will define the
0
acoustic/elastic transfer function, Ta (ω), in our LTI model for the contact case as a
ratio of force-like quantities where
with (ρ, cp) being the density and compressional wave speed of the solid in contact
with the receiving transducer. Again, such a definition is useful only if it is possible
to actually model this transfer function. This is possible since vfs is proportional to
the driving pressure, p0, so we can rewrite the transfer function as
8 9
ð
0 SR < 1 vfs ðx; ωÞ =
Ta ¼ dSðxÞ : ð9:41Þ
ST :SR p0 =ρcp ;
SR
The non-dimensional wave field in the integral of Eq. (9.41) can in principle be
calculated since it does not depend on knowing the actual pressure on the face of
the transmitting transducer. This integral can be made even easier if we note that for a
plane P-wave incident on a free surface at normal incidence, the total velocity at the free
surface is just twice the velocity of the incident wave field, vinc(x, ω), calculated with
the free surface absent, so if we make this assumption in the calculation of our transfer
function we have
8 9
< ð = 2S
0 2SR 1 vinc ðx; ωÞ R
T a ð ωÞ ¼ dSðxÞ ¼ vinc ðωÞ; ð9:42Þ
ST : S R p0 =ρcp ; ST
SR
where vinc is the normalized average incident velocity term in brackets in Eq. (9.42).
Again, we can assume vinc is modeled for an ideal material and include attenuation
398 9 Material Properties and System Function Determination
2SR
V 0 ð ωÞ ¼ sC ðωÞvinc ðωÞ ð9:43Þ
ST
Consider the pulse-echo immersion calibration setup shown in Fig. 9.6 where a
circular transducer, of radius a, is oriented normal to the plane faces of a block of
material. The fluid and solid will be modeled as ideal (non-attenuating media) with
densities (ρ1, ρ2) and wave speeds (cp1, cp2, cs2) as shown in Fig. 9.6. Treating the
transducer as a piston source and writing the solution in terms of an angular
spectrum of plane waves (see Eq. (8.83)), the incident pressure generated in the
fluid is given by
8 9
ð < þ1
ð þ1
ð =
ωρ v 0 exp ½ ip ð x y Þ
pinc ðx; ωÞ ¼ 12 1
dpx dpy dSðyÞ: ð9:45Þ
4π : p1z ;
S 1 1
Similarly, the pressure in the fluid due to the waves which reflected from the front
surface is given by
8 9
ð < þ1
ð þ1
ð P;P =
ωρ v0 R exp ½ ip ð x y Þ þ 2ip D
preflt ðx; ωÞ ¼ 12 12 1r 1z
dpx dpy dSðyÞ;
4π : p1z ;
S 1 1
ð9:46Þ
ρ1, cp1 D = D1 D2
9.2 LTI Models 399
If we introduce polar coordinates for both the angular plane wave spectrum and
0 0
the spatial variables p1 ¼ px ; py ; p1z , x y ¼ x x , y y , 0 , i.e.
0
px ¼ pρ cos ϕp , x x ¼ r cos ϕ
0 ð9:48Þ
py ¼ pρ sin ϕp , y y ¼ r sin ϕ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
then p1z ¼ k21 p2x p2y ¼ k21 p2ρ and the received pressure can be written as
8 9
ð> ð
< 2π ð
1 >
=
ωρ1 v0 RP;P
12 expð2ip1z DÞexp irpρ cos ϕ ϕp
preflt ðx; ωÞ ¼ pρ dpρ dϕp dSðyÞ:
4π 2 >: p1z >
;
S ϕp ¼0 pρ ¼0
ð9:49Þ
P;P
Since R12 and p1z are functions of pρ only, and since
2ðπ
exp irpρ cos ϕ ϕp dϕp ¼ 2πJ 0 pρ r ; ð9:50Þ
0
However, the area integration can also be written in polar coordinates as dSðyÞ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ ρ0 dρ0 dϕ with r ¼ ρ20 þ ρ2 2ρρ0 cos ϕ (see Fig. 9.7) and also
ð
2π
J 0 pρ r dϕ ¼ 2π J 0 pρ ρ J 0 pρ ρ0 ð9:52Þ
ϕ¼0
400 9 Material Properties and System Function Determination
ρ0
φ r x, x ⬘
ρ
x (x, y, 0)
ð9:53Þ
Furthermore,
ða
a J 1 pρ a
J 0 pρ ρ0 ρ0dρ0 ¼ ð9:54Þ
pρ
ρ0 ¼0
so that
ð
1
RP;P
12 J 0 pρ ρ J 1 pρ a
p reflt
ðx; ωÞ ¼ ωρ1 v0 a expð2ip1z DÞdpρ: ð9:55Þ
p1z
pρ ¼0
If we now compute the average of this pressure over the face of the transducer and
normalized as shown in Eq. (9.29), with Sf ¼ S ¼ πa2 and dSðxÞ ¼ 2πρdρ, we find
8 9
ða > ð
< 1 P;P >
=
2kp1 R12 J 0 pρ ρ J 1 pρ a
pave ðωÞ ¼ expð2ip1z DÞdpρ ρdρ: ð9:56Þ
a >
: p1z >
;
ρ¼0 pρ ¼0
But, again
ða
a J 1 pρ a
J 0 pρ ρ ρdρ ¼ ð9:57Þ
pρ
ρ¼0
9.2 LTI Models 401
ð
1 "
2 #
RP;P J 1 pρ a
pave ðωÞ ¼ 2kp1 a 12
expð2ip1z DÞdpρ : ð9:58Þ
p1z pρ a
pρ ¼0
This expression is very similar to the expression for the average pressure of a
transducer radiating into a fluid without interfaces present, where the average is
taken over a circular area in the fluid of the same size as the transducer and coaxial
to it (see problem 9.7) and where the resulting integral is called the diffraction
correction integral. Thus, Eq. (9.58) is the analogous diffraction correction integral
for the reflection off a plane surface. It can be evaluated numerically in this exact
form, but we will show that it is also possible to get an approximate analytical
expression for this integral which is much simpler to use. To begin, we use the
following Bessel function relations:
2
J 1 pρ a 1 1
¼ J 0 pρ a J 1 pρ a þ J 2 pρ a J 1 pρ a
pρ a 2 2
ð
π=2
¼ 1=π J 1 2pρ a cos θ ð cos θ cos 3θÞdθ
0
ð
π=2
¼ 4=π J 1 2pρ a cos θ sin 2 θ cos θdθ: ð9:59Þ
0
ð
π=2
8kp1 a
pave ðωÞ ¼ sin 2 θ cos θ
π
8 0 93
>
< 1 ð P;P
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi >
=
R12 J 1 2pρ a cos θ 7
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi exp 2i kp1 pρ D dpρ 5dθ:
2 2
ð9:60Þ
>
: kp1 pρ
2 2 >
;
p ¼0
ρ
Although the form of Eq. (9.60) appears more complex than our original expression
(9.58)), if we now make the paraxial assumption in p-space, i.e. assume that
(Eq.
pρ kp1 , then
P;P
RP;P
12 ffi R12 ð0 Þ " #
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p2ρ ð9:61Þ
kp1 pρ ffi kp1 1 2
2 2
2kp1
402 9 Material Properties and System Function Determination
and
ð
π=2
8a P;P
pave ðωÞ ¼ R12 ð0 Þexp 2ikp1 D sin 2 θ cos θ
π
8 0 9
>
< 1ð >
h i =
exp ip2ρ D=kp1 J 1 2pρ a cos θ dpρ dθ: ð9:62Þ
>
: >
;
pρ ¼0
[Note: in the amplitude part of the integrand we have again kept only the leading
term in the paraxial approximation for the square root but have kept both terms for
the phase]. The pρ integration can be done exactly since [11]
ð
1
1
exp iAx2 J 1 ðBxÞdx ¼ 1 exp iB2 =4A ; ð9:63Þ
B
0
so that if we let A ¼ D=kp1 and B ¼ 2a cos θ in Eq. (9.63) and use it in Eq. (9.62), we
have the average pressure expressed in terms of a finite integral as
ð
π=2
4
pave ðωÞ ¼ RP;P
12 ð0 Þexp 2ik p1 D sin 2 θ 1 exp ikp1 a2 cos 2 θ=D dθ:
π
0
ð9:64Þ
The first term in the integral of Eq. (9.64) can be integrated directly since
ð
π=2
ðπ
exp ikp1 a2 cos u=2D du ¼ πJ 0 kp1 a2 =2D
0
ð9:68Þ
ðπ
exp ikp1 a2 cos u=2D cos udu ¼ iπJ 1 kp1 a2 =2D :
0
fs
pave ðωÞ ¼ RP;P
12 ð0 Þexp 2ik p1 D
ð9:69Þ
1 exp ikp1 a2 =2D J 0 kp1 a2 =2D iJ 1 kp1 a2 =2D ;
ρ2 cp2 ρ1 cp1
RP;P
12 ð0 Þ ¼ : ð9:70Þ
ρ2 cp2 þ ρ1 cp1
Although this rather long derivation relied on the use of the paraxial approximation
to evaluate the pρ integration in Eq. (9.60), Rogers and Van Buren [12] have shown
that our final result, Eq. (9.69), actually holds for all distances D as long as the
frequency is large, i.e. kp1 a >> 1. Numerical evaluation of the exact expression has
confirmed this and also verified in fact that Eq. (9.69) is valid under even less
stringent conditions [13–15].
It is useful to consider the response from the plane front surface in Eq. (9.69)
under the following conditions:
(1) D a. In this case, Eq. (9.69) reduces to
fs
pave ðωÞ ¼ RP;P
12 ð0 Þexp 2ik p1 D ð9:71Þ
(2) D
a. In contrast, in this case
ikp1 a2
fs
pave ðωÞ ¼ RP;P
12 ð0 Þexp 2ik p1 D : ð9:72Þ
4D
for the fluid-solid-fluid path followed by a P-wave, where we have set D ¼ D1 , and
the other quantities in Eq. (9.73) are defined as
0
ðx yÞ⊥ ¼ ðx x0 Þex þ y y ey
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð9:74Þ
pαz ¼ k2pα p2ρ ðα ¼ 1, 2Þ:
ð
π=2
8a P;P P;P P;P
ave ðωÞ
pbs ¼ T 12 ð0 ÞR21 ð0 ÞT 21 ð0 Þexp 2ikp1 D1 þ 2ikp2 D2 sin 2 θ cos θ
π
0
8 9
> ð
< 1 >
h i =
exp ipρ D1 =kp1 þ D2 =kp2 J 1 2pρ a cos θ dpρ dθ;
2
>
: >
;
pρ ¼0
ð9:75Þ
ave ðωÞ is the average normalized received pressure at the transducer due to
where pbs
the waves which have been reflected once from the back surface (bs) of the solid.
The integral in Eq. (9.75) can be integrated, in exactly the same fashion as before,
so that we give here only the final result which can be written as
P;P P;P P:P
ave ðωÞ ¼ T 12 ð0 ÞR21 ð0 ÞT 21 ð0 Þexp 2ik p1 D1 þ 2ik p2 D2
pbs
1 exp ikp1 a2 =2D J 0 kp1 a2 =2D iJ 1 kp1 a2 =2D ; ð9:76Þ
where
cp2
D ¼ D1 þ D2 : ð9:77Þ
cp1
Note that the transmission and reflection coefficients in Eqs. (9.69) and (9.76) can
P;P
be written in terms of stress/pressure (for T12 ), pressure/pressure (for RP;P
21 and
P;P P;P
R12 ) and pressure/stress (for T21 ) ratios, respectively, (see Chap. 6) or all of these
9.2 LTI Models 405
coefficients can be written in terms of velocity ratios since the product of all the
factors that relate these coefficients is just unity. Since many authors prefer to use
velocity ratios for these coefficients, we will write them in terms of those ratios
henceforth in this Chapter, noting that in the case of the reflection coefficients
present here there is no difference between the use of pressure or velocity ratios so
the specification can be omitted.
Comparing Eqs. (9.69) and (9.76), we see that both the front and back surface
responses are of similar form which we rewrite here as
fs
pave ðωÞ ¼ RP;P
12 ð0 Þexp 2ik p1 D Dp k p1 a =2D
2
ð9:78Þ
where the plane wave transmission, reflection and propagation terms are shown
explicitly, and Dp is a diffraction correction term, where
Dp kp1 a2 =2D ¼ 1 exp ikp1 a2 =2D J 0 kp1 a2 =2D iJ 1 kp1 a2 =2D : ð9:79Þ
These results can also be used in the case of contact P-wave transducer setups as
long as they involve calibration setups where the paraxial approximation is valid,
which is true for the types of setups discussed here. One can see this by going back
to our starting point, which was the angular plane wave spectrum representation
model for a piston immersion transducer, Eq. (9.45), given by
8 9
ð < þ1
ð þ1
ð =
ωρ v 0 exp ½ ip ð x y Þ
pinc ðx; ωÞ ¼ 12 1
dpx dpy dSðyÞ: ð9:80Þ
4π : p1z ;
S 1 1
ð9:81Þ
This incident wave field for most NDE transducers is a highly collimated field
traveling normal to the face of the transducer (the z-direction) so the paraxial
approximation is well satisfied. In this case p1z ffi kp1 , so that comparing
Eqs. (9.80) and (9.81) we see that
showing that the transducer wave field satisfies the ordinary plane wave relation-
ship. In the paraxial approximation the same plane wave relationship is true for the
reflected and transmitted wave fields. Thus, if we consider the contact transducer
setup of Fig. 9.6b, for example, where a contact transducer reflects off the back
surface of a block, we can rewrite Eq. (9.78) for the back surface response in terms
of an averaged received velocity wave field, vr(ω), on the face of the receiving
transducer (the velocity component here is in a direction opposite to the outward
normal of the transducer face) as
P;P v P;P
vr ðωÞ ¼ v0 v T P;P
12 ð0 Þ R21 ð0 Þ T 21 ð0 Þ exp 2ik p1 D1 þ 2ik p2 D2 Dp kp1 a =2D ;
2
ð9:83Þ
where
ð
1
vr ðωÞ ¼ vr ðx; ωÞdSðxÞ ð9:84Þ
SR
SR
is the average received velocity. To use this expression, however, we must replace
v0 for the immersion case with an appropriate driving term for the contact trans-
ducer. This is easy within the paraxial approximation, since the well collimated
velocity wave field of a contact P-wave transducer is just given by (see Eq. (8.439),
which is an expression for the displacement):
ð
iωp0 exp kp1 r
vz ðx; ωÞ ¼ dS; ð9:85Þ
2πρ1 c2p1 r
ST
v r ð ωÞ P;P v P;P
¼ v T P;P
12 ð0 Þ R21 ð0 Þ T 21 ð0 Þ exp 2ik p1 D1 þ 2ikp2 D2 Dp kp1 a =2D :
2
p0 =ρ1 cp1
ð9:87Þ
Equation (9.87) can then be used directly in an LTI model for the received voltage
in this contact setup such as those given in Eq. (9.43) or Eq. (9.44).
9.2 LTI Models 407
If we consider the geometry shown in Fig. 9.6 again and use the general model
formulation of Sect. 9.2 for both the front and back surface responses, we have for
the received voltages
where αw and αs are the attenuation coefficients for the water and the solid,
respectively and we have assumed that both the front and back surface responses
are measured with the same transducer and at the same system settings so that the
efficiency factor is the same in both cases. In principle, we could obtain the
attenuation of the solid from Eq. (9.88) by simply dividing the magnitudes of the
back and front surface responses to obtain
where
bs kp1 a2
P;P
jBðωÞj ¼ V 0 ðωÞR12 Dp
2D1
ð9:90Þ
fs kp1 a2
P;P P;P P;P
jFðωÞj ¼ V 0 ðωÞT 12 R21 T 21 Dp ;
2D
assuming that the densities and wave speeds of the water and solid are known and
the distances appearing in Eq. (9.90) have been measured. However, at high and
low frequencies, this simple division process, which is really a deconvolution
carried out in the frequency domain [17], fails when the magnitudes of both the
B and F terms become small (since one is then simply dividing noise by noise),
producing wildly varying values that can mask the values in the region where the
simple division does produce reliable results. This is a standard problem encoun-
tered frequently in deconvolution procedures and there are a number of signal
processing tools available to deal with it. In ultrasonic NDE, the approach most
commonly used has been to simply modify the division process through a Wiener
filter. Then, Eq. (9.89) is replaced instead by
( )
jBðωÞj j F ð ωÞ j 2
exp½2αs D2 ¼
jFðωÞj jFðωÞj2 þ ε2
j B ð ωÞ j j F ð ωÞ j
¼ ; ð9:91Þ
j F ð ωÞ j 2 þ ε 2
where ε2 is a small constant that is chosen to stabilize the division process. Often,
n ε2
o
is simply taken as a small, fixed percentage of jFj2 such as ε2 ¼ :05 max jFj2 .
When the magnitudes of F and B are not small in comparison to the noise, then the
408 9 Material Properties and System Function Determination
0.0090
0.0060
Np/mm
0.0030 transducer
bandwidth
0.000
0.0 6.0 12.0 18.0
Frequency (MHz)
Fig. 9.8 Attenuation coefficient as measured for an aluminum block and the best-fit straight line
for values in the bandwidth of the transducer (approximately 6–10 MHz)
Wiener filter (bracket term on the right side of Eq. (9.91)) is essentially equal to one
and Eq. (9.91) is equivalent to the straightforward division process of Eq. (9.89).
When F and B are small, however, the effect of the filter is to drive the right side of
Eq. (9.91) to zero smoothly as can be seen by the second equivalent form given in
Eq. (9.91). Solving for αs from Eq. (9.91) in the range of frequencies where B and
F are not small, one can then usually fit a simple attenuation curve (as a function of
frequency) to these results. An example of the results of using such an approach for
a block of aluminum is shown in Fig. 9.8 where over a frequency range of
6–10 MHz (roughly the bandwidth of the transducer), the attenuation variation
could be fit to a linear function as
A model of the reflections of a shear wave contact transducer beam from the planar
surfaces of a specimen can be used to determine the shear wave attenuation in the
same manner as described previously for P-waves, but these details will not be
discussed further here (see problem 9.10).
In obtaining the attenuation it was not necessary to know the efficiency factor since
it canceled out during the process of deconvolution. Here, we want to use similar
deconvolution procedures to also determine the efficiency factor itself. If we use the
front surface response in Eq. (9.33), we can solve for the efficiency factor directly,
since the attenuation in the water is known (see Eq. (9.2)) to obtain
V 0fs ðωÞ
β I ð ωÞ ¼
expð2αw D1 Þpave
fs
ð ωÞ
V 0fs ðωÞ
¼ : ð9:93Þ
expð2αw D1 Þexp 2ikp1 D1 RP;P
12 Dp kp1 a =2D1
2
Again, however, this is a deconvolution process so that we resort to the Wiener filter
and replace Eq. (9.93) by
*
expð2αw D1 Þexp 2ikp1 D1 RP;P
12 Dp kp1 a =2D1
2
β I ð ωÞ ¼ V 0fs ðωÞ ; ð9:94Þ
expð2αw D1 Þexp 2ikp1 D1 RP;P Dp kp1 a2 =2D1 2 þ ε2
12
0.060
0.040
Magnitude
0.020
0.000
0.0 10.0 20.0 30.0
Frequency (MHz)
Fig. 9.9 Efficiency factors calculated for a 10 MHz, 0.25 in. diameter probe using the front
surface (solid line) and back surface (dotted line) of an aluminum block with no attenuation
corrections
0.060
0.040
Magnitude
0.020
0.000
0.0 10.0 20.0 30.0
Frequency (MHz)
Fig. 9.10 Efficiency factors calculated for the same transducer and block of Fig. 9.7 using the
front surface (solid line) and back surface (dotted line) with attenuation of the water and aluminum
included
9.2 LTI Models 411
0.0350
0.0233
Magnitude
0.0117
0.000
0.0 6.7 13.3 20.0
Frequency (MHz)
Fig. 9.11 Efficiency factor versus frequency for a 10 MHz, 0.25 in. diameter transducer using the
front surface response of a specimen made of (a) aluminum (solid line), (b) titanium (long dashes),
and (c) steel (short dashes) (Reprinted with permission from Schmerr, L.W., Song, S.J., and
H. Zhang, “Model-based calibration of ultrasonic system responses for quantitative measure-
ments,” in Nondestructive Characterization of Materials, Eds. R.E. Green, K.J. Kozaczek, and
C.O. Ruud, Plenun Press, New York, 111-118, 1994)
show that the results obtained are again the same, regardless of the setup chosen if
one carefully includes all the contributions to the total measured response.
Figure 9.11 shows some additional results of using Eq. (9.94) to measure the
efficiency factor of a planar 10 MHz, 1/4 in. diameter commercial transducer. The
multiple measurements of βI(ω) were obtained in these cases by using the echoes
(at the same system settings) from planar front surfaces of aluminum, titanium, and
steel specimens. In each case the contribution due to attenuation in the water was
included. As can be seen from Fig. 9.11, there is very good agreement between all
these measurements, demonstrating there is transferability of such results across
materials and that it is not necessary to use the same material in the reference
calibration setup as in, say, an ultrasonic flaw measurement, to obtain βI(ω).
Although all the discussion in this section has treated the calculation of effi-
ciency factors for a single planar transducer in a pulse-echo setup, exactly the same
procedures can be used for focused transducers and for a combination of trans-
ducers in a pitch-catch experiment. Again, all that is needed in these more general
cases is an explicit model for the averaged received pressure. One such model
which could be used for two planar transducers of the same radius is the diffraction
correction integral of problem 9.7. Other models, such as the pitch-catch scattering
from a small sphere in a fluid, can easily be obtained from the results given in later
chapters.
412 9 Material Properties and System Function Determination
Where aγi (ω) is the attenuation coefficient for a wave of type γ in the ith media
and di is the total path length in the ith media. To obtain the frequency
dependent attenuation coefficient for each media, we perform reference
calibration experiments on specimens of the same materials present in the
ultrasonic measurement and use a model of the average pressure received in
the reference setup to extract the attenuation coefficient through a
deconvolution procedure. The reference configurations used in this Chapter
involved the front- and back-plane surface of a body whose attenuation was to
be found. However, other reference configurations could be used instead.
(continued)
9.4 About the Literature 413
The system function is a factor that contains all the electrical and electrome-
chanical elements of an ultrasonic measurement system (pulser/receiver,
cabling, transducers). This factor is obtained in a reference calibration exper-
iment. What is needed to determine the system function is the measured
voltage response, V0(ω), in the calibration setup and an explicit model of the
acoustic/elastic transfer function, Ta(ω), (including material attenuation).
Then the system function is obtained through a deconvolution procedure,
ideally of the form
V 0 ð ωÞ
s ð ωÞ ¼
T a ð ωÞ
The book by Bhatia [6] is a good reference for more details on the sources of sound
attenuation in solids, liquids, and gases. Fitting and Adler [5] also provide more
information on ultrasonic attenuation specifically related to NDE applications.
The explicit expression (see Eq. (9.79)) we obtained for the diffraction correc-
tion integral is usable under a wide range of conditions, as shown by a number of
numerical studies (see Tang et al. [15], for example. Also see Rhyne [32]). As
discussed, this expression can also be used for contact transducers radiating directly
into elastic media [27, 28].
Deconvolution procedures to obtain ultrasonic system responses and attenuation
measurements can be accomplished both in the frequency domain, as done here,
and in the time domain as done, for example, by Tang et al. [18] and Cassereau and
Guyomar [29, 30] (see Chen and Sin [17] for another example of this approach and
some further information on Wiener filters). In either domain, care must be taken
and some sort of filtering used since the deconvolution process itself is inherently
ill-posed [31].
414 9 Material Properties and System Function Determination
9.5 Problems
Plot the behavior of the Wiener filter, based on this response, given by
j Rð f Þ j 2
Wð f Þ ¼ :
jRð f Þj2 þ c jRmax j2
(where Rmax is the maximum value of R( f )) for values of the “noise” constant
c of 0.5, 0.1, and 0.01, respectively. Based on these results, what factors might
influence your choice of c in a real experiment?
9.5 Many textbooks refer to the Wiener filter as an “optimal” filter. In what sense
is this filter optimal?
9.6 In this Chapter we found that when D
a the average pressure received from
a plane front surface of a material was given by Eq. (9.49), i.e.
ikp1 a2
fs
pave ðωÞ ¼ ρ1 cp1 v0 RP;P
12 ð 0 Þ exp 2ik p1 D :
4D
Consider now the setup shown in Fig. P9.1a where we monitor the multiple
pulses reflected off the faces of a block of material as shown in Fig. P9.1b. If
D >> a, then Eq. (P9.1) holds for the first received echo, and we expect that
the nth echo can similarly be described by
ikp1 a2
V 0 ðωÞ ¼ βðωÞ exp 2i n kp1 D
n
expð2α n DÞ: ðP9:2Þ
4nD
If we take α here to be the average (i.e. frequency independent) attenuation of
the material and invert Eq. (P9.2) back into the time domain, we would find
f t 2nD=cp1 expð2αnDÞ
v0 ðtÞ ¼
n
; ðP9:3Þ
nD
where
ð
þ1
iωa2 βðωÞ
f ðtÞ ¼ 1=2π expðiωtÞdω: ðP9:4Þ
4cp1
1
Using Eq. (P9.3) and letting vn ¼ max v0n ðtÞ be the maximum amplitude of
these echoes, determine an expression for the average attenuation coefficient
(in dB/unit length) in terms of the amplitude ratio and any two successive
echoes. Explain the physical meaning of the terms that appear in this expression.
9.7 Consider a plane piston transducer radiating into a fluid as shown in Fig. P9.2.
Following the same steps used in this chapter, show that the average pressure
over a circular area of radius a at a distance D is given by (see Eq. (9.37))
8 9
ð
π=2 > ð
< 1
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi >
=
8ωρ1 v0 a J 1 2pρ a cos θ
pave ðωÞ ¼ sin 2 θ cos θ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi exp i k2p1 p2ρ D dpρ dθ;
π >
: k 2 p2 >
;
0 pρ ¼0 p1 ρ
which is identical with Eq. (9.37) except that the reflection coefficient is now
absent and we have 2D ! D. In this case the pρ integration can be performed
exactly, since
ð
1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
J 1 2pρ a cos θ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi exp i k2p1 p2ρ D dpρ
pρ ¼0
k2p1 p2ρ
1 h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
¼ exp ikp1 D exp ikp1 D2 þ 4a2 cos 2 θ ;
2akp1 cos θ
so that we find
8 9
< ð
π=2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi =
4
pave ðωÞ ¼ ρ1 cp1 v0 exp ikp1 D exp ikp1 D2 þ 4a2 cos 2 θ sin 2 θdθ ;
: π ;
0
References
26. M. Fink, W.A. Kuperman, J.P. Montagner, A. Tourin (eds.), Imaging of Complex Media with
Acoustic and Seismic Waves (Springer, Heidelberg, 2002)
27. A. Sedov, L.W. Schmerr, Elastodynamic diffraction correction integrals. J. Acoust. Soc.
Am. 86, 2000–2006 (1989)
28. D.H. Green, H.F. Wang, Shear wave diffraction loss for circular plane-polarized source and
receiver. J. Acoust. Soc. Am. 90, 2697–2704 (1991)
29. D. Cassereau, D. Guyomar, Time deconvolution of diffraction effects—application to calibra-
tion and prediction of transducer waveforms. J. Acoust. Soc. Am. 84, 1073–1085 (1988)
30. D. Cassereau, D. Guyomar, Computation of the impulse diffraction of any obstacle by impulse
ray modeling—prediction of the signal distortions. J. Acoust. Soc. Am. 84, 1504–1516 (1988)
31. A.R. Davies, On the maximum likelihood regularization of Fredholm convolution equations of
the first kind, in Treatment of Integral Equations by Numerical Means, ed. by C.T.H. Baker,
G.F. Miller (New York, Academic Press, 1982), pp. 95–105
32. T.L. Rhyne, Radiation coupling of a disk to a plane and back to a disk: an exact solution.
J. Acoust. Soc. Am. 61, 318–324 (1977)
Chapter 10
Flaw Scattering
When a beam of ultrasound interacts with a flaw in a material, scattered waves are
generated by the flaw, traveling in all directions. The distribution of the scattered
waves is, of course, strongly dependent on the geometrical and material properties
of the flaw. In this Chapter, we will characterize the scattering responses of flaws in
terms of their far field scattering amplitudes and describe both exact and approx-
imate methods to calculate scattering amplitudes. The scattering amplitude is a
quantity of fundamental interest since it will be shown later that it is precisely this
quantity that completely characterizes the flaw response in a Thompson-Gray
model of an ultrasonic system.
First, consider the case where the material in which the flaw is embedded is
modeled as an infinite fluid and where an incident wave of pressure amplitude p0
is present (Fig. 10.1). The flaw itself can be of arbitrary shape or material, but is
assumed to be volumetric (non-crack-like) in nature. As shown in Chap. 5, the
pressure of the scattered waves, which are assumed to satisfy the Sommerfeld
radiation conditions, can be written in integral form in this case as (Eq. (5.24)):
ð
pscatt ðy; ωÞ ¼ ½pðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs Þ
S
Gðxs ; y; ωÞ∂pðxs ; ωÞ=∂nðxs ÞdSðxs Þ: ð10:1Þ
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_10) contains supplementary material, which is available to authorized users.
n(xs)
es
xs
O
expðik r s Þ
G ¼ ½expðik xs es Þ
4πrs
ð10:2Þ
∂G expðik r s Þ
¼ ½ikðes nÞexpðik xs es Þ ;
∂nðxs Þ 4πr s
expðikr s Þ
pscatt ðy; ωÞ ¼ p0 Aðei ; es Þ ; ð10:4Þ
rs
in terms of the normalized pressure p~ ¼ p=p0 and its normal derivative. As can be
seen from Eq. (10.4), the scattering amplitude defined here has the dimensions of a
length. Some authors, however, define the scattering amplitude somewhat differ-
ently by writing
~ ðei ; es Þ expðikr s Þ ;
pscatt ðy; ωÞ ¼ p0 A
kr s
From Chap. 5, we found that for a crack-like scatterer, the scattered pressure could
be written in terms of jumps in the pressure and its normal derivative on the
scatterer surface as (Eq. (5.30)):
ð
pscatt ðy; ωÞ ¼ ½Gðxs ; y; ωÞΔð∂pðxs ; ωÞ=∂nðxs ÞÞ
S
Δpðxs ; ωÞ∂Gðxs ; y; ωÞ=∂nðxs ÞdSðxs Þ: ð10:6Þ
If, again, we consider the far field of the flaw and apply Eq. (10.2), the far field
scattered pressure can be written in the spherical wave form of Eq. (10.4) where
now the scattering amplitude A(ei; es) is given by (see Fig. 10.2)
ð
1
Aðei ; es Þ ¼ ½Δð∂~
p =∂nÞ þ ikðes nÞΔ~
p expðik xs es ÞdS: ð10:7Þ
4π
S
For the crack-like scatterer, Eq. (10.7) shows that we need to be able to find the
jump in both the (normalized) pressure and its normal derivative on the flaw
surface. Again, to obtain these quantities we must in general solve the appropriate
flaw scattering boundary value problem.
n(xs)
es
xs
O
crack surface
422 10 Flaw Scattering
Consider now the case of a general volumetric flaw in an elastic solid, and the
scattered displacement components due to an incident wave of displacement ampli-
tude U0. These scattered components were obtained previously (Eq. (5.54)):
ð
un ðy; ωÞ ¼ Ckplj nk ðxs Þup ðxs ; ωÞ∂Gln ðxs ; y; ωÞ=∂xj
scatt
S
Cklpj nk ðxs ÞGln ðxs ; y; ωÞ∂up ðxs ; ωÞ=∂xj dSðxs Þ; ð10:8Þ
In the far field of the flaw, the fundamental solution Gln and its derivatives again can
be approximated (Eqs. (4.86) and (4.87)) and we find
ð
esl esn expðik1 r s Þ
unscatt ðy; ωÞ ¼ 2
τkl nk þ ik1 esj Ckplj nk up expðik1 xs es ÞdSðxs Þ
4πρc1 rs
S ð
ðδln esl esn Þexpðik2 rs Þ
τkl nk þ ik2 esj Ckplj nk up expðik2 xs es ÞdSðxs Þ;
4πρc22 rs
S
ð10:10Þ
where esn are the components of the unit vector es in the scattered wave direction.
Equation (10.10), therefore, can be expressed in the general scalar and vector forms
expðik r Þ
eiβ ; esP
1 s
unscatt ðy; ωÞ ¼ U 0 AP;β
n
rs
expðik r Þ
þ U 0 AnS;β eiβ ; esS
2 s
ðβ ¼ P, SÞ ð10:11aÞ
rs
or, equivalently,
expðik r Þ
uscatt ðy; ωÞ ¼ U0 AP;β eiβ ; esP
1 s
rs
expðik r Þ
þ U 0 AS;β eiβ ; esS
2 s
ðβ ¼ P, SÞ; ð10:11bÞ
rs
10.2 Far Field Scattering Amplitude in an Elastic Solid 423
where Aα;β
n are the components of the far-field scattering amplitude of a wave of
type α ðα ¼ P, SÞ due to an incident wave of type β ðβ ¼ P, SÞ and eiα , esα ðα ¼ P, SÞ
are unit vectors in the incident and scattered wave directions, respectively, of
type α. We use a superscript on these unit vectors to recognize the fact that both
the incident and scattered P- and S-waves that are measured in an ultrasonic system
may have directions that are different for each type of mode involved.
The scattering amplitude components are given by
P ðh i
β P eslP esn
AP;β
n ei ;es ¼ 2
~τ kl nk þ ik1 esjP Ckplj nk u~p exp ik1 xs esP dSðxs Þ
4πρc1
S
ðno sum on sÞ ðβ ¼ P,SÞ ð10:12aÞ
for the scattered S-waves, where u~k ¼ uk =U 0 and ~τ kl ¼ τkl =U 0 are displacements
and stresses, respectively, divided by the magnitude of the incident displacement.
Again, although the direction eαi associated with the incident wave does not appear
directly in these scattering amplitude expressions this dependency is shown explic-
itly in the arguments of the Anα;β , as done for the fluid case, and the dependency on
the frequency, ω, is omitted. Generally in this book we will write the scattered
S-waves in the form of Eq. (10.12b) but as shown in Chap. 4 we can introduce a set
of (s, t, v) coordinates and use Eq. (4.90) to separately write the S-wave components
in the (t, v) directions, (St, Sv), as
S S ð
β Sη eηlη eηnη h i
AnSη ;β
S
ei ; es ¼ 2
~τ kl nk þ ik2 esjη Ckplj nk u~p exp ik2 xs eSs η dSðxs Þ
4πρc2
S
ðβ ¼ P, SÞ ðη ¼ t, vÞ;
ð10:12cÞ
which we see has an identical form to the P-wave component (Eq. (10.12a)). We
can also write these scattering amplitude expressions in terms of the displacement
components and their derivatives only. We can do this by writing the stresses as
~τ kl ¼ Cklpj ∂~
u p =∂xj ¼ Clkpj ∂~
u p =∂xj
424 10 Flaw Scattering
by using the symmetry properties of the elastic constant tensor and relabeling
the dummy indices in the manner shown. Then the scattering amplitude for the
scattered P-wave becomes
ð
eslP esn
P
Clkpj
AnP;β eiβ ; esP ¼ u p =∂xj nk þ ik1 eskP np u~j exp ik1 xs esP dSðxs Þ
∂~
4πρc21
S
ðno sum on sÞ ðβ ¼ P, SÞ
ð10:13aÞ
From Eqs. (10.12a, 10.12b) and (10.13a, 10.13b), it follows that both the P- and S-
wave scattering amplitudes can be written in terms of the components of a vector
quantity, fα;β, in the form
AP;β
n eiβ ; esP ¼ eslP esn
P P;β
fl
S;β ð10:14aÞ
AnS;β eiβ ; esS ¼ δln eslS esn
S
fl ðβ ¼ P, SÞ:
where
ðh i
1
f lα;β ¼ 2
~τ kl nk þ ikα esjα Ckplj nk u~p exp ikα xs esα dSðxs Þ
4πρcα ð10:15aÞ
S
ðno sum on sÞ ðα ¼ P, SÞ ðβ ¼ P, SÞ:
f S;β
β
ei
S
es AS;β
For a crack-like scatterer in an elastic solid, we previously showed that the scattered
displacements were given by (Eq. (5.57)):
ð
un ðy; ωÞ ¼ Cklij nk ðxs ÞΔul ðxs ; ωÞ∂Gin ðxs ; y; ωÞ=∂xj dS xs
scatt
S ð
Δτkl ðxs ; ωÞnk ðxs ÞGln ðxs ; y; ωÞ dS xs ð10:17Þ
S
426 10 Flaw Scattering
in terms of displacement and traction discontinuities on the crack surface. In the far
field, the asymptotic form of Gln and its derivatives then yield
ð
esl esn expðik1 rs Þ
unscatt ðy; ωÞ ¼ 2
Δτkl nk þ ik1 esj Ckplj nk Δup expðik1 xs es ÞdSðxs Þ
4πρc1 r s
S
ð
ðδln esl esn Þexpðik2 r s Þ
Δτkl nk þ ik2 esj Ckplj nk Δup expðik2 xs es ÞdSðxs Þ
4πρc22 r s
S
which has exactly the same form as Eq. (10.10) for the volumetric flaw case if we
simply make the replacements τkl ! Δτkl , up ! Δup . Thus, all the previous results
that follow from Eq. (10.10) for the volumetric flaw case also hold for the crack-like
flaw case with this same replacement. In particular, the scattering amplitudes
become
P ðh i
β P eslP esn
AP;β
n ei ; es ¼ 2
Δ~τ kl nk þ ik1 esjP Ckplj nk Δ~
u p exp ik1 xs esP dSðxs Þ
4πρc1
S
ðno sum on sÞ ðβ ¼ P, SÞ
ð10:19aÞ
for the scattered S-waves. Similarly, the components of the f α; β vector are
ðh i
1
f lα;β ¼ 2
Δ~τ kl nk þ ikα esjα Ckplj nk Δ~
u p exp ikα xs esα dSðxs Þ
4πρcα ð10:20Þ
S
ðno sum on sÞ ðα ¼ P, SÞ ðβ ¼ P, SÞ:
As seen in Sect. 10.1 of this chapter, obtaining the scattering amplitudes normally
requires solving an appropriate boundary value problem. However, at high fre-
quencies, by making some strong assumptions about the scattering process we can
10.3 Approximate Scattering Solutions: Fluid Model 427
obtain explicit approximate expressions for the scattering amplitude for both
volumetric and crack-like flaw cases. This high frequency approximation will be
called the Kirchhoff approximation since the arguments used here for acoustic
wave scattering problems are similar to those we discussed previously in
Chaps. 6 and 8 for the transmission of waves through an interface. In the case of
electromagnetic scattering problems, the same set of assumptions is called the
physical optics approximation.
Also, explicit scattering amplitude expressions can be obtained for volumetric
flaws (inclusions) whose material properties do not differ substantially from that of
the surrounding host material by assuming that the incident wave is only slightly
affected by the presence of the flaw. This weak scattering approximation will be
called here the Born approximation in recognition of its close relationship to the
method used by Born for potential scattering problems.
Both the Kirchhoff and Born approximations are very useful because they show
how the material and geometrical properties of a flaw are related directly to the
behavior of the far field scattering amplitude. In fact, these approximate solutions
will later allow us to develop a number of quantitative model-based flaw sizing
algorithms (see Chap. 15).
can strike the flaw surface directly, this incident wave produces a reflected wave
which will be approximately the same as that produced at a planar surface whose
normal coincides with that of the surface S at xs (Fig. 10.4). On the remainder of S,
which is in the “shadow” of the incident wave, it is assumed that the shadowing is
perfect, i.e. the fields p and ∂p=∂n are identically zero. With these assumptions,
then the surface values of p and ∂p=∂n are given by
p ¼ pinc þ preflt
on Slit
∂p=∂n ¼ ∂pinc =∂n þ ∂preflt =∂n ; ð10:21Þ
p ¼ ∂p=∂n ¼ 0g otherwise on S
For the equivalent reflected wave expressions, we recall the expression for the
reflected pressure from a general planar interface is given, for an incident wave of
unit amplitude, by Eq. (6.100) as (see Fig. 10.4):
where (Eq. (6.93)) er ¼ ei 2ðei nÞn. Thus, the normal derivative of the pressure
is also given by
Placing Eqs. (10.22) and (10.25) into Eq. (10.5), the scattering amplitude becomes
ð
ik
Aðei ; es Þ ¼ ðei þ es Þ n þ Rp ðes ei Þ n exp½ikðei es Þ xs dSðxs Þ:
4π
Slit
ð10:26Þ
10.3 Approximate Scattering Solutions: Fluid Model 429
For the special case of a void Rp ¼ 1 the scattering amplitude, Av, is
ð
ik
Av ð e i ; e s Þ ¼ ðei nÞexp½ikðei es Þ xs dSðxs Þ: ð10:27Þ
2π
Slit
In contrast, the scattering amplitude for a rigid inclusion Rp ¼ 1 , Ar, is
ð
ik
Ar ðei ; es Þ ¼ ðes nÞexp½ikðei es Þ xs dSðxs Þ: ð10:28Þ
2π
Slit
In back scatter (pulse-echo) es ¼ ei and both the void and rigid scatterer cases can
be combined as
ð
ikσ β
Aβ ðei ; ei Þ ¼ ðei nÞexp½2ikðei xs ÞdSðxs Þ ðβ ¼ v, r Þ; ð10:29Þ
2π
Slit
where
1 β¼r
σβ ¼ : ð10:30Þ
1 β¼v
The pulse-echo scattering amplitude of Eq. (10.29) can be evaluated explicitly for
the special case of an ellipsoidal void or rigid scatterer with semi-major axes aj
ðj ¼ 1, 2, 3Þ (Fig. 10.5) when the incident wave direction is along one of the
principal axes of the ellipsoid. This problem includes the special case of a spherical
scatterer which is an important canonical geometry to consider. In the spherical
flaw case, however, there is no restriction on the incident wave direction since
all directions are principal axes. For the general ellipsoid, we first extend the surface
Slit to infinity along the cylindrical surface Se whose normal, n, satisfies ei n ¼ 0,
Fig. 10.6 Geometry for calculating the pulse-echo response of an ellipsoidal void or rigid
scatterer in the Kirchhoff approximation when the incident direction is along a principal axis
as shown in Fig. 10.6 and we enclose the end by a surface S1 at infinity. We then
can write
ð
ikσ β
Aβ ðei ; ei Þ ¼ ðei nÞexp½2ikðei xs ÞdSðxs Þ ðβ ¼ v, r Þ ð10:31Þ
2π
Slit þSe þS1
since the integrand vanishes on Se and the surface at infinity gives no contribution if
we add a small amount of “damping” to the exponential term in Eq. (10.29) by
setting k ! k þ iε and consider Eq. (10.31) to be the result of taking the limit as
ε ! 0. We can then apply the divergence theorem to Eq. (10.31) and replace the
closed surface integration by one over the volume V ¼ V lit þ V e within this surface
(see Fig. 10.6) given by
ð
k2 σ β
Aβ ðei ; ei Þ ¼ exp½2ikðei xÞdV ðxÞ ðβ ¼ v, r Þ: ð10:32Þ
π
V lit þV e
y1 ¼ x1 =a1
y2 ¼ x2 =a2 ð10:33Þ
y3 ¼ x3 =a3 ;
0
which transforms Vlit to one half of a sphere of unit radius (Vlit ), and Ve to an infinite
0
right cylinder, Ve , also of unit radius (Fig. 10.7) giving
ð
k 2 a1 a2 a3 σ β 0
Aβ ðei ; ei Þ ¼ exp½2ikðre yÞdV ðyÞ ðβ ¼ v, r Þ; ð10:34Þ
π
0 0
V lit þV e
φ
⬘ ρ
Vlit φ
semi-major axes are defined by the unit vectors um ðm ¼ 1, 2, 3Þ (Fig. 10.5). Since
we have assumed the incident wave direction is along one of the principal axes of
the ellipsoid, we must have (ei1, ei2, ei3) be (1, 0, 0) or (0, 1, 0) or (0, 0, 1). Now, we
set up a spherical coordinate system for Vlit and a cylindrical coordinate system for
Ve as shown in Fig. 10.8. Then the integral in Eq. (10.34) becomes explicitly
ð
0
I¼ exp½2ikðre yÞdV ðyÞ
0 0
V lit þV e
ð1 ðπ ð
1
In this case since the incident wave direction must be along one of the principal
ellipsoid axes, re is one of the corresponding ellipsoid semi-major axes values
(a1, a2, a3). However, we will see that this distance also plays an important role in
approximate scattering solutions even when the incident wave direction is not along
a principal axis. The distance re is called the equivalent or effective radius of the
ellipsoid in the ei direction. This effective radius can be interpreted geometrically as
the distance in any incident wave direction, ei, from the origin of the ellipsoid to a
plane tangent to the surface of the ellipsoid that is perpendicular to ei as shown in
Fig. 10.9 (see Appendix F).
432 10 Flaw Scattering
Both the θ and z integrations can be performed directly in Eq. (10.35) (neglecting
the upper limit at infinity in the z-integral through the introduction of a small
amount of artificial damping, as before) to obtain
ð1
π π
I¼ ðexpð2i k r e r Þ 1Þr dr þ ;
ikr e 2ikre
r¼0
where re can take on any one of the values (a1, a2, a3), depending on the axis
coinciding with the incident wave direction. Figure 10.10 shows a plot of the
magnitude of this scattering amplitude expression as a function of the
non-dimensional frequency, kre. [Note: in the first edition of this book, in
Eq. (10.37) the incident wave direction was not restricted to be along a principal
axis of the ellipsoid. That, however, was in error since in the general case the
lit-dark boundary is not properly transformed by the distorted coordinates of
Eq. (10.33) into that for the sphere.] At high frequencies we see that the scattering
amplitude is
a1 a2 a 3
lim Aβ ðei ; ei Þ ¼ expð2ikr e Þ ðβ ¼ v, r Þ: ð10:38Þ
ω!1 2r 2e
10.3 Approximate Scattering Solutions: Fluid Model 433
Although Eq. (10.37) is only valid for the special case of the incident wave along a
principal axis, as we will see later in this section, Eq. (10.38) has no such restriction
so it is valid for the general high frequency limit of the pulse-echo response of the
ellipsoid for any incident direction.
It is instructive to invert these frequency domain results into the time domain by
multiplying both sides of Eq. (10.37) by expðiωtÞ=2π and integrating over all
frequencies. In this case the incident wave pinc ¼ exp½ikðx ei Þ becomes an
impulsive incident plane wave pinc ¼ δðt x ei =cÞ, since
ð
þ1
1
δðt ei x=cÞ ¼ exp½ikðx ei Þ iωtdω: ð10:39Þ
2π
1
ð
þ1
1
að e i ; e s Þ ¼ Aðei ; es ÞexpðiωtÞdω: ð10:40Þ
2π
1
(note, however, that the time variable, t, will not be shown explicitly in the
arguments of this impulse response). Then for the ellipsoidal void or rigid inclusion
we have
a 1 a2 a3 σ β c
aβ ðei ; ei Þ ¼ δ ð t þ 2r e =c Þ U ð 2r e =c, 0; t Þ ðβ ¼ v, r Þ;
2r 2e 2re
ð10:41Þ
434 10 Flaw Scattering
where
1 a<t<b
Uða; b; tÞ ¼ : ð10:42Þ
0 otherwise
This time domain response is plotted in Fig. 10.11 for the case of a void. The earliest
response comes from the delta function term in Eq. (10.41) which gives a large
spike in the response at time t ¼ 2re =c, where t ¼ 0 corresponds to when the
impulsive incident wave front reaches the center of the ellipsoid. This spike will
be called here the leading edge response of the flaw and physically comes from a
large “specular” reflection contribution of the flaw surface at a point where the
incident wave front first touches that surface (Fig. 10.12). This leading edge
response also comes directly from a Fourier transform inversion of the high
frequency limit of the scattering amplitude, Eq. (10.38). It is in fact an exact high
frequency asymptotic result for any incident wave direction, as will be seen shortly.
The remainder of the response is contained in the U function which represents
the contribution of the rest of the lit surface (Fig. 10.12). Unlike the leading edge
response, however, this U function response is only valid for the incident wave
direction along one of the principal axes of the ellipsoid.
10.3 Approximate Scattering Solutions: Fluid Model 435
These time domain results can be obtained in a more general fashion that explicitly
illustrates the dependency of the impulse response on the flaw geometry. To see
this, we return to the results for the pulse echo response of a void or rigid inclusion
(Eq. (10.32)) and write the volume element as dV ¼ SðzÞdz where the z-axis is taken
in the incident wave direction and S(z) represents the cross sectional area of this
volume in a plane perpendicular to the incident wave direction (Fig. 10.13). Cer-
tainly, for any flaw geometry whose surface is a surface of revolution about the
incident wave direction the entire lit surface can be represented in this form,
including the pulse echo responses of the ellipsoid just considered where the
incident wave direction is along a principal axis. Since ei x ¼ z, we find
ð
k2 σ β
Aβ ðei ; ei Þ ¼ expð2ikzÞSðzÞdz ðβ ¼ v, r Þ; ð10:43Þ
π
V lit þV e
ð
þ1 ð
σβ
aβ ðei ; ei Þ ¼ 2 2 ω2 expð2ikz iωtÞSðzÞdz dω ðβ ¼ v, r Þ: ð10:44Þ
2π c
1 V lit þV e
ð
þ1
The sampling properties of the delta function allows us to evaluate the z-integration
explicitly and we finally obtain
σ β d2 Sðct=2Þ
aβ ðei ; ei Þ ¼ ðβ ¼ v, r Þ: ð10:48Þ
2πc dt2
Equation (10.48) shows that the far field pulse echo scattering impulse responses for
both voids and rigid scatterers whose surfaces are surfaces of revolution about the
incident wave direction are just proportional to the second derivative of the cross
sectional area function S(z). For simple shapes this area function can be found
explicitly and then differentiated to obtain the far field impulse response (see
Problem 10.3). Equation (10.48) could also be used for the early time response of
more general shaped scatterers but only for times less than when the plane of the
incident wave reaches the lit-dark boundary.
The leading edge response of a flaw often is larger than the remaining portion of the
response. To estimate parameters, such as flaw detectability, therefore, the leading
edge response is of particular importance. We can obtain an explicit expression for
this leading edge response in both the frequency and time domains for the general
flaw case by first returning to Eq. (10.26) and rewriting it in the general form
ð
Aðei ; es Þ ¼ f ðxs Þexpðika xs ÞdSðxs Þ; ð10:49Þ
Slit
10.3 Approximate Scattering Solutions: Fluid Model 437
where
ik
f ð xs Þ ¼ ðei þ es Þ þ ðes ei ÞRp n
4π ð10:50Þ
a ¼ ei e s :
2πi
Aðei ; es Þ ¼ ðR1 R2 Þ1=2 f xsstat exp ika xsstat ;
ð10:51Þ
k a n xsstat
where xstat
s is the stationary phase point on the lit surface and where a is anti-parallel
to the outward normal to the flaw surface at the stationary point, n(xstat s ). More
explicitly, for our case, with a ¼ ei es the stationary phase point is at the location
on the lit surface where the angle α between the outward normal and the scattered
wave direction es is equal to the angle θi between the outward normal and the
incident wave direction ei (Fig. 10.14). Therefore, we have es ei ¼ jes ei jn and
ðei þ es Þ n ¼ 0 at the stationary phase point and Eq. (10.51) evaluates to
pffiffiffiffiffiffiffiffiffiffi
R p ð θ i Þ R1 R 2
Aðe i ; e s Þ ¼ expðikjei es jd Þ; ð10:52Þ
2
where R1 and R2 are the principal radii of curvature at the stationary phase point and
d ¼ n xsstat is the distance in the n direction from the coordinate origin to the
stationary phase point (Fig. 10.14). For the case of pulse-echo, Eq. (10.52) becomes
pffiffiffiffiffiffiffiffiffiffi
Rp ð 0 Þ R1 R2
Aðei ; ei Þ ¼ expð2ikdÞ: ð10:53Þ
2
es tangent plane
n α
xsstat Slit
θi α
– es
ei
d
Fig. 10.14 Stationary phase point on a convex scatterer which is located at the point xstat
s where
θi ¼ α (i.e. es is in the same direction as a wave reflected from a planar surface P which coincides
with the tangent plane to the surface at xstat
s )
438 10 Flaw Scattering
We can evaluate this general result for an ellipsoidal void or rigid scatterer. In those
pffiffiffiffiffiffiffiffiffiffi a1 a2 a3
cases since Rp ð0 Þ ¼ σ β , R1 R2 ¼ for an ellipsoid (see Appendix F) and
r 2e
d ¼ r e when we take the coordinate origin at the center of the ellipsoid, we find
a1 a2 a3 σ β expð2ikr e Þ
Aβ ðei ; ei Þ ¼ ðβ ¼ v, r Þ: ð10:54Þ
2r 2e
But note that there is no restriction on the incident wave direction in Eq. (10.54) so
that our previous high frequency result given by Eq. (10.38) was indeed valid in
general, as previously stated.
If we invert the frequency domain results of Eq. (10.52) back into the time
domain, we obtain the pitch-catch leading edge response of a general volumetric
flaw as
pffiffiffiffiffiffiffiffiffiffi
R p ð θ i Þ R1 R 2
aðei ; es Þ ¼ δðt þ jei es jd=cÞ: ð10:55Þ
2
The general expression for the far field scattering amplitude of a crack-like flaw was
given by Eq. (10.7). For the case of a stress-free crack, the pressure vanishes on both
the front and back surfaces of the crack so that Δp ¼ 0 and Eq. (10.7) becomes
ð
1
Að e i ; e s Þ ¼ Δð∂~
p =∂nÞexpðik xs es ÞdS: ð10:56Þ
4π
S
In the Kirchhoff approximation, we again assume that on the lit (front) surface the
field corresponds to that produced at an infinite planar surface whose normal
coincides with that of the surface at xs, and on the shadowed back surface the fields
p and ∂p=∂n are identically zero (Fig. 10.15). Therefore, it follows that for an
incident plane wave of unit amplitude we have
where pinc and preflt are given by Eqs. (10.22) and (10.23) with the reflection
coefficient Rp ¼ 1, so that explicitly
Δð∂~
p =∂nÞ ¼ 2ikðei nÞexp½ikðei xÞ
10.3 Approximate Scattering Solutions: Fluid Model 439
Note that this is the same result we obtained for a void (Eq. (10.27)) except that we
have assumed here that the entire crack surface, S, is the lit surface. For highly
curved cracks this may not be the case and the surface of integration in Eq. (10.58)
must change accordingly.
For a flat crack, the unit normal n is a constant on the crack surface so that in this
case the scattering amplitude reduces to
ð
ikðei nÞ
Aðei ; es Þ ¼ exp½ikðei es Þ xs dS: ð10:59Þ
2π
S
For an elliptical flat crack the integral in Eq. (10.59) can be performed explicitly.
First, let
where eq is a unit vector in the ei es direction. Then if we have a crack with semi-
major axes a1 and a2 in the u1 and u2 directions, respectively, (Fig. 10.16) and
define a set of distorted coordinates
y1 ¼ x1 =a1
ð10:61Þ
y2 ¼ x2 =a2
we have
ð
ika1 a2 ðei nÞ
Aðei ; es Þ ¼ expðikjei es jre yÞdy1 dy2 ; ð10:62Þ
2π
0
S
440 10 Flaw Scattering
ei
n a2
a1 x2
u2
u1
x1
where S0 is now a circle of unit radius and re ¼ a1 eq u1 u1 þ a2 eq u2 u2 .
Introducing polar coordinates (r, θ) for y then gives
ð1 2ðπ
ika1 a2 ðei nÞ
Aðei ; es Þ ¼ expðikjei es jre r cos θÞrdrdθ; ð10:63Þ
2π
r¼0 θ¼0
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2ffi
re ¼ a1 eq u1 þ a2 eq u2
2 2 ð10:64Þ
is the effective radius of the crack as seen in the eq direction (Fig. 10.17). The
θ integral in Eq. (10.63) can be recognized as a Bessel function so we have
ð1
Aðei ; es Þ ¼ ika1 a2 ðei nÞ J 0 ðkjei es jre r Þrdr: ð10:65Þ
r¼0
Since
ð1
rJ 0 ðKr Þdr ¼ J 1 ðK Þ=K
0
10.3 Approximate Scattering Solutions: Fluid Model 441
we obtain, finally
i a 1 a 2 ð e i nÞ
Aðei ; es Þ ¼ J 1 ðkjei es jre Þ: ð10:66Þ
jei es jr e
In the special case of a pulse echo setup ðes ¼ ei Þ this reduces to
i a 1 a 2 ð e i nÞ
Aðei ; ei Þ ¼ J 1 ð2kr e Þ: ð10:67Þ
2re
Figure 10.18 shows a plot of the magnitude of the scattering amplitude versus
non-dimensional frequency obtained from Eq. (10.66) when the unit vector eq is not
normal to the crack surface. In this
case, one sees a series of decreasing maxima
separated by distinct nulls. When eq n ¼ 1, however, r e ¼ 0 and the scattering
amplitude in this limit becomes
ikðei nÞa1 a2
Aðei ; es Þ ¼ : ð10:68Þ
2
ika1 a2
Aðei ; ei Þ ¼ : ð10:69Þ
2
Equations (10.68) and (10.69) show that the magnitude of the scattering amplitude
simply varies linearly with frequency in these special cases (Fig. 10.19).
442 10 Flaw Scattering
We can invert these frequency domain results into the time domain to obtain
the impulse response of the crack, in the same manner as done for volumetric flaws.
In this case, from Eq. (10.66) we have
ð
þ1
aðei ; es Þ ¼ 1=2π Aðei ; es Þexp iωt dω
1
ð
þ1
i a 1 a2 ð e i n Þ
¼ J 1 ðωjei es jre =cÞexpðiωtÞdω
2π jei es jr e
1
ð
þ1
a 1 a 2 ð e i nÞ
¼ J 1 ðωjei es jre =cÞ sin ðωtÞdω: ð10:70Þ
π jei es jre
0
But, since
8
ð
1 β=α
< pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jβ j < α
J 1 ðαωÞ sin ðβωÞdω ¼ ð10:71Þ
: α β
2 2
0 0 jβ j > α
we find, if re 6¼ 0
8
>
> a1 a2 cðei nÞ t
< 2 2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi jtj < jei es jre =c
aðei ; es Þ ¼ π jei es j r e ðjei es jre =cÞ2 t2 :
>
>
:
0 otherwise
ð10:72Þ
This time domain crack response is plotted in Fig. 10.20. From that figure we see an
anti-symmetrical signal centered around t ¼ 0, with singularities occurring at times
t ¼ jei es jr e =c. These singularities are called flash points and are the earliest
and latest arriving responses from the crack edge. As we will see in Chap. 15, the
existence of such flash points gives a convenient way to size such flaws.
10.3 Approximate Scattering Solutions: Fluid Model 443
When eq n ¼ 1 and hence r e ¼ 0, Eq. (10.68) gives
ð
þ1
ðei nÞa1 a2 dδ
aðei ; es Þ ¼ ; ð10:74Þ
2c dt
a1 a2 dδ
aðei ; ei Þ ¼ : ð10:75Þ
2c dt
Schematically, we can view the responses in Eqs. (10.74) and (10.75) as that of a
“doublet” at time t ¼ 0 (Fig. 10.21) which arises from the reflection of the entire
incident wave front simultaneously from the crack surface.
444 10 Flaw Scattering
As in the volumetric flaw case, these time domain results can be obtained in a
general fashion that more closely connects the impulse response to the flaw
geometry. Returning to Eq. (10.59) for a general shaped flat crack, we can write
the impulse response as
8 9
ð þ1 ð
ð e i nÞ 1 < =
aðei ; es Þ ¼ iωexp iω jei es jeq x=c t dω dS
2πc 2π : ;
8
S 1 9
ð
ðei nÞ d < =
¼ δ jei es jeq x=c t dS : ð10:76Þ
2πc dt: ;
S
If we set up a coordinate system on the crack surface so that the x-axis is along the
same direction as the component of eq in plane of the crack and express the area
element as dS ¼ DðxÞdx in terms of the crack width function D, where D ¼ 0
outside the crack surface (Fig. 10.22), then Eq. (10.76) becomes
10.3 Approximate Scattering Solutions: Fluid Model 445
8 þ1 9
< ð =
ðei nÞ d
að e i ; e s Þ ¼ δðjei es jx cos η=c tÞDðxÞdx
2πc dt: ;
1 8 þ1 9
< ð =
ðei nÞ d
¼ δðx ct=jei es j cos ηÞDðxÞdx : ð10:77Þ
2π jei es j cos η dt : ;
1
ðei nÞS dδ
aðei ; es Þ ¼ : ð10:80Þ
2πc dt
S dδ
aðei ; ei Þ ¼ ; ð10:81Þ
2πc dt
where S is the area of the crack. Comparing these results with those for the elliptical
crack (Eqs. (10.74) and (10.75)) we see that they are identical since S ¼ πa1 a2 for
an ellipse.
Although spherical pores and circular cracks are both stress-free flaws, we have
seen that their pulse-echo far field amplitude responses in the Kirchhoff approxi-
mation are quite different. The pore has a leading edge delta function response
followed by a “box” function lit surface response while the crack has flash points at
the front and back edges of the crack. Both of these cases can be considered the
limits of an ellipsoid with semi-major axes (a1, a2, a3) going from (a, a, a) for the
sphere to (a, a, 0) for the circular crack so that considering the pulse-echo response
of a general ellipsoid, we can show how the transition occurs from a spherical pore
to a circular crack. Figure 10.23 shows an ellipsoid with semi-major axes along the
446 10 Flaw Scattering
(x1, x2, x3) axes where we wish to obtain the pulse-echo response of an ellipsoidal
void in water due to a plane wave traveling in the incident direction
pffiffiffi
ei ¼ 1= 3 ðu1 þ u2 þ u3 Þ, where (u1, u2, u3) are unit vectors along the (x1,
x2, x3) directions, respectively. We let the semi-major axes be given by (1, 1, a3)
mm for different choices of a3 ranging from a3 ¼ 1 (sphere) to a3 ¼ 0:01 (near-
crack). Note that in these cases we must perform the integration in Eq. (10.29)
numerically because the boundary between the lit and dark surface, defined by
ei n ¼ 0, is a curve whose shape is constantly changing as the ellipsoid changes
and the incident wave direction is not along a principal axis of the ellipse as
assumed in our previous explicit results. Here the integral was performed by
breaking the surface of an ellipsoid into M small triangular elements, ΔSm, and
assuming the amplitude of the integrand was constant over each element, giving
ð
ik X
Aðei ; ei Þ ¼ ðei nm Þ exp½2ikðei xs ÞdSðxs Þ: ð10:82Þ
2π ðe n Þ0
i m ΔSm
However, this integral can be expressed in edge element form through the use of
Stokes’ theorem (see Chap. 8) to turn the surface integrations into line integrals,
yielding
ð
1 X ðnm ei Þ
Aðei ; ei Þ ¼ ðe i n m Þ h i exp½2ikðei xs Þdsðxs Þ;
4π ðe n Þ0 1 ðnm ei Þ2
i m
ΔCm
ð10:83Þ
1 X X 3
em Lm
Aðei ; ei Þ ¼ ðei nm Þ ⊥ mn
4π ðe n Þ0 n¼1 sin θ
i m
h
i
h
i sin k sin θm ekm Lnm
exp 2ik sin θm ekm Dnm
; ð10:84Þ
k sin θm ekm Lnm
10.3 Approximate Scattering Solutions: Fluid Model 447
Fig. 10.24 The magnitude of the far field scattering amplitudes of ellipses versus ka for the
configuration of Fig. 10.23 where (a1, a2, a3) is given by (a) (1, 1, 1)—sphere, (b) (1, 1, 0.3),
(c) (1, 1, 0.1), (d) (1, 1, 0.01)—near-crack
where the various terms appearing in Eq. (10.84) are described in Chap. 8,
Sect. 8.6.1.1. A total of 32,768 triangles were used to mesh the ellipsoid, which
was sufficient to allow accurate results to be calculated for ka 40. We know from
our explicit expression for the leading edge response (see Eq. (10.54)) that the
amplitude of this response goes to zero as a3 ! 0 so that in the frequency domain
the constant value of the scattering amplitude at high frequencies for the sphere
must go to zero as the flaw geometry approaches that of a crack. We can see this
behavior in Fig. 10.24 which gives the magnitude of the scattering amplitude versus
ka for ellipsoids with ða1 ; a2 Þ ¼ ð1; 1Þ and a3 ¼ 1:0, 0:3, 0:1, 0:01. For the narrowest
ellipsoid ða3 ¼ 0:01Þ the high frequency constant value is very small and the
characteristic oscillations with decreasing amplitude of the circular crack have
appeared as seen in Fig. 10.24d. Since the leading edge response vanishes in the
crack limit, we expect that the flashpoint behavior of the crack in the time domain
must come from a distortion of the “box” function that characterizes the lit surface
response of the sphere. We can show this is indeed the case by subtracting the
leading edge response (which we know analytically from Eq. (10.54)) from the total
448 10 Flaw Scattering
Fig 10.25 Pulse-echo scattering amplitudes in the time domain of ellipsoids for the configuration
of Fig. 10.23 with the front surface leading edge response removed and where (a1, a2, a3) is given
by (a) (1, 1, 1)—sphere, (b) (1, 1, 0.3), (c) (1, 1, 0.2), (d) (1, 1, 0.1)
ellipsoid response and then inverting the remaining frequency domain response
back into the time domain numerically with an FFT. The results are shown in
Fig. 10.25a–d for ellipsoids with ða1 ; a2 Þ ¼ ð1; 1Þ and a3 ¼ 1:0, 0:3, 0:2, 0:1. As
seen in that figure, the box function is distorted by the appearance of a local
minimum in the lit region response which decreases in value and moves closer to
the beginning of the entire response. For the a3 ¼ 0:1 case we this distortion has
already led to flash point behavior over the lit region similar to that seen in the
limiting behavior of the crack response.
Vf
where, recall
8 8
<1 y outside V f <0 y outside V f
α ¼ 1=2 y on Sf β ¼ 1=2 y on Sf ð10:86Þ
: :
0 y inside V f 1 y inside V f
and n is the outward normal to Vf. To be precise, the fields in the integral of
Eq. (10.85) are pðxs ; ωÞ ¼ pþ ðxs ; ωÞ, and ∂pðxs ; ωÞ=∂nðxs Þ ¼ ∂pþ ðxs ; ωÞ=∂nðxs Þ,
where the plus superscript indicates that these quantities are evaluated just outside
the flaw surface (in the host material). From the continuity of pressure and normal
velocity across Sf, we have
450 10 Flaw Scattering
pþ ðxs ; ωÞ ¼ p ðxs ; ωÞ
1 ∂pþ ðxs ; ωÞ 1 ∂p ðxs ; ωÞ ð10:87Þ
¼ ;
ρ0 ∂nðxs Þ ρ1 ðxs Þ ∂nðxs Þ
where the minus superscript indicates evaluation just inside the flaw surface.
Placing these continuity conditions into Eq. (10.85), we find
ð
αpscatt ðy; ωÞ þ βpinc ðy; ωÞ ¼ ½p ðxs ; ωÞ∇Gðxs ; y; ωÞ
Sf
ðρ0 =ρ1 ÞGðxs ; y; ωÞ∇p ðxs ; ωÞ nðxs ÞdSðxs Þ;
ð10:88Þ
which by Gauss’ theorem can be written over the flaw volume Vf (assuming the flaw
properties are continuous within Vf) as
ð
αp ðy; ωÞ þ βp ðy; ωÞ ¼ f∇ ½p ðx; ωÞ∇Gðx; y; ωÞ
scatt inc
Vf
∇ ½ðρ0 =ρ1 ÞGðx; y; ωÞ∇p ðx; ωÞgdV ðxÞ;
ð10:89Þ
∇ ðϕ∇ψ Þ ¼ ϕ∇2 ψ þ ∇ϕ ∇ψ
∇ ðabÞ ¼ að∇ bÞ þ ∇a b
we obtain
ρ ω2
∇2 G þ 0 G ¼ δðx yÞ
λ0
ρ ρ ω2
∇ 0 ∇p ¼ 0 2 p
ρ1 ρ1 c 1
so that using these relations and the sampling properties of the delta function, we
obtain
ð
ρ ω2 λ0
pscatt ðy; ωÞ ¼ 1 0 ∇p ∇G 2 p G 1 dV; ð10:90Þ
ρ1 c0 λ1
Vf
10.3 Approximate Scattering Solutions: Fluid Model 451
where the α and β factors and the delta function contribution in the integral in
Eq. (10.89) all combine to give a coefficient for pscatt in Eq. (10.90) that is equal to
one for all y. If we now define the quantities
ρ1 ρ0
γρ ¼
ρ1
λ1 λ0 ð10:91Þ
γλ ¼
λ1
In the far field, we can again use the approximate forms for G and its derivatives
(see Eqs. (4.58) and (4.59)) to obtain
ð
expðik0 r s Þ
pscatt ðy; ωÞ ¼ ik0 γ ρ ðes ∇p Þ k20 γ λ p expðik0 es xÞdV;
4πrs
Vf
ð10:93Þ
where, as before, rs ¼ jyj and es ¼ y^ . From Eq. (10.4), then the scattering amplitude
for an incident wave of pressure amplitude p0 and incident direction ei is given as
the volume integral
ð
1
Að e i ; e s Þ ¼ p Þ þ k20 γ λ p~ expðik0 es xÞdV
ik0 γ ρ ðes ∇~ ð10:94Þ
4π
Vf
p~ ¼ expðik0 ei xÞ
∇~
p ¼ ik0 ei expðik0 ei xÞ
which, when placed in Eq. (10.94), gives an explicit expression for the scattering
amplitude as
452 10 Flaw Scattering
ð
k20
Aðei ; es Þ ¼ γ λ γ ρ ðei es Þ exp½ik0 ðei es Þ xdV ðxÞ: ð10:95Þ
4π
Vf
When the flaw is homogeneous, i.e. ρ1 and λ1 are constants, the first terms in
Eq. (10.95) can be taken outside the integral to obtain
ð
k20 γ λ γ ρ ðei es Þ
Að e i ; e s Þ ¼ exp½ik0 ðei es Þ xdV ðxÞ: ð10:96Þ
4π
Vf
and a geometrical shape factor, G(ei, es, ω), which is also frequency dependent:
ð
Gðei ; es ; ωÞ ¼ exp½ik0 ðei es Þ xdV ðxÞ: ð10:98Þ
Vf
ω2 f ðei ; es Þ
Aðei ; es Þ ¼ Gðei ; es ; ωÞ: ð10:99Þ
4πc20
For a homogeneous ellipsoidal flaw (see Fig. 10.5) the shape function can be
evaluated explicitly using the same transformations used previously in the Kirch-
hoff approximation. First, we write the geometrical shape function as
ð
Gðei ; es ; ωÞ ¼ exp ik0 jei es jeq x dV ðxÞ; ð10:100Þ
Vf
10.3 Approximate Scattering Solutions: Fluid Model 453
y1 ¼ x1 =a1
y2 ¼ x2 =a2
y3 ¼ x3 =a3
Then, if we set up a spherical coordinate system with the z-axis along re, we have
ð1 ðπ
Gðei ; es ; ωÞ ¼ 2πa1 a2 a3 exp½ik0 jei es jr e r cos θr 2 sin θ dr dθ; ð10:103Þ
r¼0 θ¼0
where now
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 2ffi
r e ¼ j r e j ¼ a 1 e q u1 þ a 2 e q u2 þ a 3 e q u3
2 2 2 ð10:104Þ
is the distance in the eq direction from the origin of the ellipsoid to a plane tangent
to the ellipsoid surface that is also perpendicular to eq. Thus, re is the equivalent
radius of the ellipsoid in the eq direction. The θ integration can be done directly to
yield
ð1
4πa1 a2 a3
Gðei ; es ; ωÞ ¼ sin ðk0 jei es jre r Þr dr ð10:105Þ
k0 jei es jre
r¼0
Placing this result back into the scattering amplitude expression (Eq. (10.99)) we
find, finally
454 10 Flaw Scattering
a1 a2 a3 f ð e i ; e s Þ sin ðk0 jei es jre Þ k0 jei es jre cos ðk0 jei es jre Þ
Að e i ; e s Þ ¼ :
jei es j2 r 2e k0 jei es jre
ð10:107Þ
From Eqs. (10.37) and (10.108) we see there is a strong similarity between the Born
and Kirchhoff approximations for the pulse echo response of an ellipsoid. This
similarity can be made even closer by noting that for a weak scatterer
f ðei ; ei Þ ¼ γ λ þ γ ρ
Δλ Δρ
ffi þ
λ0 ρ0
Now, consider transforming the frequency domain results of Eq. (10.108) into
the time domain to obtain the impulse response, a(ei; es), of the ellipsoid in the Born
approximation. Multiplying Eq. (10.108) by expðiωtÞ=2π and integrating over all
frequencies, the integrations again lead directly to delta functions and the
U function defined previously (Eq. (10.42)) in the form
10.3 Approximate Scattering Solutions: Fluid Model 455
a1 a2 a3 f ðei ; es Þ
aðei ; es Þ ¼ fδðt jei es jre =c0 Þ þ δðt þ jei es jr e =c0 Þ
2jei es j2 r 2e
c0
U ðjei es jr e =c0 , jei es jre =c0 ; tÞ ð10:112Þ
jei es jre
a 1 a 2 a 3 Rp ð 0 Þ
aðei ; ei Þ ¼ fδðt 2r e =c0 Þ þ δðt þ 2r e =c0 Þ
2r 2e
c0
Uð2r e =c0 , 2r e =c0 ; tÞ : ð10:113Þ
2r e
A graph of Eq. (10.113) is shown in Fig. 10.27. A leading edge delta function
response exists when the incident wave front first touches the flaw, followed by a
constant amplitude response as the incident wave travels through the flaw, and,
finally, a trailing edge delta function response when the incident wave front last
touches the flaw. Even though a void or rigid scatterer are certainly not weak
scatterers, comparing Eqs. (10.113) and (10.41) for these cases shows that both
the Born and Kirchhoff approximations predict exactly the same early time
response, which includes the leading edge delta function response and the subse-
quent immediate step change in amplitude (note that this equivalency is only
strictly true if we let the reflection coefficient appearing in the Born approximation
be the exact reflection coefficient as opposed to only its weak scattering limit). The
reason these results coincide is that both the Born and Kirchhoff approximations are
“single scattering” theories, i.e. the early time interactions of the incident wave with
the flaw boundaries are accounted for, at least approximately, in both cases. In the
Kirchhoff approximation, these early time interactions are obtained by approximat-
ing the fields on the lit surface by the incident and reflected fields due to a plane
interface. In the Born approximation, the interactions arise from letting the trans-
mitted wave (approximated as just the incident wave) sweep through the flaw
undisturbed.
In Fig. 10.28 the experimentally obtained pulse echo scattering amplitude for a
polystyrene sphere embedded in Lucite is shown. Both the front and back surface
spikes are clearly visible, although the intermediate constant portion of the response
has been “washed out” by the effects of having a finite bandwidth system.
456 10 Flaw Scattering
(x10–4)
.78
measured far field scattering
amplitude time domain
response for a 320 μm
diameter polystyrene sphere .33
in Lucite
Amplitude (v)
–.11
–.55
.00 .50 1.00 1.49 1.99
(x10–6)
Time (sec)
As in the Kirchhoff approximation, we can give the time domain results here in a
more general form that explicitly illustrates the dependency of the far field scatter-
ing amplitude on the flaw geometry. Multiplying Eq. (10.99) by expðiωtÞ=2π and
integrating over all frequencies, we obtain
8 þ1 9
ð
f ðei ; es Þ < 1 2 =
að e i ; e s Þ ¼ ω G ð e i ; e s ; ωÞ exp ð iωt Þdω
4πc20 :2π ;
8 1 þ1 9
ð
f ðei ; es Þ < 1 d 2 =
¼ Gð e ;
i se ; ω Þexp ð iωt Þdω : ð10:114Þ
4πc20 :2π dt2 ;
1
Placing the volume integral expression for the geometrical shape function
(Eq. (10.100)) into Eq. (10.114) and interchanging orders of integration, gives
8 2 þ1 3 9
>
< 2 ð ð >
=
f ðei ; es Þ d 4 1 5
aðei ; es Þ ¼ exp ½iωT dω dV ð x Þ
4πc20 > :dt
2 2π >
;
1
8 Vf
9
> ð >
f ðei ; es Þ < d2 =
¼ δ ðT ÞdV ð x Þ ; ð10:115Þ
4πc0 >2
:dt
2 >
;
Vf
where T ¼ jei es j eq x =c0 t. Setting up a coordinate system so the x-axis
is along the eq direction (Fig. 10.29), we can write the volume element as dV ðxÞ
¼ SðxÞdx and δðT Þ ¼ c0 δðx c0 t=jei es jÞ=jei es j so that
10.4 Approximate Scattering Solutions: Elastic Solid Model 457
dx
8 þ1 9
ð
f ðei ; es Þ < d2 =
að e i ; e s Þ ¼ δ ðx c 0 t= j ei e s j ÞS ð x Þdx : ð10:116Þ
4πc0 jei es j :dt2 ;
1
Using the sampling properties of the delta function, Eq. (10.116) becomes finally
Consider now the case when a plane elastic wave strikes an elastic inclusion of
volume Vf embedded in an infinite homogeneous surrounding medium (see
458 10 Flaw Scattering
Fig. 10.1). From Eq. (10.13a) the far field scattering amplitude of the scattered
P-waves is given by
ð
e P e P Clkpj
AnP;β eiβ ; esP ¼ sl sn 2 u p =∂xj nk þ ik1 eskP np u~j exp ik1 xs esP dSðxs Þ
∂~
4πρc1
S
ðno sum on sÞ ðβ ¼ P, SÞ:
ð10:119Þ
in terms of the total displacement components and their derivatives on the surface
of the flaw. In the Kirchhoff approximation, we again assume the flaw surface is
separated into a lit (Slit) and shadowed ðS Slit Þ portions (Fig. 10.4) so that for an
incident wave of unit displacement amplitude we have
p þ up
uinc reflt
on Slit
u~p ¼ ; ð10:121Þ
0 otherwise
where uinc
p and up
reflt
are the incident and reflected wave displacements as calculated
for an infinite plane interface whose normal, n, coincides with the outward normal
of the surface at each point xs on Sf (Fig. 10.30). Recall from Chap. 6 that when a
plane P- or S-wave strikes such a planar interface, the incident wave direction eβi ,
and the normal to the interface, n, define a “plane of incidence” whose normal is
perpendicular to both eβi and n (Fig. 10.31). In the case where the incident wave is a
S
Vf
O
10.4 Approximate Scattering Solutions: Elastic Solid Model 459
O⬘
plane P
shear wave whose polarization is in the plane of incidence (Fig. 10.32a), we will
call the incident S-wave a “vertically polarized” shear wave or SV-wave. For either
an incident P- or SV-wave, both reflected and transmitted P- and SV-waves are
generated at the interface in the plane of incidence and the total displacement on the
interface can be written as
h
i X m;β h
i
u~p ¼ d ipβ exp ikβ eiβ x
0 0
þ m
R12 d rp exp ikm erm x
m¼P, SV ð10:122Þ
ðno sum on i, r Þ ðβ ¼ P, SV Þ;
We can combine both of these cases and write in general for an incident wave of
any type and any polarization (P, SV, SH):
h
i X h
i
u~p ¼ dipβ exp ikβ eiβ x
0 m;β m 0
þ R12 drp exp ikm erm x
m¼P, SV , SH ð10:124Þ
ðno sum on i, r Þ ðβ ¼ P, SV, SH Þ
m;SH SH;β
with the understanding that R12 ¼ 0 ðm 6¼ SHÞ and R12 ¼ 0 ðβ 6¼ SH Þ. The
point x0 in Eqs. (10.122)–(10.124) is an arbitrary point on the plane P as measured
from some fixed origin O0 on the plane (Fig. 10.30).
From Eq. (10.124) it also follows then that the derivatives of the displacements
needed in Eqs. (10.119) and (10.120) are given by
h
i X h
i
u p =∂xj ¼ ikβ d ipβ eijβ exp ikβ eiβ x
0 0 m;β m m 0
∂~ þ ikm R12 d rp erj exp ikm erm x
m¼P, SV , SH
ðno sum on i, r Þ ðβ ¼ P, SV, SH Þ:
ð10:125Þ
Before placing these results into those equations, however, we would like to move
the origin O0 to a general point O, as shown in Fig. 10.30, that is not on the plane
P and consider an incident plane wave of unit amplitude given by
h
i
β β
p ¼ d ip exp ik β ei x
uinc ðno sum on iÞ ðβ ¼ P, SV, SHÞ: ð10:126Þ
(Note that the phase of this incident wave is now such that the wave front of a pulse
generated from this frequency domain solution is at the origin O at time t ¼ 0
whereas for the original incident wave just considered the wave front would be at
origin O0 at t ¼ 0.)
In this case, we will show that the displacement components of the reflected
waves are given by
10.4 Approximate Scattering Solutions: Elastic Solid Model 461
X h
i
upreflt ¼ R12 drp exp ikβ eiβ x
m;β m
ðβ ¼ P, SV, SH Þ ð10:127Þ
m¼P, SV , SH
To prove Eq. (10.127) (and, hence, Eq. (10.128)), we note that an incident wave
given by Eq. (10.126) can be written as
h
i h
i
β
d ipβ exp ikβ eiβ x
0
p ¼ exp ik β ei d
uinc
ðno sum on iÞ ðβ ¼ P, SV, SH Þ ð10:130Þ
so that the displacements of the reflected waves, from Eq. (10.127), are given by
X h
i
upreflt ¼ exp ikβ eiβ d R12
m;β m
drp exp ikm erm d exp ikm erm x
m¼P, SV , SH
ðno sum on i, r Þ ðβ ¼ P, SV, SHÞ:
ð10:131Þ
so that
h
i
kβ eiβ d km erm d ¼ kβ eiβ n km erm n ðn dÞ
h
i
¼ kβ eiβ n km erm n ðn xÞ ð10:133Þ
km erm x ¼ kβ eiβ x þ km erm n ðn xÞ kβ eiβ n ðn xÞ: ð10:134Þ
m¼P, SV , SH ð10:135Þ
ðβ ¼ P, SV, SHÞ;
which proves Eq. (10.127). In an entirely similar fashion then Eq. (10.128) can be
also shown to be true.
Placing Eqs. (10.128) and (10.129) into Eqs. (10.119) and (10.120) then yields
ð h
i
β P iω eslP esn
P
Clkpj P;β β
AP;β
n e i ; e s ¼ Dpjk exp ik e
β i ik e
1 s
P
x s dSðxs Þ
4πρc21
Slit
for the scattered S-wave far field scattering amplitude (for which we do not
decompose the polarization into SV- or SH-components), where
α;β
Dpjk ¼ dipβ eijβ nk =cβ þ eskα d ijβ np =cα
X
þ m;β
R12 drp erj nk =cm þ eskα drjm np =cα :
m m
ð10:138Þ
m¼P, SV , SH
ðα ¼ P, SÞ ðβ ¼ P, SV, SH Þ
Comparing Eqs. (10.136) and (10.137) with Eq. (10.26) shows that a strong
similarity exists between the fluid and elastic wave cases. This similarity becomes
even stronger for the pulse echo response of stress-free voids where, as discussed
later (in Sect. 10.4.2.3), the elastic solid case yields results that are identical to
Eq. (10.29) for the fluid. Thus, the explicit results we obtained for the pulse echo
response of ellipsoidal voids and the relationship we found between the impulse
response and the flaw cross sectional area function in a fluid carry over directly to
the pulse echo elastic wave case.
10.4 Approximate Scattering Solutions: Elastic Solid Model 463
To obtain the leading edge response, consider first the scattering amplitude for the
scattered P-wave which we rewrite as
ð h
i
β P iω β
An e i ; e s ¼
P;β
I P;β
exp ik β e i ik 1 e P
x s dSðxs Þ
4πρc21 n s
ð10:139Þ
S
ðno sum on sÞ ðβ ¼ P, SV, SH Þ;
where
P;β
n ¼ esl esn Clkpj Dpjk :
I P;β ð10:140Þ
P P
From Eq. (10.139), the stationary phase point of the integrand occurs when eiβ =cβ
esP =c1 is parallel to the normal to the surface (and plane P). This implies that ePs is
in the direction of the specularly reflected P-wave, i.e. esP ¼ erP , and the integrand
IP;β
n in Eq. (10.139) must be evaluated under this condition. To evaluate InP;β , it
is convenient to consider separately the incident P- and S-wave cases and break
P;β
up the Dpjk coefficient into its incident and reflected wave contributions. First,
consider the case when β ¼ P, i.e. we have an incident P-wave. The incident
P-wave contribution to InP;P is then (from the first two terms in Eq. (10.138))
n ¼ ern erl Clkpj eip eij nk þ erk eij np =c1 ;
I P;P ð10:141Þ
P P P P P P
where we have used the fact that dijP ¼ eijP in this case. But, the elastic constant
tensor is given explicitly as
Clkpj ¼ λδlk δpj þ μ δlp δkj þ δlj δkp ð10:142Þ
o
þ μ erp eip eij nj þ erlP eilP eipP np þ erp
P P P
erj eij np þ erjP eijP erp
P P P P
np =c1
n
o
¼ ern
P
λ erk
P
nk þ eijP nj þ 2μ erp eip eij nj þ erjP eijP erp
P P P P
np =c1
¼ 0; ð10:143Þ
where we have used the fact that eijP eijP ¼ erjP erjP ¼ 1 and erjP nj ¼ eijP nj . Now,
consider the reflected P-wave contribution (m ¼ P) to InP;P :
464 10 Flaw Scattering
P;P
I P;P
n ¼ e P P
e
rn rl C lkpj R 12 e P P
e
rp rj n k þ e P P
e
rk rj n p =c1 ; ð10:144Þ
where we have drjP ¼ erjP . Again, expanding this expression out explicitly, we find
n
o
P P;P
I P;P
n ¼ e R
rn 12 2λ e P P P
e e
rp rp rk n k þ 4μ e P P P
e e
rj rj rp n p =c1
¼ 2ρc1 ern
P P
erk nk RP;P
12 ðno sum on r Þ: ð10:145Þ
Finally, consider the reflected SV-wave contribution (note that there is no reflected
SH-wave term since RP;SH 12 ¼ 0 ). Here m ¼ SV and dSVrj is a unit vector that is
perpendicular to erj , i.e. drj erj ¼ 0. Thus, we have in this case
SV SV SV
P;SV
n ¼ ern erl Clkpj R12
I P;P rp erj nk =c2 þ erk d rj np =c1 : ð10:146Þ
P P
d SV SV P SV
so that
P
SV
erp drp ¼ erm
P SV
nm ðc1 =c2 ÞeSV
r m nm drp np
P
SV ð10:149Þ
erjP eSV
rj ¼ c =c
1 2 þ e n
rm m ð c =c Þe
1 2 rm m
SV
n erp np
Collecting all these results then the total value of InP;P at the stationary phase point is
just the contribution from Eq. (10.145).
If we had considered an incident SV- or SH-wave instead, following exactly the
same procedures outlined above to evaluate I P;β n ðβ ¼ SV, SH Þ gives
P P;β
n ¼ 2ρc1 ern erk nk R12
I P;β ðno sum on r Þ ðβ ¼ SV, SH Þ; ð10:153Þ
P
where
(
eiP erP ðβ ¼ PÞ
gP;β ¼ ; ð10:155Þ
ðc1 =c2 ÞeSV
i erP ðβ ¼ SV Þ
and again R1 and R2 are the principal radii of curvature at the stationary phase
point xstat on the lit surface and n is the unit outward normal to the flaw surface.
The scattered S-wave case (SV- or SH-polarization) can be treated in exactly the
same manner as for the scattered P-waves, starting with Eq. (10.137). After
expanding the integrands again and using generalized Snell’s law, as before, we
obtain
466 10 Flaw Scattering
α;β pffiffiffiffiffiffiffiffiffiffi α
erjα nj R12 R1 R2 d rn
Anα;β eiβ ; esα ¼ exp iks gα;β xstat ð10:156Þ
jgα;β
nj
ðno sum on r Þ ðα ¼ SV, SH Þ ðβ ¼ P, SV, SH Þ;
Transforming these frequency domain results into the time domain, where
ð
þ1
anα;β eiβ ; esα ¼ 1=2π Anα;β eiβ ; esα expðiωtÞdω; ð10:158Þ
1
for the scattered SV- or SH-waves and d ¼ n xstat is again the distance from the
coordinate origin to the stationary phase point in the n direction.
When an open crack is imbedded in an elastic solid, the traction is zero on both
faces of the crack so that Eqs. (10.19a, 10.19b) for the scattering amplitude become
ð
β P ik1 eslP esn
P P
esj Ckplj
AP;β
n e i ; e s ¼ 2
nk Δ~
u p exp ik1 xs esP dSðxs Þ
4πρc1 ð10:161Þ
S
ðno sum on sÞ ðβ ¼ P, SÞ
10.4 Approximate Scattering Solutions: Elastic Solid Model 467
u p ¼ uinc
Δ~ p þ up
reflt
h
i X h
i
¼ dipβ exp ikβ eiβ x þ drp exp ikβ eiβ x
m;β m
R12 ð10:163Þ
m¼P, SV , SH
ð h
i
ik1 eslP esn
P P
esj Ckplj
AnP;β eiβ ; esP ¼ 2
nk Dpβ exp i kβ eiβ k1 esP xs dSðxs Þ
4πρc1
S
ðno sum on sÞ ðβ ¼ P, SV, SHÞ
ð10:164Þ
For a flat crack both Dβp and n are constants and Eqs. (10.164) and (10.165) can
be written
ð h
i
ik1 eslP esn
P P
esj Ckplj Dpβ nk
AnP;β eiβ ; esP ¼ exp i k e
β i
β
k e
1 s
P
x s dSðxs Þ
4πρc21
S
ðno sum on sÞ ðβ ¼ P, SV, SHÞ
ð10:167Þ
and
S ð
ik2 δln eslS esn
S
esj Ckplj Dpβ nk h
i
AnS;β eiβ ; esS ¼ exp i kβ eiβ k2 esS xs dSðxs Þ
4πρc22
S
ðno sum on sÞ ðβ ¼ P, SV, SH Þ:
ð10:168Þ
These results are of an identical form with those obtained previously for a flat crack
in the fluid model (Eq. (10.59)) if we make the replacements
eslP esn
P P
esj Ckplj Dpβ nk
ð e i nÞ $
2ρc21
ð10:169Þ
ðei es Þ $ c1 =cβ eiβ esP ¼ gP;β
k $ k1
Using these replacements and Eq. (10.66) for the general scattering amplitude of an
elliptical crack in a fluid, in this elastic wave case we find
10.4 Approximate Scattering Solutions: Elastic Solid Model 469
ia1 a2 eslP esn
P P
esj Ckplj Dpβ nk
β
β P
P P;β
AP;β
n e i ; e s ¼ J 1 k 1 c =c
1 β ie e s r e
2ρc21 c1 =cβ eiβ esP r P;β e
for the scattered S-waves, where (see Figs. 10.16 and 10.17)
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
2 2
r eα;β ¼ a21 eqα;β u1 þ a22 eqα;β u2 ðα ¼ P, SÞ ðβ ¼ P, SV, SH Þ
ð10:173Þ
eqα;β ¼ gα;β =gα;β :
These results can also be inverted into the time domain as done for the fluid case.
When the incident wave is a P- or SH-wave, the reflection coefficients contained in
Dβp are real so that we find directly, as before
a1 a2 eslP esn
P P
esj Ckplj Dpβ nk
β P
aP;β e ; e ¼ β 2
2 F ðtÞ
P;β
n i s
P;β ð10:174Þ
2πρc1 c1 =cβ ei es r e
P
ðno sum on sÞ ðβ ¼ P, SH Þ
and
S
a1 a2 δln eslS esn
S
esj Ckplj Dpβ nk S;β
β S
anS;β ei ; es ¼ 2
2 F ðt Þ
2πρc2 c2 =cβ eiβ esS r eS;β ð10:176Þ
for the scattered S-waves. Equations (10.174) and (10.176) are also valid for an
incident SV-wave as long as the incident angle is less than the first critical angle
where the reflected P-wave direction is parallel to the surface. However, when this
critical angle is exceeded, the reflection coefficient in DSV p becomes frequency
SV
dependent (see Chap. 6) and we must break Dp into its real and imaginary parts,
DSV SV
Rp and DIp , respectively, as
When we go to invert Eqs. (10.171) and (10.172) then, we need to consider terms of
the form
ð
þ1 ð
þ1
DSV
Rp iJ 1 ðσωÞexpðiωtÞdω DSV
Ip sgnωJ 1 ðσωÞexpðiωtÞdω
1 1
ð
1 ð
1
¼ 2DSV
Rp J 1 ðσωÞ sin ðωtÞdω 2DSV
Ip J 1 ðσωÞ cos ðωtÞdω: ð10:178Þ
0 0
The first integral is just proportional to the Fα;β function appearing in Eq. (10.175)
while for the second integral we find
ð
1 (
1=σ jtj < σ
J 1 ðσωÞ cos ðωtÞdω ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð10:179Þ
σ= t2 σ 2 jtj þ t2 σ 2 jtj > σ
0
SV P a1 a2 eslP esn
P P
esj Ckplj nk n o
aP;SV
n ei ; es ¼
2
DSV P;SV
Rp F ð t Þ D SV P;SV
Ip G ð tÞ
2πρc1 ðc1 =c2 ÞeSV i es
P 2 r P;SV
e
ðno sum on sÞ
ð10:180Þ
ð10:181Þ
10.4 Approximate Scattering Solutions: Elastic Solid Model 471
and
S
a1 a2 δln eslS esn
S
esj Ckplj nk n SV S;SV o
i ; es ¼ 2
S;SV 2 DRp F ðtÞ DIp G ðtÞ
S;SV SV S;SV
an eSV S
SV
2πρc2 ei esS r e
ðno sum on sÞ
ð10:182Þ
for the scattered S-waves. Beyond the critical angle for incident SV-waves,
the existence of the Gα;SV(t) function shows the effects of pulse distortion on the
scattering amplitude which is now a linear combination of both Fα;SV(t) and Gα;SV(t)
as shown in Fig. 10.33. Below the critical angle, the scattered wave form is only a
function of Fα;SV(t).
Pulse distortion also exists when we consider the general impulse response of a
crack. Thus, if we return to Eqs. (10.167) and (10.168), we can rewrite those
equations as
e P e P e P Ckplj nk n o
β sl sn sj β P;β β P;β
AP;β e ; e P
¼ D I ð ω Þ þ D I ðω Þ
n i s
4πρc31 Rp 1 Ip 2
ð10:183Þ
ðno sum on sÞ ðβ ¼ P, SV, SHÞ
where
ð h
i
I 1α;β ðωÞ ¼ iωexp i kβ eiβ kα esα xs dSðxs Þ
S
ð h
i ð10:185Þ
I 2α;β ðωÞ ¼ sgn ω exp i kβ eiβ kα esα xs dSðxs Þ
S
β
and DIp ¼ 0 except when β ¼ SV and the SV wave is above the critical angle. If we
invert Eq. (10.184) into the time domain, the inversion of the I1α;β term follows
exactly the same procedures as in the fluid case and we find
ð
þ1
1
I 1α;β ðωÞexpðiωtÞdω
2π
1
8 þ1 9
< ð =
d ;
¼ δ gα;β x cos ηα;β =cα t Dα;β ðxÞdx
dt: ;
1
cα dDα;β cα t=gα;β cos ηα;β
¼ α;β ð10:186Þ
jg j cos ηα;β dt
where Dα;β is the crack width function along a line perpendicular to the component
of eqα;β in the plane of the crack and ηα;β is the angle between eα;β
q and the crack
surface (see Fig. 10.22). For the Iα;β
2 we have instead
ð
þ1
1
ω sgn ω I 2α;β ðωÞexpðiωtÞdω
2π
1
8 þ12 1 3 9
< ð ð =
d
¼ 42 sin ω gα;β cos ηα;β =cα t dω5Dα;β ðxÞdx : ð10:187Þ
dt: ;
1 0
The ω integration can be done directly (the limit at infinity can be ignored if we add
a small amount of damping to the system and then take the limit as this damping
goes to zero) to obtain
10.4 Approximate Scattering Solutions: Elastic Solid Model 473
ð
þ1
1
ω sgn ω I 2α;β ðωÞexpðiωtÞdω
2π
1
8 þ1 9
ð
d <1 D ðxÞ α;β =
¼ dx
dt:π t jgα;β j cos ηα;β x=cα ;
1
8 þ1 9
ð
cα d <1 Dα;β ðxÞ =
¼ α;β dx
jg j cos ηα;β dt:π x cα t=jgα;β j cos ηα;β ;
1
cα dH DðxÞ; cα t=gα;β cos ηα;β
¼ α;β ; ð10:188Þ
jg j cos ηα;β dt
where H is the Hilbert transform. Thus, Eqs. (10.183) and (10.184), when inverted
into the time domain become
(
eslP esn
P P
esj Ckplj nk dDP;β c1 t=gP;β cos ηP;β
aP;β eiβ ; esP ¼ β
DRp
n
4πρc21 jgP;β j cos ηP;β dt
P;β
cos ηP;β
β dH DðxÞ; c1 t= g
þ DIp
dt
ðno sum on sÞ ðβ ¼ P, SV, SH Þ ð10:189Þ
for the scattered S-waves. The solutions given in Eqs. (10.189) and (10.190) are
valid as long as the unit vector eα;β
q is not parallel to the crack normal. In that case,
Eqs. (10.189) and (10.190) become, simply
ik1 eslP esn
P P
esj Ckplj Dpβ nk S
AnP;β eiβ ; esP ¼
4πρc21
ðno sum on sÞ ðβ ¼ P, SV, SHÞ ð10:191Þ
474 10 Flaw Scattering
and
S
ik2 δln eslS esn
S
esj Ckplj Dpβ nk S
AnS;β eiβ ; esS ¼
4πρc22 ð10:192Þ
ðno sum on sÞ ðβ ¼ P, SV, SH Þ;
where S is the area of the crack. Inverting these expressions into the time domain,
we find, respectively
e P e P e P Ckplj D β nk S dδðtÞ
β P sl sn sj p
aP;β e ; e ¼
n i s
4πρc31 dt ð10:193Þ
ðno sum on sÞ ðβ ¼ P, SV, SH Þ
and
S
δln eslS esn
S
esj Ckplj Dpβ nk S dδðtÞ
anS;β eiβ ; esS ¼
4πρc32 dt ð10:194Þ
ðno sum on sÞ ðβ ¼ P, SV, SH Þ:
For a same mode, pulse echo setup in the case of incident P-waves,
m;P P;P
R12 ¼ 0 ðm 6¼ PÞ, R12 ¼ 1, with d ipP ¼ drpP
¼ eipP , and esn
P
¼ einP , nk ¼ eikP ,
so that expanding out the coefficient in Eqs. (10.191) and (10.193) we find
P P
eslP esn
P P
esj Ckplj Dpβ nk ¼ eilP einP eijP λδkp δlj þ μ δkl δpj þ δkj δpl 2eip eik
¼ 2ðλ þ 2μÞeinP
¼ 2ρc21 einP : ð10:195Þ
P ik1 SeinP
AP;P
n ei ; eiP ¼ ð10:196Þ
2π
P Se P dδðtÞ
aP;P
n ei ; eiP ¼ in : ð10:197Þ
2πc1 dt
have to distinguish between polarizations here and we simply let the S superscript
10.4 Approximate Scattering Solutions: Elastic Solid Model 475
denote an S-wave of any polarization. Again esjS ¼ eijS and the coefficients in
Eqs. (10.192) and (10.194) are
S
δln eslS esn
S
esj Ckplj DpS nk ¼ δln eilS einS eijS λδkp δlj þ μ δkl δpj þ δkj δpl
2dipS eikS
¼ 2μd inS
ð10:198Þ
ik2 S dinS
AnS;S eiS ; eiS ¼ ð10:199Þ
2π
S d inS dδðtÞ
anS;S eiS ; eiS ¼ : ð10:200Þ
2πc2 dt
The Kirchhoff approximation has been covered extensively because it has some key
properties that make it useful for NDE applications. As we have seen, the Kirchhoff
approximation predicts the leading edge response of volumetric flaws and the
flashpoint signals of cracks. The leading edge response in particular is important
because the large specular reflection from the front surface of a flaw is often the
largest scattered response present in the flaw signal and we have obtained a simple
analytical model of this signal that can be placed into an ultrasonic measurement
model to predict directly the measured voltage of the flaw. In recent studies of
imaging with phased arrays, the leading edge response has been shown to be the
flaw response responsible for the images generated with methods such as the
Synthetic Aperture Focusing Technique (SAFT), the Total Focusing method
(TFM), and the Physical Optics Far Field Inverse Scattering (POFFIS) method
[1]. This fact allows us to connect the images seen with these methods to the
physical properties of the flaw. The flash points predicted for cracks with
the Kirchhoff approximation are also key elements in sizing flaws, either with the
time of flight diffraction (TOFD) method or the equivalent flaw sizing methods
discussed in Chap. 15.
Although we have discussed the Kirchhoff approximation separately for fluids
and solids, for the pulse echo (same mode) response of stress-free flaws such as
voids and cracks it has been shown [2] through a combination of analytical and
476 10 Flaw Scattering
numerical evaluations that that the response in the fluid (see Eq. (10.29), for
example, for the volumetric case):
ð
ik
Aðei ; ei Þ ¼ ðei nÞexp½2ikei xs dS ð10:201aÞ
2π
Slit
and a particular component of the vector scattering amplitude for an elastic solid
given by (see Eqs. (10.136)–(10.138) for the underlying expressions):
h
i iω d β C ð h i
Aβ;β eiβ ; eiβ diβ ¼ il lkpj
D β;β
pjk exp 2ik e
β i
β
x s dSðxs Þ
4πρc2β
Slit
ðno sum on iÞ ðβ ¼ P, St , Sv Þ
ð10:201bÞ
are in fact identical, where dβi is the polarization vector of the incident wave of type
β. This component of the scattering amplitude is the one that is present in a
Thompson-Gray model of an ultrasonic flaw measurement (see Chap. 12), so that
for all flaws with stress free surfaces within the Kirchhoff approximation we can
compute the pulse echo voltage response in the Thompson-Gray model with the
fluid model of Eqs. (10.201a, 10.201b).
Since the publication of the first edition of this book it has also been shown that
the leading edge pulse echo response of a volumetric flaw, which was derived here
for an isotropic solid, is in fact also valid for a general anisotropic flaw in a general
anisotropic media [2]. From Eq. (10.156) (with drβ ¼ diβ ) this pulse echo case
yields
h
i Rβ;β ð0 Þpffiffiffiffiffiffiffiffiffiffi
R 1 R2
β β β
A β;β
ei ; ei di ¼ 12 exp 2ikβ eiβ xstat ð10:202Þ
2
for the isotropic case, which is identical for anisotropic materials for quasi-P and
quasi-S waves [2]. [Note that the reflection coefficient (at normal incidence) is
calculated in Eq. (10.202) consistent with the choice of reflected wave polarization
drβ ¼ diβ .]
For cracks, some studies have suggested that the Kirchhoff approximation is
only valid for the pulse echo responses of flat cracks when the incident wave
direction is approximately normal to the crack. However, recent studies have
shown that this is true only at single frequencies or for very narrow band transducer
responses [2]. For wide band width crack responses, the Kirchhoff approximation
has been shown to accurately predict the flashpoint signals over a wide range of
angles.
10.4 Approximate Scattering Solutions: Elastic Solid Model 477
In summary, while the Kirchhoff approximation does not capture all the physics
of the flaw scattering process, it does correctly identify some important elements of
that process that make it a very valuable modeling tool for NDE inspections.
ð10:203Þ
where, recall, the plus and minus superscripts are used to indicate quantities that are
evaluated on the host or flaw side of the surface S, respectively. From the continuity
of displacements and tractions across S we have
uþ
p ðxs ; ωÞ ¼ up ðxs ; ωÞ
ð10:204Þ
nk C0lkpj ∂uþ
p ðxs ; ωÞ=∂xj ¼ nk Cklpj ðxs Þ∂up ðxs ; ωÞ=∂xj
1
ρ1 (x), Cklpj
1 (x)
478 10 Flaw Scattering
Applying the divergence theorem to the right side of Eq. (10.205) then gives
ðn h i
αun ðy; ωÞ þ βun ðy; ωÞ ¼
scatt inc
∂ C0lkpj uj ðx; ωÞ∂Gln ðx;y;ωÞ=∂xk =∂xp
Vf
h i o
∂ C1lkpj ðxÞGln ðx; y; ωÞ∂up ðx; ωÞ=∂xj =∂xk dV ðxÞ;
ð10:206Þ
where we have dropped the minus superscript since the integration points are all
within Vf. However, we have
h i
∂ C0lkpj ∂Gln ðx; y; ωÞ=∂xk =∂xp þ ρ0 ω2 Gnj ðx; y; ωÞ ¼ δnj δðx yÞ
h i ð10:207Þ
∂ C1lkpj ðxÞ∂up ðx; ω Þ=∂x j =∂xk ¼ ρ1 ω2 u
l ðx; ωÞ
where
Δρ ¼ ρ1 ðxÞ ρ0
ð10:209Þ
ΔCklpj ¼ C1klpj ðxÞ C0klpj :
P ð
eslP esn
AP;β
n eiβ ; esP ¼ Δρω2 u~l ðx; ωÞ
4πρ0 c2p0
Vf
þ ikp0 eskP ΔCklpj ∂~
u p ðx; ωÞ=∂xj exp ikp0 x esP dV ðxÞ
ðno sum on sÞ ðβ ¼ P, SÞ ð10:210Þ
where the tilde indicates that the displacements have been normalized by the
magnitude of the incident displacement amplitude. Combining both of these
forms, we write
ð
Blnα
Anα;β eiβ ; esα ¼ 2
Δρω2 u~l ðx; ωÞ
4πρ0 cα0
Vf
þ ikα0 eskα ΔCklpj ∂~
u p ðx; ωÞ=∂xj exp ikα0 x esα dV ðxÞ
ðno sum on sÞ ðα ¼ P, SÞ ðβ ¼ P, SÞ; ð10:212Þ
where
P P
esn esl S S ð α ¼ PÞ
Blnα ¼ : ð10:213Þ
δln esn esl ð α ¼ SÞ
If the material properties of the flaw are again only slightly different from the host
material, then for an incident
h i wave of type β and unit (displacement)
plane
β β
l ¼ d il exp ik β0 ei x
amplitude, uinc we approximate the normalized displace-
ments and their derivatives inside the flaw in Eq. (10.212) as
h
i
u~l ¼ dilβ exp ikβ0 eiβ x
h
i ð10:214Þ
u l =∂xk ¼ ikβ0 eikβ d ilβ exp ikβ0 eiβ x :
∂~
480 10 Flaw Scattering
ðα ¼ P, SÞ ðβ ¼ P, SÞ; ð10:215Þ
with
f l eiβ ; esα ¼ Δρdilβ =ρ0 eskα eijβ dipβ ΔCklpj =ρ0 cα0 cβ0 : ð10:216Þ
For a homogeneous flaw fl is just a constant and can be taken outside the integral to
find
ω2 Blnα f l eiβ ; esα ð
Anα;β eiβ ; esα ¼ exp ikα0 gα;β x dV ðxÞ
4πc2α0 ; ð10:217Þ
Vf
ðα ¼ P, SÞ ðβ ¼ P, SÞ
where here
gα;β ¼ cα0 =cβ0 eiβ esα : ð10:218Þ
Comparing Eqs. (10.217) and (10.96) for the fluid case, we see that they are of
identical form if we make the replacements
f ðei ; es Þ $ Blnα f l eiβ ; esα
ei es $ gα;β : ð10:219Þ
c0 $ cα0
Thus, all the previous fluid case results can be carried over directly to the elastic
case. In all the subsequent expressions the coefficients Bαln fl(eβi ; eαs ) can be evaluated
explicitly by expanding out the terms in Eq. (10.216) where
ΔCklpj ¼ Δλδkl δpj þ Δμ δkp δlj þ δkj δlp ð10:220Þ
and we have
Also, for any vector, x, we define the vector component of x, transverse to eSs , as
ðxÞT ¼ x x esS esS : ð10:222Þ
From the results in the fluid case (see Eq. (10.106)) and the above replacements
(Eq. (10.219)) the scattering amplitude for a homogenous ellipsoidal inclusion can
be given explicitly as
a1 a2 a3 Blnα f l eiβ ; esα
Anα;β eiβ ; esα ¼
2
jgα;β j r eα;β
( )
sin kα0 gα;β r eα;β kα0 gα;β r eα;β cos kα0 gα;β r eα;β
kα0 jgα;β jr eα;β
ðα ¼ P,SÞ ðβ ¼ P,SÞ; ð10:225Þ
482 10 Flaw Scattering
where
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
2 2 2
r eα;β ¼ a21 eqα;β u1 þ a22 eqα;β u2 þ a23 eqα;β u3
ð10:226Þ
eα;β
q ¼ gα;β =gα;β :
Inverting Eq. (10.226) into the time domain, the impulse response function then is
a1 a2 a3 Blnα f l eiβ ; esα
anα;β eiβ ; esα ¼
2
2jgα;β j2 r eα;β
(
δ t gα;β r α;β =cα0 þ δ t þ gα;β r α;β =cα0
e e
)
cα0
U gα;β r α;β =cα0 , gα;β r α;β =cα0 , t
jgα;β jr eα;β
e e
ðα ¼ P, SÞ ðβ ¼ P, SÞ; ð10:227Þ
Using the replacements in Eq. (10.219) in Eq. (10.117) from the fluid case, the
general impulse response in the elastic solid case becomes
Blnα f l eiβ ; esα d2 Sα;β cα0 t=gα;β
α;β β α
an e i ; e s ¼ ð10:228Þ
4πcα0 dt2
ðα ¼ P, SÞ ðβ ¼ P, SÞ;
where Sα;β is the cross sectional area of the flaw for a plane that is perpendicular to
the eα;β
q direction (see Fig. 10.29).
In the pulse echo case, the Born approximation is in relatively simple form.
From Eqs. (10.217) and (10.224) we have
10.4 Approximate Scattering Solutions: Elastic Solid Model 483
ð
β β ω2 Δρ Δcβ0 β β
Anβ;β ei ; ei ¼ þ d exp 2ik β0 e x dV ðxÞ
2πc2β0 ρ0 cβ0 in i
ð10:229Þ
Vf
ðβ ¼ P, SÞ;
where dβin are the components of the polarization vector of the incident wave. Apart
from a sign and the polarization vector, Eq. (10.229) is identical to the fluid case
(see Eq. (10.96) for the pulse echo case combined with Eqs. (10.97) and (10.109)).
As mentioned in the Kirchhoff approximation discussion, the component of the
vector far field scattering amplitude that contributes to the measured pulse echo
voltage
response
in the
Thompson-Gray
LTI model of an ultrasonic measurement is
A eiβ ; eiβ ¼ Aβ;β eiβ ; eiβ diβ so that
ð
ω2 Δρ Δcβ0
A eiβ ; eiβ ¼ þ exp 2ikβ0 eiβ x dV ðxÞ
2πcβ0 ρ0
2 cβ0 ð10:230Þ
Vf
ðβ ¼ P, SÞ;
which is now identical to the scalar far field scattering amplitude in the fluid case.
Thus, like with the case of stress-free flaws in the Kirchhoff approximation we see
that we can calculate the pulse echo P- and S-wave flaw responses of inclusions in a
solid (in the Born approximation) with a fluid model.
Unfortunately, comparison of the Born approximation with an “exact” separa-
tion of variables solution for the pulse echo response of a spherical inclusion [3]
shows that, as the difference between properties of the host and the inclusion
increase, the Born approximation quickly loses accuracy for two reasons: (1) the
weak scattering amplitude assumption causes the delta function leading edge
response to have an incorrect amplitude and (2) the assumption that the wave inside
the inclusion travels with the wave speed of the surrounding host material causes
the time interval between the back and front surface delta functions response to be
incorrect for stronger scattering flaws. As noted in the fluid case (see Eq. (10.110))
the material coefficient appearing in Eq. (10.230) is proportional to the weak
scattering approximation of the plane wave reflection coefficient at normal
incidence between the flaw and the host material. Thus, if we replace this weak
scattering coefficient by the exact plane wave coefficient, Rβ(0 ) for a P- or S-wave,
we have
ω2 ð
A eiβ ; eiβ ¼ 2 Rβ ð0 Þ exp 2ikβ0 eiβ x dV ðxÞ
πcβ0 ð10:231Þ
Vf
ðβ ¼ P, SÞ
484 10 Flaw Scattering
and this change allows the Born approximation to correctly model the delta
function leading edge response predicted by the separation of variables solution
for a sphere (and also agrees with the Kirchhoff approximation for other inclusion
shapes) even for inclusions with properties quite different from that of the host
material [3]. For a spherical inclusion of radius a in particular we have
β β β sin 2kβ0 a 2kβ0 a cos 2kβ0 a
A ei ; ei ¼ aR ð0 Þ
2kβ0 a ð10:232aÞ
ðβ ¼ P, SÞ;
In the time domain either Eq. (10.232a) or Eq. (10.232b) produce the response of
the spherical inclusion shown in Fig. 10.35a, which is identical in general form with
the fluid case, i.e.
aRp ð0 Þ
aðei ; ei Þ ¼ fδðt 2a=c0 Þ þ δðt þ 2a=c0 Þ
2
c0 o
Uð2a=c0 , 2a=c0 ; tÞ : ð10:233Þ
2a
Fig. 10.35 (a) Born approximation for a spherical flaw with the correct front surface delta
function response, (b) modification of the Born approximation for incorrect wave speed when
traveling in the flaw, and (c) correction for the time of arrival of the front surface delta function
10.4 Approximate Scattering Solutions: Elastic Solid Model 485
To correct the error in the time separation between front and back surface delta
functions, one can try to simply replace the wave speed of the host material by that
of the flaw, causing Eq. (10.231) to become
ω2 ð
A eiβ ; eiβ ¼ 2 Rβ ð0 Þ exp 2ikβ1 eiβ x dV ðxÞ
πcβ1 ð10:234Þ
Vf
ðβ ¼ P, SÞ:
For the spherical inclusion in particular this change replaces Eq. (10.232b) by
j 2kβ1 a
A eiβ ; eiβ ¼ 4k2β1 a3 Rβ ð0 Þ 1
2kβ1 a ð10:235Þ
ðβ ¼ P, SÞ:
This makes sense from a physical standpoint since any waves traveling in the
inclusion must travel at the wave speed of the inclusion, not the host. This change
does produce the proper time interval between front and back delta function
responses, as shown in Fig. 10.35b for the spherical inclusion but it also gives an
incorrect arrival time for the front surface delta function, since that response comes
from waves that only travel in the host material up to the time it strikes the flaw so
that it must arrive at a time t ¼ 2a=cβ0 relative to the time t ¼ 0, which is when the
incident wave, traveling in the host material, arrives at the center of the sphere, not
the time t ¼ 2a=cβ1 obtained from Eq. (10.235) and shown in Fig. 10.35b. Thus
we also need to make a phase correction to make this arrival time correct while
keeping the proper time between the front and back surface responses. In the
general case this can be accomplished by the modified form
ω2 ð
A eiβ ; eiβ ¼ 2 Rβ ð0 Þexp 2ikβ1 d 1 cβ1 =cβ0 exp 2ikβ1 eiβ x dV ðxÞ
πcβ1
Vf
ðβ ¼ P, SÞ;
ð10:236Þ
where d is the distance from a fixed point in the flaw (normally taken at the flaw
center for simple shapes) to the front surface as shown in Fig. 10.36. For the
spherical inclusion, of course, d ¼ a, the radius of the sphere, producing the
response
j1 2kβ1 a
A eiβ ; eiβ ¼ 4k2β1 a3 Rβ ð0 Þexp 2ikβ1 a 1 cβ1 =cβ0
2kβ1 a ð10:237Þ
ðβ ¼ P, SÞ;
whose time domain response shown in Fig. 10.35c. We will call Eq. (10.236) the
Modified Born Approximation (MBA) [3]. It was obtained through some plausible
486 10 Flaw Scattering
but non-rigorous arguments so it is not a fundamental result but one which “fixes”
the ordinary Born approximation in a number of ways that make it more useful for
describing the inclusions often seen in NDE tests, which may not be weak scat-
terers. Note, however, that except for weak scattering flaws the MBA does not
predict the amplitude of the back surface delta function correctly, nor does it model
other flaw responses such as internal reflections, for example, or creeping waves [3]
since like the Kirchhoff approximation the Born approximation is a single scatter-
ing approximation, meaning that it only accounts for single, direct interactions of
the incident wave with the flaw geometry. For some flaw materials and geometries
those other responses may in fact be larger than the leading edge delta function
response, which is that part of the total response predicted correctly by the MBA.
In this section we will obtain some general reciprocal properties of plane wave far
field scattering amplitudes that are valid for both fluid and elastic media. There are
several ways to derive these reciprocal properties. One method (see Varadan
et al. [47]) considers the asymptotics of the far field explicitly. Here, however, we
will use a simpler approach which is based on a special form of the reciprocal
theorem [4].
ð
pb ∂pa =∂n pa ∂pb =∂n dS ¼ 0: ð10:238Þ
S
For example, for a void or rigid scatterer, Eq. (10.238) follows directly from the
boundary conditions which both solutions must satisfy, which are pa ¼ pb ¼ 0 for
the void and ∂pa =∂n ¼ ∂pb =∂n ¼ 0 for the rigid scatterer. For a general inclusion,
since both pa and pb satisfy homogeneous Helmholtz equations inside the inclusion,
the reciprocal theorem (Eq. (5.7)) applied to the volume Vf of the flaw (with zero
body force) again gives Eq. (10.238).
We will now use Eq. (10.238) to derive a reciprocal theorem that can be used to
obtain the desired reciprocal relations for the scattering amplitude. First, we break
up the total pressure for problem b into incident and scattered wave parts, i.e.
If we consider the right hand side of Eq. (10.240) and decompose pa into its incident
and scattered parts
Consider the integrand of the second integral in Eq. (10.242). If we integrate that
same integrand over both the surface of the flaw, S, and a large sphere, SR, which
surrounds the flaw, then since the scattered waves satisfy the homogeneous
Helmholtz equation in the volume between S and SR, the reciprocal theorem
(Eq. (5.7)) again gives
ð
scatt;b scatt;a
p ∂p =∂n pscatt;a ∂pscatt;b =∂n dS ¼ 0: ð10:243Þ
SþSR
However, since both pscatt;a and pscatt;b satisfy the Sommerfeld radiation conditions,
as the radius of the sphere becomes infinitely large, the integral over SR vanishes
and we obtain
ð
scatt;b scatt;a
p ∂p =∂n pscatt;a ∂pscatt;b =∂n dS ¼ 0: ð10:244Þ
S
Now, if we replace the scattered wave terms for solution b in Eq. (10.245) with the
total and incident waves for this problem, we have
ð
I ¼ pb ∂pinc;a =∂n pinc;a ∂pb =∂n dS
S
ð
þ pinc;b ∂pinc;a =∂n pinc;a ∂pinc;b =∂n dS: ð10:246Þ
S
But, both pinc;a and pinc;b are solutions of homogeneous Helmholtz equations over
the volume of the flaw, so reciprocity (Eq. (5.7)) would require this integral also to
vanish, leaving simply
ð
I ¼ pb ∂pinc;a =∂n pinc;a ∂pb =∂n dS: ð10:247Þ
S
S S
ð10:248Þ
10.5 The Far Field Scattering Amplitude and Reciprocity 489
Equation (10.248) is a reciprocity relation that is valid for any type of incident
wave. However, if we assume the incident waves for solutions a and b are plane
waves, then the pressure and normal derivative of the pressure terms for these
incident waves are explicitly
pinc;a ¼ p0a exp ik x eia
∂pinc;a =∂n ¼ ikp0a eia n exp ik x eia
ð10:249Þ
pinc;b ¼ p0b exp ik x eib
∂pinc;b =∂n ¼ ikp0b eib n exp ik x eib ;
where p~a ¼ pa =p0a and p~b ¼ pb =p0b are normalized total pressures due to incident
waves of unit pressure amplitude. From the definition of the scattering amplitude
(Eq. (10.5)), which is of the same general form as both sides of Eq. (10.250), it
follows that Eq. (10.250) can be rewritten in terms of plane wave scattering
amplitudes as
A eia ; eib ¼ A eib ; eia : ð10:251Þ
Equation (10.251) states that the amplitude of waves scattered in the eib direction
due to a plane wave incident on the flaw in the eai direction is the same as the
amplitude of waves scattered in the eia direction due to a plane wave incident on
the flaw in the ebi direction. It is in this sense that the plane wave scattering
amplitude itself can be said to be “reciprocal”.
ð
τklb nk ula τkla nk ulb dS ¼ 0; ð10:252Þ
S
since for a void, for example, τkla nk ¼ τklb nk ¼ 0, or for a rigid immobile scatterer
ula ¼ ulb ¼ 0. Alternatively, the reciprocal theorem (Eq. (5.42)) yields Eq. (10.252)
directly for a general inclusion, since both solutions a and b satisfy homogeneous
equations of motion within the flaw. Following the same procedure as in the fluid
case we can decompose solutions a and b into incident and scattered waves, i.e.
ula ¼ uinc;a
l þ uscatt;a
l
τkla ¼ τinc;a
kl
scatt;a
þ τkl
ð10:253Þ
ulb ¼ uinc;b
l þ uscatt;b
l
τklb ¼ τinc;b
kl
scatt;b
þ τkl
which is the reciprocal relationship for the elastic solid case equivalent to
Eq. (10.248). When the incident waves are plane waves, we can write
with u~la ¼ ula =u0α;a , ~τ kla ¼ τkla =u0α;a and similarly u~lb ¼ ulb =u0β;b , ~τ klb ¼ τklb =u0β;b being
the displacements and stresses due to incident waves of unit (displacement) ampli-
tude. Note that these normalized displacements are dimensionless but the normal-
ized stresses are not. From Eqs. (10.12a, 10.12b) it follows that
ðh i
4πρc2α Anα;β eiβ ; esα dnα ¼ ~τ kl nk d lα þ ikα esjα diα Cklij nk u~l exp ikα x esα dS
S
ð10:257Þ
so that the left and right sides of Eq. (10.256) can be written equivalently in terms of
scattering amplitudes as
c2β Anβ;α eα;a ; eβ;b dnβ;b ¼ c2α Anα;β eβ;b ; eα;a d nα;a ; ð10:258Þ
where a common factor of 4πρ has been canceled out. When expanded out for all
mode types, Eq. (10.258) is equivalent to the three explicit reciprocity relations
P;a
AP;P
n e ; eP;b dP;bn ¼ An e ; eP;a dnP;a
P;P P;b
AnS;S eS;a ; eS;b dnS;b ¼ AnS;S eS;b ; eS;a dnS;a ; ð10:259Þ
S;a
c2p AP;S
n e ; eP;b d P;bn ¼ c s An
2 S;P P;b
e ; eS;a dnS;a
First, consider the case where a plane harmonic pressure wave (of expðiωtÞ time
dependency) is incident on a spherical fluid flaw of radius a imbedded in an
otherwise homogeneous fluid medium (Fig. 10.38). If we let ρj and cj be the
densities and wave speeds for the flaw ðj ¼ 1Þ and surrounding medium ðj ¼ 2Þ,
respectively, then in either material the pressure satisfies the homogeneous Helm-
holtz equation
∇2 p þ k2j p ¼ 0: ð10:261Þ
To solve Eq. (10.262) by the method of separation of variables, we assume that the
pressure, p, can be written in the form of a product of terms, each dependent only on
a single coordinate, i.e. p ¼ Rðr ÞΘðθÞΦðϕÞ, which then gives
1 d 2 dR 1 d dΘ 1 d2 Φ
r þ sin θ þ kj r sin 2 θ ¼
2 2
: ð10:263Þ
Rdr dr Θ sin θdθ dθ Φ dϕ2
d2 Φ
þ γ 2 Φ ¼ 0: ð10:264Þ
dϕ2
Furthermore, since the pressure must be single valued in ϕ, Eq. (10.264) shows that
the constant γ must satisfy γ ¼ m, where m is an integer, and the corresponding
solutions for Φ are then
Φ ¼ Φm ¼ expðimϕÞ: ð10:265Þ
and from g ¼ l2
1 d dΘ m2
sin θ þ nð n þ 1 Þ Θ ¼ 0: ð10:268Þ
sin θ dθ dθ sin 2 θ
rffiffiffiffiffiffiffiffi
π
jn kj r ¼ J nþ1=2 kj r
2kj r
ð10:269Þ
rffiffiffiffiffiffiffiffi
π
nn k j r ¼ Y nþ1=2 kj r :
2kj r
Or, in terms of spherical Hankel functions of the first and second kinds, we have the
solutions:
hðn1Þ kj r ¼ jn kj r þ i nn kj r
ð10:270Þ
hðn2Þ kj r ¼ jn kj r i nn kj r :
494 10 Flaw Scattering
Only solutions in terms of the spherical Bessel functions of the first kind are
permitted if the solutions cannot be singular as r ! 0. This can be seen from the
expansion of these functions for small argument:
2n n! kj r n
lim jn kj r ¼
kj r!0 ð2n þ 1Þ!
ð2nÞ!
lim nn kj r ¼ nþ1
kj r!0 n
2 n! kj r
ð10:271aÞ
ið2nÞ!
lim hðn1Þ kj r ¼ nþ1
kj r!0 n
2 n! kj r
ið2nÞ!
lim hðn2Þ kj r ¼ nþ1 :
kj r!0 n
2 n! kj r
Similarly, only spherical Hankel functions of the first kind can be present (for exp
(iωt) time dependency) if the solutions are to satisfy the Sommerfeld radiation
conditions as kj r ! 1 since for kj r
1 all these solutions have the asymptotic
behaviors:
1
jn k j r cos kj r ðn þ 1Þπ=2
kj r
1
nn k j r sin kj r ðn þ 1Þπ=2
kj r
ð10:271bÞ
ðiÞnþ1
hðn1Þ kj r exp ikj r
kj r
ðiÞnþ1
hðn2Þ kj r exp ikj r :
kj r
This outgoing property of the spherical Hankel functions of the first kind can also be
seen in the following representation of these functions as a finite sum:
exp ikj r ðiÞnþ1 Xn
ðn þ lÞ! i l
hðn1Þ kj r ¼ : ð10:272Þ
kj r l¼0
l!ðn lÞ! 2kj r
One important property shared by all these solutions is that they satisfy the
recurrence relations
0 n
f n ðxÞ ¼ f ðxÞ f nþ1 ðxÞ
x n
ðn þ 1Þ
¼ f n ðxÞ þ f n1 ðxÞ; ð10:273Þ
x
10.6 Scattering by a Sphere: Separation of Variables 495
ð1Þ ð2Þ
where fn can be any of the functions jn, nn, hn , hn and the prime superscript
indicates differentiation on x.
Equation (10.268) also has two types of solutions which are the associated
Legendre functions Pm m
n and Qn of the first and second kinds, respectively. However,
m
the Qn functions must normally be discarded because they are singular in the range
0 θ π. The Pm n functions are given by
m=2 dm Pn ðxÞ
Pnm ðxÞ ¼ 1 x2 ð10:274Þ
dxm
in terms of Legendre polynomials Pn, where Pn ð cos θÞ ¼ P0n ð cos θÞ and explicitly
1 dn 2 n
P n ðxÞ ¼ n n
x 1 ; ð10:275Þ
2 n! dx
X
1
pinc ¼ expðik2 zÞ ¼ in ð2n þ 1Þjn ðk2 r ÞPn ð cos θÞ: ð10:276Þ
n¼0
Since this incident wave generates scattered waves that are axisymmetric
ð∂=∂ϕ ¼ 0Þ we need only retain outgoing waves in medium two for m ¼ 0,
i.e. the scattered pressure, pscatt, can be written as
X
1
pscatt ¼ Cn hðn1Þ ðk2 rÞPn ð cos θÞ: ð10:277Þ
n¼0
Similarly, within the flaw (medium one) the pressure field, p flaw, must be axisym-
metric and non-singular so that we have
X
1
p flaw ¼ Dn jn ðk1 rÞPn ð cos θÞ: ð10:278Þ
n¼0
At the spherical interface between the flaw and the surrounding medium ðr ¼ aÞ,
the pressure and normal velocity must be continuous so these fields must satisfy the
boundary conditions
496 10 Flaw Scattering
and
( 0 0 )
X1
in ð2n þ 1Þk2 ½jn ðk2 aÞ þ Cn k2 hðn1Þ ðk2 aÞ Dn k1 ½jn ðk1 aÞ
0
Pn ð cos θÞ ¼ 0; ð10:281Þ
n¼0
ρ2 ρ1
where the prime superscript denotes differentiation of the enclosed function with
respect to its entire argument.
Since the coefficients of each term in Eqs. (10.280) and (10.281) must vanish, we
obtain a set of simultaneous equations for the unknown coefficients Cn and Dn
which can be written in matrix form as
2 3 2 3
hðn1Þ ðk2 aÞ jn ðk1 aÞ in ð2n þ 1Þjn ðk2 aÞ
6 ð1Þ 0 0 7
Cn 6 n 0 7
4 h ð k 2 aÞ ½jn ðk1 aÞ 5 D ¼ 4 i ð2n þ 1Þ½jn ðk2 aÞ 5: ð10:282Þ
n
n
ρ2 c 2
ρ2 c 2 ρ1 c 1
where
0 0
jn ðk1 aÞ hðn1Þ ðk2 aÞ hðn1Þ ðk2 aÞ½jn ðk1 aÞ
Δ¼ : ð10:284Þ
ρ2 c2 ρ1 c 1
These results are for a general fluid inclusion. We can obtain the special case of a
rigid inclusion by letting ρ1 c1 ! 1. In this case the scattered wave coefficients
reduce to
10.6 Scattering by a Sphere: Separation of Variables 497
0
in ð2n þ 1Þ½jn ðk2 aÞ
Cn ¼ ð1Þ 0
h n ð k 2 aÞ
ð10:285Þ
in ð2n þ 1Þ n jn ðk2 aÞ k2 a jnþ1 ðk2 aÞ
¼ h i ;
ð1Þ
n hðn1Þ ðk2 aÞ k2 a hnþ1 ðk2 aÞ
where use has been made of the recursion relations (Eq. (10.273)). Similarly, we
can obtain the limiting case of a void by letting ρ1 c1 ! 0 giving
In the far field of the flaw, the asymptotic values of the spherical Hankel function
(Eq. (10.271b)) can be used in the expression for the scattered waves to obtain the
form
( )
expðik2 r Þ 1 X 1
pscatt ¼ ðiÞnþ1 Cn Pn ð cos θÞ ; ð10:287Þ
r k2 n¼0
which shows that the plane wave far field scattering amplitude for the sphere is
given by
1X 1
Að e z ; e s Þ ¼ ðiÞnþ1 Cn Pn ð cos θÞ: ð10:288Þ
k2 n¼0
1 X1
Aðez ; ez Þ ¼ ðiÞnþ1 Cn ; ð10:289Þ
k2 n¼0
which for the case of a rigid sphere, for example, becomes explicitly
i X1
ð1Þn ð2n þ 1Þ n jn ðk2 aÞ k2 a jnþ1 ðk2 aÞ
Ar ðez ; ez Þ ¼ h i : ð10:290Þ
ð1Þ
k2 n¼0 n hð1Þ ðk2 aÞ k2 a h ðk2 aÞ
n nþ1
Recall, we previously found a high frequency leading edge approximation for the
pulse echo response of a rigid ellipsoid (see Eq. (10.54)). Specializing that equation
to a sphere gives
498 10 Flaw Scattering
a expð2ik2 aÞ
Ar ðez ; ez Þ ¼ : ð10:291Þ
2
If we go back the general fluid inclusion case and consider the behavior of the far
field scattering amplitude at low frequencies, then one needs to keep, to first order,
only the first two terms in the series of Eq. (10.288), i.e.
i C0 C1
Aðe i ; e s Þ ffi cos θ; ð10:292Þ
k2 k2
where C0 and C1 are given by Eq. (10.283). At low frequencies the arguments of all
the functions appearing in Eq. (10.283) are small, so that from the properties of
these functions, we have, approximately
where use has also been made of the recursion relations (Eq. (10.273)). Placing
these approximations into Eq. (10.283) and collecting terms, one then finds
500 10 Flaw Scattering
i 3 λ2
C0 ffi ð k 2 aÞ 1
3 λ1
ð10:294Þ
3 ρ2 ρ1
C1 ffi ðk2 aÞ ;
2ρ1 þ ρ2
where λi ði ¼ 1, 2Þ are the bulk moduli for the flaw and surrounding medium,
respectively. Using these low frequency approximations in Eq. (10.292), the scat-
tering amplitude at low frequencies becomes explicitly
1 2 3 Δλ 3Δρ
Aðei ; es Þ ffi k2 a cos θ ; ð10:295Þ
3 λ1 2ρ1 þ ρ2
f ðei ; es Þ
Aðei ; es Þ ¼ f sin ðk2 jei es jaÞ k2 jei es ja cos ðk2 jei es jaÞg;
k2 jei es j3
ð10:296Þ
with
Δλ Δρ
f ðei ; es Þ ¼ ðei es Þ; ð10:297Þ
λ1 ρ1
Comparing Eqs. (10.298) and (10.295) we see that in the weak scattering limit
ðρ1 ffi ρ2 , λ1 ffi λ2 Þ both the exact solution and the Born approximation results are
identical. For strong scatterers, however, significant differences can exist. For
example for a rigid scatterer ðλ1
λ2 , ρ1
ρ2 Þ the exact solution gives
1 3
Ar ðei ; es Þ ffi k22 a3 1 cos θ ; ð10:299Þ
3 2
1
Ar ðei ; es Þ ffi k22 a3 f1 cos θg: ð10:300Þ
3
10.6 Scattering by a Sphere: Separation of Variables 501
In the back scatter case ðθ ¼ π Þ the Born and exact results differ by only 25 % for
the rigid sphere, but in the forward scattering case ðθ ¼ 0Þ, the Born approximation
fails completely, incorrectly predicting a zero amplitude. A similar type of behavior
was found for the scattering of a void in an elastic medium in the Born approxi-
mation [6]. These results suggest that in general if one wants to apply the Born
approximation to cases where the flaw properties differ substantially from the host
material then the scattered wave direction should not be too far from back scatter
(pulse-echo) conditions.
The results obtained previously for the scattering of the sphere assumed that the
incident wave direction was along the z-axis. In Chap. 13, we will need to evaluate
scattering responses where both the incident and scattered wave directions are both
in general directions with respect to a fixed (x, y, z) coordinate system (Fig. 10.41).
If we use spherical coordinates (θi, ϕi) to represent the incident wave direction and
coordinates (θs, ϕs) to represent the scattered wave direction, with β being the angle
between the two directions, then from Eq. (10.288) we can write
1X 1
Aðei ; es Þ ¼ ðiÞnþ1 Cn Pn ð cos βÞ; ð10:301Þ
k2 n¼0
where ei ¼ ei ðθi ; ϕi Þ and es ¼ es ðθs ; ϕs Þ. But from the addition theorem for the
Legendre polynomials [7]
c z
er
θr
y
es φr
x
502 10 Flaw Scattering
X n
ðn mÞ! m
Pn ð cos βÞ ¼ Pn ð cos θi ÞPnm ð cos θs Þexp½imðϕi ϕs Þ ð10:302Þ
m¼n
ð n þ mÞ!
1X 1 X n
ðn mÞ!
Að e i ; e s Þ ¼ ðiÞnþ1 Cn Pnm ð cos θi ÞPnm ð cos θs Þexp½imðϕi ϕs Þ:
k2 n¼0 m¼n ðn þ mÞ!
ð10:303Þ
1X 1 X n
ðn mÞ!
Aðei ðθi ; ϕi Þ; er ðθr ; ϕr ÞÞ ¼ ðiÞnþ1 ð1Þnm Cn
k2 n¼0 m¼n ðn þ mÞ!
Pnm ð cos θi ÞPnm ð cos θr Þexp½imðϕi ϕr Þ; ð10:304Þ
which is in a general pitch-catch form that will be used later in Chap. 13.
The method of separation of variables can also be used to find the waves scattered
by spherical flaws in an elastic medium. This problem has been considered by a
number of authors including Ying and Truell [8], Pao and Mow [9], and Varadan
et al. [10]. Here, we will follow the general approach laid out by Graff [11]. Spe-
cifically, we will consider the case where a plane compressional wave is incident on
a spherical inclusion of radius a in an otherwise infinite homogeneous elastic
medium. The density and longitudinal and shear wave speeds of the flaw and
surrounding medium will be taken to be (ρ1, cp1, cs1) and (ρ2, cp2, cs2), respectively.
If, as in the fluid case, we take the incident wave along the z-axis, then the scattered
waves disturbances will again be axisymmetric. In this case, as Graff [11] shows,
we can represent the displacement field in spherical coordinates in terms of two
scalar potentials, i.e.
∂ψ
u ¼ ∇ϕ þ ∇ eϕ ; ð10:305Þ
∂θ
where the potentials both satisfy (for harmonic disturbances of expðiωtÞ time
dependency) scalar Helmholtz equations
10.6 Scattering by a Sphere: Separation of Variables 503
∇2 ϕ þ k2pj ϕ ¼ 0
ð10:306Þ
∇2 ψ þ k2sj ψ ¼ 0
for either the flaw ðj ¼ 1Þ or the host medium ðj ¼ 2Þ. The incident plane P-wave in
this case will be taken to be
X1
ϕinc ¼ Φ0 exp ikp2 z ¼ in ð2n þ 1Þjn kp2 r Pn ð cos θÞ ð10:307Þ
n¼0
and the scattered P- and S-waves, since they both must satisfy radiation conditions
at infinity, can be represented in spherical coordinates as
X
1
ϕscatt ¼ An hðn1Þ kp2 r Pn ð cos θÞ
n¼0
ð10:308Þ
X
1
ψ scatt
¼ Bn hðn1Þ ðks2 r ÞPn ð cos θÞ:
n¼0
Inside the spherical flaw, since the fields must be finite at the origin, the refracted
P- and S-waves can be represented in the form
X
1
ϕflaw ¼ Cn jn kp1 r Pn ð cos θÞ
n¼0
ð10:309Þ
X
1
ψ flaw
¼ Dn jn ðks1 r ÞPn ð cos θÞ;
n¼0
where, following Graff [11] minus signs are introduced to make some later expres-
sions more symmetric.
The displacements associated with these compressional and shear wave poten-
tials are given in spherical coordinates by
∂ϕ 1 1 ∂ϕ ∂
ur ¼ þ ðDθ ψ Þ, uθ ¼ ðDr ψ Þ; ð10:310Þ
∂r r r ∂θ ∂θ
where
1 ∂ðrf Þ 1 ∂ ∂f
Dr f ¼ , Dθ f ¼ sin θ ð10:311Þ
r ∂r sin θ ∂θ ∂θ
k2s 2 ∂ϕ 1 Dθ ∂ψ ψ
τrr ¼ 2μ ϕ Dθ ϕ þ
2 r ∂r r 2 r ∂r r
8
9
< k2s 2k2p 1 ∂ϕ 1 cot θ ∂ϕ=
τθθ ¼ 2μ ϕþ þ 2 Dθ ϕ 2
: 2 r ∂r r r ∂θ ;
cot θ ∂ 1 ∂ψ
þ 2μ Dr ψ Dθ ð10:312Þ
r ∂θ r ∂r
8
9
< k2s 2k2p 1 ∂ϕ cot θ ∂ϕ 1 cot θ ∂ =
τϕϕ ¼ 2μ ϕþ þ 2 þ 2 Dθ ψ Dr ψ
: 2 r ∂r r ∂θ r r ∂θ ;
∂ 2 ∂ϕ 2 2 ∂ψ 2 2
τrθ ¼ μ 2 ϕ þ k2s ψ þ þ 2 ψ þ 2 Dθ ψ ;
∂θ r ∂r r r ∂r r r
where ϕ and ψ can be for either medium and where μ, k2p , k2s ¼ μ1 , k2p1 , k2s1 for the
flaw and μ, k2p , k2s ¼ μ2 , k2p2 , k2s2 for the surrounding medium. If we place the
separation of variables solutions for the two media into these displacement and
stress expressions, we obtain, for medium one (the flaw)
ϕ1 ¼ ϕ flaw , ψ 1 ¼ ψ flaw
1X 1
ð ur Þ 1 ¼ ðCn ε13 þ Dn ε14 ÞPn
r n¼0
1X 1
dPn
ðuθ Þ1 ¼ ðCn ε23 þ Dn ε24 Þ
r n¼0 dθ ð10:313Þ
2μ X1
ðτrr Þ1 ¼ 21 ðCn ε33 þ Dn ε34 ÞPn
r n¼0
2μ X 1
dPn
ðτrθ Þ1 ¼ 21 ðCn ε43 þ Dn ε44 Þ
r n¼0 dθ
for the incident wave contributions and the functions εij for the scattered waves in
medium two are
ð1Þ
ε11 ¼ n hðn1Þ kp2 r kp2 r hnþ1 kp2 r
ε21 ¼ hðn1Þ kp2 r
ð1Þ
ε31 ¼ n2 n k2s2 r 2 =2 hðn1Þ kp2 r þ 2kp2 r hnþ1 kp2 r
ð1Þ
ε41 ¼ ðn 1Þhðn1Þ kp2 r kp2 r hnþ1 kp2 r
ð10:316Þ
ε12 ¼ nðn þ 1Þhðn1Þ ðks2 r Þ
ð1Þ
ε22 ¼ ðn þ 1Þhðn1Þ ðks2 r Þ þ ks2 r hnþ1 ðks2 r Þ
n o
ð1Þ
ε32 ¼ nðn þ 1Þ ðn 1Þhðn1Þ ðks2 r Þ ks2 r hnþ1 ðks2 r Þ
ð1Þ
ε42 ¼ n2 1 k2s2 r 2 =2 hðn1Þ ðks2 r Þ ks2 r hnþ1 ðks2 r Þ;
while for the waves in medium one (the flaw) the functions εij are
ε13 ¼ n jn kp1 r kp1 r jnþ1 kp1 r
ε23 ¼ jn kp1 r
ε33 ¼ n2 n k2s1 r 2 =2 jn kp1 r þ 2kp1 r jnþ1 kp1 r
ε43 ¼ ðn 1Þjn kp1 r kp1 r jnþ1 kp1 r
ð10:317Þ
ε14 ¼ nðn þ 1Þjn ðks1 r Þ
ε24 ¼ ðn þ 1Þjn ðks1 r Þ þ ks1 r jnþ1 ðks1 r Þ
ε34 ¼ nðn þ 1Þ ðn 1Þjn ðks1 r Þ ks1 r jnþ1 ðks1 r Þ
ε44 ¼ n2 1 k2s1 r 2 =2 jn ðks1 r Þ ks1 r jnþ1 ðks1 r Þ:
506 10 Flaw Scattering
ður Þ1 ¼ ður Þ2
ð u θ Þ 1 ¼ ð uθ Þ 2
ð10:318Þ
ðτrr Þ1 ¼ ðτrr Þ2
ðτrθ Þ1 ¼ ðτrθ Þ2 :
X
1
ðΦ0 E1 þ An E11 þ Bn E12 þ Cn E13 þ Dn E14 ÞPn ¼ 0
n¼0
X1
dPn
ðΦ0 E2 þ An E21 þ Bn E22 þ Cn E23 þ Dn E24 Þ ¼0
n¼0
dθ
ð10:319Þ
X
1
ðΦ0 E3 þ An E31 þ Bn E32 þ μr Cn E33 þ μr Dn E34 ÞPn ¼ 0
n¼0
X1
dPn
ðΦ0 E4 þ An E41 þ Bn E42 þ μr Cn E43 þ μr Dn E44 Þ ¼ 0;
n¼0
dθ
where
Ei ¼ εi jr¼a , Eij ¼ εij r¼a ð10:320Þ
and where μr ¼ μ1 =μ2 is a non-dimensional ratio of shear moduli. Since both Pn and
dPn/dθ form complete sets of functions, each coefficient of these functions must
themselves vanish in Eq. (10.319), giving a set of four equations for An, Bn, Cn, Dn.
In matrix form, these equations are
2 32 3 2 3
E11 E12 E13 E14 An E1
6 E21 E22 E23 E24 7 6 Bn 7 6 E2 7
6 76 7 ¼ Φ0 6 7: ð10:321Þ
4 E31 E32 μr E33 μr E34 54 Cn 5 4 E3 5
E41 E42 μr E43 μr E44 Dn E4
10.6 Scattering by a Sphere: Separation of Variables 507
From the general formulation leading to Eq. (10.321) one can also derive the
solution for other special cases. For a spherical void, for example, the boundary
conditions on r ¼ a are
ðτrr Þ2 ¼ 0
ð10:322Þ
ðτrθ Þ2 ¼ 0
In this case, for the fluid inclusion we can still use the potentials appearing in
Eqs. (10.310)–(10.312), if we take ψ 1 ¼ 0 (no shear waves in the fluid) and consider
the limit as μ1 ! 0. Then we find that the pressure, ( p)1, in the fluid is given by
ðpÞ1 ¼ ðτrr Þ1 ¼ ðτθθ Þ1 ¼ τϕϕ 1 ¼ ρ1 ω2 ϕ1 ð10:324Þ
ður Þ1 ¼ ður Þ2
ðpÞ1 ¼ ðτrr Þ2 ð10:325Þ
ðτrθ Þ2 ¼ 0;
where
f ρ1 k2s2 a2
E33 ¼ jn kp1 a : ð10:327Þ
2ρ2
508 10 Flaw Scattering
Once the coefficients are obtained in these problems (Eqs. (10.321), (10.323), and
(10.326)) the far field displacements of the scattered waves can be obtained as
( 1 )
exp ikp2 r X nþ2
ur ffi ðiÞ An Pn ð cos θÞ
r n¼0
( ) ð10:328Þ
expðiks2 r Þ X 1
nþ2 dPn ð cos θÞ
uθ ffi ðiÞ Bn :
r n¼0
dθ
Note that in the problems considered in this section the incident wave has the
displacement (see Eqs. (10.305) and (10.307)) given by
uz ¼ ikp2 Φ0 exp ikp2 z : ð10:329Þ
Since we previously defined the far field plane wave scattering amplitude, Aα;β(eβi ; eαs ),
as the coefficients of a spherically spreading wave of type α due to an incident plane
wave of type β having a unit incident displacement amplitude (see Eqs. (10.11a,
10.11b)), it follows that for the scattered P-waves
( )
1 X 1
nþ1
A P;P
ðez ; er Þ ¼ ðiÞ An Pn ð cos θÞ er ð10:330Þ
kp2 Φ0 n¼0
where er and eθ are unit vectors in the spherical coordinate r- and θ-directions,
respectively, and we have used the relationship dPn ð cos θÞ=dθ ¼ P1n ð cos θÞ in
Eq. (10.331).
There are more special cases considered by Graff and others [11, 12] that could
also be analyzed but the ones described previously (elastic inclusion, void, and fluid
inclusion) are those likely most useful from an NDE standpoint. It is also possible to
solve for the scattering of a sphere due to an incident plane shear wave [13–15] in a
fashion similar to what was done here for an incident P-wave (see Problem 10.13).
As in the fluid case, some explicit low frequency results can also be obtained from
the separation of variables solutions, as shown, for example, by Harker [16].
Separation of variables solutions are powerful tools for solving scattering prob-
lems when the shape of the boundary of the scatterer matches a coordinate surface
of an appropriate coordinate system. For fluid (scalar) scattering problems, there are
a total of eleven coordinate systems in which the Helmholtz (wave) equation is
10.6 Scattering by a Sphere: Separation of Variables 509
separable [17]. These include (1) rectangular, (2) circular cylindrical, (3) elliptic
cylindrical, (4) parabolic cylindrical, (5) spherical, (6) conical, (7) parabolic,
(8) prolate spheroidal, (9) oblate spheroidal, (10) ellipsoidal, and (11) paraboloidal.
For elastic scattering problems, which are governed in general by vector Helmholtz
(wave) equations, separation of variables is possible only in the first five of the
above cases. Of those five cases, only the rectangular and spherical cases are three-
dimensional in nature, and of those two shapes only the sphere is an appropriate
shape for representing a “real” isolated flaw. Thus, the sphere is typically the only
canonical 3-D shape of interest in NDE applications that can be solved exactly by
the method of separation of variables.
V0 ( w) = s( w) P(w ) M( w) C1 (w ) T1 (w ) A( w) T2 (w ) C2 (w )
A ð ωÞ ¼ A ð e i ; e s Þ
We have examined a large variety of flaw scattering examples in this chapter. In this
section we will discuss MATLAB implementations of the pulse-echo far field
scattering of spherical scatterers and circular cracks. These examples were chosen
since the flaw response is available in analytical form for these cases and because
these canonical geometries are highly important in calibration and sizing applica-
tions. Note that for the Kirchhoff approximation in all these cases the fluid and
elastic cases are identical so that we can use the simpler fluid results directly in
NDE applications. In Schmerr and Song [18], one can find additional MATLAB
functions for scatterers such as side-drilled holes, and for some of the separation of
variables solutions presented in this chapter. The only separation of variables
solution discussed below is for a rigid sphere in a fluid.
The pulse-echo response of a spherical void is a special case of Eq. (10.37) for
an ellipsoid. This case was implemented in the MATLAB function, A_void_pe
which has the calling sequence
The function returns the far field scattering amplitude of the void, A, (in mm) at a
frequency, f, (in MHz) for a void of radius, b, (in mm) in a material with wave
speed, c, (in m/s) as found in the Kirchhoff approximation. The MATLAB function
A_crack_pe with the calling sequence
likewise uses the Kirchhoff approximation (see Eq. (10.67)) to model the far field
scattering amplitude, A, (in mm) of a circular crack of radius b, (in mm) where the
incident direction is at an angle, angd, (in degrees) with respect to the crack normal
and f and c are again the same frequency parameters found in A_void_pe. The third
MATLAB function is A_incl_pe. It has the calling sequence
This function returns the far field scattering amplitude, A, (in mm) of a weakly
scattering inclusion of radius, b, (in mm) using the modified Born approximation
(Eq. (10.237)). The frequency, again is f (in MHz) and (c1, c0) are the wave speeds
in the flaw and host material, respectively, (in m/s) and (d1, d0) are the densities of
the flaw and host (in arbitrary but consistent units).
Finally, we have included the function sphere_rigid_pe. It has the calling
sequence
This function returns the pulse-echo far field scattering amplitude, A, (in mm)
for a rigid sphere of radius, b, (in mm) in a fluid whose wave speed is c (in m/s). The
function uses the method of separation of variables (see Eq. (10.290)) so it is an
“exact” case that can be used to compare to approximate solutions.
We will use these frequency domain functions and Fourier analysis to generate
the time domain far field scattering amplitudes of these flaws. First we will generate
the necessary frequency domain and time domain values using the MATLAB
function s_space (see Appendix A)
Here, we have generated 1024 frequencies, ranging from 0 to 100 MHz, with a
sampling spacing of approximately df ¼ 0.1 MΗz. The reciprocal of this spacing is
also the length of the time domain window (10.24 μs) which is generated again with
s_space and the sampling interval in the time domain, dt, is .01 μs. For the void we
will let the radius b ¼ 2 mm, and take the wave speed c ¼ 5900 m/s:
>> b ¼ 2;
>> c ¼ 5900;
This filter simply has values of one below the frequency fstart and then smoothly
tapers those values to zero for all frequencies at or above fend. In this case we
choose fstart ¼ 10 MHz and fend ¼ 20 MHz to generate very wide bandwidth
signals. Our filtered scattering amplitude values, Af, then are generated as:
We can invert these filtered values back into the time domain with the MATLAB
function IFourierT (see Appendix A). Since we have only generated scattering
amplitudes for positive frequencies we must in general first take one half the zero
frequency value of the scattering amplitude (which is already zero in these cases)
and then take twice the real part of the inverse Fourier transform (see Schmerr and
Song [18]), giving the time domain scattering amplitude, at:
512 10 Flaw Scattering
In problems like this the time domain function often appears split between the
beginning and ending time values of the window, so our last processing step is to
plot the time domain function, using the MATLAB functions c_shift and t_shift:
The c-shift function simply moves the last n components of the vector y into the
first n components and shifts the remaining components of y to follow those n
components and places the result in the vector ys. The function t_shift does the
same circular shift on the time axis and changes those values so that the arrival
times remain correct. In our case, the resulting plot of the time domain scattering
amplitude of the void produces the centered wave form shown in Fig. 10.42. We see
a bandlimited leading edge delta function response followed by the constant
response over the lit surface (as discussed previously—see Fig. 10.11 for the
ellipsoid). In a real measured flaw signal there will also be a loss of low frequency
signals, which will effectively make the lit region response very small (as well as
other signals not predicted by the Kirchhoff approximation, such as creeping
waves), leaving primarily the leading edge delta function.
The simulation of the crack far field scattering amplitude proceeds in exactly
the same fashion as for the void. Here the only difference is that we need to specify
Fig. 10.43 The time domain pulse-echo far field scattering amplitude of a circular 2 mm radius
crack in the Kirchhoff approximation
the incident wave direction with respect to the crack normal by giving the parameter
angd. In this case we let angd ¼ 30 and calculated the time domain scattering
amplitude with all the other parameters and steps unchanged:
The result is shown in Fig. 10.43, which clearly shows the flashpoints for
this 2 mm radius circular crack.
For the spherical inclusion, again we kept the same parameters as for the void
except now that we have to specify the wave speed and density values for both the
flaw and the host material before computing the frequency domain scattering
amplitude. Here we chose
>> c1 ¼ 6000;
>> c2 ¼ 5900;
>> d1 ¼ 7.9
>> d2 ¼ 7.0
>> Ai ¼ A_incl_pe( f, b, c1, d1, c0, d0);
Fig. 10.45 The time domain pulse-echo far field scattering amplitude of a rigid sphere in water as
calculated with (a) the method of separation of variables, and (b) with the Kirchhoff
approximation
travel multiple times around the sphere, but these occur at much later times and are
generally not easily seen as the creeping waves radiate energy as they propagate
around the sphere and become very small. The creeping waves are absent in the
Kirchhoff solution (Fig. 10.45b) which only has the leading edge response followed
by a constant lit surface response until t ¼ 0. In both cases the leading edge
responses extend beyond the maximum vertical values shown in Fig. 10.45 but
they have been truncated to better show the additional small flaw responses.
The Kirchhoff approximation has been widely used for acoustic, electromagnetic,
and elastic wave scattering problems so that we have given below only a small
sampling of the available literature. The general approach and notation used in this
chapter for elastic waves follows that of Sedov and Schmerr [19].
The Born approximation also has a long history in a variety of fields. Gubernatis
et al. [6] applied this approximation to elastic wave scattering problems, where it
has since seen a number of applications, including the development of flaw sizing
algorithms, which will be discussed later in Chap. 15. Actually, the Born approx-
imation we presented here is only the first term in a complete series of terms in the
low frequency, weak scattering limit [6]. Higher order terms, however, must
typically be obtained numerically and can be used to assist in the solution of
scattering problems via other methods, such as boundary elements [20].
Although we are not considering numerical methods for solving scattering prob-
lems in this book, there are a number of references available that outline the use of
methods such as boundary elements [21, 22], finite differences [23], finite elements,
[24], the method of optimal truncation (MOOT) [25] and the closely related T-matrix
methods [26], and the elastodynamic finite integration technique [27].
Reciprocity has been used here to derive some basic properties of far field
scattering amplitudes, following the approach of Tan [4]. As we will see in later
chapters, reciprocity will also play a key role in modeling the entire ultrasonic
measurement process. As mentioned in Sect. 10.6, the spherical geometry is
essentially the only type of 3-D shape of interest in NDE applications for which
an exact elastic wave scattering solution is possible through the use of the method of
separation of variables. In two-dimensions, the problem of scattering from an
infinitely long circular cylinder can similarly be solved exactly. Such canonical
shapes are extremely important because they provide independent checks of the
numerical methods that are needed to solve for the scattering from more compli-
cated shapes, and because spheres and cylinders are commonly used as reference
scatterers in NDE calibration setups. We have given here only a limited number of
references on the scattering of spheres and cylinders, primarily for elastic wave
problems, but there is also an extensive literature on analogous acoustic and
electromagnetic problems [10, 28, 29]. We have also included a number of refer-
ences [30–47] that can be used for additional reading on many of the topics
discussed in this Chapter.
516 10 Flaw Scattering
10.9 Problems
10.1. Consider a plane harmonic pressure wave (of unit amplitude and expðiωtÞ
time dependency) incident on a spherical void of radius a in a fluid
(Fig. P10.1). From Eq. (10.29), in the Kirchhoff approximation the far
field scattering amplitude in pulse-echo is given by
ð
ik
Av ðei ; ei Þ ¼ ðei nÞexp½2ikðei xs ÞdSðxs Þ: ðP10:1Þ
2π
Slit
θ spherical scatterer
a
y
φ
x
ei
incident wave
10.2. For the spherical void scattering problem of Problem 10.1, show that the
pulse echo scattering amplitude can be written as
8 9
>
< ðπ >
=
ka d
Av ðei ; ei Þ ¼ expð2ika cos θÞ sin θ dθ : ðP10:2Þ
2 dk >
: >
;
π=2
10.3. For a rigid sphere in a fluid, write down the cross sectional area function S(z)
explicitly and by differentiating it, obtain the pulse-echo response directly.
Show that your result agrees with Eq. (10.41).
10.4. For a circular “crack” in a fluid, write down the crack width function D(z)
explicitly and by differentiating it, obtain the pulse-echo response directly.
Show that your results agree with Eq. (10.72).
10.5. Consider a plane harmonic pressure wave in a fluid (of unit amplitude and
expðiωtÞ time dependency) incident on a void in the shape of an ellipsoid,
cylinder, cone, or general surface of revolution, as shown in Fig. P10.2,
where the incident wave direction is along the z-axis.
(a) In the Kirchhoff approximation, evaluate the pulse-echo response for
these voids by setting up an explicit expression for the area function
S and differentiating (see Eq. (10.48)). Show that when the general
surface of revolution is a cone, the results agree with those obtained for
the cone directly.
(b) From the results of part (a), determine the leading edge responses for
the ellipsoid, cylinder, and cone. What can you say about the leading
edge response for the general surface of revolution?
(c) For the ellipsoid, cylinder, and cone, determine, in the Kirchhoff
approximation, the far field scattering amplitude as a function of
frequency and sketch the magnitude of the far field scattering ampli-
tude behavior for each of these cases.
a b
x x
a1 a3
z z
a2 a
y L
y
c x d x
r = f(z)
θ
z z
y y
L L
expðikRÞ
p ¼ p0 Aðei ; es Þ pffiffiffi ; ðP10:4Þ
R
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where R ¼ ðx x0 Þ2 þ ðy y0 Þ2 and the far field scattering amplitude is
given by
1=2 ð
i
Aðei ; es Þ ¼ p =∂n þ ik p~ðes nÞexpðik xs es Þds;
½∂~ ðP10:5Þ
8πk
C
where n is the unit outward normal and the line integration is in the
clockwise direction around C (Fig. P10.3).
y
n(xs )
incident wave
ei xs
x
C
cylindrical scatterer
10.8. In the Kirchhoff approximation, determine the far field scattering amplitude
for the 2-D scattering of a rigid cylinder, using the results of Problem 10.7.
Based on this solution, consider the following cases:
(a) For the pulse-echo case and a circular cylinder of radius a, write the
scattering amplitude expression in terms of polar coordinates
(Fig. P10.4) and do the θ integration exactly (Hint: use the method of
Problem 10.2)
(b) For the pulse-echo case of part (a), do the θ integration instead by the
method of stationary phase. Show at high frequencies ðka
1Þ the
“exact” results of part (a) agree with the stationary phase result
(c) Consider the more general pitch-catch case and a circular cylinder of
radius a. As in part (a), write the scattering amplitude expression in
terms of polar coordinates. In this case, do the θ integration by the
method of stationary phase and show that it agrees with (b) in the
special case of pulse-echo scattering. Interpret physically the meaning
of the stationary phase solution.
y
incident wave n
a
ei
θ
x
circular scatterer
Fig. P10.4 Plane wave incident on an infinitely long cylindrical scatterer in a fluid
10.9. Do the 2-D problem of the plane wave scattering of a rigid circular cylinder
of radius a exactly by the method of separation of variables (Fig. P10.4),
i.e. write a series expansion for the incident plane wave as
X
1
expðikzÞ ¼ εm im J m ðkRÞ cos ðmθÞ ðP10:6Þ
m¼0
X
1
pscatt ¼ εm im Cm Hðm1Þ ðkRÞ cos ðmθÞ; ðP10:7Þ
m¼0
ð1Þ
where Jm is a Bessel function of order m, Hm is a Hankel function of the first
kind of order m, and
520 10 Flaw Scattering
1 m¼0
εm ¼ : ðP10:8Þ
2 m>0
(a) Using this solution, obtain an exact series expansion for the far field
scattering amplitude for the general pitch-catch case. For a pulse-echo
setup, plot the magnitude of the normalized exact far field scattering
pffiffiffi
amplitude, jAðei ; ei Þ= aj versus ka, and compare with the same
results based on the Kirchhoff approximation (see Problem 10.8(a)).
10.10. Invert Eq. (10.93) back into the time domain and use that result to obtain an
expression for the far field scattered pressure from an arbitrary homoge-
neous volumetric flaw in a fluid (in the Born approximation) due to an
incident plane wave of arbitrary time dependency.
10.11. Invert Eq. (10.59) back into the time domain and use that result to obtain an
expression for the far field scattered pressure from an arbitrary shaped plane
crack in a fluid (in the Kirchhoff approximation) due to an incident plane
wave of arbitrary time dependency.
10.12. Consider the two dimensional harmonic
motion of an elastic solid, i.e.
u ¼ ux ðx; y; ωÞex þ uy ðx; y; ωÞey expðiωtÞ for the displacement. Show
that in this 2-D case the far field displacement components for the scattering
from a two dimensional flaw in an elastic solid (whose dimensions are
uniform in the z-direction) can be written as
expðik RÞ
expðik RÞ
eiβ ; esP pffiffiffi þ U 0 AαS;β eiβ ; esS
1 2
uαscatt ¼ U0 AP;β
α pffiffiffi
R R ðP10:9Þ
ðβ ¼ P, SV Þ;
where the Greek indices here only range over values of (1, 2) and
AαP;β eiβ ; esP ¼ esηP esα
P P;β
fη
ðP10:10Þ
AαS;β eiβ ; esS ¼ δαη esηS esα S
f ηS;β :
In this case
1=2 ð
i Cηγδσ h ν
i
f ην:β ¼ ð ∂~
u δ =∂x σ Þn γ þ ik ν e n δ ~
u σ exp ikν xs esν dsðxs Þ
8πkν ρcν
2 sγ
C
ðν ¼ P, SV Þ ðβ ¼ P, SV Þ:
ðP10:11Þ
10.13. Use the method of separation of variables to set up and solve the problem of
the scattering of an incident shear wave by a spherical elastic inclusion.
From this general solution obtain the results for the scattering from a void or
References 521
fluid-filled inclusion. Determine formal expressions for the far field scatter-
ing amplitudes of the scattered P- and S-waves.
10.14. Use the method of separation of variables to set up and solve the 2-D
problem of the scattering of an incident P-wave by a cylindrical elastic
inclusion. From this general solution obtain the results for the scattering
from a void or fluid-filled inclusion. Determine formal expressions for the
far field scattering amplitudes of the scattered P- and S-waves.
10.15. Use the method of separation of variables to set up and solve the 2-D
problem of the scattering of an incident SV-wave by a cylindrical elastic
inclusion. From this general solution obtain the results for the scattering
from a void or fluid-filled inclusion. Determine formal expressions for the
far field scattering amplitudes of the scattered P- and S-waves.
10.16. Obtain low frequency far field scattering amplitude results analogous to
Eq. (10.278) for the case of the scattering of a compressional wave by a
spherical elastic inclusion. Compare these results with those obtained from
the Born approximation.
References
1. L.W. Schmerr, Fundamentals of Ultrasonic Phased Arrays (Springer, New York, 2014)
2. R. Huang, L.W. Schmerr, A. Sedov, T.A. Gray, Kirchhoff approximation revisited—some new
results for scattering in isotropic and anisotropic elastic solids. Res. NDE 17, 137–160 (2006)
3. R. Huang, L.W. Schmerr, A. Sedov, A modified Born approximation for scattering in isotropic
and anisotropic elastic solids. J. NDE 25, 139–154 (2006)
4. T.H. Tan, Reciprocity relations for scattering of plane, elastic waves. J. Acoust. Soc. Am. 61,
928–931 (1977)
5. G.S. Kino, Acoustic Waves: Devices, Imaging, and Analog Signal Processing (Prentice-Hall,
Englewood Cliffs, 1987)
6. J.E. Gubernatis, E. Domany, J.A. Krumhansl, M. Hubermann, The Born approximation in the
theory of scattering of elastic waves by flaws. J. Appl. Phys. 48, 2812–2819 (1977)
7. A. Ben-Menahem, S.J. Singh, Seismic Waves and Sources (Springer, New York, 1981)
8. C.F. Ying, R. Truell, Scattering of a plane compressional wave by a spherical obstacle in an
isotropically elastic solid. J. Appl. Phys. 27, 1086–1097 (1956)
9. Y.H. Pao, C.C. Mow, Scattering of a plane compressional wave by a spherical obstacle.
J. Appl. Phys. 34, 493–499 (1963)
10. V.V. Varadan, Y. Ma, V.K. Varadan, A. Lakhtakia, Chapter 5: Scattering of Waves by Spheres
and Cylinders, in Field Representations and Introduction to Scattering, ed. by V.V. Varadan,
A. Lakhtakia, V.K. Varadan (Elsevier Science, Amsterdam, 1991)
11. K.F. Graff, Wave Motion in Elastic Solids (Dover, New York, 1991)
12. Y.H. Pao, C.C. Mow, Diffraction of Elastic Waves and Dynamic Stress Concentrations (Crane,
Russak, New York, 1973)
13. N. Einspruch, E. Witterholt, R. Truell, Scattering of a plane transverse wave by a spherical
obstacle in an elastic medium. J. Appl. Phys. 31, 806–818 (1960)
14. R.J. McBride, D.W. Kraft, Scattering of a transverse elastic wave by an elastic sphere in a solid
medium. J. Appl. Phys. 43, 4853–4861 (1972)
15. L. Knopoff, Scattering of shear waves by spherical obstacles. Geophysics 24, 209–219 (1959)
16. A.H. Harker, Elastic Waves in Solids (Adam Hilger, Philadelphia, 1988)
522 10 Flaw Scattering
17. P.M. Morse, H. Feshbach, Methods of Theoretical Physics, Parts I and II (McGraw-Hill,
New York, 1953)
18. L.W. Schmerr, S.-J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and Mea-
surements (Springer, New York, 2007)
19. A. Sedov, L.W. Schmerr, Pulse distortion and the elastodynamic Kirchhoff approximation for
cracks: the direct and inverse problems. SIAM J. Appl. Math. 47, 1201–1215 (1987)
20. G.T. Schuster, A hybrid BIE and Born series modeling scheme: generalized Born series.
J. Acoust. Soc. Am. 77, 865–879 (1985)
21. J. Dominguez, Boundary Elements in Dynamics (Elsevier Applied Science, Amsterdam, 1994)
22. M. Kitahara, Boundary Integral Equation Methods in Eigenvalue Problems in Elastodynamics
and Thin Plates. Studies in Applied Mechanics, vol. 10 (Elsevier, Amsterdam, 1985)
23. L.J. Bond, M. Punjani, N. Safari, Ultrasonic Wave Propagation and Scattering Using Explicit
Finite Difference Methods, in Mathematical Modelling in Nondestructive Testing, ed. by
M. Blakemore, G.A. Georgiou (Clarendon Press, Oxford, 1988)
24. R. Ludwig, W. Lord, A finite-element formulation for the study of ultrasonic NDT systems.
IEEE Trans. Ultrason. Ferroelectr. Freq. Control UFFC-35, 809–820 (1988)
25. J.L. Opsal, W.M. Visscher, Theory of elastic wave scattering: applications of the method of
optimal truncation. J. Appl. Phys. 58, 1102–1115 (1985)
26. V.K. Varadan, V.V. Varadan (eds.), Acoustic, Electromagnetic, and Elastic Wave Scattering—
Focus on the T-matrix Method (Pergamon Press, New York, 1980)
27. P. Fellinger, K.J. Langenberg, Numerical Techniques for Elastic Wave Propagation and
Scattering, in Elastic Waves Ultrasonic Nondestructive Evaluation, ed. by S.K. Datta,
J.D. Achenbach, Y.S. Rajapakse (North Holland, Amsterdam, 1990)
28. J.J. Bowman, T.B.A. Senior, P. Uslenghi, Electromagnetic and Acoustic Scattering by Simple
Shapes (North Holland, New York, 1969)
29. P.M. Morse, K. Uno Ingard, Theoretical Acoustics (Princeton University Press, Princeton,
1968)
30. K.J. Langenberg, V. Schmitz, Numerical modelling of ultrasonic scattering by cracks. Nucl.
Eng. Des. 94, 427–445 (1986)
31. D.M. Johnson, Model for predicting the reflection of ultrasonic pulses from a body of known
shape. J. Acoust. Soc. Am. 59, 1319–1323 (1976)
32. N.F. Haines, D.B. Langston, Reflection of ultrasonic pulse from surfaces. J. Acoust. Soc.
Am. 67, 1443–1454 (1980)
33. W.G. Neubauer, A summation formula for use in determining the reflection from irregular
bodies. J. Acoust. Soc. Am. 35, 279–285 (1963)
34. J.S. Avestas, The physical optics method in electromagnetic scattering. J. Math. Phys. 21,
290–299 (1980)
35. J.S. Avestas, Physical optics and the direction of maximization of the far field average power.
IEEE Trans. Ant. Prop. AP-34, 1459–1460 (1986)
36. J.A. Hudson, J.R. Heritage, The use of the Born approximation in seismic scattering problems.
Geophys. J. Royal Astron. Soc. 66, 221–240 (1981)
37. J.E. Gubernatis, E. Domany, J.A. Krumhansl, Formal aspects of the theory of scattering of
ultrasound by flaws in elastic materials. J. Appl. Phys. 48, 2804–2811 (1977)
38. J.H. Rose, J.M. Richardson, Time domain Born approximation. J. NDE 3, 45–53 (1982)
39. R.P. Kanwal, Boundary value problems of composite media. Comput. Struct. 16, 471–478
(1983)
40. D. Quak, A.T. de Hoop, H.J. Starn, Time-domain farfield scattering of plane acoustic waves by
a penetrable object in the Born approximation. J. Acoust. Soc. Am. 80, 1228–1234 (1986)
41. V.V. Varatharajulu, Reciprocity relations and forward amplitude theorems for elastic waves.
J. Math. Phys. 18, 537–543 (1977)
42. A.T. deHoop, Time-domain reciprocity theorems for acoustic wave fields in fluids with
relaxation. J. Acoust. Soc. Am. 84, 1877–1882 (1988)
References 523
43. N. Einspruch, R. Truell, Scattering of a plane longitudinal wave by a spherical fluid obstacle in
an elastic medium. J. Acoust. Soc. Am. 32, 214–220 (1960)
44. R. Truell, C. Elbaum, B.B. Chick, Ultrasonic Methods in Solid State Physics (Academic,
New York, 1969)
45. D.W. Kraft, M. Franzblau, Scattering of elastic waves from a spherical cavity in a solid
medium. J. Appl. Phys. 42, 3019–3024 (1971)
46. L. Knopoff, Scattering of compressional waves by spherical obstacles. Geophysics 24, 30–39
(1959)
47. V.V. Varadan, A. Lakhtakia, V.K. Varadan (eds.), Field Representations and Introduction to
Scattering (North Holland, New York, 1992)
Chapter 11
The Transducer Reception Process
From the results obtained in Chaps. 8 and 10, we now know how to model the
incident sound beam produced by a transducer and the waves scattered from a flaw
due to such an incident wave field. In this Chapter, we will model the reception of
those scattered waves at the face of a receiving transducer. It will be shown that in
the paraxial approximation the reception process, like the transmission process, can
be characterized by plane wave propagation terms and a diffraction correction term.
This diffraction correction term and an additional factor due to averaging the waves
over the receiver face will constitute the reception term that appears in a Thompson-
Gray measurement model of an ultrasonic system.
Consider first the case where a scatterer exists in a fluid medium and there are
separate transmitting and receiving transducers in a pitch-catch setup (Fig. 11.1).
As shown in Chap. 10, the scattered pressure from the “flaw,” pscatt, can be written,
in the far field of the scatterer, as (see Eq. (10.4)):
expðikr s Þ
pscatt ðy; ωÞ ¼ p0 Aðei ; es Þ ; ð11:1Þ
rs
where p0 is the incident wave pressure and ei and es are unit vectors in the incident
and scattered wave directions, respectively. In the paraxial approximation, the
direction of the incident waves can be taken to be a fixed direction, ei0, (see
Fig. 11.1). The incident pressure term p0 will then contain the plane wave propa-
gation terms and the appropriate diffraction coefficient for the incident beam. At the
receiving transducer, we will assume that the measured response is proportional to
es0 es ri
ei
the average pressure, pave(ω), received over the face of the transducer (which could
be either a planar or focused probe), where
ð
1
pave ðωÞ ¼ pscatt ðy; ωÞ dSðyÞ ð11:2Þ
Sr
Sr
However, if the scattering amplitude is slowly varying over the transducer face with
respect to a particular fixed scattering direction es0, we can also remove that
amplitude from the integrand of Eq. (11.3) to obtain
ð
p0 Aðei0 ; es0 Þ exp ðikr s Þ
pave ðωÞ ¼ dSðyÞ: ð11:4Þ
Sr rs
Sr
or, equivalently,
ð
expðikr i Þ 2π
dS ¼ exp ðikz1 Þ C1 ðat ; R0t ; ω; z1 Þ; ð11:6Þ
ri ik
St
where ri is the distance from an arbitrary point y on the transducer surface to the
point x in the fluid at which the pressure p0 is being calculated and at, R0t and z1 are
the radius, focal length and transducer-flaw distance, respectively, for the transmit-
ter. In exactly the same manner, therefore, we have at the receiver (paraxial
assumption on reception)
ð
exp ðikr s Þ 2π
dS ¼ exp ðikz2 Þ C1 ðar ; R0r ; ω; z2 Þ ð11:7Þ
rs ik
Sr
where
2π
C2 ðar ; R0r ; ω; z2 Þ ¼ C1 ðar ; R0r ; ω; z2 Þ : ð11:9Þ
ikSr
Examining Eq. (11.8), we see that the first term on the right side of this equation is
just the pressure incident on the flaw from the transmitter, while the second term is
the influence of the flaw in converting this incident pressure to the appropriate
scattered waves, and the third term is just a plane wave propagation term due to the
scattered waves traveling from the flaw to the receiver. Thus, the last term is due to
the reception process. From Eq. (11.9) we see that this reception process term
consists of the product of a beam diffraction correction, of the same form as that at
the transmitter, and a term which is due to the averaging process at the receiver. In a
pulse-echo setup where the same transducer is used as both transmitter and receiver,
we have ar ¼ at ¼ a, R0r ¼ R0t ¼ R0 , Sr ¼ St ¼ S, and z1 ¼ z2 ¼ z to give
2π
pave ðωÞ ¼ p0 Aðei0 , ei0 Þ exp ðikzÞC1 ða; R0 ; ω; zÞ : ð11:10Þ
ikS
When interfaces are present between the scatterer and both the transmitter and
receiver, as shown in Figs. 11.2 and 11.3, the influence of the transmission process
through the interface must be included in the transmission and reception processes.
528 11 The Transducer Reception Process
θ1
θ2
θ3
D3
es
ei D2
D
x
medium 2
(fluid)
θ10
θ20
θ30
D30 0
es D20
e0i
medium 2
(fluid)
In this section we will model both media as fluids and treat the interfaces as planar.
Making again the paraxial approximation at the transmitter, the pressure of the
scattered waves in medium 2 (containing the flaw) is given by
exp ðik2 r s Þ
pscatt ðy; ωÞ ¼ p0 Aðei0 ; es Þ : ð11:11Þ
rs
as done in Chap. 8 when considering the transmission process for the transmitter.
Here we have
ð þ1
þ1 ð
i p Aðei0 ; es Þ exp ½ip2 ðy xÞ
p scatt
ðy; ωÞ ¼ 0 dpx dpy: ð11:12Þ
2π p2z
1 1
(see Eq. (8.80) where now the roles of points x and y are interchanged and the
“incident” medium is now medium 2). Following the same procedures as before
(see Sect. 8.3.1, for example) the incident plane wave exp½ip2 ðy xÞ becomes,
on transmission through the interface, T 21 exp½ip1 ðy xÞ þ p2z D p1z D and the
scattered pressure in medium 1 is, therefore, given by
ð þ1
þ1 ð
i p Aðei0 ; es Þ T 21 exp ½ip1 ðy xÞ þ p2z D p1z D
p scatt
ðy; ωÞ ¼ 0 dpx dpy;
2π p2z
1 1
ð11:13Þ
ð11:14Þ
The remaining angular plane wave spectrum integrals can then be evaluated by the
method of stationary phase, giving
ð
p0 Aðei0 ; es ÞT 21 ðθ3 Þ exp ½iðk1 D4 þ k2 D3 Þ
pave ðωÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dS; ð11:15Þ
Sr D3 þ c1 D4 =c2 D3 þ c1 cos 2 θ3 D4 =c2 cos 2 θ4
Sr
where the distances D3, D4 and angles θ3, θ4 are as shown in Fig. 11.2. In the
paraxial approximation, the scattering amplitude, transmission coefficient, and
square root terms are all slowly varying functions and can be removed from the
surface integral to yield
where the distances and angles in Eq. (11.16) are for a fixed received wave “ray”
(Fig. 11.3). The remaining integration is then directly related to our previous results
on transmission through an interface to the scatterer (see Eq. (8.201)) where we
found that the pressure incident on the scatterer could be expressed, in the paraxial
approximation, as
iω ρ1 v0 T 12 ðθ10 Þ
p0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2π D10 þ c2 D20 =c1 D10 þ c2 cos 2 θ10 D20 =c1 cos 2 θ20
ð
exp ½iðk1 D1 þ k2 D2 ÞdS
St
¼ ρ1 c1 v0 T 12 ðθ10 Þ exp ½iðk1 D10 þ k2 D20 Þ C1 ðat ; R0t ; ω; θ10 ; θ20 ; D10 ; D20 Þ;
ð11:17Þ
where now the distances and angles are for an incident wave “ray” (Fig. 11.3) and
T12 is the plane wave transmission coefficient along this fixed ray in going from
material 1 to material 2. Comparing Eqs. (11.16) and (11.17), we see that they are of
a very similar form except that the ratios of wave speeds and angles in the square
root terms are reversed (where θ10 $ θ40 and θ20 $ θ30 ). However, we can make
them of identical form by rewriting Eq. (11.16) as
ð11:18Þ
where
2π cos θ40
C2 ðar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 Þ ¼ C1 ðar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 Þ :
ik2 Sr cos θ30
ð11:20Þ
We can identify the first four terms in Eq. (11.19) as due to the incident waves on
the scatterer, the scatterer response, the transmission through the interface on
reception, and the propagation from the scatterer to the receiver. Thus, the C2
11.3 Reception across a Plane Fluid-Solid Interface 531
term, as in the single medium case, contains the contributions due to diffraction
effects and averaging on reception.
It is interesting to note that through the use of Stokes’ relations for the transmission
coefficient (Eq. (6.82a)), we can also express the average received pressure as
p
pave ðωÞ ¼ p0 Aðei0 ; es0 ÞT 12 ðθ40 Þ exp ½iðk1 D40 þ k2 D30 Þ
2π ρ1 c1
C1 ðar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 Þ ; ð11:21Þ
ik2 Sr ρ2 c2
which is of a form identical to that during the transmission of waves from the
transmitter to the scatterer except for the scattering amplitude and the last term in
brackets. Note, however, that the form of expressions such as Eq. (11.21) are also
dependent on the type of transmission coefficient one chooses. In Eq. (11.21) and all
previous equations in this section, the transmission coefficients were based on pressure
ratios. We have indicated this fact in Eq. (11.21) by explicitly putting a superscript “p”
on the transmission coefficient. If, however, we had used transmission coefficients
(Tv12 , Tv21 ) based on velocity ratios (in the direction of propagation) instead, such as
p ρ2 c 2 v
T 12 ¼ T
ρ1 c1 12
ð11:22Þ
p ρ c1 v
T 21 ¼ 1 T 21 ;
ρ2 c 2
different impedance ratios will appear in the last bracketed terms of expressions
such as those in Eqs. (11.20) and (11.21). We will come back to this point in the
next chapter when we set up our entire LTI model of the ultrasonic measurement
process via reciprocity relations.
Finally, consider the received response of Eq. (11.19) for a pulse-echo setup. In
this case θ30 ¼ θ20 , θ40 ¼ θ10 , D30 ¼ D20 , D40 ¼ D10 , ar ¼ at ¼ a, R0r ¼ R0t ¼ R0 ,
Sr ¼ St ¼ S and Eq. (11.19) becomes
p
pave ðωÞ ¼ p0 Aðei0 ; ei0 ÞT 21 ðθ20 Þexp½iðk1 D10 þ k2 D20 Þ
2π cos θ10
C1 ða; R0 ; ω; θ10 ; θ20 ; D10 ; D20 Þ : ð11:23Þ
ik2 S cos θ20
Now, consider the case where two planar fluid-solid interfaces exist between a
scatterer and transmitting and receiving transducers, i.e. a model of a typical immer-
sion setup (Fig. 11.4). In this case, the displacements in the solid were given for a
scattered wave of type α due to a wave of type β as (Eqs. (10.11a and 10.11b)):
532 11 The Transducer Reception Process
medium 2
(solid)
exp ðik r Þ
α s
uscatt ðy; ωÞ ¼ U0 Aα;β ei0β ; esα ; ð11:24Þ
rs
where U0 is the incident wave displacement amplitude and eαs is scattered wave
direction. As in our all previous cases, we have taken the incident wave direction in
Eq. (11.24), eαi0 , to be along a fixed incident ray so that U0 implicitly contains the
diffraction correction terms. Using the angular spectrum of plane waves represen-
tation for the spherical wave then we have
iU 0 Aα;β ei0β ; esα ð ð exp ip α ðy xÞ
þ1 þ1
u ðy; ωÞ ¼
scatt 2
α dpx dpy: ð11:25Þ
2π p2z
1 1
During the reception process (Fig. 11.5) it will only be the P- and SV-components
of the scattered waves that will produce a pressure on the face of the receiver (where
“SV” refers to a shear wave with polarization in the plane of incidence at the
interface). The other remaining polarization component of uscatt, is an SH compo-
nent perpendicular to the plane of incidence that is not transmitted through the fluid-
solid interface (Fig. 11.5). We can choose the polarization directions
dsα ðα ¼ P, SV, SH Þ of the scattered waves to be in the directions shown in
Fig. 11.5 or in opposite directions. However, for either the scattered P- or SV-waves
that reach the receiving transducer, the received pressure, which is a scalar, must be
related to a specified component of the vector scattering amplitude contained in
Eq. (11.25). In the next chapter, using reciprocity relations we will show that that this
specified scattering amplitude component is given by Aα;β liα where the
polarization direction lαi is defined as follows: let the receiving transducer act as a
transmitter and consider a bulk wave ray path from the transducer to the flaw which
11.3 Reception across a Plane Fluid-Solid Interface 533
n x
SV
P ds
ds
planar
SH
interface y ds
flaw
Fig. 11.6 Definition of the polarization vectors for the receiving transducer acting as a transmitter
is just the reverse of the path taken by a bulk wave scattered from the flaw to the
transducer (Fig. 11.6). Let eαr be the “incident” direction in the fluid for this reversed
ray path that generates a wave of type α ðα ¼ P, SV Þ at the flaw with a polarization
vector, lαi , that lies in the plane of incidence formed between the incident direction,
eαr , and the unit vector n, normal to the planar surface between the flaw and the
receiver. Thus, the specified component of the scattered displacement field, in the
medium surrounding the flaw that we will use to describe the reception process is
h
i exp ðikα r s Þ
U s ¼ uscatt ðy; ωÞ liα ¼ U 0 Aα;β ei0β ; esα liα : ð11:26aÞ
rs
[Note: the directions we have defined for these polarizations lαi when solving for the
transmitted waves in Chap. 6 are shown in Fig. 11.6. As with the scattered wave
polarizations shown in Fig.11.5 these choices are arbitrary. But, we will see in
534 11 The Transducer Reception Process
Chap. 12, this choice is in the end immaterial since a change in sign on the
polarization will also produce a canceling sign change in a corresponding trans-
mission coefficient that appears in the measurement model.]
At high frequencies, we can use plane wave relations to write this displacement
component in terms of a stress amplitude, Σ s, where
h
i exp ðikα r s Þ
Σ s ¼ iωρ2 cα2 Us ¼ iωρ2 cα2 U 0 Aα;β ei0β ; esα liα : ð11:26bÞ
rs
Then, using Weyl’s representation to turn the spherical wave exp(ikαrs/rs) into an
angular spectrum of plane waves, we can transmit this spectrum across the plane
interface and write the pressure in the fluid at the receiver, pα;β, due to a wave
scattered from the flaw of type α in the solid (where the incident wave on the flaw
was of type β in the solid) as (see Eqs. (8.80) and (6.126))
h
i
ωρ2 cα2 U 0 Aα;β ei0β ; esα liα
pα;β ðy; ωÞ ¼
2π
ð þ1
þ1 ð α
T P;α
21 exp ip1 ðy xÞ þ p2z D p1z D
P P
ð11:27Þ
α dpx dpy ;
p2z
1 1
where β, like α, ranges over (P, SV) only since there is also no transmitted SH-wave
polarization component across a fluid-solid interface when going from the fluid to
the solid. Note that the polarization directions of the incident waves, however, are
in general with respect to a different plane of incidence for each mode ray path from
P;α
the transmitting transducer to the flaw. Also note that in Eq. (11.27) T21 is the
transmission coefficient for the interface between the flaw and receiver based on a
pressure/stress ratio. Averaging this pressure over the face of the receiver:
ð
pave ðωÞ ¼ 1=Sr pα;β ðy; ωÞ dSðyÞ;
α;β
ð11:28Þ
Sr
we find
ð
α;β ωρ2 cα2 U 0 n α;β
β α α
pave ðωÞ ¼ A ei0 ; es li
2πSr
Sr
P 9
ð þ1
þ1 ð α =
T P;α exp ip ð y x Þ þ p D p P
D
y dSðyÞ:
21 1 2z 1z
α dp dp
p2z x
;
1 1
ð11:29Þ
11.3 Reception across a Plane Fluid-Solid Interface 535
Performing the angular plane wave spectrum integrals by the method of stationary
phase yields
ð Aα;β e β ; e α l α T P;α θ α expikP1 D α þ kα2 D α
α;β iωρ2 cα2 U 0 i0 s i 21 3 4 3
pave ð ωÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dS:
Sr D3 þ cp1 D4 =cα2 D3 þ cp1 cos θ3 D4 =cα2 cos θ4α
α α α 2 α α 2
Sr
ð11:30Þ
ð11:31Þ
where the “0” subscript indicates, as before, quantities as measured with respect to a
fixed ray in the paraxial approximation and the distances on reception. We can
define the quantities
α 2 α α 2 α
x0 ¼ D40 þ cα2 cos θ 40 D30 =cp1 cos θ 30
Δrα
α α
ð11:32Þ
Δrα
y0 ¼ D40 þ cα2 D30 =cp1
β β
i
U0 ¼ T 12 exp i kp1 D10 þ kβ2 D20 C1β ð11:33Þ
ρ2 cβ2 iω
in terms of a diffraction coefficient, Cβ1 , given by (see both Eqs. (8.223) and
(8.224)):
h
i
β β
iω exp i kp1 D10 þ kβ2 D20
C1β at ; R0t ; ω; θ10
β β
; θ20 β
; D10 β
; D20 ¼ qffiffiffiffiffiffiffiqffiffiffiffiffiffiffi
β β
2πcp1 Δx0 Δy0
ð h
i ð11:34Þ
exp i kp1 D1β þ kβ2 D2β dS:
St
536 11 The Transducer Reception Process
Following the same steps as in the fluid-fluid case, Eq. (11.31) gives
h
i α
α;β
pave ðωÞ ¼ iωρ2 cα2 U 0 Aα;β ei0β ; es0
α
li0α T P;α21 θ 30
α α
α α α α α
exp i kp1 D40 þ kα2 D30 C2 ar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 ; ð11:35Þ
where
C2α ar ; R0r ; ω; θ40
α α
; θ30 α
; D40 α
; D30 ¼ C1α ar ; R0r ; ω; θ40
α α
; θ30 α
; D40 α
; D30
α
2π cos θ40
α ; ð11:36Þ
ikα2 Sr cos θ30
which is of a very similar form to the reception term found for the previous fluid
problems. As in the fluid problems, the choice of the transmission coefficient
appearing in Eq. (11.35) will affect the factor multiplying the diffraction coefficient
in Eq. (11.36).
Where Cα1 (SR, ω) is the diffraction correction term obtained when the receiv-
ing transducer acts as a sender and generates a wave of type α. The exact
form of the K term depends on the choice of other variables in the LTI model,
such as the transmission coefficients. In the case of pulse-echo setups, the
diffraction correction term is identical to the one calculated for the sending
transducer (whose area is ST), i.e.
Papers in the literature do not discuss the reception process by itself, but normally
consider it only as part of more complete transmission, scattering and reception
models. The cases considered by Freedman [1], Lhemery [2] and Sedov et al. [3],
for example, are relatively general ultrasonic models that do include reception.
Other authors have also treated the reception process as part of more specific
models (see [4–6]). The identification of the reception process as a product of
diffraction and averaging terms in the paraxial approximation, as obtained in this
chapter, was first noted by Thompson and Gray [7].
11.5 Problems
11.1. Consider the problem of the reception across a fluid-fluid interface where it
assumed that the transducer (plane or focused) responds to the average
normal velocity on the transducer face. How does the expression for this
average received velocity compare with our results for the average pressure?
Assume the paraxial assumption is valid during reception, where the fixed
received wave direction is taken to be along the transducer central axis.
11.2. Consider a circular receiver of radius a which responds in a non-uniform
fashion to the scattered pressure. Assume that this non uniform response can
be modeled by weighting the received pressure by the factor 2ða2 ρ2 Þ=a2 ,
i.e. assume the weighted response, pw(ω), is given by
ð
2ða2 ρ2 Þ scatt
pw ðωÞ ¼ 1=Sr p ρdρdϕ;
a2
Sr
where ρ is the radial distance from the transducer central axis. Determine an
expression for the on-axis weighted response due to a scatterer in a fluid. How
is the received response related to the diffraction coefficient for a similar
transmitting transducer? (see Problem 8.3).
References
4. V.S. Yamshchikov, V.N. Danilov, V.L. Shkuratnik, A.A. Ermolin, Reflection by spherical
inhomogeneity in half-space of longitudinal waves excited by disk radiator. Sov. J. NDT 19,
291–296 (1983)
5. M. Ueda, E. Morimatsu, Analysis of echoes from a sphere which includes the directivity of a
transmitter and receiver. J. Acoust. Soc. Am. 87, 1903–1910 (1990)
6. T. Li, M. Ueda, Analysis of echoes from a cylinder that includes the directivity of the transmitter
and receiver. J. Acoust. Soc. Am. 87, 1880–1884 (1990)
7. R.B. Thompson, T.A. Gray, Analytical diffraction corrections to ultrasonic scattering measure-
ments, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 2A, ed. by
D.O. Thompson, D.E. Chimenti (1983), pp. 567–586
Chapter 12
Ultrasonic Measurement Models
The first ultrasonic setup for which we will derive a complete ultrasonic measure-
ment model is the pitch-catch immersion response of a scatterer in a single fluid
medium (see Fig. 11.1). Although this is a simple configuration, it is one commonly
used for calibration purposes, so that it is also a problem of practical importance.
Actually, all of the elements needed to describe a complete model are present in our
previous discussion of the reception process in Sect. 11.1. However, rather than
assembling the components from those results (an approach followed in the next
section for the more complex models involved in immersion testing) we will go
through a brief but explicit derivation of all the elements in this simple model.
First, from Chap. 8 we know that the pressure field, p(x0 , ω), due to a piston
transducer (which can be either planar or spherically focused) can be given by the
Rayleigh-Sommerfeld integral as
ð
expðikr Þ
pðx0 ; ωÞ ¼ iωρv0 =2π dSðy0 Þ: ð12:1Þ
r
ST
expðikr Þ expðikr i Þ
ffi expðik ei x0 Þ: ð12:3Þ
r ri
If we let A(ei; es) be the plane wave far field scattering amplitude of the scatterer due
to an incident plane wave of unit pressure amplitude, then the scattered pressure at
point y due to the transmitting transducer will be given by (see Fig. 12.2)
flaw
12.1 LTI Model for a Single Fluid Medium 541
rs z1
z2 ei 0
es0 es ri
ei
8 9
< ð = exp ðikr Þ
exp ðikr i Þ
Aðei ; es ÞdSðy0 Þ
s
pðy; ωÞ ¼ iωρv0 =2π : ð12:4Þ
: ri ; rs
ST
At the receiving transducer (which can also be either planar or focused), we will
assume the measured response is proportional to the average pressure on the
transducer face, pave, where
ð
pave ðωÞ ¼ 1=SR pðy; ωÞ dSðyÞ ð12:5Þ
SR
and SR is the area of the transducer. From Eq. (12.4) this average pressure can be
written explicitly as
ðð
exp ðikr i Þ exp ðikr s Þ
pave ðωÞ ¼ iωρv0 =2πSR Aðei ; es Þ dSðy0 Þ dSðyÞ: ð12:6Þ
ri rs
SR ST
However, if the angle subtended by the bundle of waves from the scatterer to both
the transmitter and receiver is small enough so that the scattering amplitude in
Eq. (12.6) is approximately a constant, it can be removed from both integrations to
yield
ð ð
iωρv0 exp ðikr i Þ exp ðikr s Þ
pave ðωÞ ¼ Aðei0 ; es0 Þ dSðy0 Þ dSðyÞ; ð12:7Þ
2πSR ri rs
ST SR
where the unit vectors ei0 and es0 are fixed directions from the flaw to the transmit-
ting and receiving transducers, respectively (Fig. 12.2). From Chap. 8, it follows
that the two integrals appearing in Eq. (12.7) can be written in terms of propagation
terms and diffraction correction terms as
542 12 Ultrasonic Measurement Models
ð
exp ðikr i Þ 2πexp ðikzi Þ
dSðy0 Þ ¼ C1 ðat ; R0t ; ω; zi Þ
ri ik
ST
ð ð12:8Þ
exp ðikr s Þ 2πexp ðikzs Þ
dSðyÞ ¼ C1 ðar ; R0r ; ω; zr Þ
rs ik
SR
and we described these diffraction corrections explicitly for circular sending and
receiving transducers having radii (at, ar), respectively. But note that within the
paraxial approximation these diffraction correction terms simply represent the
pressure wave fields ( pi, ps) of the transmitting and receiving transducers as the
quasi-plane waves
pi ðx; ωÞ
¼ C1 ðST ; ω; . . .Þexp ðikzi Þ
ρcv0
ps ðx; ωÞ
¼ C1 ðSR ; ω; . . .Þexp ðikzs Þ
ρcv0
so they are applicable to any shaped of sending and receiving transducer that is put
into this quasi-plane wave form. Placing Eq. (12.8) into Eq. (12.7) the average
pressure expression becomes
To take the attenuation of the fluid into account, we merely add in the appropriate
attenuation terms on transmission and reception (which involve the distances zi and
zs, respectively) to obtain
where pave ðωÞ ¼ pave ðωÞ=ρcv0 ðωÞ and the attenuation terms are now included in
this normalized average. Using Eq. (12.9) then we find
Or, in terms of the system function, sI(ω), since from Eq. (9.21)
SR
β I ð ωÞ ¼ 2 sI ðωÞ; ð12:13Þ
ST
we have
Comparing Eq. (12.14) with the general LTI model presented in Chap. 2, we see the
following correspondence exists between that LTI model and the pitch-catch model
of this section:
sðωÞ $ sI ðωÞ
PðωÞ $ exp ½ikðzi þ zs Þ
MðωÞ $ exp½αðωÞðzi þ zs Þ
C1 ðωÞ $ C1 ðat ; R0t ; ω; zi Þ
T 1 ðωÞ $ 1 ð12:15Þ
AðωÞ $ Aðei0 ; es0 Þ
T 2 ðωÞ $ 1
4π
C2 ðωÞ $ C1 ðar ; R0r ; ω; zs Þ ;
ikST
where both the transmission terms are unity since there is only one medium
involved. In order to arrive at this model result in the form of products of LTI
systems, however, note that it was necessary to assume (1) both the sender and
receiver were in the far field of the scatterer, (2) both the transmitted and received
waves could be approximated as quasi-plane waves through the paraxial assump-
tion, and (3) the scattering amplitude response was slowly varying over the bundle
of received waves. If these assumptions are violated we cannot expect that a model
of the entire measurement process will be in terms of a merely a series of individual
contributions multiplied together. In Chap. 13, we will develop a more general
544 12 Ultrasonic Measurement Models
Most of the elements for the fluid-fluid case (Fig. 12.3) are already in place from our
discussion of the reception problem in Sect. 11.2. There, we found that the average
received pressure across a fluid-fluid interface could be written in the paraxial
approximation as
pave ðωÞ ¼ p0 Aðei0 ; es0 ÞT 21 exp ½iðk1 D40 þ k2 D30 ÞC2 ; ð12:16Þ
and the diffraction correction coefficient on reception, C2, had the form
2π cos θ40
C2 ¼ C1 : ð12:18Þ
ik2 Sr cos θ30
Writing these diffraction coefficients, for example, as those obtained for the circular
piston transducers and geometry terms discussed in Chap. 8, the average pressure
received is explicitly given by the terms
pave ðωÞ ¼ ρ1 c1 v0 exp ½iðk1 D10 þ k2 D20 þ k1 D40 þ k2 D30 ÞT 12 ðθ10 ÞT 21 ðθ30 ÞAðei0 ; es0 Þ
C1 ðat ; R0t ; ω; θ10 ; θ20 ; D10 ; D20 ÞC1 ðar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 Þ
2π cos θ40
:
ik2 SR cos θ30
ð12:19Þ
If we again normalize the average pressure and include the efficiency factor, the
frequency components of the output voltage, V0(ω), is given by
Comparing Eq. (12.21) with the general LTI model presented in Chap. 2, in this
case we have the following correspondence:
sðωÞ $ sI ðωÞ
PðωÞ $ exp½iðk1 D10 þ k2 D20 þ k1 D40 þ k2 D30 Þ
MðωÞ $ exp½ðα1 D10 þ α2 D20 þ α1 D40 þ α2 D30 Þ
C1 ðωÞ $ C1 ðat ; R0t ; ω; θ10 ; θ20 ; D10 ; D20 Þ
T 1 ðωÞ $ T 12 ðθ10 Þ ð12:22Þ
AðωÞ $ Aðei0 ; es0 Þ
T 2 ðωÞ $ T 21 ðθ30 Þ
4π cos θ40
C2 ðωÞ $ C1 ðar ; R0r ; ω; θ40 ; θ30 ; D40 ; D30 Þ :
ik2 ST cos θ30
12.2 LTI Model for Immersion Testing 547
For an immersion model where a fluid surrounds an elastic solid, again we need to
merely collect all the appropriate elements from previous chapters to obtain a
complete ultrasonic measurement model. From Sect. 11.3, the average received
pressure for a wave scattered of type α (in the solid), that was of type β (in the solid)
when incident on the flaw, pα;β ave (ω), ðα ¼ P, SV Þ , ðβ ¼ P, SV Þ is given by (see
Fig. 12.4)
h
i P;α α
β α;β β α α α
pα;β
ave ð ω Þ ¼ iωρ c U
2 α2 0 A e ;
i0 s0e α
l i0 T 21 θ30 exp i kp1 D40 þ kα2 D30
C2α ar ; R0r ; ω; θ40
α
; θ30α α
; D40 α
; D30 ;
ð12:23Þ
Setting the wave type α ¼ γ, where γ ¼ P, S (for notational convenience since the
symbol α will now be used to denote the attenuation) and adding the efficiency
factor and attenuation terms, as before, the frequency components of the received
voltage, V0(ω), becomes (for a particular γ and β):
h
i
V 0 ðωÞ ¼ βI ðωÞT β;P β β β α
12 θ 10 exp i kp1 D10 þ kβ2 D20 þ kp1 D40 þ kγ2 D30
γ
h
ih
i
β
exp α1P D10 þ α2β D20
β
þ α1P D40 γ
þ α2γ D30
γ
Aγ;β ei0β ; es0
γ
li0γ
P;γ γ β β β β β γ
T 21 θ30 C1 at ; R0t ; ω; θ10 ; θ20 ; D10 ; D20 C1γ ar ; R0r ; ω; θ40
γ γ
; θ30 γ
; D40 ; D30
γ
2π ρ2 cγ2 cos θ40
γ ;
ikγ2 SR ρ2 cβ2 cos θ30
ð12:27Þ
Identifying all the terms in this case with the general LTI model, we obtain
12.2 LTI Model for Immersion Testing 549
s ð ωÞ $ s I ð ωÞ
h
i
β β γ γ
PðωÞ $ exp i kp1 D10 þ kβ2 D20 þ kp1 D40 þ kα2 D30
h
i
β
MðωÞ $ exp α1P D10 þ α2β D20 β
þ α1P D40 γ
þ α2γ D30
γ
C1 ðωÞ $ C1β at ; R0t ; ω; θ10β β
; θ20 β
; D10 β
; D20
ð12:29Þ
β;P β
T 1 ðωÞ $ T 12 θ10
h
i
AðωÞ $ Aγ;β ei0β ; es0 γ
li0γ
γ
T 2 ðωÞ $ T P;γ21 θ30
γ
γ 4π ρ2 cγ2 cos θ40
C2 ðωÞ $ C1γ ar ; R0r ; ω; θ40γ γ
; θ30 γ
; D40 ; D30 γ :
ikγ2 SR ρ2 cβ2 cos θ30
In Chap. 11, we mentioned that the exact form of results such as Eq. (12.29)
depends on the choices made for the transmission coefficients contained in T1 and
T2. For example, if we use Stokes’ relations to transform the transmission coeffi-
P;γ γ;P
cient T21 into T12 , from Chap. 6 Eq. (6.150a) we have
γ
γ ρ1 cp1 cos θ30 γ;P γ
T P;γ
21 θ 30 ¼ γ T 12 θ 40 ; ð12:30Þ
ρ2 cγ2 cos θ40
so that the T2 and C2 terms in the LTI model of Eq. (12.29) would be changed to
γ
T 2 ðωÞ $ T γ;P
12 θ 40
γ 4π ρ1 cp1 ð12:31Þ
C2 ðωÞ $ C1γ ar ; R0r ; ω; θ40
γ γ
; θ30 γ
; D40 ; D30 ;
ikγ2 ST ρ2 cβ2
with all other terms remaining the same in Eq. (12.29). The transmission coeffi-
cients used in both Eqs. (12.27) and (12.29) are for stress/pressure or pressure/stress
ratios. If we had chosen instead to write T1 and T2 in the LTI model in terms of
transmission coefficients (going from material 1 to material 2) based on velocity
ratios, since
ρ2 cγ2 v γ;P
T γ;P
12 ¼ T
ρ1 cp1 12
ρ cβ2 β;P ð12:32Þ
β;P
T 12 ¼ 2 v T 12 ;
ρ1 cp1
550 12 Ultrasonic Measurement Models
where the “v” denotes a velocity ratio coefficient, we would find the following
changes in the LTI model:
β;P β
T 1 ðωÞ $ v T 12 θ10
γ
T 2 ðωÞ $ v T γ;P
12 θ 40 ð12:33Þ
γ 4π ρ2 cγ2
C2 ðωÞ $ C1γ ar ; R0r ; ω; θ40
γ γ
; θ30 γ
; D40 ; D30 ;
ikγ2 ST ρ1 cp1
ð12:35Þ
and
γ γ γ γ
V R;γ ðωÞ ¼ v T γ;P α γ γ γ
12 θ 40 exp i kp1 D40 þ kγ2 D30 C1 ar ; R0r ; ω; θ 40 ; θ 30 ; D40 ; D30
γ γ
exp α1P D40 þ α2γ D30 :
ð12:36Þ
The first two terms on the right hand side of Eq. (12.35) represents the velocity of a
plane P-wave that has started from the sending transducer with unit velocity and
traveled to the flaw across the interface as a wave of type β. The third term on the
12.2 LTI Model for Immersion Testing 551
right side of Eq. (12.35) represents the diffraction correction which accounts for the
beam effects in the paraxial approximation. Recall that earlier in this chapter we
related this coefficient to the pressure wave field, pi, of the transmitting transducer as
pi ðx; ωÞ
¼ C1 exp ikp1 zi ;
ρ1 cp1 v0
Thus, the first three terms in Eq. (12.36) represent the quasi-plane wave velocity at
the flaw due to a sending transducer with a unit velocity acting on its face. We can
compute this velocity directly with one or more ultrasonic beam models, as shown
extensively in Chap. 8. The remaining terms on the right side of Eq. (12.35) simply
accounts for the material attenuation from the transmitting transducer to the flaw. In
a similar fashion, the first three terms on the right side of Eq. (12.36) is just the
quasi-plane wave velocity at the flaw for a wave of type γ due to the receiving
transducer when it acts as a P-wave transmitter with a unit velocity acting on its
face and the remaining term includes attenuation effects in going from the receiving
transducer to the flaw. In terms of these velocity fields our ultrasonic measurement
model is simply written as
h
i 4π ρ c
γ;β β γ γ 2 γ2
V 0 ðωÞ ¼ sI ðωÞV ðωÞV ðωÞ A
T;β R;γ
ei0 ; es0 li0 : ð12:37Þ
ikγ2 ST ρ1 cp1
[Note that while we have used a setup with plane interfaces to obtain the results of
this section they can be used for much more general problems involving curved
interfaces as long as we have a beam model, such as the multi-Gaussian model
discussed in Chap. 8, capable of predicting the appropriate diffraction corrections.]
An ultrasonic measurement model of this type was first developed by Bruce
Thompson and Tim Gray in 1983 [4]. The Thompson-Gray measurement model
552 12 Ultrasonic Measurement Models
does rely on a number of assumptions which we will discuss in more depth later in
this chapter, but it is directly applicable to many NDE flaw inspections so that it
represents a significant advance in ultrasonic NDE modeling. A key element in the
Thompson-Gray model, like the fluid models discussed previously, is that it sepa-
rates explicitly the flaw response from the transduction terms present due to the
instrumentation (sI(ω) or βI(ω)), and the wave propagation effects present (VT;β(ω)
and VR;γ(ω)). This separation facilitates the extraction of information on the flaw
itself, an important feature for flaw sizing and flaw characterization studies, as
mentioned previously. We will show in the next sections how the Thompson-Gray
measurement model arises naturally from a more general reciprocity-based formu-
lation originally developed by Auld [1].
In the previous sections, ultrasonic measurement models were developed for single
media and interface problems by explicitly considering the transmission, flaw
scattering, and reception processes and combining them appropriately. While this
direct approach is feasible for such cases, it is also possible to develop a more
general ultrasonic measurement model for the immersion testing configuration
shown in Fig. 12.5 by the use of the reciprocity relations discussed in Chap. 5. In
such a configuration a flaw and other internal surfaces can be present in the part
being inspected and the part itself can be a complex geometry (planar surfaces are
shown in Fig. 12.5 and subsequent figures merely for convenience of illustration).
Fig. 12.5 Geometry of a general pitch-catch immersion testing setup where a flaw is being
interrogated by the transmitting transducer, T. The surface Sf is the surface of the flaw, ST is the
ð1Þ
surface of the transmitting transducer, which has a driving velocity vT on its face and SR is the
surface of the receiving transducer, R. Besides the flaw, there can be other surfaces in the
component being inspected, such as Si. This is the geometry of state (1)
12.3 Reciprocity-Based Model for Immersion Testing 553
where the normal derivative can be taken either in the inward or outward normal
direction. Since the normal derivatives ∂pð1Þ =∂n, ∂pð2Þ =∂n of the pressure are
ð1Þ ð2Þ
proportional to the normal components of the velocity vectors (vn , vn ) we can
also write this reciprocity relation as
ð
pð1Þ vðn2Þ pð2Þ vðn1Þ dS
S ð
¼ pð1Þ vð2Þ pð2Þ vð1Þ ndS ¼ 0: ð12:40Þ
S
Note that this relation is also satisfied for an open surface, S, if that surface is a free
surface, since then for both solutions we have pð1Þ ¼ pð2Þ ¼ 0, or if S is a surface
with no normal velocity vðn1Þ ¼ vðn2Þ ¼ 0 . For a volume, V, of an elastic solid
contained within a closed surface, S, we also have the reciprocity relation (Chap. 5)
ð
ð1Þ ð2Þ
tk nk uð2Þ tk nk uð1Þ dS ¼ 0; ð12:41Þ
S
which can also be written in terms of the traction vectors (t(1)(n), t(2)(n)) and, for
harmonic waves, the velocities vðmÞ ¼ iωuðmÞ (m ¼ 1, 2) as
ð
tð1Þ vð2Þ tð2Þ vð1Þ dS ¼ 0; ð12:42Þ
S
ðmÞ ðmÞ
or in terms of the stresses, τji , and velocity components, vi , (m ¼ 1, 2) as
ð
ð1Þ ð2Þ ð2Þ ð1Þ
τji vi τji vi nj dS ¼ 0: ð12:43Þ
S
Equations (12.41)–(12.43) are also true for any stress-free surface tð1Þ ¼ tð2Þ ¼ 0
or rigid (motionless) surface vð1Þ ¼ vð2Þ ¼ 0 .
554 12 Ultrasonic Measurement Models
In using these reciprocity relations we will take the fields in state (1) as the fields
of our original problem (Fig. 12.5), where a transmitting transducer generates a
ð1Þ
uniform normal velocity vT over a surface area, ST, which produces a sound beam
that interacts with a flaw surface, Sf, in an elastic solid. Scattered waves from this
flaw produce fields on the surface, SR, of the receiving transducer which are
converted to electrical energy and amplified to produce an output voltage. There
also may be scattered waves from other surfaces in the solid, such as Si but our
objective is to develop an explicit expression for the measured voltage due only to the
flaw. We will take state (2) to be the problem shown in Fig. 12.6, where the receiving
ð2Þ
transducer acts as a transmitter, generating a uniform normal velocity, vR , on its
surface, SR, and where the flaw is absent. Note that all of the surfaces present in state
(1), such as Si, for example, are also present in state (2) except for the flaw, which we
show as the dotted surface, Sf, of the unflawed elastic solid in Fig. 12.6.
First, apply the reciprocity relation, Eq. (12.40), to the fluid region in this
immersion problem, as shown in Fig. 12.7. We have
ð
pð1Þ vð2Þ pð2Þ vð1Þ ndS ¼ 0; ð12:44Þ
Sfs þSw þST þSR þSe þSother
where Sfs is the free surface of the fluid, Sw is the surface of the immersion tank
walls, ST is the active area of the face of the transmitting transducer, SR is the area of
the face of the receiving transducer, Se is the surface of the elastic solid in contact
with the fluid, and Sother is a collection of all the other surfaces present such as the
surfaces of the cables and casings of the transducers, the surfaces of the supports of
the solid in the tank, etc., which are not labeled explicitly in Fig. 12.7. The unit
normals for most of these surfaces are also shown in Fig. 12.7. The integral in
Eq. (12.44) over the free surface, Sfs, vanishes since pð1Þ ¼ pð2Þ ¼ 0 there. Likewise,
the tank wall, Sw, and all the other surfaces, Sother, will be taken as motionless where
vð1Þ n ¼ vð2Þ n ¼ 0 so that the integrals in Eq. (12.44) over those surfaces also
will vanish. Thus, we have
ð
ð
pð1Þ vð2Þ pð2Þ vð1Þ ndS ¼ pð1Þ vð2Þ pð2Þ vð1Þ ndS: ð12:45Þ
ST þSR Se
Since we have assumed the transmitting transducers in states (1) and (2) act as
piston transducers we can remove those velocity terms from the integrals in
Eq. (12.45), leaving integrals of pressure terms over the transducer faces. Those
pressure integrals can then be written as forces, transforming Eq. (12.45) into
ð1Þ ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð1Þ
FT vT FT vT þ FR vR FR vR
ð
¼ pð1Þ vð2Þ pð2Þ vð1Þ ndS; ð12:46Þ
Se
ðmÞ ðmÞ
where FT (ω), vT (ω) (m ¼ 1, 2) are the compressive forces and normal velocities
ðmÞ ðmÞ
on the face of the transmitting transducer is states (1) and (2), and FR (ω), vR (ω)
(m ¼ 1, 2) are the corresponding forces and normal velocities acting on the face of
the receiving transducer for those states. Note that these normal velocities are all
positive when pointing outwards from the face of the transducer into the fluid. On
the surface Se of the elastic solid component being inspected the traction vector and
the normal velocity must be continuous, i.e., we have the boundary conditions
pðmÞ nfluid ¼ tðmÞ solid
ð12:47Þ
vðmÞ nfluid ¼ vðmÞ nsolid
Note that the normal in Eq. (12.48) acts outward from the surface of the solid
(Fig. 12.7).
Now, consider the volume of the elastic solid contained between the external
surface of that solid, Se, and the internal surfaces such as the flaw surface, Sf and any
other internal surfaces such as Si (Fig. 12.8). Our reciprocal relation for this source
free volume gives
ð
tð1Þ vð2Þ tð2Þ vð1Þ dS ¼ 0: ð12:49Þ
Se þSf þSi
But the integral over Si, which is present in both states (1) and (2), is itself an
integral over a source-free region if it is an inclusion or if it is a hole in the
component, for example, it is an integral over a stress-free surface. In either case,
the integral over Si itself vanishes in Eq. (12.49) and
ð
ð
tð1Þ vð2Þ tð2Þ vð1Þ dS ¼ tð1Þ vð2Þ tð2Þ vð1Þ dS: ð12:50Þ
Se Sf
In Eq. (12.51) the unit normal of the flaw surface is directed inwards as shown in
Fig. 12.8. If we write Eq. (12.51) in terms of stress and velocity components and
change the unit normal to be directed outwards from the flaw into the surrounding
material of the elastic solid, we have
ð1Þ ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð1Þ
FT vT FT vT þ FR vR FR vR
ð
ð1Þ ð2Þ ð2Þ ð1Þ
¼ τji vj τji vj nj dS; ð12:52Þ
Sf
12.3 Reciprocity-Based Model for Immersion Testing 557
where now nj are components of the outward normal (pointing into the surrounding
solid). At this point, it must be pointed out that in Eq. (12.52) the forces and
velocities present at the transducers in state (1) are not just the fields due to the
presence of flaw since they contain other responses such as reflections from the
surface, Si, for example, or from other interactions that do not involve the flaw such
as propagation directly from the transducer T to the surface Se and then to the
receiver, R. To remove these non-flaw parts of the response we can define state
(3) which is the same as state (1) except where the flaw is absent. Obviously, in state
(3) the forces and velocities at the transducer faces are due to all those interactions
that do not involve the flaw. Since states (1) and (3) differ only in terms of the
presence of the flaw, we see that following the same steps that led to Eq. (12.52)
with states (1) and (2) will lead to the same equation for state (3) but with the
integral over the flaw surface absent, i.e.,
ð3Þ ð2Þ ð2Þ ð3Þ ð3Þ ð2Þ ð2Þ ð3Þ
FT vT FT vT þ FR vR FR vR ¼ 0: ð12:53Þ
where
ð1Þ ð3Þ
FTf ¼ FT FT
ð1Þ ð3Þ
vTf ¼ vT vT
ð12:55Þ
ð1Þ ð3Þ
FRf ¼ FR FR
ð1Þ ð3Þ
vRf ¼ vR vR :
The differences appearing in Eq. (12.55) are indeed only forces and velocities
appearing due the flaw being present so it is appropriate to call them the flaw
ð2Þ
responses. However, at the transmitting transducer both FfT and FT are forces
generated on that transducer face when that transducer is in a purely passive mode
since the same driving voltage sources are present in both states (1) and (3) so FfT is
the force on the transmitter in the absence of such sources and, by the definition of
ð2Þ
state (2), the force FT is calculated when the transmitting transducer is not active.
When a transducer is in a passive mode, we can characterize it as an equivalent
acoustic impedance, Zain (ω) that relates the total force acting on the transducer,
Ftotal, to the total velocity, vtotal, through Ftotal ¼ Zina vtotal (Fig. 12.9). For state
ð2Þ
(1) minus state (3), vTf ¼ vtotal , Ftotal ¼ FTf while in state (2) vT ¼ vtotal ,
558 12 Ultrasonic Measurement Models
Fig. 12.9 (a) The transmitting transducer T when it is in a passive mode and waves are incident on
it, and (b) a model of this passive transducer as an acoustic impedance where the total compressive
force, Ftotal, on the face of the transducer is due to all the waves (incident and scattered) and is
related to the total velocity of these waves, vtotal, through an acoustic input impedance, Zain
ð2Þ ð2Þ
Ftotal ¼ FT . The minus signs for the velocities are present because both vfT and vT
are positive in the outward normal direction, n, on the face of the transducer. Thus,
we have
As shown in [5] if one has models of the electrical impedance of the pulser, the
transfer matrix of the attached cabling, and the transfer matrix of the transducer, one
could in principle calculate an explicit model for this acoustic input impedance.
This is really not necessary since placing Eq. (12.56) into Eq. (12.54) shows that all
the terms at the transmitting transducer cancel, leaving
ð
f ð2Þ ð2Þ f ð1Þ ð2Þ ð2Þ ð1Þ
F R v R FR v R ¼ τji vj τji vj nj dS: ð12:57Þ
Sf
Now, note that in state (2) we have at the receiving transducer (which is acting as a
ð2Þ ð2Þ
transmitter) FR ¼ Z R;ar vR where Zr
R;a
is the acoustic radiation impedance of the
receiving transducer, so that we can rewrite Eq. (12.57) as
ð
ð2Þ ð1Þ ð2Þ ð2Þ ð1Þ
FRf ZR;a
r v f
R v R ¼ τji vj τji vj nj dS: ð12:58Þ
Sf
But the terms in parentheses on the left side of Eq. (12.58) is just the blocked force
at the receiver due to the flaw, FfB , (see Eq. (9.22)) so Eq. (12.58) can be written as
ð
f 1
ð1Þ ð2Þ ð2Þ ð1Þ
FB ¼ ð2Þ τji vj τji vj nj dS: ð12:59Þ
vR
Sf
12.3 Reciprocity-Based Model for Immersion Testing 559
Equation (12.59) shows that the blocked force appears naturally as the quantity at
the receiving transducer one should use to characterize a flaw response. Since we
defined (see Chap. 9) the acoustic/elastic transfer function, Ta(ω), as the ratio
F B ð ωÞ
T a ð ωÞ ¼ ð1Þ
; ð12:60Þ
ρ1 cp1 ST vT ðωÞ
where (ρ1, cp1) are the density and compressional wave speed of the fluid at the
ð1Þ
transmitting transducer and vT (ω) is the velocity acting on the transducer surface,
ST, this transfer function is
ð
1 ð1Þ ð2Þ ð2Þ ð1Þ
T a ð ωÞ ¼ ð1Þ ð2Þ
τ v
ji j τ v
ji j nj dS ð12:61Þ
ρ1 cp1 ST vT ðωÞvR ðωÞ
Sf
and we can use this transfer function and the system function to obtain the measured
output voltage, V0(ω), as
ð
s I ð ωÞ ð1Þ ð2Þ ð2Þ ð1Þ
V 0 ð ωÞ ¼ ð1Þ ð2Þ
τji vj τji vj nj dS: ð12:62Þ
ρ1 cp1 ST vT ðωÞvR ðωÞ
Sf
Equation (12.62) is a very general measurement model for an immersion testing but
that generality does come at a price, namely the acoustic fields involved are a
complex combination of the incident ultrasonic beam, attenuation effects, and the
scattering properties of the flaw and it is difficult to extract any physical meaning
from this general model in terms of the flaw being examined, as found in the
Thompson-Gray measurement model. In this section, we will demonstrate how the
560 12 Ultrasonic Measurement Models
ð2Þ
where vR (ω) is the velocity on the surface of the receiving transducer (acting as a
transmitter), l2;γ
0j is the polarization of the incident wave of type γ ðγ ¼ P, SÞ in the
solid (see Fig. 12.11), and e2;γ
0 is a unit vector in the direction of propagation in the
Sr n
solid for a fixed ray path as shown in Figs. 12.10 and 12.11. The plane wave
γ;P
transmission coefficient, vT12 , is calculated based on a ratio of velocities, and all of
the distances and angles appearing in Eq. (12.63) are measured along a ray
satisfying Snell’s law from the center of the transducer to a fixed point, x0, near
the flaw. The position vector, xs, is measured from x0 to the flaw surface. As shown
ð2Þ ð2Þ;inc
earlier in this chapter this incident velocity expression vj vj can be written
more compactly as
h
i
ð2Þ;inc ð2Þ 2;γ
vj ðxs ; ωÞ ¼ vR V R;γ l0j exp i kγ2 e02;γ xs ; ð12:64Þ
where obviously
γ;P γ γ γ
V R;γ ¼ v T 12 cos θ40 exp i kp1 D40 þ kα2 D30
γ γ γ γ
exp α1P D40 þ α2γ D30 γ
C1 ar ; R0r ; ω; θ40 γ
; θ30 γ
; D40 ; D30 : ð12:65Þ
ð2Þ
vT R;γ h
i
ð2Þ;inc 2;γ 2;γ
τij ðxs ; ωÞ ¼ V Cijkl l0k e0l exp i kγ2 e02;γ xs : ð12:66Þ
cγ2
Now, consider the waves in state (1). In this case, the flaw is present and the
ð1Þ
transmitting transducer is firing with a velocity vT (ω) on its face and the compo-
ð1Þ
nents of the velocity field on the flaw surface Sf, vj , consist of a combination of the
ð1Þ ð1Þ;inc ð1Þ;scatt
incident and scattered waves, i.e., vj ¼ vj þ vj . For an incident wave of
type β in the solid
h
i
ð1Þ;inc ð1Þ 1;β
vj ðxs ; ωÞ ¼ vT V T;β l0j exp i kβ2 e01;β xs ; ð12:67Þ
where
h
i
β;P β β β
V T;β ¼ v T 12 θ10 exp i kp1 D10 þ kβ2 D20
h
i
β
exp α1P D10 þ α2β D20
β
C1β at ; R0t ; ω; θ10
β β
; θ20 β
; D10 β
; D20 ð12:68Þ
and the incident wave stresses for state (1) are then given by
ð1Þ
vT T;β h
i
ð1Þ;inc 1;β 1;β
τij ðxs ; ωÞ ¼ V Cijkl l0k e0l exp i kβ2 e01;β xs ; ð12:69Þ
cβ2
562 12 Ultrasonic Measurement Models
ð1Þ;β ð1Þ;β
where e0 ei0 is a unit vector along an incident wave paraxial ray path from the
center of the transmitter to point x0 (near the flaw), and l1;β
0j are the components of
the polarization vector for this incident wave (Fig. 12.11). The total velocity and
stresses for state (1) can be written in normalized form as
iωv1j
^v 1j ¼ ð1Þ
vT V T;β
ð12:70Þ
iωτ1ij
^τ 1ij ¼ ð1Þ
;
vT V T;β
where these normalized fields physically are the velocity and stress components in
state (1) due to an incident plane wave of unit displacement amplitude. Note,
however, that neither ^v 1j or ^τ 1ij are dimensionless.
Now, consider Auld’s measurement model, Eq. (12.62), for a flaw small enough
so that the amplitude of the incident quasi-plane waves in states (1) and (2) do not
vary significantly over the flaw surface. Then we can take the amplitudes of these
waves out of the integral over the flaw surface and write the remaining integral in
terms of the normalized velocity and stress fields of state (1) as
ðn o h
i
sI ðωÞV T;β V R;γ 2:γ 2;γ 2;γ 1
V 0 ðωÞ ¼ ^τ 1ij l0j þ Cijkl l0k e0l ^v j =cγ2 ni exp ikγ2 e02;γ xs dSðxs Þ:
iω ρ1 cp1 ST
Sf
ð12:71Þ
In Chap. 10 (see (10.12 a, b, c)) we found expressions for the plane wave far field
vector scattering amplitude of a flaw for a scattered wave of type γ due to an
incident plane wave of type β. The component of the vector scattering amplitude
appearing in Eq. (12.71) is easily seen to be (in terms of our current notation):
h
i
4πρ2 c2γ2 Aγ;β e01;β ; e02;γ l02;γ
ðn o h
i
2;γ 2;γ 2:γ 1
¼ ^τ 1ij l0j þ Cijkl l0k e0l ^v j =cγ2 ni exp ikγ2 e02;γ xs dSðxs Þ ð12:72Þ
Sf
which can be rewritten in terms of a scattered wave direction for state (1), defined
1;γ
here as es0 e02;γ , to yield, finally
h
i 4π ρ2 cγ2
γ;β
1;β 1;γ 2;γ
V 0 ðωÞ ¼ sI ðωÞV T;β
ðωÞV R;γ
ð ωÞ A ei0 ; es0 l0 :
ikγ2 ST ρ1 cp1
ð12:74Þ
Equation (12.74) is just the Thompson-Gray measurement model (see Eq. (12.37)).
In their original derivation of their model from reciprocity principles, Thompson
and Gray [4] replaced the flaw surface by a spherical surface Sr surrounding the flaw
that was located in the flaw’s far field. They obtained a form similar to Eq. (12.74)
by approximately evaluating the integral on that spherical surface, assuming it was
in the far field of the flaw. That evaluation required making a number of assump-
tions [4]. However, as the above derivation has shown, the only assumptions truly
required to turn Auld’s general measurement model into their LTI model is the
assumption that the incident waves are quasi-plane waves and the flaw is small
enough so that the amplitudes of those waves can be considered to be constants over
the flaw surface, as stated earlier. Our reciprocity-based derivation has the added
advantage of showing explicitly that the scalar component
of
the flaw far field
scattering amplitude that must be calculated is Aγ;β l02;γ , a result that was
missing in the original discussions in [4].
Auld’s measurement model of Eq. (12.62) was derived for an immersion setup. The
basic ideas that led to that model, however, are quite general and also applicable to
contact testing problems. However, there are modeling differences between the
immersion and contact cases that need to be addressed. In this section we will
develop the equivalent Auld’s model explicitly for contact testing. We will use the
angle beam shear wave inspection setup of Figure 12.12 to discuss the contact case,
where contact P-wave sending and receiving transducers are placed on wedges,
which are in smooth contact with the flawed elastic solid to be tested. The wedges
are cut so that the transducer face is along a direction which is beyond the critical
angle for P-waves in the underlying solid. Thus, the primary wave generated by a
transmitter in this configuration is a SV-wave traveling at oblique incidence to the
interface in the solid. In this setup, unlike the immersion case where we assumed the
transducers acted as piston sources, we will assume that when either the contact
transmitter or receiver is driven by an electric signal, a uniform pressure distribu-
tion is generated over the face of the transducer on the wedge surface, and that there
are no active sources in any of the materials. In that case, when the reciprocity
564 12 Ultrasonic Measurement Models
relations of Chap. 5 are employed for the transmitting wedge (see (5.42)) we obtain
(Fig. 12.13)
ð n o
ð1Þ ð2Þ ð2Þ ð1Þ
τij vj τij vj ni dS ¼ 0; ð12:75Þ
S1 þSi
where S1 is the surface of the wedge except for the interface Si between the wedge
and the flawed solid and we have used the form of the reciprocal relation in
Eq. (12.75) in terms of the stress components τij and velocity components vj.
Again, we take state (1) to be the problem when the transmitting transducer is
firing and the flaw is present and take state (2) to be the case where the receiver acts
as a transmitter and the flaw is absent. Similarly, if we apply the reciprocal theorem
to the volume of the elastic solid between the flaw surface, Sf, and the other real
surfaces and interfaces of the part being inspected, we find
12.4 Reciprocity-Based Model for Contact Testing 565
ð n o
ð1Þ ð2Þ ð2Þ ð1Þ
τij vj τij vj ni dS ¼ 0; ð12:76Þ
Si þSf þSj þSother
where Sj is the interface with the receiving transducer wedge and Sother is the
contribution from all the other remaining surfaces in the flawed solid exterior to Sf
(excluding, of course, Si and Sj). However, since Sother consists either of stress-free
surfaces, where the traction terms in Eq. (12.71) vanish for both states (1) and (2), or
interfaces with inclusions for which both states (1) and (2) satisfy the same homo-
geneous equations, as in the immersion case such Sother contributions vanish and
need not be included in Eq. (12.76). Also, for the receiving wedge itself, we have
ð n o
ð1Þ ð2Þ ð2Þ ð1Þ
τij vj τij vj ni dS ¼ 0; ð12:77Þ
S2 þSj
where S2 is the surface of the wedge of the receiving transducer except for the
interface Sj. Note that in Eq. (12.75) and Eq. (12.77) the integrals over the fields in
those equations are the fields in the wedges while the integrals in Eq. (12.76)
involve fields in the elastic solids so when considering the integrals over Si and
Sj, the surfaces of contact between the wedges and the solid being inspected, we
must in principle acknowledge that difference. However, since the wedges and the
solid being inspected are in smooth contact through a thin fluid layer (whose
thickness we will neglect), we have
ðn o ðn o
τij vj τij vj ni dS pð2Þ vðn1Þ pð1Þ vðn2Þ dS
ð1Þ ð2Þ ð2Þ ð1Þ
¼
Sm Sm
wedge
fluid
ðn o
τij vj τij vj ni dS
ð1Þ ð2Þ ð2Þ ð1Þ
¼ ;
Sm inspected solid
ð12:78Þ
where ( p, vn) are the pressure and normal velocity in the thin fluid layer and Sm can
be either Si or Sj, so that in writing the integrals over those contacting surfaces there
is in fact no difference. If we combine the results of Eqs. (12.75)–(12.77), we have
ð n o ðn o
ð1Þ ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð1Þ
τij vj τij vj ni dS ¼ τij vj τij vj ni dS; ð12:79Þ
ST þSR Sf
where the normals n for all the surfaces involved are shown in Fig. 12.13 so that for
Eq. (12.79) the normal is the outward normal to flaw surface (pointing into the
surrounding solid) and we have replaced the integrals over S1 and S2 by integrals
only over the faces of the transmitting and receiving transducers, ST and SR,
566 12 Ultrasonic Measurement Models
respectively, since the remaining portions of S1 and S2 are stress free in both states
ðmÞ
and so their integrals are zero. Since the stresses τij ¼ pðmÞ δij (m ¼ 1, 2) on the
transducers, Eq. (12.79) becomes
ð n o ðn o
ð1Þ ð2Þ ð2Þ ð1Þ
pð2Þ vðn1Þ pð1Þ vðn2Þ dS ¼ τij vj τij vj ni dS; ð12:80Þ
ST þSR Sf
where the normal velocities on the transducers are in the direction of the normal to
the wedges (Fig. 12.13) so they are positive when directed inwards to the transducer
faces. In the contact case we assume the pressure acting on a transducer face are
uniform, so we can use that fact to write Eq. (12.80) in terms of compressive forces
and average velocities, similar to what was done for the immersion case. For
ð2Þ
example, if the uniform pressure acting on the receiver in state (2) is p0R (ω), we
have
8 9
ð <1 ð =
ð2Þ ð2Þ ð1Þ
pð2Þ vðn1Þ dS ¼ p0R ðωÞSR vðn1Þ dS ¼ FR ðωÞvR ðωÞ: ð12:81Þ
:SR ;
SR SR
ð2Þ
In terms of a compressive force, FR , acting on the receiver in state (2) and an
ð1Þ ð1Þ
average velocity, vR , acting on the face of the receiver in state (1), where again vR
is positive directed inwards to the transducer face. Following the same procedure
for all the other terms on the left side of Eq. (12.80) we have
ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ
FR vR FR vR þ FT vT FT vT
ðn o
ð1Þ ð2Þ ð2Þ ð1Þ
¼ τij vj τij vj ni dS; ð12:82Þ
Sf
which is identical to Eq. (12.52) for the immersion case except for a minus sign
because of the difference in positive directions assumed for the velocities on the
faces of the sending and receiving transducers. At this point we can follow exactly
the same steps outlined in the immersion case and eliminate the non-flaw responses
contained in state (1) by defining a state (3) where the sending transducer is firing
and the flaw is absent, leading to the reciprocity relation
ð2Þ ð3Þ ð3Þ ð2Þ ð2Þ ð3Þ ð3Þ ð2Þ
F R v R FR v R þ FT vT FT vT ¼ 0; ð12:83Þ
which, when subtracted from Eq. (12.82), gives and expression for the flaw
response at the receiving transducer only as
12.4 Reciprocity-Based Model for Contact Testing 567
ðn o
ð2Þ ð2Þ ð1Þ ð2Þ ð2Þ ð1Þ
FR vRf FRf vR ¼ τij vj τij vj ni dS; ð12:84Þ
Sf
since, as in the immersion case, the terms at the transmitting transducer are all
obtained when that transducer is in a passive mode and hence cancel. In Eq. (12.84)
we can relate the force in state (2) to the average velocity in the state through that
ð2Þ ð2Þ
transducer’s radiation impedance, i.e., vR ¼ FR =Z R;a r , where the minus sign is
ð2Þ
present because of the positive direction of vR is opposite to the outward normal of
the transducer face. Using this relation we find
ðn o
ð2Þ ð1Þ ð2Þ ð2Þ ð1Þ
vRf þ FRf =Z R;a
r F R ¼ τ v
ij j τ v
ij j ni dS; ð12:85Þ
Sf
where in terms of the average free surface received velocity of the flaw, v fsf (ω) (see
Eq. (9.39)), we have
ðn o
f 1 ð1Þ ð2Þ ð2Þ ð1Þ
vfs ðωÞ ¼ ð2Þ τij vj τij vj ni dS: ð12:86Þ
F R ð ωÞ
Sf
Thus, in contact testing setups, our reciprocity relations naturally leads to a rela-
tionship between the average free surface velocity at the receiver in terms of the
fields on the flaw surface. In Chap. 9 we defined the acoustic/elastic transfer
0
function for the contact case, Ta (ω), as (see Eq. (9.40))
where (ρ, cp) are the density and compressional wave speed of the wedge material in
ð1Þ
contact with the receiving transducer, p0T is the pressure on the face of the
transmitting transducer, and (ST, SR) are the areas of the sending and receiving
transducers, respectively. The acoustic/elastic transfer function in the contact case
is then
ðn o
0 ρcp SR ð1Þ ð2Þ ð2Þ ð1Þ
T a ðωÞ ¼ ð1Þ ð2Þ
τij vj τij vj ni dS ð12:88Þ
FT ðωÞFR ðωÞ
Sf
and introducing the system function, sC(ω), the measured output voltage, V0(ω), is
ðn o
sC ðωÞρcp SR ð1Þ ð2Þ ð2Þ ð1Þ
V 0 ð ωÞ ¼ ð1Þ ð2Þ
τij vj τij vj ni dS: ð12:89Þ
FT ðωÞFR ðωÞ
Sf
568 12 Ultrasonic Measurement Models
This is our Auld type of measurement model for contact testing. We can compare
Eq. (12.89) with the immersion case (Eq. (12.62)) given again here as
ð
s I ð ωÞ ð1Þ ð2Þ ð2Þ ð1Þ
V 0 ð ωÞ ¼ ð1Þ ð2Þ
τji vj τji vj nj dS: ð12:90Þ
ρ1 cp1 ST vT ðωÞvR ðωÞ
Sf
We see in both cases the fields on the surface of the flaw are normalized by the
driving forces or velocities in states (1) and (2) so that those normalized fields can
be obtained from theoretical transducer beam models and flaw scattering models
without a knowledge of the driving forces or velocities. We can make the form of
the contact case identical to the immersion
case
by defining for the contact case a
_ð1Þ _ð2Þ
set of fictitious driving velocities, v T , v R as
_ð1Þ ð1Þ
vT ¼ FT =ρ1 cp1 ST
_ð2Þ
ð12:91Þ
ð2Þ
vR ¼ FR =ρcp SR ;
Since Eq. (12.92) is identical in form with the immersion measurement model we
can assume, as before, that the waves in the solid are in the form of a quasi-plane
wave incident on the flaw and that the flaw is small enough so that the amplitude of
this incident wave does not vary over the flaw surface, leading again to the
Thompson-Gray measurement model of Eq. (12.74) in the form
_ _ h
i 4π ρ2 cγ2
V 0 ðωÞ ¼ sC ðωÞV T;β ðωÞV R;γ ðωÞ Aγ;β ei0
1;β 1;γ
; es0 l02;γ ;
ikγ2 ST ρ1 cp1
ð12:93Þ
T;β R;γ
_ _
where the only difference is that the velocity fields V ;V are normalized by
_
ð1Þ _ð2Þ ð1Þ ð2Þ
v T , v R , respectively, instead of the immersion driving velocities (vT , vR ).
In the angle beam probe case, however, the wedge angles are set up so that, in the
paraxial approximation, only shear waves are generated in the solid and picked up
12.4 Reciprocity-Based Model for Contact Testing 569
by the receiver and these shear waves are polarized such that they lie in the planes
of incidence defined by the propagation directions of P-waves in the wedges and the
unit normal to the solid surface. Thus, we need only consider the case γ ¼ β ¼ SV
and Eq. (12.93) becomes for our angle beam shear wave setup
_ _ h
i 4π ρ2 cs2
V 0 ðωÞ ¼ sC ðωÞV T;SV R;SV
V A SV;SV
i0 ; es0
e1;SV 1;SV 2;SV
l0 :
iks2 ST ρ1 cp1
ð12:94Þ
Since in the paraxial approximation an angle beam shear wave transducer model
produces velocity components in the material being inspected that are identical in
form to those in an equivalent fluid-solid problem, we have (assuming the wedges
are both made of the same material)
_ SV
V T;SV ¼ v T SV;P
12 θ10 exp i kp1 DSV
10 þ ks2 D20
SV
SV
exp α1P DSV 10 þ α2 D20 C1 at ; ω; θ 10 ; θ 20 ; D10 ; D20
S SV SV SV SV SV
ð12:95Þ
and
_ SV
V R;SV ¼ v T SV;P
12 θ40 exp i kp1 DSV
40 þ ks2 D30
SV
SV
exp α1P DSV 40 þ α2 D30 C1 ar ; ω; θ 40 ; θ 30 ; D40 ; D30 ; ð12:96Þ
S SV SV SV SV SV
where vTSV;P
12 is the plane wave transmission coefficient (based on a velocity ratio)
for two solids in smooth contact, CSV
1 is a diffraction correction term for a fluid-solid
problem, and the distances and angles are defined in Fig. 12.14. The final LTI model
then also is in exactly the same form as obtained for an immersion case, namely
SV;P SV v SV;P SV
V 0 ðωÞ ¼ sC ðωÞv T 12 θ10 T 12 θ40 exp i kp1 DSV
10 þ k s2 D20
SV
P SV P SV
exp i kp1 DSV40 þ k s2 D30 exp α1 D10 þ α2 D20 exp α1 D40 þ α2 D30
SV S SV S SV
SV
1 ar ; ω; θ 40 ; θ 30 ; D40 ; D30 C1 at ; ω; θ 10 ; θ20 ; D10 ; D20
CSV SV SV SV SV SV SV SV SV
h
i 4π ρ c
1;SV 1;SV 2 s2
ASV;SV ei0 ; es0 l02;SV :
iks2 ST ρ1 cp1
ð12:97Þ
magnetic fields in the sending and receiving cables (surfaces Sa and Sb). Thus, Eq.
(5.73) gives
ð h
i ð h
i
Eð2Þ Hð1Þ n Eð1Þ Hð2Þ n dS þ Eð2Þ Hð1Þ n Eð1Þ Hð2Þ n dS
Sa ð
Sb
ð 1Þ ð 2Þ
¼ tk nkI vð2Þ tk nkI vð1Þ dS:
Sr
ð12:98Þ
As done previously, we will take state (1) to correspond to the excitation of the
system with an electrical power P in cable a attached to the transmitting transducer
T1, when the flaw inside Sr is present. Similarly, state (2) is taken to correspond to
the excitation of the system with the same power P in the line b that feeds the
receiving transducer T2 when the flaw is absent. Furthermore, we assume that the
electrical disturbances in the cables at the surfaces Sa and Sb can be taken as
traveling fundamental modes in these lines. Since at these cable locations we
have the electromagnetic fields of both the incident and reflected fundamental
wave modes, in state (1) the solution is of the form [1]
for line a:
þ þ
E1 ¼ 1 þ Γaa
f
E , H1 ¼ 1 Γaa
f
H ð12:99aÞ
for line b:
f
E1 ¼ Γba Eþ , f
H1 ¼ Γba Hþ ; ð12:99bÞ
where Γfaa and Γfba are reflection and transmission coefficients for the fundamental
mode when the flaw is present and Eþ and Hþ are the electric and magnetic fields of
the fundamental mode that transmits a power P to the transducer. Since in both
cables the unit vector n points outward, across either Sa or Sb we have
ð
ðEþ Hþ Þ n dS ¼ 2P: ð12:100Þ
Sa or Sb
for line b:
where Γbb and Γab are the reflection and transmission coefficients for the funda-
mental mode in the cable when the flaw is absent. Placing Eqs. (12.99a, b) and
(12.101 a, b) into Eq. (12.98), and using Eq. (12.100) gives
ð
ð1Þ ð2Þ
4P Γba Γab ¼ tk nkI vð2Þ tk nkI vð1Þ dS:
f
ð12:102Þ
Sr
If, instead, we had taken the flaw to be absent in both states (1) and (2) (this is
equivalent to using states (3) and (2) as defined earlier), since these solutions both
satisfy the same homogeneous equations of motion within Sr it follows
ð
ð1Þ ð2Þ
4PðΓba Γab Þ ¼ tk nkI vð2Þ tk nkI vð1Þ dS ¼ 0; ð12:103Þ
Sr
f
where now δ Γba ¼ Γba Γba is the difference in transmission coefficients for the
case where transmitter a is firing and the flaw is present and the case where
transmitter a is firing and the flaw is absent. If we change the unit normal on Sr to
be the outward normal n ¼ nI then Eq. (12.96) becomes
ð
ð1Þ ð2Þ
δ Γba ¼ 1=4P tk nk vð2Þ tk nk vð1Þ dS: ð12:105Þ
Sr
However, there are no sources between the surface Sr and the flaw surface Sf so that
we can replace the integration to one over the flaw surface, giving
ð
κðωÞ
ð1Þ ð2Þ ð2Þ ð1Þ
V 0 ð ωÞ ¼ τkl vl τkl vl nk dS: ð12:108Þ
4P
Sf
We can make Eq. (12. 108) identical to our previous contact result (Eq. (12.89)) by
letting
h i
ð1Þ ð2Þ
κðωÞ FT ðωÞ FR ðωÞ=ρcp SR
s C ð ωÞ ¼ ; ð12:109Þ
4 PðωÞ
Both Eq. (12.109) and Eq. (12.110) contain a ratio of a mechanical power term to
the input electrical power, P, provided to the transducer. This ratio is an efficiency
and is the reason why the term system efficiency factor was used in previous studies
when describing βC(ω).
The models obtained in this Chapter (and their extensions) can be used to represent
many practical ultrasonic NDE systems. To legitimately apply these models,
however, one must be aware of the assumptions that went into their derivation.
For the Auld models we used basically four assumptions:
1. linearity and causality of the entire system, which is assumed noise-free.
2. reciprocity of the fields in the media involved.
3. planar and focused transducers that were assumed to act as piston sources during
transmission and reception for immersion and as constant pressure sources for
contact problems.
4. Use of a system function (or system efficiency factor) to characterize all the
electrical and electroacoustic parts of the system (pulser/receiver, cabling, trans-
ducers) that must be obtained experimentally in a reference experiment.
Linearity and causality are reasonable assumptions for most ultrasonic NDE
systems, since the amplitudes of the waves generated are small and not likely to
produce nonlinear material responses, and the linearity of the associated elec-
tronics is a characteristic that codes and standards often require to be validated.
574 12 Ultrasonic Measurement Models
Noise, however, is another matter. All ultrasonic NDE systems contain both
electronic and “metal” noise (due to grain scattering) and those noise sources are
not modeled here. If one assumes that these noise contributions can be treated as
additive, then typical noise sources can be measured experimentally and simply
summed with our noise-free model predictions. Reciprocity relations are satis-
fied for the fields present in ideal fluids and elastic solids, so with the appropriate
transducer beam models and flaw scattering models Auld’s measurement models
can be applied to complex problems such as anisotropic and inhomogeneous
elastic solids. The assumption that the transducers behaved as piston or constant
pressure sources is a convenient but not a necessary assumption. For
non-uniform velocity or pressure distributions one needs to redefine the
“lumped” force and velocity parameters appropriately at the transducer faces
used in the development of our expressions for the acoustic/elastic transfer
function. Auld’s measurement models in the forms we have developed here
have incorporated either a system function or a system efficiency factor. Either
of these factors lumps all the electrical-to-mechanical and mechanical-to-elec-
trical conversions that occur in an ultrasonic system into a single term, so that
they are complicated composite functions of the properties of the pulser, trans-
ducers, cabling, and receiver. In the approach outlined in Chap. 9, either the
system function or the system efficiency factor must be obtained experimentally
from a reference scattering configuration where the same pulser/receiver and
transducers are used at the same system settings as in the ultrasonic flaw
measurement. This approach, therefore, requires a new reference measurement
whenever any component or component setting is changed in an ultrasonic
setup. This restriction could be removed, however, if models were available
for the electromechanical aspects of the ultrasonic system just as they are for the
purely mechanical parts that have been modeled in this book. In research efforts
conducted since the publication of the first edition of this book, explicit models
of the pulser, cabling, transducers, and receiver have been developed and it has
been shown how the properties of these individual components can be measured
with purely electrical measurements and combined to synthesize a system
function which agrees with direct measurements [5]. This allows one to evaluate
how changes of many of the individual components in an ultrasonic measure-
ment system affect the measured signal. A challenge remains with characteriz-
ing the pulser/receiver, however, since this is still done in [5] at a set of specific
settings and so must be redone when those settings change.
Reduction of Auld’s models to models of the Thompson-Gray type rests on
the following additional assumptions:
5. incident fields of the sending and receiving transducers that can be represented
as high frequency quasi-plane waves incident on the flaw (paraxial
approximation).
6. transmission/reflection processes modeled by plane waves at smooth interfaces
(surface roughness ignored).
12.6 Measurement Models: A Discussion 575
7. a small flaw where the amplitude of the incident quasi-plane waves does not vary
significantly over the flaw surface.
8. an isolated flaw response that comes from only “direct” path contributions.
The paraxial assumption (or quasi-plane wave assumption) is fundamental to
obtaining a final Thompson-Gray model that is (for a given mode) simply a single
series of products of LTI systems. Since most ultrasonic NDE transducers produce
highly directional, well-collimated beams, this assumption is often well satisfied in
practice but as discussed in Chap. 8, there are testing situations where the paraxial
approximation may not be appropriate. The high frequency approximation also has
the consequence of allowing us to introduce the effects of interfaces in terms of
plane wave transmission coefficients for a planar surface. In all the examples we
have considered in this book, any roughness of the surface is neglected. If rough-
ness effects are important, the present models must, therefore, be modified. This is a
non-trivial task and still an open area of research. In the Thompson-Gray measure-
ment model, the flaw response is in terms of a plane wave far field scattering
amplitude. In our direct modeling approach of Sects. 12.1 and 12.2 this scattering
amplitude arose because (a) the flaw was assumed to be small enough to approx-
imate the amplitude of the incident quasi-plane wave fields at a fixed point near the
flaw- a point often taken to be at the flaw “center”, (b) the receiver-flaw distance
was assumed large enough that far field scattering conditions could be assumed, and
(c) the variations of the scattering amplitude over the face of the receiver could be
neglected. When using reciprocity principles to develop the Thompson-Gray model
from the Auld relations, however, we saw we only needed the two assumptions of
(a) and (c). As will be seen in this next chapter, the slowly varying scattering
amplitude assumption can be relaxed for some problems in which case the total
measured response will be in terms of multiple series of products of LTI systems.
However, for sufficiently large flaws, where there are significant incident field
variations over its surface, or for a flaw close enough to the receiver, even the
extended models of the next chapter may be in error. In many ultrasonic inspections
of high performance systems a “small-flaw-not-too-close-to-the receiver” assump-
tion is likely acceptable since the critical flaws one needs to find in such systems are
indeed quite small and most ultrasonic setups are arranged so that the flaw is well
separated from the transmitter and receiver. For applications where these assump-
tions are unacceptable, however, one will be forced to use Auld’s measurement
model, thus losing much of the simplicity and explicitness of the Thompson-Gray
result. Finally, the Thompson-Gray models previously described have all treated
the flaw as if it was isolated (i.e., well removed from any other surfaces or
interfaces) and have considered only the response coming from those waves that
pass directly into the flawed material, interact with the flaw, and then pass out
directly back to the receiving transducer. Incorporating multiple interactions of an
isolated flaw with other surfaces or interfaces during both transmission and
reception is possible within the context of the Thompson-Gray model once the
576 12 Ultrasonic Measurement Models
appropriate wave fields from those interactions are calculated. Note that Auld’s
measurement model can also be used for flaws at interfaces or surface-breaking
flaws if the appropriate fields on the surface of the flaw due to interactions of the
incident transducer beam with the flaw and associated boundary are included in the
integrals over the flaw surface.
In Schmerr and Song [5] the Thompson-Gray measurement model was
implemented in MATLAB® by combining a multi-Gaussian beam model similar
to the one described in Chap. 8 with the far field scattering amplitudes of
Chap. 10 and experimentally measured or simulated system functions. Details can
be found there on the MATLAB® functions used and the results of comparing
directly measured voltage signals with those predicted by the Thompson-Gray
model for common reference scatterers such as voids, flat-bottom holes, and side-
drilled holes.
Once ultrasonic measurement models are in hand, they can, of course, be used
for many practical applications. Some of those applications will be described in
Chap. 14. Reciprocity obviously plays a key role in the development of these
measurement models so we have also included a number of references that discuss
reciprocity in different contexts [19–24].
12.8 Problems
whereδxn indicates there is a point source at xn. Similarly show that two immersion
piston transducers, T1 and T2, both acting in state C are reciprocal, i.e.,
v02 ST 2 pave T 2 T 1 ; C ¼ v01 ST 1 pave T 1 T 2 ; C ;
where pave T m T n ; C is the average pressure received by transducer Tm due to
transducer Tn in state C.
12.3. Consider a more general immersion testing case model where the transducers
T1 and T2 involved are not piston sources but have a normal velocity
distribution generated over their faces given by vnα ¼ v0α wα ðxα Þ ðα ¼ 1, 2Þ,
respectively, where v0α are reference velocities (such as the maximum or
average velocity present in the distribution) and the wα are non-dimensional
weighting factors (on transmission) that define the velocity distributions.
First, define the response due to these non-uniform transducers at an arbitrary
point x, for a given geometry and material configuration C, as
ð
w
f x, ωT β β ; C ¼ 2iωρ1 v0α wβ xβ f x, ωδxβ ; C dS xβ ;
ST β
ST α
578 12 Ultrasonic Measurement Models
0
where wα is a non-dimensional weighting factor (on reception). Using our
0
reciprocity theorems, show that if wα ¼ wα these non-uniform transducers are
also reciprocal in the same manner as for a piston probe, i.e.,
v02 ST 2 p T w2 2 T w1 1 ; C ¼ v01 ST 1 p T w1 1 T w2 2 ; C :
Fig. P12.1 A testing setup in a “delta” configuration where an angle beam shear wave transducer
is used as a transmitter and a contact P-wave transducer is the receiver
12.6. Set up an LTI model for contact testing where a contact shear wave trans-
ducer is used on the plane surface of an elastic solid. Assume that the
transducer generates a uniform shearing stress on the surface over a circular
area of radius a (Fig. P12.2).
y z
Fig. P12.2 Uniform shear stress acting on a circular area of an otherwise stress-free plane surface
as a model of a shear wave contact transducer
12.8 Problems 579
12.7. Model an immersion testing setup with a scalar (fluid) model where a flaw is
located across fluid-fluid planar interfaces from the transmitter and receiver.
Following the same steps as carried out in Sect. 12.3, develop a general
reciprocity-based model for the received response and show all the steps
needed to reduce this general model to an LTI model.
12.8. A steel block contains a spherical pore of radius b as shown in Fig. P12.3a.
An immersion P-wave transducer is located so that the pore is on its central
axis. The efficiency of the system is as shown in Fig. P12.3b. Assume the
material attenuation to be a constant in the steel (neglect the attenuation of the
water) and assume the pore is in the far field of the transducer so that the
diffraction coefficient is C1 ¼ ikp1 a2 =2z where z ¼ z1 þ cp2 z2 =cp1 (see
Chap. 8).
(a) Using our LTI ultrasonic measurement model and Fourier inversion,
determine an explicit expression for the voltage received as a function of
time, V0(t), from the leading edge response of the pore in terms of the
parameters of the system.
(b) If a ¼ 0:25 in:, b ¼ 1 mm, f 1 ¼ 1 MHz, f 2 ¼ 5MHz, z1 ¼ 1 in:,
z2 ¼ 4 in:, β0 ¼ 0:2 106 volt sec , and the attenuation of the steel
is a constant 6 dB/m, determine the magnitude of the maximum voltage
in the received wave form. Show the numerical contributions separately
for each term in the LTI measurement model that produces your final
answer.
Fig. P12.3 (a) Immersion testing geometry for an on-axis spherical void, and (b) the efficiency
factor for the system
580 12 Ultrasonic Measurement Models
a b
ρ1 , cp 1
2a 2a
β( f )
z1 z1
β0
z2 z2
f
2bs 2bh –fmax fmax
ρ2 , cp 2 , cs 2
Fig. P12.4 (a) Pulse-echo immersion testing of a spherical void and a flat-bottom hole, and
(b) the efficiency factor for the system
(a) Using our LTI ultrasonic measurement model determine explicit expres-
sions for the voltage received as a function of frequency, V0(ω), from the
leading edge responses of both the pore and the flat-bottom hole in terms
of the parameters of the system.
(b) Using the results of part (a), and assuming that the efficiency factor is as
shown in Fig. P12.4b, determine through Fourier inversion explicit
expressions for the voltages received as a function of time, Vs(t) and
Vh(t), from the leading edge responses of the pore and flat-bottom hole,
respectively, in terms of the parameters of the system (see problem 2.8).
References 581
References
1. B.A. Auld, General electromechanical reciprocity relations applied to the calculation of elastic
wave scattering coefficients. Wave Motion 1, 3–10 (1979)
2. J.N. Gray, T.A. Gray, N. Nakagawa, R.B. Thompson, Models for predicting NDE reliability,
Metals Handbook, Vol. 17 (ASM International, Materials Park, 702–715, 1989)
3. J.A. Ogilvy, Model for predicting ultrasonic pulse-echo probability of detection. NDT & E Int.
26, 19–29 (1993)
4. R.B. Thompson, T.A. Gray, A model relating ultrasonic scattering measurements through
liquid-solid interfaces to unbounded medium scattering amplitudes. J. Acoust. Soc. Am. 74,
1279–1290 (1983)
5. L.W. Schmerr, S.-J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and Mea-
surements (Springer, New York, 2007)
6. A. Freedman, The high frequency echo structure of some simple body shapes. Acustica 12,
61–70 (1962)
7. A. Freedman, A mechanism of acoustic echo formation. Acustica 12, 10–21 (1962)
8. A. Freedman, The use of linear system theory in acoustic radiation and scattering analysis.
J. Sound Vib. 52, 265–276 (1977)
9. A.V. Malinka, Measurement of the size and shape of flaws by the method of ultrasonic
spectrum, I. Soviet J. NDT 15, 13–23 (1979)
10. A.V. Malinka, Measurement of the size and shape of flaws by the method of ultrasonic
spectrum, II. Soviet J. NDT 15, 23–27 (1979)
11. W.G. Neubauer, A summation formula for use in determining the reflection from irregular
bodies. J. Acoust. Soc. Am. 35, 279–285 (1963)
12. D.M. Johnson, Model for predicting the reflection of ultrasonic pulses from a body of known
shape. J. Acoust. Soc. Am. 59, 1319–1323 (1976)
13. J. Krautkramer, Determination of the size of defects by the ultrasonic impulse method.
Br. J. Appl. Phys. 19, 240–245 (1959)
14. A. Lhemery, Impulse-response method to predict echo-responses from targets of complex
geometry, part I: theory. J. Acoust. Soc. Am. 90, 2799–2807 (1991)
15. R.B. Thompson and T.A. Gray, Erratum: A model relating ultrasonic scattering measurements
through liquid-solid interfaces to unbounded medium scattering amplitudes [J. Acoust. Soc.
Am., 74, 1279–1290, 1983], J. Acoust. Soc. Am., 75, 1645–1646 (1984)
16. G.S. Kino, The application of reciprocity theory to scattering of acoustic waves by flaws.
J. Appl. Phys. 49, 3190–3199 (1978)
17. D. Bennink, Modeling of ultrasonic scattering experiments with application to system and
transducer characterization, (PhD thesis, Iowa State University, 1990)
582 12 Ultrasonic Measurement Models
18. J.M. Coffey, R.K. Chapman, Application of elastic scattering theory for smooth flat cracks to
the quantitative prediction of ultrasonic defect detection and sizing. Nucl. Energy 22, 319–333
(1983)
19. A.D. Yaghjian, Generalized reciprocity relations for electroacoustic transducers. J. Res. Nat.
Bur. Stds. 78B, 17–39 (1975)
20. L.L. Foldy, H. Primakoff, A general theory of passive linear electroacoustic transducers and
the electroacoustic reciprocity theorem: I. J. Acoust. Soc. Am. 17, 109–120 (1945)
21. H. Primakoff, L.L. Foldy, A general theory of passive linear electroacoustic transducers and
the electroacoustic reciprocity theorem: II. J. Acoust. Soc. Am. 19, 50–58 (1947)
22. N.F. Haines, D.B. Langston, The reflection of ultrasonic pulses from surfaces. J. Acoust. Soc.
Am. 67, 1443–1454 (1980)
23. G.S. Kino, B.T. Khuri-Yakub, Application of the reciprocity theorem to nondestructive
evaluation. Res. Nondestr. Eval. 4, 193–204 (1992)
24. R.B. Thompson, Interpretation of Auld’s electromechanical reciprocity relation via a
one-dimensional example. Res. Nondestr. Eval. 5, 147–155 (1994)
25. A.H. Harker, Elastic Waves in Solids (Adam Hilger, Philadelphia, PA, 1988)
Chapter 13
Near Field Measurement Models
ri
y⬘ ei
ðð
0
expðikr i Þ expðikr r Þ
pave ðωÞ ¼ iωρv0 =2πSR Aðei ; er ÞdS y dSðyÞ; ð13:1Þ
ri rr
SR ST
where the distances ri, rr and unit vectors ei, er are given in Fig. 13.1. Note that
here we have taken the unit vectors ei, er to both point from the transducer to
the scatterer. This is done for convenience of notation in some of the later
results. However, this means that during the reception process the unit vector er is
opposite to the scattered unit vector direction, es, that we have used consistently in
previous chapters to be along a scattered wave path pointing away from the
scatterer, i.e. er ¼ es .
For Eq. (13.1) to be valid, we require that the scatterer be small (so that the
incident waves in the transmitted beam behave near the flaw like a quasi-plane
wave with an amplitude which does not vary significantly over the scatterer
volume) and that both kr i
1, kr r
1. These conditions, however, still allow
the scatterer to be essentially anywhere in the transducer wave field (except, of
course, in a region immediately adjacent to the transducer face). From a numerical
standpoint, Eq. (13.1) is not very practical an expression to evaluate directly
because of the multiple surface integrations present. Fortunately, some further
simplifications of this equation are still possible under a relatively mild set of
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
conditions. To see this, first note that since r i ¼ ρ2i þ z2 and r r ¼ ρ2r þ z2 , we
have ρi dρi ¼ r i dr i and ρr dρr ¼ r r dr r so that the area elements in Eq. (13.1) become
0
dS y ¼ r i dr i dϕi and dSðyÞ ¼ rr dr r dϕr (Fig. 13.2). Then Eq. (13.1) becomes
ðð
pave ðωÞ ¼ iωρv0 =2πSR Aðei ; er Þexpðikr i Þexpðikr r Þdr i dr r dϕi dϕr : ð13:2Þ
To evaluate Eq. (13.2), we will examine separately the following two cases:
1. The scatterer in the main beam of the transducer
When point x (usually taken as the center of the scatterer) is such that it lies in
the cylinder formed by extending the face of the transducer normal to its own
surface, the double integral on the radial coordinates ri, rr becomes explicitly
ð r ri ¼r
rr ¼r e
ði
e
er
y
ρr
rr
φr
j
i
k z x
where the “e” superscript on a quantity indicates that it is evaluated on the edge
of the transducer.
We have already assumed kri
1, kr r
1. Thus, if the scatterer is of a
small characteristic dimension b, where both kr i
kb, kr r
kb, then the
phase of the scattering amplitude is slowly varying in comparison to the large
phases of the exponentials in Eq. (13.3) and both integrals can be integrated by
parts to obtain an asymptotic expression for I given (to leading order) by
where (i, j, k) are unit vectors along the (x, y, z) coordinate axes (Fig. 13.2).
Placing Eq. (13.4) into Eq. (13.2) some of the ϕi, ϕr integrations can be done
explicitly to yield
586 13 Near Field Measurement Models
pave ðωÞ ¼ ρcv0 =2πikSR 4π 2 A½ei ð0; Þ; er ð0; Þexp½2ikz
2ðπ
2π A ei ð0; Þ; er θre ; ϕr exp ik r re þ z dϕr
0
2ðπ
2π A ei θie ; ϕi ; er ð0; Þ exp ik r ie þ z dϕi ð13:6Þ
0
2ðπ 2ðπ
9
e e e =
þ A ei θi ; ϕi ; er θr ; ϕr exp ik r i þ r re dϕi dϕr :
;
0 0
pave ðωÞ ¼ ρcv0 =2πikSR A ei θ
i ; ϕi ; er θ r ; ϕr exp ik r i þ r r
>
:
ϕi ¼ϕ1i ϕr ¼ϕ1
r
þ
A ei θ þ
i ; ϕi ; er θ r ; ϕr exp ik r i þ r r
þ
A ei θ þ
i ; ϕi ; er θ r ; ϕr exp ik r i þ r r
þ þ
þ A ei θ þ þ
i ; ϕi ; er θ r ; ϕr exp ik r i þ r r dϕi dϕr :
ð13:7Þ
If the directions of the integrations over the edge of the transducer involving the
functions r
i , r r are changed so that these integrations continue in the same
Both Eqs. (13.8) and (13.9) can then be combined into a single expression as
pave ðωÞ ¼ ρcv0 =2πikSR 4π 2 Θ2 A½ei ð0; Þ; er ð0; Þexp½2ikz
ð
2πΘ A ei ð0; Þ; er θre ; ϕr exp ik r re þ z dϕr
C
ð
2πΘ A ei θie ; ϕi ; er ð0; Þ exp ik r ie þ z dϕi
C
ð13:10Þ
9
ðð =
þ A ei θie ; ϕi ; er θre ; ϕr exp ik r ie þ r re dϕi dϕr ;
;
CC
Equation (13.10) can be shown to be valid for x located anywhere in the wave field
of the transducer, including on the edge of the main beam. When x is on the beam
edge the integrals must then be interpreted in a principal value sense (see the
discussions involving similar expressions in Chap. 8). By adding the appropriate
attenuation terms and introducing either the efficiency factor or system function,
Eq. (13.10) leads directly to a near field measurement model for the received
voltage. Since the details follow directly those of Chap. 12, we will not give them
here, but merely refer to Eq. (13.10) as our near field measurement model. Note,
however, that when placing the attenuation terms into Eq. (13.10), there are now
four different distances involved, so that there will also be four separate attenuation
contributions as well.
Physically, Eq. (13.10) says that the total received response from a small flaw in
the near field of a transducer is composed of four terms when the scatterer is in the
main beam: (1) a term involving a “direct-direct” wave that travels from the face of
the transducer to the flaw and directly back, (2) a “direct-edge” wave that travels
from the face of the transducer to the flaw and then back to the transducer edge,
(3) an “edge-direct” wave that travels from the transducer edge to the flaw and then
directly back to the transducer face, and (4) an “edge-edge” wave that travels from
the edge of the transducer to the flaw and then back to the edge. These four paths
are shown in Fig. 13.4a–d. Outside the main beam, all the direct wave terms are
absent and the response consists only of an edge-to-edge wave type of term. These
results show that in the near field of a transducer the transducer sending and
receiving responses are inherently intermixed with the flaw scattering response
itself. However, in some special cases it is possible to further simplify these model
results.
13.1 Model for a Single Fluid Medium 589
If the scatterer is located on the central transducer axis and the transducer is
circular, then both distances rei , rer are independent of ϕi and ϕr, respectively so
that setting r ie ¼ r re ¼ r e and θie ¼ θre ¼ θe Eq. (13.10) becomes
pave ðωÞ ¼ 2πρcv0 =ikSR fA½ei ð0; Þ; er ð0; Þexp½2ikz
exp½ikðr e þ zÞhA½ei ð0; Þ; er ðθe ; ϕr Þiϕr
exp½ikðr e þ zÞhA½ei ðθe ; ϕi Þ; er ð0; Þiϕi
o ð13:12Þ
þ exp½2ikr hhA½ei ðθ ; ϕi Þ; er ðθ ; ϕr Þiiϕi , ϕr ;
e e e
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
with re ¼ a2 þ z2 , where a is the radius of the transducer, and the angled brackets
denote an angular averaging, i.e.
2ðπ
If we further assume that the on-axis scatterer has an axis of symmetry that
coincides with the transducer axis, all the angular averages appearing in Eq. (13.11)
disappear except for one and we obtain
pave ðωÞ ¼ 2πρcv0 =ikSR fA½ei ð0; Þ; er ð0; Þexp½2ikz
exp½ikðr e þ zÞA½ei ð0; Þ; er ðθe ; Þ
exp½ikðr e þ zÞA½ei ðθe ; Þ; er ð0; Þ
o ð13:14Þ
þ exp½2ikre hA½ei ðθe ; Þ; er ðθe ; ϕr Þiϕr :
In Chap. 10, we showed that with the use of the method of separation of variables,
we could obtain an exact expression for the plane wave far field scattering ampli-
tude from a sphere. In the general case for a plane wave incident in the ei(θi, ϕi)
590 13 Near Field Measurement Models
direction and scattered in the es ¼ er ðθr ; ϕr Þ direction from a spherical scatterer,
the far field scattering amplitude was given by
1X 1 X n
ðn mÞ!
A½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þ ¼ ðiÞnþ1 ð1Þnm Cn
k n¼0 m¼n ðn þ mÞ!
Pnm ð cos θi ÞPnm ð cos θr Þexp½imðϕi ϕr Þ: ð13:15Þ
From this general expression, all the terms appearing in Eq. (13.14) can be
calculated explicitly. For example, using Eq. (13.15) and one of the properties of
the associated Legrendre function, Pm n , namely
(
1 m¼0
Pnm ð1Þ ¼ ; ð13:16Þ
0 otherwise
we find
1X 1
1. A½ei ð0; Þ; er ð0; Þ ¼ ik ðiÞn Cn ð13:17aÞ
n¼0
1X 1
2. A½ei ð0; Þ; er ðθe ; Þ ¼ ðiÞn Cn Pn ð cos θe Þ ð13:17bÞ
ik n¼0
1X 1
3. A½ei ðθe ; Þ; er ð0; Þ ¼ ðiÞn Cn Pn ð cos θe Þ: ð13:17cÞ
ik n¼0
2ðπ (
2π m¼0
exp½imðϕi ϕr Þdϕr ¼
0 otherwise
0
so that
1X 1
4. hA½ei ðθe ; Þ; er ðθe ; ϕr Þiϕr ¼ ðiÞn Cn Pn ð cos θe ÞPn ð cos θe Þ: ð13:17dÞ
ik n¼0
These scattering amplitude terms are exact results for any sphere. The Cn
coefficients, of course, depend on the specific boundary conditions to be satisfied.
For a rigid sphere, for example, the Cn were given in Chap. 10 by
ðiÞn ð2n þ 1Þ njn ðkbÞ kbjnþ1 ðkbÞ
Cn ¼ h i ; ð13:18Þ
ð1Þ
nhðn1Þ ðkbÞ kbhnþ1 ðkbÞ
where b is the radius of the sphere, jn is a spherical Bessel function of order n, and
ð1Þ
hn is a spherical Hankel function of the first kind of order n.
At high frequencies, these series expressions converge slowly so many terms
must be included in the series. In that case, however, one can use the Kirchhoff
13.1 Model for a Single Fluid Medium 591
approximation for the leading edge response of the sphere to calculate the scattering
amplitude terms instead. For example, combining Eq. (10.52) and Eq. (13.5) the
leading edge response of an arbitrary sphere can be written as
b
A½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þ ¼ Rp ðes ns Þ exp½ikbjei þ er j
2
b
¼ R p ð es n s Þ
2
h i
exp ikbf2 þ 2 cos θi cos θr þ 2 sin θi sin θr cos ðϕi ϕr Þg1=2 ;
ð13:19Þ
where
ðe i þ er Þ
ns ¼ : ð13:20Þ
j ei þ er j
For the special case of a rigid sphere, Rp ¼ 1 in all four of the above terms. Even
for a rigid sphere the last (averaged) scattering amplitude term is in the form of an
integral that must be done numerically unless the angle θe is small. In the small
angle case, we have
h
1=2 i
exp ikb 2 þ 2 cos 2 θe þ 2 sin 2 θe cos ϕr
h
1=2 i h n oi
1=2
ffi exp ikb 2 þ 2 cos 2 θe exp ikb sin 2 θe cos ϕr =ð2 þ 2 cos 2 θe Þ
592 13 Near Field Measurement Models
and
2ðπ
h n 1=2 oi
exp ikb sin 2 θe cos ϕr = 2 þ 2 cos 2 θe dϕr
0
1=2
¼ 2πJ 0 kb sin 2 θe = 2 þ 2 cos 2 θe
so that for the edge-edge wave contribution for the rigid sphere we obtain, explicitly
b h
i
2 e 1=2
40 . hA½ei ðθ ; Þ; er ðθ ; ϕr Þiϕr ¼ 2 exp ikb 2 þ 2 cos θ
e e
1=2
J 0 kb sin 2 θe = 2 þ 2 cos 2 θe : ð13:21d0 Þ
ik
A½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þ ¼ ðer ei Þ n þ ðer þ ei Þ nRp
4π
ð
exp½ikðei þ er Þ xs dSðxs Þ; ð13:22Þ
S
where Rp is the reflection coefficient. Using the expressions for ei, er given in
Eq. (13.5) and writing a general point on the flat end in polar coordinates as
the argument of the exponential term in the integral of Eq. (13.22) becomes
where
and therefore
Thus, if we let
A B
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ cos ψ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ sin ψ ð13:27Þ
A þ B2
2
A þ B2
2
ðb 2ðπ h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
ð13:28Þ
¼ exp ik A2 þ B2 ρ cos ðϕ ψ Þ dϕρdρ:
0 0
ðb
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I ¼ 2π J 0 kρ A2 þ B2 ρdρ ð13:29Þ
0
ib
A½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þ ¼ ð cos θi cos θr Þ Rp ð cos θi þ cos θr Þ
2
1=2
J 1 kb½ sin 2 θi þ sin 2 θr þ 2 sin θi sin θr cos ðϕi ϕr Þ
:
½ sin 2 θi þ sin 2 θr þ 2 sin θi sin θr cos ðϕi ϕr Þ1=2
ð13:31Þ
2ðπ
1
hA½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þiϕr ¼ A½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þdϕr : ð13:32Þ
2π
0
594 13 Near Field Measurement Models
ik
hA½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þiϕr ¼ ð cos θi cos θr Þ Rp ð cos θi þ cos θr Þ
4π
2ðπ ð
b
1=2
J 0 kρ sin 2 θi þ sin 2 θr þ 2 sin θi sin θr cos ðϕi ϕr Þ ρdρdϕr
0 0
ð13:33Þ
and the ϕr integration can be done, since
2ðπ
1=2
J0 x2 þ y2 þ 2xy cos ϕ dϕ ¼ 2πJ 0 ðxÞJ 0 ðyÞ: ð13:34Þ
0
(see Gradshteyn and Ryzhik [6], p. 741) where the ϕi term can be neglected because
the integral can be shown to be independent of it (see the discussion in Chap. 8
following Eq. (8.140)). Thus, we find
ik
hA½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þiϕr ¼ ð cos θi cos θr Þ Rp ð cos θi þ cos θr Þ
2
ðb ð13:35Þ
J 0 ðkρ sin θi ÞJ 0 ðkρ sin θr Þρdρ;
0
where the ρ integration can also be done since (see N.N. Lebedev [5], p. 128)
ðb
c1 J 0 ðc2 bÞJ 1 ðc1 bÞ c2 J 0 ðc1 bÞJ 1 ðc2 bÞ
J 0 ðc1 ρÞJ 0 ðc2 ρÞρdρ ¼ b 2 ð13:36Þ
c1 c22
0
to give, finally
b ð cos θi cos θr Þ Rp ð cos θi þ cos θr Þ
hA½ei ðθi ; ϕi Þ; er ðθr ; ϕr Þiϕr ¼
2ik ð sin 2 θr sin 2 θi Þ
½k sin θi J 0 ðkb sin θr ÞJ 1 ðkb sin θi Þ
k sin θr J 0 ðkb sin θi ÞJ 1 ðkb sin θr Þ:
ð13:37Þ
13.1 Model for a Single Fluid Medium 595
From Eqs. (13.31) and (13.37) all the scattering amplitude terms needed in
Eq. (13.14) can be obtained explicitly as
1. A½ei ð0; Þ; er ð0; Þ ¼ ikb Rp ð0 Þ
2
ð13:38aÞ
2
ikb2
2. A½ei ð0; Þ; er ðθe ; Þ ¼ ð1 cos θe Þ Rp ð0 Þð1 þ cos θe Þ
2
J 1 ðkb sin θe Þ ð13:38bÞ
kb sin θe
ikb2
3. A½ei ðθ ; Þ; er ð0; Þ ¼ 2 ð cos θ 1Þ Rp ðθ Þð cos θ þ 1Þ
e e e e
where l ’ Hôpital ’ s rule has been used on Eq. (13.37) to obtain Eq. (13.38d).
Again, these results are also valid for a rigid cylinder if we take Rp ¼ 1 in all the
above terms. One difference between these results and our previous ones for the
sphere is that the Kirchhoff approximation for the cylinder gives scattering
amplitude terms that are not reciprocal. From our results in Chap. 10 (see
Eq. (10.252)) we should have
A½ei ð0; Þ; er ðθe ; Þ ¼ A½ei ðθe ; Þ; er ð0; Þ; ð13:39Þ
which is satisfied by both the exact scattering amplitude results for the sphere
(Eqs. (13.17b) and (13.17c)) as well as by the high frequency leading edge
results (Eqs. (13.21b) and (13.21c)), but is not satisfied here for the cylinder
unless the angle θe is small. Fortunately, the flat end of the cylinder acts as a very
specular reflector so that its response falls off rapidly at angles away from
normal incidence. Thus, this violation of reciprocity is probably not as serious
as it may seem.
If we consider again our general result (Eq. (13.10)) and make the paraxial
approximation, then both the incident and scattered waves are traveling approxi-
mately in the plus or minus z-directions, respectively. In this case the scattering
amplitude becomes merely its direct back scattered value, A½ei ð0; Þ; er ð0; Þ,
which is independent of ϕi, ϕr, and the edge integrals are independent of one
596 13 Near Field Measurement Models
another so we do not have to distinguish between ϕi, ϕr (ϕi ¼ ϕr ¼ ϕ) and rei , rer
(r ie ¼ r re ¼ r e ). Then Eq. (13.10) reduces to
2π
pave ðωÞ ¼ ρcv0 A½ei ð0; Þ; er ð0; Þexp½2ikz
ikSR
8 92
< ð =
Θ 2π exp½ikðr e zÞdϕ
1
: ; ð13:40Þ
C
2π
¼ ρcv0 A½ei ð0; Þ; er ð0; Þexp½2ikz ½C1 ða; ω; xÞ2 ;
ikSR
which is of the same form as our previous results (see Eq. (12.9)) for this problem.
Thus, the near field model (Eq. (13.10)) reduces to a Thompson-Gray type of
measurement model as a special case.
The near field model presented in the previous section treated the transducer as a
piston (constant velocity) source and used the average pressure as the quantity to
measure on reception. Both these assumptions are physically reasonable but they
are not the only possible models we could have taken. For example, we could have
taken the average velocity, (vz)ave, on reception instead. In that case, since
1 ∂pðx; ωÞ
ðvz Þave ¼ ð13:41Þ
iωρ ∂z ave
where we have assumed high frequencies so that terms of O(1/krr) or smaller have
been neglected. The only differences between Eq. (13.42) and Eq. (13.1) is that the
constant outside the integral has been changed by a factor of 1/ρc, and a directivity
term z/rr appears in the integral on reception. We will call our original model the
velocity-pressure model (piston velocity taken on the sound generation side and
average pressure on reception) and this new model the velocity-velocity model.
These two models, however, are not the only possibilities. We could have
originally modeled our transducer as a constant pressure source instead of the
piston velocity model used in the above two cases. To see the effects of this
13.2 Other Models for a Single Fluid Medium 597
which eliminated the second term in the integral of Eq. (13.43) since we showed
∂G* =∂n ¼ 0 on the plane Sp. This process then lead directly to the Rayleigh-
Sommerfeld equation (Eq. (8.8)). For a piston (constant velocity source) on an area
S in an infinite baffle, we then found
ð
iωρv0 expðikr Þ
pðx; ωÞ ¼ dS ð13:45Þ
2π r
S
which was the starting point for our above two models. However, we could have
taken instead
0 expðikr Þ exp ikr *
G ¼ ð13:46Þ
4πr 4πr*
Thus, if we assume that the transducer produces a constant pressure p0 over a finite
aperture S and the pressure is zero on the remainder of Sp, we would find (letting
r ! r i to correspond to our previous notation)
ð
ikp0 z expðikr i Þ
pðx; ωÞ ¼ dS: ð13:49Þ
2π ri ri
S
The coefficient outside the integrand of Eq. (13.49) can be made identical to that of
the Rayleigh-Sommerfeld equation if we let p0 ¼ ρcv0 . The integrand itself is
different from the Rayleigh-Sommerfeld equation only in the additional directivity
function z/ri. Thus, if this constant pressure source model was used in place of the
piston velocity source, Eq. (13.1) of our near field model would be changed to
ðð
0
z expðikr i Þ expðikr r Þ
pave ðωÞ ¼ ikp0 =2πSR Aðei ; er ÞdS y dSðyÞ:
ri ri rr
SR ST
ð13:50Þ
We will call this model the pressure-pressure model. Similarly, if we had used a
constant pressure source and calculated average velocity on reception instead, we
would find
ðð
0
z expðikr i Þ z expðikr r Þ
ðvz Þave ¼ ikp0 =2πρcSR Aðei ; er ÞdS y dSðyÞ;
ri ri rr rr
SR ST
ð13:51Þ
which we refer to as the pressure-velocity model. Table 13.1 lists each of these four
models and their corresponding coefficients and integrand directivity terms.
In the paraxial approximation, all four of these models are identical in form since
in that approximation both z=ri ! 1, z=r r ! 1 and the different external coeffi-
cients can be made the same by merely redefining the incident pressure and/or
velocity terms appropriately. Thus, these models will all lead to the same Thomp-
son-Gray type of LTI measurement model and it does not make any difference what
Table 13.1 Four possible near field models and their differences both in their corresponding
coefficient terms and the directivity functions that appear in the integrands during transmission
(integrand-1) or reception (integrand-2). The coefficient column indicates what the input velocity
term v0 in the velocity-pressure model must be replaced by in order to obtain the corresponding
coefficient for the other models
Model Coefficient Integrand-1 Integrand-2
Velocity-pressure v0 1 1
Velocity-velocity v0 ! v0 =ρc 1 z/rr
Pressure-pressure v0 ! p0 =ρc z/ri 1
Pressure-velocity v0 ! p0 =ðρcÞ2 z/ri z/rr
13.2 Other Models for a Single Fluid Medium 599
average quantity one uses on reception. Similarly, one sees that in such cases it also
makes no difference whether one uses a constant pressure or velocity source on the
input side.
For the more general near field model case, of course, differences do exist
between the various models. Since the analysis follows the same steps in going
from Eq. (13.1) to Eq. (13.10), we will write down here only the final results in a
form analogous to Eq. (13.10), which was for the velocity-pressure model. The
velocity-velocity model gives
ðvz Þave ¼ v0 =2πikSR 4π 2 Θ2 A½ei ð0; Þ; er ð0; Þexp½2ikz
ð
z
2πΘ A ei ð0; Þ; er θre ; ϕr exp ik r re þ z dϕr
rre
C
ð
2πΘ A ei θie ; ϕi ; er ð0; Þ exp ik r ie þ z dϕi
C
ð13:52Þ
9
ðð =
z
þ A ei θie ; ϕi ; er θre ; ϕr exp ik r ie þ r re dϕi dϕr :
rre ;
CC
Since all these near field models were derived under a high frequency approx-
imation, one cannot expect them to necessarily give accurate results as ω ! 0.
Normally, this is not a problem since the scattering amplitude terms can be
shown to be of O(ω2) as ω ! 0 so that the response at low frequencies is
small. However, if one uses approximate high frequency scattering amplitude
expressions in these models which do not go to zero as ω ! 0 (such as the
leading edge response of the sphere given in Eqs. (13.21a)–(13.21d), then
because of the 1/ikSR term present in all these models, the average pressure or
velocity expressions will typically go to infinity at zero frequency. There are
some special cases (on-axis response of a sphere to a circular transducer for all
models but the pressure-velocity model, for example) where the combination of
four direct and edge wave terms themselves vanish (see Problem 13.6), thus
canceling out this infinity, but this behavior is the exception rather than the rule.
In general, simply the use of an exact (or approximate) scattering amplitude
expression that vanishes at low frequency will eliminate this singular behavior.
In the Thompson-Gray models based on the paraxial approximation, this type of
behavior is not exhibited because although the 1/ikSR term also appears in those
models (see Eq. (13.40)), the product of diffraction coefficient terms, (C1)2,
containing the four direct and edge wave terms in that approximation, always
vanishes as O(ω2) as ω ! 0.
Although all four models are different, those differences arise solely from
terms such as z/rei and/or z/rer , which are functions only of the closeness of the
scatterer to the transducer. Thus, one expects that all the models will agree
substantially when the geometric conditions z=r re ffi z=r ie ffi 1 are met, regardless
of the frequency. For example, if the scatterer is located no closer than one
transducer diameter away from the transducer, then the above geometric condi-
tions will always be satisfied to within approximately 10 % (z=rre ¼ z=r ie ffi 0:894).
If we use this one-diameter distance, D, as an approximate estimate of the limit
of the domain of agreement between all these theories, then where this position is
located in the near field of the transducer depends on the frequency and the wave
speed of the material that the transducer is radiating into (see Fig. 13.5). For a
5 MHz, 1/2 in. diameter planar transducer radiating into water, the near field
distance is approximately 5.36 in. so that D ¼ 0:09N (Fig. 13.5a). In contrast, at
1 MHz, D ¼ 0:47N for this same transducer (Fig. 13.5b). If we also use these fluid
models to estimate similar domains of common agreement for a transducer
radiating into a solid, for a 5 MHz, 1/2 in. diameter P-wave transducer radiating
into steel we would have D ¼ 0:37N (Fig. 13.5c), while at 1 MHz, D ¼ 1:85N for
this transducer (Fig. 13.5d). Thus, at frequencies of 5 MHz and higher, the
distance D lies well within the near field of the transducer, and for scatterers
located at a distance z, where z
D, we can consider all the theories to produce a
single equivalent near field model. At lower frequencies, particularly as
Fig. 13.5d shows, one may have to actually be greater than a near field distance
before the theories will coincide. Thus, calling these theories “near field” theo-
ries is somewhat of a misnomer. However, we will stay with that description with
the understanding that for z < D any of the above models will need some
13.3 Model for a Fluid-Solid Interface (Normal Incidence) 601
Fig. 13.5 Near field distance (N ) and a location (D) one diameter away from the face of a 0.5-in.
diameter circular transducer for: (a) 5 MHz and the transducer radiating into water, (b) 1 MHz and
the transducer radiating into water, (c) 5 MHz and the transducer radiating into steel, and
(d) 1 MHz and the transducer radiating into steel
independent validation. Such validation can likely only come from comparison
with detailed experiments. However, at such close distances, other model
assumptions (such as the assumption of piston behavior) may also need to be
carefully examined.
The generalized fluid models considered in this Chapter can be easily extended to
the case where an immersion transducer is oriented at normal incidence to a planar
fluid-solid interface and an arbitrary scatterer is located in the solid (pulse-echo
setup) as shown in Fig. 13.6. To set this problem up, recall from Eq. (8.192) that the
displacement vector in the solid can be expressed in terms of boundary diffraction
waves as
(
X ρ1 cp1 v0 ð β;P
uðx; ωÞ ¼ T 12 θ1β dβ
β¼P, S
2πiωρ2 cβ2
Cr
9
h
i
0 >
>
β
exp i kp1 D1 þ kβ2 D2 β β
n er1 ds y >
=
2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffir
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
> ;
β
1 n er1 D1β þ cβ2 D2β =cp1 D1β þ cβ2 cos 2 θ1β D2β =cp1 cos 2 θ2β > >
;
ð13:55Þ
602 13 Near Field Measurement Models
and Stokes’ theorem, the integral over the receiver surface in Eq. (13.57) can be
written in terms of an integral over the contour Cr (see Eq. (8.79)) to obtain
(
X X iρ cp1 cα2 U β ð
pave ðωÞ ¼ 2 0
Aα;β eiβ ; esα liα
α¼P, S β¼P, S
2πSr
Cr
9
ð T P;α n e P expip P ðy xÞ þ p α D p P D
ð þ1
þ1 >
>
=
21 p1 1 2z 1z
2 dp dp dsðyÞ:
x y
>
>
1 1 1 n ep1
P α
p2z ;
ð13:59Þ
13.3 Model for a Fluid-Solid Interface (Normal Incidence) 603
The angular plane wave spectrum integral can then be evaluated approximately by
the method of stationary phase, as done previously (see Sect. 8.3.1), giving
(
X X ρ2 cp1 cα2 U β ð
pave ðωÞ ¼ 0
Aα;β eiβ ; esα liα
α¼P, S β¼P, S
S r
Cr
9
P;α α =
α
n ur1 T 21 θ3 exp i kP1 D4α þ kα2 D3α
h ipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dsðyÞ;
α 2
1 n ur1 D3α þ cp1 D4α =cα2 D3α þ cp1 cos 2 θ3α D4α =cα2 cos 2 θ4α ;
ð13:60Þ
where uαr1 is a unit vector along a stationary ray path (in going from the scatterer to
the receiver), for a wave of type α in the solid, as shown in Fig. 13.6. If we place the
explicit expression for Uβ0 into Eq. (13.60), then we have, finally
ð13:61Þ
Although Eq. (13.61) is fairly compact in form, because the extended edge integrals
include both direct wave and edge wave contributions, there are a total of sixteen
possible edge wave and direct wave contributions contained in Eq. (13.61). Only
nine of those combinations, however, actually exist since there are no direct shear
waves generated in the second medium. It is interesting to compare the form of this
generalized model to that for the same problem in the paraxial approximation (see
Eq. (12.26)). For a planar transducer at normal incidence Eq. (12.26) reduces to:
pave ðωÞ ¼ ρ1 cp1 v0 T P;P
12 ð0 Þexp i 2k p1 z1 þ 2kp2 z2
AP;P ei0P ð0 , Þ; es0 P
ð0 , Þ ðnÞ T P;P
21 ð0 ÞC1 ða, ω, 0 , 0 , z1 , z2 Þ
P
2π
C1P ða, ω, 0 , 0 , z1 , z2 Þ :
ikp2 SR
ð13:62Þ
We see some similar structure in both Eqs. (13.61) and (13.62). The propagation
terms, transmission coefficient terms, scattering amplitude term, and a term due to
averaging are all present in each expression. In the paraxial approximation, all such
604 13 Near Field Measurement Models
terms were evaluated only for sets of fixed ray paths to and from the transducer and
were present in simply a product fashion with each other and with diffraction
correction terms for the transmission and reception processes. In contrast, in the
generalized model, the diffraction terms appear as integral operators acting on all
the other terms, which must be evaluated along multiple ray paths. Also, in the
paraxial case there are only P-wave contributions in both media, since all the mode-
converted terms are zero (to first order) in that approximation. Thus, Eq. (13.62)
contains only four P-wave contributions (direct-direct wave, direct-edge wave,
edge-direct wave and edge-edge wave) in contrast to the nine waves present in
Eq. (13.61). Those four contributions are not directly apparent in Eq. (13.62)
because they are contained in the diffraction coefficient term, CP1 , (which appears
twice in Eq. (13.62)) where
ð
C1P ða, ω, 0 , 0 , z1 , z2 Þ ¼ Θ 1=2π exp ikp1 ρ2e =2 z1 þ cp2 z2 =cp1 dϕ: ð13:63Þ
C
The first term in this coefficient represents the direct wave contribution, which is
present only in the main beam of the transducer, while the second term is the
edge wave.
Another difference between the two models is that the general paraxial result
(Eq. (12.26)) was derived for the case of a general oblique incidence pitch-catch
immersion setup while the generalized model here is only valid when a planar
transducer is at normal incidence to the interface in a pulse echo setup. This
restriction is present in the near field model case because, as mentioned previously
in Chap. 8, the boundary diffraction wave decomposition has not yet been able to be
developed in oblique incidence cases. Of course, it is possible to write the diffrac-
tion integrals in Eq. (13.61) as surface integrals (which could be done via an edge
element approach as discussed in Chap. 8) for the more general pitch-catch, oblique
incidence problems.
A generalized measurement model of the type discussed in Sect. 13.1 was first
described by Sedov et al. [1] and Song et al. [2]. In both of those papers a velocity-
velocity formulation was used instead of the velocity-pressure model considered in
Sect. 13.1. Although there have not been any direct comparison of measurement
models of the types discussed in Sect. 13.2, other authors (see Wolf and Marchand
[3], [4] and Stamnes [7], for example) have compared different scalar diffraction
models for radiation through an aperture (considered were a constant velocity
source (Rayleigh-Sommerfeld) model, a constant pressure source model, and a
Kirchhoff-based model which uses a combination of pressure and velocity sources).
Little difference between the various models was found except in a region close to
the aperture, as would be expected from our previous discussions.
13.5 Problems 605
13.5 Problems
13.1 If we use the integral representation theorem (Eq. (8.1)) for a fluid half space
ðz
0Þ and the ordinary fundamental solution, G, (Eq. (8.2)), then the
pressure at any point in the fluid is given by (see Fig. 8.1):
ð
pðx; ωÞ ¼ ½Gðx; y; ωÞ∂pðy; ωÞ=∂nðyÞ pðy; ωÞ∂Gðx; y; ωÞ=∂nðyÞdSðyÞ:
Sp
ðP13:1Þ
Also, the fundamental solution and its normal derivative are, at high frequen-
cies, on the aperture
expðikr Þ
G¼
4πr
ðP13:3Þ
∂G z expðikr Þ
¼ ik
∂n r 4πr
so that if we use these incident wave values in Eq. (P13.1), which is the
Kirchhoff approximation, the pressure in the fluid is
ð
iωρv0 expðikr Þ z expðikr Þ
pðx; ωÞ ¼ þ dS: ðP13:4Þ
4π r r r
S
Using this model (to represent the incident wave field) and the approach of
this chapter, determine an expression for the received average pressure at the
same transducer due to an arbitrary scatterer in the fluid. Compare this
Kirchhoff-based near field model and the velocity-pressure model (derived
earlier from the Rayleigh-Sommerfeld integral) for the case where the scat-
terer is an on-axis sphere in the fluid. Where do these theories agree and where
do they break down?
13.2 Consider the case of a spherical void in a solid which is on the axis of a planar
piston immersion transducer (Fig. P13.1). Using the leading edge response of
the sphere, as predicted by the Kirchhoff approximation (see Chap. 10),
606 13 Near Field Measurement Models
determine, from Eq. (13.61), an expression for the average pressure, pave(ω),
received by only the direct and edge P-wave components only, i.e. neglect all
mode conversion. Plot the magnitude of the normalized pressure, pave(ω)/
ρ1cp1v0, versus the distance, z, of the transducer in the fluid from the interface
for a 1/2 in. diameter, 5 MHz transducer radiating normally on a 1 mm
diameter void in steel located 5 mm beneath the interface as shown in
Fig. P13.1.
ρ2 , cp 2 , cs 2
ρ1 , cp 1 z 5 mm
(a) Compare these near field model results with the same plot for a near field
“fluid-fluid” model of this problem where all the media are treated as
fluids.
(b) Compare these near field model results with the same plot based on the
use of the paraxial approximation (Eq. (13.62)). Is there any difference in
the paraxial approximation, between a fluid-solid model and a fluid-fluid
model?
13.3 Repeat the comparisons of problem 13.2 for the case where the scatterer is an
on-axis 1 mm diameter circular crack whose normal is parallel to the trans-
ducer axis. What differences, if any, are there between these comparisons and
those of problem 13.2?
13.4 Consider the pulse echo response of an on-axis 3 mm diameter spherical void
in a fluid due to a 1/2 in. diameter, 5 MHz planar transducer.
(a) Using Eq. (13.14) for the velocity-pressure model, and (1) the scattering
amplitude expressions determined from the leading edge response of the
sphere in the Kirchhoff approximation (Eqs. (13.21a)–(13.21d)), plot the
magnitude of the normalized pressure, pave(ω)/ρcp1v0, versus the dis-
tance, z, of the center of the sphere from the transducer.
(b) Repeat the plot of part (a) using the exact scattering amplitude expres-
sions (Eqs. (13.17a)–(13.17d)) instead. Compare your two results.
13.5 Invert the velocity-pressure near field model (Eq. (13.10)) into the time
domain to obtain an explicit expression for the average received pressure as
a function of time (impulse-response). Do the same inversion for the paraxial
result (Eq. (13.40)) and discuss the similarities and differences between the
two models.
13.6 Using all four near field theories developed in this chapter for a scatterer in a
fluid, consider the case where the scatterer is an on-axis rigid sphere. Use the
References 607
References
1. A. Sedov, L.W. Schmerr, S.J. Song, A bounded beam solution for the pulse-echo transducer
response of an arbitrary on-axis scatterer in a fluid. Wave Motion 19, 159–169 (1994)
2. S.J. Song, L.W. Schmerr, A. Sedov, A frequency domain ultrasonic model for the pulse-echo
transducer response of an arbitrary scatterer in a fluid. Res. Nondestruct. Eval. 5, 111–122
(1993)
3. E. Wolf, E.W. Marchand, Comparison of the Kirchhoff and the Rayleigh-Sommerfeld theories
of diffraction at an aperture. J. Opt. Soc. Am. 54, 587–594 (1964)
4. E.W. Marchand, E. Wolf, Consistent formulation of Kirchhoff’s diffraction theory. J. Opt. Soc.
Am. 56, 1712–1722 (1966)
5. N.N. Lebedev, Special Functions and Their Applications (Dover Publications, New York, 1972)
6. I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products (Academic, New York,
1980)
7. J.J. Stamnes, Waves in Focal Regions (Adam Hilger, Boston, 1986)
Chapter 14
Quantitative Ultrasonic NDE with Models
In previous Chapters we have shown that it is possible to model or measure all the
important factors that enter into an ultrasonic NDE measurement setup. These
ultrasonic measurement models (and similar models that are being developed for
other techniques such as eddy currents and x-rays [1]) are forming the basis for a
new, quantitative NDE technology where entire NDE tests can be simulated and
test parameters (including those associated with flaws) can be manipulated for
engineering design and analysis purposes. This new technology has a number of
important practical applications. For example, models are important in applications
where one wants to guarantee the sensitivity of a given NDE measurement setup.
Typically, in the past this has involved the very costly process of (a) manufacturing
flawed calibration specimens that closely mimic the material and other conditions
of the actual setup, and (b) carefully specifying the test procedures so that variations
due to different pulser/receiver and transducer choices and settings can be mini-
mized. However, with a model-based approach, it is possible to develop calibration
procedures that are independent of a particular measurement setup and also trans-
ferable across different materials and geometries, making multiple calibration
samples unnecessary.
In Sect. 14.1, we will demonstrate model-based approaches that allow one to
obtain “effective” geometrical parameters of both planar and focused probes in
addition to the overall system efficiency factor. It will be shown that with these
effective parameters and the efficiency factor, it is possible to accurately predict the
complete voltage versus time waveforms received. In Sect. 14.2, the near field
models of Chap. 13 will be used to generate an explicit scattering model for a flat
bottom hole, including the development of model-based distance-gain-size (DGS)
curves. A flat bottom hole is a highly specular reference reflector that is commonly
found in standard ASTM reference block specimens. DGS curves, which were
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_14) contains supplementary material, which is available to authorized users.
originally developed by Krautkramer [2, 3], have also been used for sizing flaws in
terms of flat-bottom hole diameters that produce equivalent amplitude signals.
A Thompson-Gray measurement model relates the frequency components of the
received voltage directly to the far field scattering amplitude of the flaw.
As mentioned in Chap. 12 this explicit relationship is important since the far field
scattering amplitude (unlike the received voltage itself) is a function only of the
properties of the flaw and not the other parts of the measurement setup such as the
pulser/receiver, transducer(s), etc. Since in most NDE flaw measurements we are
primarily interested in obtaining the important characteristics of an unknown flaw
(flaw type, size, orientation, etc.) from its ultrasonic response, this can be done in a
rational manner if the scattering amplitude can be extracted from the measurement
process and then used in appropriate flaw classification and/or sizing methods. In fact,
in the next Chapter several model-based flaw sizing methods that use flaw scattering
amplitudes will be described. In Sect. 14.3 of this Chapter we will demonstrate that
scattering amplitudes can be obtained from the received signals by combining the
Thompson-Gray models developed previously with deconvolution procedures.
In the past, NDE inspectability requirements have rarely been considered at the
very early design stages of components. This is understandable since in current
NDE practice one can look at inspectability issues only by performing NDE tests on
actual manufactured parts. However, if models such as the ones discussed in this
book are interfaced to computer-aided design (CAD) tools, materials data, and
failure (flaw) information, NDE inspections conducted during both manufacture
and in-service use can be simulated entirely in software, even at the initial stages of
a design. This capability can make NDE a true partner in the design/manufacturing
process and help to prevent the making of parts that are difficult or impossible to
inspect. In Sect. 14.4 we will describe ultrasonic NDE simulators that have been
developed for these purposes.
Consider the calibration setup where a planar circular piston transducer of radius
a is radiating into a fluid and where a reflector is located on the transducer central
axis. In Chap. 12 we showed that the measured pulse-echo response of this
configuration can be modeled as (see Eq. 12.13):
2π
V 0 ðz; ωÞ ¼ βðωÞexp½2ikzexp½2αðωÞzC1 Aðez ; ez Þ
2
; ð14:1Þ
ikSr
where
h
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
C1 ða; ωÞ ¼ 1 exp ik z 2 þ a2 z
R p ð 0∘ Þ b
Aðez ; ez Þ ¼ expð2ikbÞ: ð14:4Þ
2
Alternatively, if we reflect instead off the flat end of a small steel cylinder, of radius
b, the pulse-echo far field scattering amplitude can be obtained (see Chap. 10) as
612 14 Quantitative Ultrasonic NDE with Models
i k Rp ð0∘ Þ b2
Aðez ; ez Þ ¼ : ð14:5Þ
2
For a water-steel interface, Rp ð0∘ Þ ffi 1 so that both these scatterers can be consid-
ered to be rigid. Equation (14.3) shows that for either of these scatterers any nulls in
the on-axis pressure response of the transducer will also be nulls of the measured
voltage response, V0(z, ω). In particular, in Chap. 8 we showed that the last on-axis
null in the on-axis near field pressure was located at a distance, znull, given, for ka
large, by
a2
znull ¼ : ð14:6Þ
2λ
Equation (14.6) gives us a direct means for determining an effective radius for the
transducer in the following manner. If the scatterer is moved along the central axis
and the received waveforms captured and stored for different locations, a fast
Fourier transform can be performed on those time domain signals and a particular
frequency component of the transformed signals, V0(z, ωc), can be found as a
function of the distance z, where ωc is normally taken at some fixed value near
the transducer’s center frequency. If the location of the last on-axis null, znull, is
obtained from this plot, then the effective radius of the transducer, aeff, can be
obtained by solving Eq. (14.6) for a, i.e.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
aeff ¼ 2 znull λc ; ð14:7Þ
Equation (14.3) is also directly applicable to the case where a scatterer is on the axis
of a spherically focused probe where the on-axis pressure is a function of both the
radius, a, and geometrical focal length, R0:
14.1 Transducer/System Characterization 613
with
Solving Eq. (14.11) for a, which we again define as the effective radius, aeff, we find
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2zmin λc R0
aeff ¼ ; ð14:12Þ
R0 zmin
which reduces to the planar case (Eq. 14.7) when R0 ! 1. Unlike the planar case,
however, the effective radius also depends on the geometric focal length, which is a
separate parameter whose effective value must also be obtained. One way to
independently obtain the focal length parameter is to also measure the location,
zmax, of the maximum on-axis pressure response (i.e. the location of the true focus).
From Chap. 8, this maximum location should satisfy the transcendental equation
(Eq. 8.111)
Recall that
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δ¼ ðzmax hÞ2 þ a2 zmax
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð14:14Þ
h ¼ R0 R20 a2
and kc ¼ ωc =c is the wavenumber at the center frequency. Thus, through the use of
Eq. (14.12), Eq. (14.13) can be written entirely in terms of R0 and the known values
614 14 Quantitative Ultrasonic NDE with Models
ωc, zmin, zmax. If one attempts to solve Eq. (14.13) directly for R0 in this manner,
unfortunately the root of this equation was found to be very sensitive to errors in
locating zmax, making it difficult to determine an accurate value for the effective
focal length. In fact, in some cases direct solutions of Eq. (14.13) for R0 can give
values where R0 < zmax , which is clearly incorrect since the true focus always
occurs at a distance less than that of the geometrical focus. To avoid this problem,
therefore, an alternate two-step procedure was suggested [5] where in the first step
we rewrite Eq. (14.13) in the form
ðπ 2 x2 Þ 2ð1 þ 2x=ðkc zmax ÞÞ zmax ðπ 2 x2 Þ
x cos ðxÞ π x þ ¼ sin ðxÞ π x þ ;
kc zmin ð2 þ 2x=ðkc zmax ÞÞ zmin kc zmin
ð14:15Þ
with x ¼ kc δ=2. Solving for the root, x, of Eq. (14.15) is well-behaved. In fact, in
most cases kc zmin
1, kc zmax
1 and Eq. (14.15) reduces to
which is very simple in structure and has a well-defined root 0 < x < π=2. Once the
value, x, of the root of Eq. (14.15) is obtained, in the second step of this process an
expression for R0 in terms of x, zmax, and zmin can be obtained directly from
Eqs. (14.10) and (14.14) which we will take as the definition of the effective
focal length, (Ro)eff. It follows after some considerable algebra that this expression
is given exactly by
π x þ ðπ 2 x2 zmin =zmax Þ=ðkc zmin Þ
ðR0 Þeff ¼ zmax ð14:17Þ
π x zmax =zmin þ ðπ 2 x2 Þ=ðkc zmin Þ
Equation (14.18) shows that ðR0 Þeff > zmax in general and as the frequency
increases, since zmax =zmin ! 1, we have ðR0 Þeff ! zmax , as expected from geomet-
rical ray theory.
In summary, we can find the effective radius and focal length of a spherically
focused probe by measuring experimentally the location of both the on-axis null
before the true focus and the location of the true focus itself. Then solving
Eq. (14.15) for the intermediate parameter x, the effective focal length can be
obtained from Eq. (14.17) and this value used in Eq. (14.12) to find the
corresponding effective radius. Note that in locating zmax, what one wants to
measure is the location where the on-axis pressure is a maximum. If we solve the
14.1 Transducer/System Characterization 615
Thompson-Gray model (Eq. 14.3) in terms of this on-axis pressure, at the center
frequency we can write that expression in the form
pða; R0 ; z; ωc Þ 2
¼ Γðωc ÞV 0 ðz; ωc Þexp½2αðωc Þz; ð14:19Þ
ρcv0
which shows that the maximum of the non-dimensional pressure term ( p/ρcv0)2
(which occurs when the pressure p itself is a maximum) can be determined by
locating where the quantity V0(z, ωc)exp[2α(ωc)z] is a maximum since all the terms
contained in Γ are independent of z. In some cases, the material attenuation may be
small enough so that maxima of V0 and p are indistinguishable. Otherwise, the
product of the measured voltage response with the attenuation term of Eq. (14.19)
must be used when determining the zmax location. Of course, the location of zmin does
not require any such compensation since the null locations of V0 and p coincide. As in
the planar probe case the effective parameters obtained in this manner do depend
somewhat on the frequency chosen, so that one can again compensate for this
variability by taking an average of values obtained about the center frequency.
Two MATLAB® functions, eff_parameters and transeq, have been written to
determine effective parameters for planar and focused transducers. If Eq. (14.15) is
placed in the form f ðx; zmin ; zmax ; kc Þ ¼ 0 then the function transeq returns the values
of that function at an arbitrary value of x with the call
where (zmin, zmax) are the measured locations (in mm) and k is the wave number
(in rad/mm). The MATLAB® function eff_parameters determines the value of
x where f ¼ 0 using the built-in MATLAB function fzero and then calculates the
effective radius and focal length with Eqs. (14.17) and (14.12). The calling
sequence is
where (aeff, Reff) are the effective radius and focal length (in mm), c is the wave
speed (in m/s), f is the frequency (in MHz), and (zmin, zmax) are again the
measured locations (in mm). If the input value zmax is omitted the functions returns
Reff ¼ inf and calculates the effective radius of a planar transducer using
Eq. (14.7). For example, if one simulates the on-axis response of a 6.35 mm radius,
76 mm focal length transducer radiating into water (c ¼ 1480 m/s) at 5 MHz with
the MATLAB function onaxis_foc then the zmin and zmax values are easily found
as approximately zmin ¼ 35.89 mm, zmax ¼ 59.31 mm, respectively. In this case
eff_parameters returns (approximately) aeff ¼ 6.35 mm and Reff ¼ 75.9 mm. Find-
ing the minimum location is generally easier that the maximum as the pressure
function is rather broad at the maximum location and the effective values are
somewhat sensitive to the value chosen for this location. How to handle this
sensitivity is discussed in Sect. 14.1.4.
616 14 Quantitative Ultrasonic NDE with Models
If one instead uses the function onaxis_foc to generate the on-axis response of a
6.35 mm radius planar transducer radiating into water at 10 MHz, the zmin location
can easily be found as approximately zmin ¼ 136 mm and calling eff_parameters
with only this value returns aeff ¼ 6.34 mm, Reff ¼ inf.
will be completely known once the effective parameters of the transducer and the
attenuation are obtained. If we want to find the system function instead then note that
in pulse echo setups such as this one we have simply sðωÞ ¼ βðωÞ=2. For water, the
attenuation at room temperature has previously been given (see Eq. 9.2) and can be
used directly in Eq. (14.21). If the voltage versus time waveform received from the
scatterer is captured when the scatterer is at a particular on-axis location, zr, and the
frequency components, V0(zr, ω), computed from this waveform via a fast Fourier
transform, then β(ω) can in principle be determined by a division process, i.e.
V 0 ðzr ; ωÞ
βðωÞ ¼ : ð14:22Þ
pave ðzr ; ωÞ
As discussed in Chap. 9, however, when V0 and pave become small, as they typically
do at both large and small frequencies, this division process is contaminated by
noise. To stabilize this process a Wiener filter was employed in Chap. 9. Applying
the same Wiener filter process to the deconvolution process of Eq. (14.22), we
would find instead
14.1 Transducer/System Characterization 617
2 3
*
½p ð z ; ωÞ
βðωÞ ¼ V 0 ðzr ; ωÞ4 ave r
n o5; ð14:23Þ
jpave ðzr ; ωÞj þ ε2 max jpave ðzr ; ωÞj2
2
where ε is a small stabilization constant and ½∗ denotes the complex conjugate. It
can be seen that when V0 and jpave j become small, the Wiener filter effectively
forces the value for β(ω) to go smoothly to zero but where maxfjpave jg >> ε the
Wiener filtered response reduces to just the original straight division process.
Examples of the determination of the efficiency factor in this manner will be
discussed in the next section. For more explicit examples of the models used to
experimentally determine a system function, see Schmerr and Song [6].
Table 14.1 shows the effective radii and focal lengths obtained for four commercial
transducers by following the previously outlined procedures [7]. Figure 14.1 also
Table 14.1 Effective parameters obtained for four commercial spherically focused ultrasonic
transducers
Nominal Values Effective Values Center frequency
Probe R0 (cm) a (cm) R0 (cm) a (cm) fc (MHz)
A 10.2 0.64 9.7 0.66 5.00
B 15.2 0.64 19.5 0.70 10.00
C 5.1 0.64 5.4 0.66 15.00
D 10.2 0.95 10.6 0.99 2.25
1.00
0.67
Beta, V-s
xE-6
0.33
0.00
0.0 5.0 10.0 15.0 20.0
Freq, MHz
Fig. 14.1 Magnitude of the system efficiency factor calculated for probe A when the scatterer was
on-axis and 9.6 cm from the transducer. Reprinted with permission from: T. Lerch, L.W. Schmerr,
and A. Sedov, Characterization of spherically focused transducers using an ultrasonic measure-
ment model approach, Res. Nondestr. Eval., 8, 1-21, (1996)
618 14 Quantitative Ultrasonic NDE with Models
0.080
R0=8.7 cm
R0=9.0 cm
R0=9.4 cm
R0=9.7 cm
R0=10.1 cm
R0=10.6 cm
0.053 experiment
IVrl, V-s
xE-6
0.027
0.000
0.0 5.3 10.7 16.0
z, cm
Fig. 14.2 Family of 5 MHz on-axis plots for probe A that demonstrates the effects of changing the
values of the effective parameters (R0 values shown only). Reprinted with permission from:
T. Lerch, L.W. Schmerr, and A. Sedov, Characterization of spherically focused transducers
using an ultrasonic measurement model approach, Res. Nondestr. Eval., 8, 1-21, (1996)
shows the magnitude of the system efficiency factor versus frequency calculated for
one of those transducers (probe A). In all these cases, the reference scatterer was a
0.25 in. diameter steel ball bearing. Although the effective radii obtained were very
close to that of the nominal radii specified by the manufacturer, in one case the
effective focal length differed substantially from the nominal value. However, it
should be realized that for a spherically focused probe, the on-axis wave field is
sensitive to the choice of effective values. Figure 14.2 shows a comparison of the
predicted and measured on-axis pressure for transducer D in Table 14.1. Although
the nominal and effective values differed only by about 4 %, the on-axis pressures
predicted can differ by 10 % or more. This sensitivity was also reflected in the
manner in which the effective values were obtained for these focused transducers,
as discussed in Sect. 14.1.2. Although in principle one needs only to measure two
quantities—the location of zmin and zmax—in practice the location of zmax in
particular was found to be difficult, particularly for lower frequency transducers
where the peak of the on-axis pressure profile can be rather broad. In most cases,
there were a number of competing values for zmax, all of which produced substan-
tially different values of the effective focal length. For example, six slightly
different zmax values were found for probe A in Table 14.1 resulting in calculated
effective focal lengths varying from 8.7 to 10.6 cm. To choose the “best” value,
therefore, the following procedure was used. For all the potential zmax values (and
also, in some cases, potential zmin values as well) the effective parameters were
determined by the procedures described previously and the system efficiency factor
also calculated from the on-axis response as measured at the geometrical focus. The
theoretical on-axis voltage, V0(z, ωc), was then computed from these parameters
and compared to the actually measured voltage frequency components as shown in
Fig. 14.3. In this case, the curve that fit the on-axis experimental results best in a
14.1 Transducer/System Characterization 619
0.0300
effective
nominal
experiment
0.0200
IVrl, V-s
xE-6
0.0100
0.0000
0.0 5.3 10.7 16.0
z, cm
Fig. 14.3 Comparison of the theoretical and measured on-axis voltage magnitudes for probe D
when (1) the nominal parameters are used (dashed line), and (2) the effective parameters are used
(solid line). Reprinted with permission from: T. Lerch, L.W. Schmerr, and A. Sedov, Character-
ization of spherically focused transducers using an ultrasonic measurement model approach, Res.
Nondestr. Eval., 8, 1-21, (1996)
least squares sense, was for a focal length R0 ¼ 9:7 cm (Fig. 14.3), the value that is
reported in Table 14.1. A similar fitting procedure was also used to determine the
other best effective values in that table. The resulting parameters obtained in this
manner then typically predicted the entire wave field of the transducer accurately.
Figure 14.4 shows, for instance, the close agreement between theory and experi-
ment obtained for probe A when the effective values were used to predict the cross
axis voltage profile (at the center frequency of 5 MHz) for an axial distance
z ¼ 6:9 cm which was close to the plane normal to the transducer containing the
true focus. Similarly, Fig. 14.5 shows the experimentally observed and theoretically
predicted time domain waveforms predicted for probe A at z ¼ 6:9 cm. The
theoretical waveform in Fig. 14.5 was calculated by placing the effective transducer
values and the measured β(ω) into the Thompson-Gray measurement model and
inverting the resulting frequency response, V0, into the time domain by an inverse
fast Fourier transform. The close correspondence between the magnitude and shape
of the predicted and measured waveforms of Fig. 14.5 illustrates the ability to
quantitatively predict measured signals when the appropriate effective transducer
values and system efficiency are obtained in this manner. These same values could
be used to obtain similar quantitative pulse-echo immersion measurements in other
NDE tests with the same transducer provided the pulser/receiver settings are not
changed. If the pulser/receiver settings do have to be changed, the reference
scattering setup used here is simple enough that it could be easily implemented in
the same immersion tank used for the actual experiment and the procedure outlined
previously followed to obtain the appropriate β(ω) at the chosen system settings.
620 14 Quantitative Ultrasonic NDE with Models
0.060
predicted
experiment
0.040
IVrl, V-s
xE-6
0.020
0.000
–1.20 –0.40 0.40 1.20
x, cm
Fig. 14.4 Comparison of the theoretical and measured cross-axis voltage magnitudes when the
scatterer was 6.9 cm from the transducer (using the effective parameters for probe A at 5 MHz).
Reprinted with permission from: T. Lerch, L.W. Schmerr, and A. Sedov, Characterization of
spherically focused transducers using an ultrasonic measurement model approach, Res. Nondestr.
Eval., 8, 1-21, (1996)
0.80
predicted
experiment
0.27
Volt., V
–0.27
–0.80
0.00 0.67 1.33 2.00
time, s ⫻ 10–6
Fig. 14.5 Comparison of the theoretical (solid line) and measured (dashed line) time domain
wave forms for probe A where an on-axis sphere is located 6.9 cm from the transducer. Wave
forms are shown offset for comparison purposes. Reprinted with permission from: T. Lerch,
L.W. Schmerr, and A. Sedov, Characterization of spherically focused transducers using an
ultrasonic measurement model approach, Res. Nondestr. Eval., 8, 1-21, (1996)
14.1 Transducer/System Characterization 621
0.3
Magnitude 0.2
0.1
0.0
0.0 8.3 16.7 25.0
Frequency
Fig. 14.6 Magnitude of the efficiency factor versus frequency calculated for a 10 MHz, 0.25 in.
diameter planar transducer using (1) the planar front surface of a steel specimen (solid line), (2) the
backscatter response of a 5/64 in. diameter flat-bottom hole in an ASTM 4340-5-0038 specimen
(long dashes), (3) the backscatter response of a 4/64 in. diameter steel cylinder in an immersion
setup (short dashes), and (4) the backscatter response of a 6/64 in. diameter steel cylinder in an
immersion setup (dash-dots). Reprinted with permission from: L.W. Schmerr, S.J. Song, and
H. Zhang, Model-based calibration of ultrasonic system responses for quantitative measurements,
(Nondestructive Characterization of Materials, VI, R.E. Green Jr., K.J. Kozaczek, and C.O. Ruud,
Eds., Plenum Press, N.Y., 111-118, 1994)
Table 14.2 Effective radii of two commercial planar immersion transducers obtained with four
different scatterers/receivers
Transducer Scatterer/receiver zmin (cm) aeff (cm)
A: 5 MHz 0.25 in. sphere 6.63 0.648
0.5 in. (1.27 cm) diameter 0.125 in. sphere 6.69 0.651
0.088 in. cylinder 7.65 0.696
1 mm hydrophone 6.68 0.651
B: 2.25 MHz 0.25 in. sphere 3.21 0.645
0.5 in. (1.27 cm) diameter 0.125 in. sphere 3.13 0.630
0.088 in. cylinder 3.27 0.644
1 mm hydrophone 2.67 0.583
From a theoretical standpoint, there is also another reason to prefer the sphere
as a reference scatterer. Consider, for example, the determination of the effective
radius of a planar probe. While the Thompson-Gray model predicts that the null
location of the measured voltage response and the pressure are the same, there
may be modeling errors in this determination since the Thompson-Gray model
relies on both the paraxial approximation and the approximation of a slowly
varying scattering amplitude for the scatterer, as discussed in Chap. 13, and
these assumptions break down in the near field where znull is being located.
However, the near field models of Chap. 13 could be used to simulate more
accurately the expected on-axis pulse-echo pressure response received from an
on-axis scatterer, and the znull found from those near field models used to test the
validity of the Thompson-Gray model-based procedures outlined previously for
determining the effective radius. When such simulations are conducted for a
5 MHz, 0.5 in. diameter piston transducer (a model probe similar to probe A in
Table 14.2) reflecting from different diameter rigid spheres and cylinders, the results
obtained for the extracted effective radius versus sphere/cylinder diameter are as
shown in Fig. 14.7. As can be seen from that figure, the near field model predicted
that for a spherical scatterer there was less than a 2.5 % variation in the estimated
effective radius over a wide range of sphere sizes while the flat end of the rigid
cylinder exhibited a very strong size effect, with the results approaching that of the
sphere only for very small diameter cylinders. These results are understandable
since the flat end of the cylinder is a highly directional specular reflector while the
specular front surface reflection from the sphere acts more like that for an omnidi-
rectional point scatterer even for relatively large diameter spheres. These results are
also consistent with what was observed for probe A in Table 14.2, where the
estimated effective radius obtained for that transducer are also plotted (as discrete
points) in Fig. 14.7 for the cylinder and two different size spheres used in those tests.
Note also that both for very small spheres and cylinders the near field model
predicted the estimated effective transducer diameter approached a common value
of approximately 0.614 cm, which is 3.5 % lower than the actual radius of 0.635 cm
assumed in the model. This small bias, however, is likely negligible since it would
typically be less than the experimental error.
14.2 Flat-Bottom Hole Models and DGS Diagrams 623
1.0
cylinder
0.8
0.6
aeff sphere
(cm)
0.4
0.2
Fig. 14.7 The effective radius found for a 0.5 in. (1.27 cm) diameter planar transducer using the
location of the last on-axis null predicted by the near field theories of Chap. 13, where the on-axis
scatterer was taken to be either a rigid cylinder or a rigid sphere. Experimental results shown: for a
0.088 in. diameter cylinder (square); and for 0.25 and 0.125 in. diameter spheres (circles),
(transducer A in Table 14.2)
z1 z2
ð ð (
1 cos θ4P
pave ðωÞ ¼ ρ1 cp1 v0 AP;P ðei ; es Þ liP
2πikp2 SR cos θ3P
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
Cri Crs
P;P P
T 21 θ3 exp i kp1 D4P þ kp2 D3P D4P þ cp2 D3P =cp1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D4P þ cp2 cos 2 θ4P D3P =cp1 cos 2 θ3P
9
P qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
>
T P;P
12 θ 1 exp i k D
p1 1
P
þ k p2 2D P
D P
1 þ c D
p2 2
P
=c p1 =
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dϕi dϕs ;
>
;
D1P þ cp2 cos 2 θ1P D2P =cp1 cos 2 θ2P
ð14:24Þ
where the extended edge integrals are Cri, Crs (and corresponding angular
integration variables ϕi, ϕs) during the transmission and reception processes,
respectively. These two extended edge integrals are actually four separate terms
(see Chap. 8) since they include both edge waves arising from integrations around
the actual edge of the transducer, Ci, Cs, and direct wave contributions which
appear here in the form of limits, as ε ! 0, of integrations around small circular
contours of radius ε, Cεi, Cεs, during the transmission and reception processes,
respectively, i.e.
8 9
ð ð <ð ð ð ð ð ð ð ð >
> =
¼ lim þ :
ε!0 >
: >
;
Cri Crs Ci Cs Cs Cεi Ci Cεs Cεi Cεs
Fortunately, because of the symmetry of this configuration, the notation for all the
edge wave contributions contained in Eq. (14.24) can be simplified considerably
(Fig. 14.9) by letting
D4P ¼ D1P ¼ D1
D3P ¼ D2P ¼ D2
ð14:25Þ
θ4P ¼ θ1P ¼ θ1
θ3P ¼ θ2P ¼ θ2 :
14.2 Flat-Bottom Hole Models and DGS Diagrams 625
z1 z2
This symmetry can also be exploited to perform most of the integrations appearing
in Eq. (14.24) so that the four edge and direct wave contributions reduce to
(
2π ∘ P;P ∘
pave ðωÞ ¼ ρ1 cp1 v0 exp 2i kp1 z1 þ kp2 z2 T P;P 12 ð0 ÞT 21 ð0 Þ
ikp2 Sr
AP;P eiP ð0∘ Þ; esP ð0∘ Þ liP ð0∘ Þ T P;P ∘ P;P
12 ð0 ÞT 21 ðθ 2 Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D1 þ cp2 D2 =cp1
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
cos θ1
exp i kp1 z1 þ kp2 z2 þ kp1 D1 þ kp2 D2 AP;P eiP ð0∘ Þ; esP ðθ2 Þ
cos θ2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P P;P ∘ D1 þ cp2 D2 =cp1
li ðθ2 Þ T 12 ðθ1 ÞT 21 ð0 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P;P
) Cs
P
li ðθ2 ; ϕs Þ dϕs :
ð14:26Þ
To completely evaluate Eq. (14.26) the scattering amplitude terms for the flat-
bottom hole must also be obtained. In the Kirchhoff approximation, the scattering
of the flat, stress-free end of the hole is identical to that of a circular crack. In
Chap. 10, the components of the far field scattering amplitude for a circular crack of
radius b were obtained as
e P esP
i
q ¼ P
eP;P ; ð14:29Þ
ei esP
where DPp is given in terms of (1) the reflection coefficients for a stress-free planar
surface, (2) the polarization unit vector, dPi , of an incident plane P-wave and, (3) the
polarization unit vectors, dPr and dSVr , of the reflected P- and SV-waves, respec-
tively, that such an incident P-wave produces at a planar free surface.
In spherical coordinates, with the z (u3) axis taken normal to the flat end of the
hole pointing away from the interface (Fig. 14.10), the incident and scattered unit
vectors, ePi , ePs , are
Note that in writing Eq. (14.26), for economy of notation the ϕi or ϕs arguments of
these unit vectors were omitted whenever the scattering amplitude term
was independent of these arguments, e.g. AP;P eiP ðθ2 Þ; esP ð0∘ Þ AP;P
P
ei ðθ2 ; Þ; esP ð0∘ ; Þ etc. We will also use this abbreviated notation for quantities
such as lPi and DPp when appropriate. Then the scattering amplitude terms appearing
in Eq. (14.26) can be written as
ð14:32a dÞ
Using the expression for the elastic constants tensor, Ckplj, this coefficient can also
be written as
h
i
C ¼ ρ2 c2p2 DkP nk þ 2ρ2 c2s2 eslP nl esjP DjP DkP nk : ð14:34Þ
Placing the appropriate values for ePi , ePs into Eq. (14.34) then gives the four
coefficients appearing in Eq. (13.32a–d) as
C1 eslP ð0∘ ÞesjP ð0∘ ÞCkplj DpP ð0∘ Þnk ¼ 2ρ2 c2p2
C2 eslP ð0∘ ÞesjP ð0∘ ÞCkplj DpP ðθ2 Þnk ¼ 2ρ2 c2p2 cos θ2 f 1 ðθ2 Þ
h i
C3 eslP ðθ2 ÞesjP ðθ2 ÞCkplj DpP ð0∘ Þnk ¼ 2ρ2 c2p2 1 2c2s2 sin 2 θ2 =c2p2 ð14:35Þ
C4 eslP ðθ2 ; ϕs ÞesjP ðθ2 ; ϕs ÞCkplj DpP ðθ2 Þnk ¼ 2ρ2 c2p2 cos θ2 ½f 1 ðθ2 Þ
i
2c2s2 sin 2 θ2 f 1 ðθ2 Þ=c2p2 2c2s2 sin 2 θ2 f 2 ðθ2 Þ cos ϕs =c2p2 ;
where
1
f 1 ðθ 2 Þ ¼ 1 RP;P ðθ2 Þ cs2 sin θ2 RSV;P ðθ2 Þ=cp2 cos θ2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð14:36Þ
1h i
f 2 ðθ2 Þ ¼ 1 þ RP;P ðθ2 Þ 1 c2s2 sin 2 θ2 =c2p2 RSV;P ðθ2 Þ= sin θ2 :
2
628 14 Quantitative Ultrasonic NDE with Models
ikp2 b2
AP;P eiP ð0∘ Þ; esP ð0∘ Þ liP ð0∘ Þ ¼
2
P ∘
P ∘
ikp2 b2 cos θ2 f 1 ðθ2 Þ 2J 1 kp2 b sin θ2
A P;P P
ei ðθ2 Þ; es ð0 Þ li ð0 Þ ¼
2 kp2 b sin θ2
ikp2 b2 1 2c2s2 sin 2 θ2 =c2p2 2J 1 kp2 b sin θ2
AP;P eiP ð0∘ Þ; esP ðθ2 Þ liP ðθ2 Þ ¼
2 kp2 b sin θ2
2ðπ
1 ikp2 b2 cos θ2
AP;P eiP ðθ2 ; 0Þ; esP ðθ2 ; ϕs Þ liP ðθ2 ; ϕs Þ dϕs ¼ ½f 1 ðθ2 Þ
2π 2
0
1 2ðπ J k b sin θ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ð1 cos ϕs Þ
2c2 sin 2 θ2 f 2 ðθ2 Þ
1 p2 2
1 2c2s2 sin 2
θ2 =c2p2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dϕs s2
π kp2 b sin θ2 2ð1 cos ϕs Þ c2p2
0
2ðπ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3
J k
1 1 p2 b sin θ 2 ð 1 cos ϕ Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffis cos ϕs dϕs 5:
2
π kp2 b sin θ2 2ð1 cos ϕs Þ
0
ð14:37a dÞ
To evaluate the two integrals appearing in Eq. (14.37d), we first write them
symbolically as the integrals In where
2ðπ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 J 1 kp2 b sin θ2 2ð1 cos ϕs Þ
In ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi cos nϕs dϕs ðn ¼ 0, 1Þ: ð14:38Þ
π kp2 b sin θ2 2ð1 cos ϕs Þ
0
We then use the integral relations of Eqs. (13.29) and (13.30) and a change of
variables to transform them as
2ðπ ð
1
b
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In ¼ 2 J 0 kp2 sin θ2 u 2ð1 cos ϕs Þ cos nϕs ududϕs
πb
0 0
2 2π 3
ðb ð
1
¼ 4 J 0 2kp2 sin θ2 u sin ðϕs =2Þ cos nϕs dϕs 5udu ð14:39Þ
πb2
0 0
2π 3
ðb ð
2
¼ 4 J 0 2kp2 sin θ2 u sin β cos ð2nβÞdβ5udu:
πb2
0 0
14.2 Flat-Bottom Hole Models and DGS Diagrams 629
ðπ
J 0 2kp2 sin θ2 u sin β cos ð2nβÞdβ ¼ πJ 2n kp2 sin θ2 u : ð14:40Þ
0
ðz
z2 2
J 2o ðtÞ t dt ¼ J 0 ðzÞ þ J 21 ðzÞ ð14:41Þ
2
0
Alternatively, we could also have used Eqs. (13.31), (13.32) and (13.38d) with
θi ¼ θr ¼ θ2 to obtain this same result.
Now, for the case n ¼ 1, I1 is given by
ðb
2
I 1 ¼ 2 J 21 kp2 sin θ2 u udu; ð14:43Þ
b
0
which is also an integral that can be expressed in closed form since (Jeffrey
[9, p. 278])
ðz
z2 h i
J 21 ðtÞ t dt ¼ fdJ 1 ðzÞ=dzg2 þ 1 1=z2 fJ 1 ðzÞg2 : ð14:44Þ
2
0
Combining Eq. (14.44) with the recurrence relation for Bessel functions
J 1 kp2 sin θ2 u þ u dJ 1 kp2 sin θ2 u =du ¼ kp2 sin θ2 u J 0 kp2 sin θ2 u : ð14:45Þ
Now, placing these values for I0, I1 into Eq. (14.37d), we obtain
630 14 Quantitative Ultrasonic NDE with Models
2ðπ
1 ikp2 b2 cos θ2
AP;P eiP ðθ2 ; 0Þ; esP ðθ2 ; ϕs Þ liP ðθ2 ; ϕs Þ dϕs ¼ ½ f 1 ðθ 2 Þ
2π 2
0
!
2c2s2 sin 2 θ2 2
2c2 sin 2 θ2 f 2 ðθ2 Þ
1 2
J 0 k2p b sin θ2 þ J 21 k2p b sin θ2 s2
cp2 c2p2
2J 1 kp2 b sin θ2
J 0 k2p b sin θ2
2
J 0 kp2 b sin θ2 þ J 21 k2p b sin θ2 :
kp2 b sin θ2
ð14:47Þ
Using Eqs. (14.37a–c) and (14.47) in Eq. (14.26) then gives an explicit model for
the scattering from the flat-bottom hole as
2
b ∘ P;P ∘
pave ðωÞ ¼ ρ1 cp1 v0 2
exp 2i kp1 z1 þ kp2 z2 T P;P 12 ð0 ÞT 21 ð0 Þ
a
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ1 P;P ∘ P;P D1 þ cp2 D2 =cp1
T 12 ð0 ÞT 21 ðθ2 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ2 D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
2J k b sin θ
1 p2 2
exp i kp1 z1 þ kp2 z2 þ kp1 D1 þ kp2 D2 1 2c2s2 sin 2 θ2 =c2p2
kp2 b sin θ2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P;P ∘ D1 þ cp2 D2 =cp1
T P;P
12 ð θ ÞT
1 21 ð 0 Þ p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
2J 1 kp2 b sin θ2
exp i kp1 z1 þ kp2 z2 þ kp1 D1 þ kp2 D2 cos θ2 f 1 ðθ2 Þ
kp2 b sin θ2
cos θ1 P;P D1 þ cp2 D2 =cp1
þ T ðθ1 ÞT P;P21 ðθ 2 Þ
cos θ2 12 D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
" !
2c2s2 sin 2 θ2
exp 2i kp1 D1 þ kp2 D2 f 1 ðθ2 Þ 1
c2p2
2c2 sin 2 θ2 f 2 ðθ2 Þ
J 20 kp2 b sin θ2 þ J 21 kp2 b sin θ2 s2
c2p2
2J 1 kp2 b sin θ2
J 0 kp2 b sin θ2
2
J 0 kp2 b sin θ2 þ J 1 kp2 b sin θ2
2
:
kp2 b sin θ2
ð14:48Þ
The functions f1(θ2) and f2(θ2) appearing in Eq. (14.48) were given in Eq. (14.36) in
terms of the velocity-based reflection coefficients vRP;P and vRSV;P obtained previ-
ously in Chap. 6 for a planar stress-free interface. Those coefficients, in terms of the
present notation, (see Eq. 6.177b) are
14.2 Flat-Bottom Hole Models and DGS Diagrams 631
sin 2θ2 sin 2θs2 c2p2 =c2s2 cos 2 2θs2
R ¼
v P;P
sin 2θ2 sin 2θs2 þ c2p2 =c2s2 cos 2 2θs2
ð14:49Þ
2 cp2 =cs2 sin 2θ2 cos 2θs2
v SV;P
R ¼
;
sin 2θ2 sin 2θs2 þ c2p2 =c2s2 cos 2 2θs2
with
The first complete flat-bottom hole scattering model in the form of Eq. (14.48) was
obtained by Sedov et al. [10] for the case where the shear strength of the solid was
neglected (fluid-fluid model). We can recover the fluid-fluid case from the present
results by letting cs2 =cp2 ! 0. In that limit f 1 ðθ2 Þ ! 1, f 2 ðθ2 Þ ! 0 and so we find
2
b P;P ∘ P;P ∘
pave ðωÞ ¼ ρ1 cp1 v0 exp 2i kp1 z1 þ kp2 z2 T 12 ð0 ÞT 21 ð0 Þ
a2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ1 P;P ∘ P;P D1 þ cp2 D2 =cp1
T ð0 ÞT 21 ðθ2 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cos θ2 12 D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
2J 1 kp2 b sin θ2
exp i kp1 z1 þ kp2 z2 þ kp1 D1 þ kp2 D2
kp2 b sin θ2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P;P ∘ D1 þ cp2 D2 =cp1 ð14:51Þ
P;P
T 12 ðθ1 ÞT 21 ð0 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
D1 þ cp2 cos θ1 D2 =cp1 cos θ2
2 2
2J 1 kp2 b sin θ2
exp i kp1 z1 þ kp2 z2 þ kp1 D1 þ kp2 D2 cos θ2
kp2 b sin θ2
cos θ1 P;P D1 þ cp2 D2 =cp1
þ T ðθ1 ÞT 21 ðθ2 Þ
P;P
cos θ2 12 D1 þ cp2 cos 2 θ1 D2 =cp1 cos 2 θ2
exp 2i kp1 D1 þ kp2 D2 J 20 kp2 b sin θ2 þ J 21 kp2 b sin θ2 :
Actually, the fluid-fluid model obtained in [10] was for the average velocity
(normal to the transducer face) received, (vn)ave. That model differs from this one
only in the appearance of additional cos θ1 factors in the returning edge wave terms
and the replacement ρ1 cp1 v0 ! v0 for the leading coefficient term in Eq. (14.51)
(see the discussion in Chap. 13—the minus sign shows up here since on reception
the received waves propagate in a direction with a component opposite to that of the
transducer normal). As discussed in Chap. 13, such model differences are not
632 14 Quantitative Ultrasonic NDE with Models
significant except in a region near the transducer face, which often lies in the very
near field.
In many cases the terms that are neglected in going from the fluid-solid model to
the fluid-fluid model are also negligible since they are all of O(c2s2 sin2θ2/c2p2 ) or
smaller. Thus, it is not surprising that the fluid-fluid model has given results that
agree with experiments with solid ASTM specimens [11].
Since Eq. (14.48) is an explicit model for the scattering of a flat-bottom hole, it
is easily evaluated for a wide set of transducer and/or hole parameters. However,
it is instructive to examine some of the limiting cases of this general equation.
Case (a): z1
a or z2
a
In both of these cases θ1 , θ2 ffi 0 so that if we neglect these angles in all the
amplitude terms in Eq. (14.48) our near field model reduces to that of the Thomp-
son-Gray model, i.e.
2 n
b ∘ P;P ∘
pave ðωÞ ¼ ρ1 cp1 v02
exp 2i kp1 z1 þ kp2 z2 T P;P
12 ð0 ÞT 21 ð0 Þ
a
ð14:52Þ
2 o
1 exp ikp1 ðD1 z1 Þ þ ikp2 ðD2 z2 Þ :
which gives
∘ P;P ∘ πa
2
πb2
pave ðωÞ ¼ ρ1 cp1 v0 exp 2i kp1 z1 þ kp2 z2 T P;P
12 ð0 ÞT 21 ð0 Þ ; ð14:55Þ
λp1 z1 λp1 z1
where λpm ðm ¼ 1, 2Þ are the wavelengths for P-waves in the respective media.
14.2 Flat-Bottom Hole Models and DGS Diagrams 633
Case (c): z2
a, kp2 a2 =2z2 1
Similarly, for the case where z2
a while z1/a is fixed, r I ffi a. If also kp2 a2 =2z2
1 the phase terms in Eq. (14.52) again simplify to
1 exp ikp1 ðD1 z1 Þ þ ikp2 ðD2 z2 Þ ffi ikp2 a2 =2z2 ð14:56Þ
and we have
Both Eqs. (14.55) and (14.57) can be viewed as generalizations of the results found
by Krautkramer [3] for the amplitude response of a transducer radiating into a
single fluid medium when the flat-bottom hole is in the far field spherical wave
spreading region of the transducer wave field. These generalized results, however,
also includes both the transmission coefficients and propagation phase terms that
properly account for the effects of transmission twice through the interface.
Krautkramer [2, 3] first formulated and experimentally determined DGS curves for
on-axis rigid circular reflectors in water. Krautkramer showed that in the far field
spherical wave region of the transducer the amplitude of the average received
pressure should be given in a form such as
Sb St
pave ¼ p0 ; ð14:58aÞ
λz λz
where Sb and St are the areas of the rigid reflector and the transducer, respectively,
and p0 ¼ ρcv0 is a reference pressure. When the reflector is very close to the
transducer, Krautkramer argued that the amplitude of the average pressure received
should also be
Sb
pave ¼ p0 : ð14:58bÞ
St
Since the form of the response between these two limits was unknown, Krautkramer
proposed obtaining the response experimentally in this intermediate region in the
form of DGS (distance-gain-size) diagrams, where the voltage received (“gain”) is
plotted versus z (“distance”) for a series of different diameter (“size”) reflectors.
Equation (14.48) for the fluid-solid model or Eq. (14.51) for the fluid-fluid model
are generalizations of Krautkramer’s results and applicable to the interrogation of a
standard ASTM flat-bottom hole specimen. These flat-bottom hole models, how-
ever, are valid into the near field so that they can be used to generate complete
634 14 Quantitative Ultrasonic NDE with Models
101
100
Amplitude, <vz >/v0
10–1
10–2
10–3
10–4
10–2 10–1 100 101
Water Depth, h1 /N
Fig. 14.11 Single frequency DGS diagram (at 5 MHz) for flat-bottom holes in a steel specimen
immersed in water; metal distance (h2): 0.38 in.; transducer: 5 MHz, 0.5 in. diameter; solid line:
front surface reflection; dashed lines: for b/a ¼ 1.0, 0.8, 0.5, 0.3, 0.2, 0.1, respectively, from
the top. Reprinted with permission from: A. Sedov, L.W. Schmerr, and S.J. Song, Ultrasonic
scattering by a flat-bottom hole in immersion testing: an analytical model, J. Acoust. Soc. Am.,
92, 478-486, (1992)
theoretical DGS diagrams. This allows one to obtain such calibration curves
without resorting to tedious experiments. Figure 14.11 shows a theoretical set of
DGS curves obtained by Song et al. [11], using the fluid-fluid model, and Fig. 14.12
shows that this model does agree well with experiments (note that in Figs. 14.11,
14.12, 14.13, and 14.14 the distances h1 and h2 correspond to our z1 and z2,
respectively). As pointed out by Song et al. [11], the explicit near field limit used
by Krautkramer in Eq. (14.58b) is not strictly speaking valid for a transmitting
transducer modeled as a constant velocity (piston) source when the mechanical
quantity being measured on reception is the average pressure, but the form is valid
in this case when one assumes the appropriate received mechanical response is
instead the average normal velocity. As pointed out in Chap. 13, however, in the
very near field one cannot claim validity of any of these near field model formu-
lations without further careful experimental validation.
When DGS curves are obtained experimentally, the response plotted is typically
a peak-to-peak voltage response of a received A-scan waveform, whereas the
results shown in Fig. 14.11 are for the average received pressure response at a
single frequency. Therefore, the theoretical diagrams of Fig. 14.11 might be called
single frequency DGS diagrams. However, it is possible to synthesize A-scan
voltage versus time waveforms from these models by evaluating the response at
different frequencies, multiplying by the efficiency factor (assuming it has been
obtained experimentally), and inverting the result into the time domain through the
use of a fast Fourier transform. In this manner, one can produce “true” DGS
14.2 Flat-Bottom Hole Models and DGS Diagrams 635
101
100
Amplitude, <vz >/v0
10–1
10–2
10–3
10–4 –2
10 10–1 100 101
Water Depth, h1 /N
Fig. 14.12 Single frequency DGS diagram (at 5 MHz) for a flat-bottom hole in an ASTM 4340-5-
0038 reference block immersed in water (dashed line) compared to experiment (circles); metal
distance (h2): 0.38 in.; transducer: 5 MHz, 0.5 in. diameter; b/a ¼ 0.156. The solid line is the
theoretical response from the front surface and corresponding experimental results (squares).
Reprinted with permission from: A. Sedov, L.W. Schmerr, and S.J. Song, Ultrasonic scattering
by a flat-bottom hole in immersion testing: an analytical model, J. Acoust. Soc. Am., 92, 478-486,
(1992)
101
100
Amplitude, V
10–1
10–2
10–3
10–4 –2
10 10–1 100 101
Water Depth, h1 /N
Fig. 14.13 Time domain DGS diagram for flat-bottom holes in a steel specimen immersed in
water; metal distance (h2): 0.38 in.; transducer: 5 MHz, 0.5 in. diameter; solid line: front surface
reflection, dashed lines: hole responses for b/a ¼ 1.0, 0.8, 0.5, 0.3, 0.2, 0.1, respectively, from the
top. Reprinted with permission from: S.J. Song, L.W. Schmerr, and A. Sedov, DGS diagrams and
frequency response curves for a flat-bottom hole; a model-based approach, Res. Nondestr. Eval.,
3, 201–219, (1991)
636 14 Quantitative Ultrasonic NDE with Models
101
100
Amplitude, V
10–1
10–2
10–3
10–4 –2
10 10–1 100 101
Water Depth, h1 /N
Fig. 14.14 Time domain DGS diagram for a flat-bottom hole in an ASTM 4340-5-0038 reference
block immersed in water (dashed line) compared to experiment (circles); metal distance (h2):
0.38 in.; transducer: 5 MHz, 0.5 in. diameter; b/a ¼ 0.156. The solid line is the theoretical
response from the front surface and corresponding experimental results (squares). (Reprinted
with permission from: Song, S.J., Schmerr, L.W., and A. Sedov, “DGS diagrams and frequency
response curves for a flat-bottom hole; a model-based approach,” Res. Nondestr. Eval.,
3, 201–219, 1991)
pave ¼ p0 : ð14:59Þ
This same reflector, when placed in the spherical wave far field region of the
transducer should, Krautkramer argued, produce a response given by
St
pave ¼ p0 : ð14:60Þ
2λ z
As shown in Chap. 9, the diffraction correction term obtained there (Eq. 9.69) is a
direct generalization of Eqs. (14.59) and (14.60) for a planar interface located
anywhere in the transducer wavefield. Thus, the procedures outlined in Chap. 9 to
obtain the efficiency factor using such a reflector are very close in spirit to the
calibration methods suggested by Krautkramer to obtain the reference pressure, p0.
14.3 Deconvolution and the Determination of Far Field Scattering Amplitudes 637
0.06 360
0.05 270
Phase (degrees)
0.04
|A(f)| (cm)
180
0.03
90
0.02
0
0.01
0.00 –90
0 5.0 10.0 15.0 20.0 0 5.0 10.0 15.0 20.0
Frequency (MHz) Frequency (MHz)
Fig. 14.15 Comparison of the theoretical (solid line) and experimental (dashed) magnitude and
phase of the L ! L backscattered scattering amplitude for a 200 μm 400 μm oblate spheroidal
cavity in a Ti-6Al-4 V disk at normal incidence to the sample surface (along the minor axis of the
void). Reprinted with permission from: R.B. Thompson, and T.A. Gray, A model relating
ultrasonic scattering measurement through liquid-solid interfaces to unbounded medium scattering
amplitudes, J. Acoust. Soc. Am., 74, 1279-1290, (1983)
0.010 360
270
Phase (degrees)
0.008
|A(f)| (cm)
0.006 180
0.004 90
0.002 0
0.000 –90
0 5.0 10.0 15.0 20.0 0 5.0 10.0 15.0 20.0
Frequency (MHz) Frequency (MHz)
Fig. 14.16 Comparison of theoretical (solid line) and experimental magnitude and phase of the
L ! L backscattered scattering amplitude for a 114 μm radius spherical tin-lead solder sphere in
thermoplastic disk. Experimental data are for normal incidence (dashed line) and for a 15.7
(dotted line) incident angle (30 in the solid). Reprinted with permission from: R.B. Thompson,
and T.A. Gray, Erratum: A model relating ultrasonic scattering measurement through liquid-solid
interfaces to unbounded medium scattering amplitudes, J. Acoust. Soc. Am., 75, 1645-1646,
(1984)
spheroidal void in a titanium disk and Fig. 14.16 shows similar backscatter P-wave
results for a 114 μm radius spherical tin-lead solder sphere in a thermoplastic disk.
For the void, the theoretical results were calculated numerically using MOOT (the
method of optimal truncation) [14]. For the spherical inclusion, the theoretical
scattering amplitudes were obtained using a separation of variables solution (see
Sect. 10.6). Thompson and Gray also gave a variety of pitch-catch P-wave results
for the tin-lead solder sphere at different angular separations of the sender and
receiver (Figs. 14.17 and 14.18).
14.3 Deconvolution and the Determination of Far Field Scattering Amplitudes 639
0.010 360
0.008 270
Phase (degrees)
|A(f)| (cm)
0.006 180
0.004 90
0.002 0
0.000 –90
0 5.0 10.0 15.0 20.0 0 5.0 10.0 15.0 20.0
Frequency (MHz) Frequency (MHz)
Fig. 14.17 Comparison of theoretical (solid line) and experimental (dashed) magnitude and phase
of the L ! L pitch-catch scattering amplitude for a 114 μm radius tin-lead solder sphere in a
thermoplastic disk. Illumination was at normal incidence and reception at an 8 angle (15 in the
solid). Reprinted with permission from: R.B. Thompson, and T.A. Gray, Erratum: A model
relating ultrasonic scattering measurement through liquid-solid interfaces to unbounded medium
scattering amplitudes, J. Acoust. Soc. Am., 75, 1645-1646, (1984)
0.010 360
0.008 270
Phase (degrees)
|A(f)| (cm)
0.006 180
0.004 90
0.002 0
0.000 –90
0 5.0 10.0 15.0 20.0 0 5.0 10.0 15.0 20.0
Frequency (MHz) Frequency (MHz)
Fig. 14.18 Comparison of theoretical (solid line) and experimental (dashed) magnitude and phase
of the L ! L pitch-catch scattering amplitude for a 114 μm radius tin-lead solder sphere in a
thermoplastic disk. Illumination was at normal incidence and reception at a 27.7 angle (60 in the
solid). (Reprinted with permission from: Thompson, R.B., and T.A. Gray, Erratum: “A model
relating ultrasonic scattering measurement through liquid-solid interfaces to unbounded medium
scattering amplitudes, [J. Acoust. Soc. Am., 74, pp. 1279-1290, 1983]” J. Acoust. Soc. Am.,
75, pp. 1645-1646, 1984, Copyright 1984, Acoustical Society of America)
The close agreement found between theory and experiment in these cases dem-
onstrates the ability of the Thompson-Gray models to serve as the basis for obtaining
high quality, quantitative flaw information directly from NDE measurements. As
Thompson and Gray point out, such models are also useful for solving inverse
problems (obtaining flaw properties), correcting for systematic errors in ultrasonic
measurements, and in computing the probability of detecting flaws in a noisy
environment [12]. As shown in the next section, these Thompson-Gray models
(or more general Auld measurement models) can also be linked to computer-aided
design (CAD) information in a sophisticated ultrasonic simulation environment.
640 14 Quantitative Ultrasonic NDE with Models
One application of LTI models that has far reaching consequences is their incor-
poration into a simulation environment that allows the consideration of
inspectability issues at the design stage of components, even before parts are
available for NDE tests. This capability is important since in the past NDE
considerations have rarely been addressed in the design and manufacturing pro-
cesses on an equal par with other design parameters. In some cases, this situation
has led to the production of difficult-to-inspect parts and limited NDE techniques to
much less than their full potential. Over the last decade this situation has changed
with the appearance of NDE simulators that can model inspections in particular
geometries or are coupled directly to a CAD environment. Here we will briefly
touch on some of the models known to the author.
14.4.1 UTSIM
UTSIM [15, 16] is an ultrasonic simulation package developed at the Center for
NDE, Iowa State University. UTSIM initially incorporated a paraxial beam model
(based on Gauss-Hermite expansions) in conjunction with approximate scattering
models such so that it could perform very rapid simulations. These models were tied
to an interactive CAD interface. Later versions of UTSIM have replaced the Gauss-
Hermite beam model with a Multi-Gaussian beam model since the latter is even
faster and more versatile than the Gauss-Hermite model. The multi-Gaussian beam
model has allowed UTSIM to simulate many different transducer types, both
focused and unfocused, and to consider inspection simulations through curved
interfaces. Material attenuation and grain scattering effects have also been incor-
porated into the propagation models.
In UTSIM, the flaw scattering models used in most cases are based on the
Kirchhoff approximation. This scattering model is used for volumetric flaws,
side-drilled holes, flat-bottom holes, and flat cracks. A separation of variables
flaw scattering model is also available for spherical voids.
In UTSIM the electrical and electromechanical components of the ultrasonic
measurement system are lumped into a system efficiency factor term. This term can
be put into UTSIM parametrically by the user (as a spectrum in the form of a cosine-
squared function) or as an experimentally determined factor obtained in a well-
characterized reference experiment. This latter capability allows UTSIM to predict
in absolute terms the signals seen in ultrasonic tests.
Unlike most other simulators UTSIM implements the Thompson-Gray measure-
ment model. This model is less general than Auld’s reciprocity relationship, but
results in a modular form where the beam model, flaw scattering model, and
efficiency factor terms are all separated, allowing one to efficiently perform various
parametric studies. The CAD interface of UTSIM allows the user to input geom-
etries in standard CAD and geometry formats.
14.4 Model-Based Ultrasonic Simulation 641
14.4.2 GPSS
14.4.3 GB
components are included in UTDefect, but UTDefect does allow the user to input a
transducer spectrum parametrically in terms of center frequency and bandwidth
(using a cosine-squared spectrum shape) and simSUNDT allows one to supply the
data from an experimentally measured transducer spectrum.
Because the geometry that UTDefect, SUNDT, and simSUNDT were designed
to model consists of two planar surfaces, there is no CAD interface present in those
packages, but simSUNDT does have a Windows based graphical interfaces that
allows one to enter parameters and display results in the form of A-, B-, and C-scans
and plot various echo-dynamics of the flaw signals.
Some extensive validation studies of simSUNDT have been carried out [29–32],
and simSUNDT has also been used in POD studies [33, 34].
14.4.5 EFIT
EFIT stands for Elastodynamic Finite Integration Technique. The method was
developed at the University of Kassel by Marklein, Langenberg, and others. It
uses a discretized form of the elastodynamic equations written in integral form so
that it is very general and has been developed for both inhomogeneous and
anisotropic materials [35–38]. With this generality, of course, comes a price in
terms of computational efficiency. However, EFIT has been used to simulate a
variety of ultrasonic setups. Note that the flaw model is not separate from the beam
model in EFIT since the flaw is merely part of the discretized geometry (with
appropriate boundary conditions). EFIT, like UTDefect, treats a transducer in terms
of an equivalent source acting over an area of the surface of the solid and on
reception uses an averaging of the waves over the same area.
EFIT has been coupled to a CAD-based ultrasonic imaging tool called ULIAS
(for Ultrasonic Inspection Applying Simulation). This coupling allows the visual-
ization of the ultrasonic fields within the CAD geometry in conjunction with the
synthesized A-scan responses.
14.4.6 CIVA
The French Atomic Energy Commission (CEA) has developed an ultrasonic soft-
ware modeling package, Champ-Sons, and has integrated that software with pre-
and post processing CAD and imaging tools to form an overall software simulation
platform called CIVA [39, 40]. CIVA is now a commercially available product with
extensive capabilities. Updates of those capabilities occur regularly, so we will just
outline here some of the basic ingredients.
The ultrasonic beam model in CIVA is based on an asymptotic Green’s function
type of model (pencil method) [41]. This model allows the simulation of a wide
range of transducers, including immersion transducers, angle beam probes, arrays,
644 14 Quantitative Ultrasonic NDE with Models
and focused as well as planar types. These transducers can transmit through or
reflect from planar or curved surfaces/interfaces in anisotropic media. Some simple
forms of continuous material inhomogeneity are also possible.
There are a number of flaw scattering models in CIVA. Models based on the
Kirchhoff approximation and the high frequency geometrical Theory of Diffraction
(GTD) as well as a modified form of the Born approximation have all been
incorporated. More exact separation of variables solutions and scattering models
that use the finite element method (FEM) have also been developed.
The reception of the scattered waves is handled through the use of Auld’s
reciprocity relationship. The user can input a transducer/system spectrum paramet-
rically or from experimental data. CIVA contains a computer-aided design (CAD)
interface that allows it to treat 2D and 3D geometries represented in standard CAD
library formats.
The methods described in Sect. 14.1 for characterizing planar and focused trans-
ducers are based on the approach described by Lerch et al. [7]. There are also a
number of other methods available for obtaining effective transducer parameters,
for both planar and focused probes, using a variety of model-based features (nulls or
maxima, etc.) expected in the on-axis and/or cross-axis wave field of a transducer
[4, 42–48]. In other cases, the effective parameters were obtained directly from a
non-linear least squares fitting of on- or off-axis data to an LTI model [49, 50]. The
methods described in Sect. 14.1, however, are a complete characterization of a
transducer and the system to which it is attached in the sense that they can be used to
quantitatively predict the actual waveforms expected in ultrasonic tests. To obtain
such a complete characterization, of course, requires that one obtain the system
function or efficiency factor as a function of frequency as well as determine the
effective transducer parameters.
The flat-bottom hole model of Sect. 14.2 is a generalization of the fluid-fluid
model described by Sedov et. al. [10] (see also Schmerr and Sedov [71]) which has
been successfully used to generate DGS diagrams and near field frequency response
curves of planar transducers [11]. This model has also led to the development of
more general near field models, of the type discussed in Chap. 13, for both on- and
off-axis scatterers [51, 52].
The use of a Thompson-Gray measurement model to obtain scattering ampli-
tudes, as outlined in Sect. 14.3, is but one example of how such models can serve as
the basis for quantitative NDE tests. Other uses of Thompson-Gray models include
qualifying the performance of NDE systems in terms of their probability of
detection of flaws [53, 54], characterizing the sources of variability seen in ultra-
sonic tests [55], the calculation of distance-amplitude correction (DAC) curves for
railway axle inspections [56], and the correction for surface curvature effects in
contact testing [57].
14.6 Problems 645
The more general Auld reciprocity model (see Eq. 12.62) has also been used for
a number of quantitative NDE applications. Some of those applications have been:
the evaluation of inter-granular stress corrosion cracking [58, 59], the inspection of
composites [60], the calibration of transducers and systems [61, 62], and the
modeling of Rayleigh wave scattering by distributions of surface-breaking
cracks [63].
Other models have also formed the basis for obtaining quantitative NDE infor-
mation. Temple, for example, describes some of the related modeling approaches
used in the United Kingdom [64]. Schmerr and Lerch [70] describe a focused
transducer model with a scatterer present. Lhemery [65] used a time domain
approach, called the impulse response method, as part of the original development
of the modeling package CIVA mentioned in Sect. 14.4 and Marklein describes
some complex NDE simulations using the EFIT modeling package [66].
Gray [67] was the first to couple Thompson-Gray models to a CAD environment
and demonstrate the ability to use models to quantify inspectability issues for
complex parts. Similar couplings of x-ray and eddy current measurement model
to CAD have been made and incorporated into a general life-cycle engineering
approach to design which includes NDE issues [68, 69]. This new technology is
now being extended into the generation of sophisticated CAD-based simulators of
the type described in Sect. 14.4.
14.6 Problems
14.1 In Sect. 14.1, it was shown that the location of the last on-axis minimum in the
response of highly specular reflector does not occur at the same location as the
last on-axis null in the transducer pressure response except for very small size
reflectors, leading to inaccurate estimates of the effective radius. Using the
flat-bottom hole scattering model of Sect. 14.2, determine a calibration cor-
rection curve (true effective radius versus apparent effective radius for various
size holes) that would allow one to measure the last on-axis null in the
response as seen with a standard ASTM flat-bottom hole calibration block
and, with the use of a calibration correction curve and experimental data,
determine the effective radius of a probe.
14.2 In the near field scattering model of Sect. 14.2, all the mode-converted
responses from the flat bottom hole were neglected. Using the same approach
as taken in that section, determine explicitly expressions for all the mode-
converted responses that were neglected in that section. Compare the magni-
tudes of these various neglected terms with the P-wave contributions retained
in Eq. (14.48).
14.3 As the size of an on-axis scatterer becomes comparable to the transducer size,
there are significant variations in incident field amplitude on the face of the
scatterer that are not accounted for in the near field theory used in Sect. 14.2.
However, a large scatterer could be simulated as an assemblage of small (and
646 14 Quantitative Ultrasonic NDE with Models
in general, off-axis) planar scattering elements for which the near field theory
of Chap. 13 would apply. If the scattering amplitudes for these elements are
again obtained within the Kirchhoff approximation, it would then be possible
to obtain a more exact model for such large scatterers. Use this concept to
generate DGS-like diagrams for the scattering from various size on-axis rigid
cylinders in water (single medium problem). Compare these DGS diagrams
with those obtained directly from the near field theory of Chap. 13. At what
size scatterer do the near field theories begin to break down?
14.4 Repeat the large flaw modeling simulation approach described in problem
14.3 for the case of an on-axis rigid sphere in water and compare DGS
diagrams obtained via that approach with those generated from the near
field theory of Chap. 13.
14.5 Repeat the large flaw modeling simulation approach described in problem
14.3 for the case of an on-axis flat-bottom hole in an ASTM block (fluid-solid
problem) and compare DGS diagrams obtained via that approach with those
generated from the near field theory of Chap. 13. As in Sect. 14.2, only
consider the P-wave contributions in the models.
14.6 Consider the case of an on-axis spherical void in a solid which is interrogated
by a planar immersion transducer in a configuration similar to that described
for the flat bottom hole model of Sect. 14.2. Using the near field theory of
Chap. 13, develop a model for this scattering configuration and plot DGS-like
diagrams for the sphere pulse-echo response versus distance at different void
sizes when using a 1/2 in. diameter, 5 MHz planar transducer. Use the leading
edge response of the void only when calculating the scattering amplitude
terms needed in the model.
14.7 In problem 14.6, replace the scattering amplitude expressions, as calculated
by the leading edge response, with calculations based on the exact separation
of variables solution for a void given in Chap. 10. Compare the DGS-like
diagrams obtained in both cases.
References
1. J.N. Gray, T.A. Gray, N. Nakagawa, R.B. Thompson, Models for predicting NDE reliability,
Metals Handbook, 9th edn, Vol. 17: Nondestructive Evaluation and Quality Control (ASM
International, Materials Park, 1989), pp. 702–715
2. J. Krautkramer, Determination of the size of defects by the ultrasonic impulse echo method.
Br. J. Appl. Phys. 10, 240–245 (1959)
3. J. Krautkramer, H. Krautkramer, Ultrasonic Testing of Materials, 4th edn. (Springer,
New York, 1990)
4. R.C. Chivers, L. Bosselaar, P.R. Filmore, Effective area to be used in diffraction correction.
J. Acoust. Soc. Am. 68, 80–84 (1980)
5. L.W. Schmerr, S.J. Song, H. Zhang, Model-based calibration of ultrasonic system responses
for quantitative measurements, in Nondestructive Characterization of Materials, vol. VI,
ed. by R.E. Green Jr., K.J. Kozaczek, C.O. Ruud (Plenum Press, New York, 1994),
pp. 111–118
References 647
6. L.W. Schmerr, S-J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and Mea-
surements (Springer, New York, 2007)
7. T. Lerch, L.W. Schmerr, A. Sedov, Characterization of spherically focused transducers using
an ultrasonic measurement model approach. Res. Nondestruct. Eval. 8, 1–21 (1996)
8. M. Abramowitz, I.A. Segun (eds.), Handbook of Mathematical Functions (Dover, New York,
1965)
9. A. Jeffrey, Handbook of Mathematical Formulas and Integrals (Academic, San Diego, 1995)
10. A. Sedov, L.W. Schmerr, S.J. Song, Ultrasonic scattering by a flat-bottom hole in immersion
testing: an analytical model. J. Acoust. Soc. Am. 92, 478–486 (1992)
11. S.J. Song, L.W. Schmerr, A. Sedov, DGS diagrams and frequency response curves for a flat-
bottom hole; a model-based approach. Res. Nondestruct. Eval. 3, 201–219 (1991)
12. R.B. Thompson, T.A. Gray, A model relating ultrasonic scattering measurements through
liquid-solid interfaces to unbounded medium scattering amplitudes. J. Acoust. Soc. Am. 74,
1279–1290 (1983)
13. R.B. Thompson, T.A. Gray, Erratum: A model relating ultrasonic scattering measurements
through liquid-solid interfaces to unbounded medium scattering amplitudes [J. Acoust. Soc.
Am., 74, 1279-1290, 1983], J. Acoust. Soc. Am. 75, 1645–1646 (1984)
14. J.L. Opsal, W.M Visscher, Theory of elastic wave scattering: applications of the method of
optimal truncation, J. Appl. Phys. 58, 1102–1115 (1985)
15. A. Turnbull, M. Garton, Ultrasound ray tracing in arbitrary complex geometries, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 14A, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1995), pp. 1105–1110
16. M. Garton, Refining automated ultrasonic inspections with simulation models, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 17B, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1998), pp. 1825–1829
17. M. Spies, Semi-analytical elastic wave field modeling applied to arbitrarily oriented
orthotropic media. J. Acoust. Soc. Am. 110, 68–79 (2001)
18. M. Spies, Efficient optimization of single and multiple element transducers for the inspection
of complex-shaped components. NDT&E Int. 37, 455–459 (2004)
19. M. Spies, Kirchhoff evaluation of scattered elastic wave fields in anisotropic media. J. Acoust.
Soc. Am. 107, 2755–2759 (2000)
20. M. Spies, Transducer field modeling in anisotropic media by superposition of Gaussian base
functions. J. Acoust. Soc. Am. 105, 633–638 (1999)
21. M. Spies, Modeling of transducer fields in inhomogeneous anisotropic materials using Gauss-
ian beam superposition. NDT&E Int. 33, 155–162 (2000)
22. M. Spies, Ultrasonic field modeling for immersed components using Gaussian beam superpo-
sition. Ultrasonics 46, 138–147 (2007)
23. M. Spies, W.D. Feist Application and validation of the Gaussian beam superposition technique
to simulate the inspection of aero engine components, in Review of Progress in Quantitative
NDE, vol. 26, ed. by D.O. Thomson, D.E. Chimenti (American Institute of Physics, Melville,
2007), pp. 886–893
24. H. Wirdelius, A. Blomquist, The application of mathematical modeling of ultrasonic NDT in
the qualification process, NDT.net, vol. 3, no. 11, pp. 1–6 (1998)
25. H. Wirdelius, The simSUNDT software and validation of the grain noise model, in Pro-
ceedings of the World Congress on Ultrasonics (2003), pp. 101–104
26. A., Bostrom, H. Wirdelius, Ultrasonic probe modeling and nondestructive crack detection,
J. Acoust. Soc. Am. 97, 2836–2848 (1995)
27. A. Bostrom, H. Wirdelius, Ultrasonic probe modeling and nondestructive crack detection.
J. Acoust. Soc. Am. 97, 2836–2848 (1995)
28. A. Bostrom, Review of hypersingular integral equation method for crack scattering and
application to modeling of ultrasonic nondestructive evaluation, Appl. Mech. Rev. 56,
383–405 (2003)
648 14 Quantitative Ultrasonic NDE with Models
50. T.A. Gray, Model-based characterization of planar and focused immersion ultrasonic trans-
ducers, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 14A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1995), pp. 1021–1028
51. A. Sedov, L.W. Schmerr, S.J. Song, Ultrasonic scattering models for standards: flat-bottom
holes and spherical reflectors, in Review of Progress in Quantitative Nondestructive Evalua-
tion, vol. 10A, ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1991),
pp. 59–66
52. L.W. Schmerr, A. Sedov, A near-field measurement model for ultrasonic reference standards,
in Review of Progress in Quantitative Nondestructive Evaluation, vol. 11A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1992), pp. 1191–1198
53. T.A. Gray, R.B. Thompson, B.P. Newberry, Model-based ultrasonic NDE system qualification
methodology, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 6A,
ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1987), pp. 93–100
54. T.A. Gray, F. Amin, R.B. Thompson, Application of ultrasonic POD models, in Review of
Progress in Quantitative Nondestructive Evaluation¸vol. 7B, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1988), pp. 1737–1744
55. D.D. Bennink, A.L. Pate, Investigation of scatter in ultrasonic responses caused by variability
in transducer and material properties, in Review of Progress in Quantitative Nondestructive
Evaluation, vol. 7A, ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1988),
pp. 621–628
56. D. Utrata, T.A. Gray, Use of ultrasonic modeling in the inspection of railway axles, in Review
of Progress in Quantitative Nondestructive Evaluation, vol. 10A, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1991), pp. 621–628
57. T.A. Gray, Ultrasonic measurement model for contact transducers on curved surfaces, in
Review of Progress in Quantitative Nondestructive Evaluation, vol. 13B, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1994), pp. 2191–2198
58. J.D. Achenbach, D.E. Budreck, 3-D modeling of ultrasonic scattering from intergranular stress
corrosion cracks, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 6A,
ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1987), pp. 87–92
59. T.A. Gray, R.B. Thompson, B.P. Newberry, Applications of models for IGSCC inspection, in
Review of Progress in Quantitative Nondestructive Evaluation, vol. 6A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1987), pp. 93–100
60. F.J. Margetan, T.A. Gray, R.B. Thompson, B. P. Newberry, A model for ultrasound transmis-
sion through graphite composite plates containing delaminations, in Review of Progress in
Quantitative Nondestructive Evaluation, vol. 7B, ed. by D.O. Thompson, D.E. Chimenti
(Plenum Press, New York, 1988), pp. 1083–1092
61. D.D. Bennink, A.L. Pate, Reciprocity-based measurement models for ultrasonic NDE, in
Review of Progress in Quantitative Nondestructive Evaluation, vol. 8A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1989), pp. 843–848
62. D.D. Bennink, A.L. Pate, Ultrasonic transducer calibration for reciprocity-based methods, in
Review of Progress in Quantitative Nondestructive Evaluation, vol. 8A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1989), pp. 849–856
63. A.S. Cheng, Ultrasonic reflection by a planar distribution of surface-breaking cracks, in
Review of Progress in Quantitative Nondestructive Evaluation, vol. 14A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1995), pp. 59–65
64. A. Temple, UK development in theoretical modeling for NDT, in Review of Progress in
Quantitative Nondestructive Evaluation, vol. 6A, ed. by D.O. Thompson, D.E. Chimenti
(Plenum Press, New York, 1987), pp. 21–35
65. A. Lhemery, Impulse-response method to predict echo-responses from defects in solids: a first
approach, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 14A, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1995), pp. 115–122
66. R. Marklein, R. Barmann, K.J. Langenberg, The ultrasonic modeling code EFIT as applied to
inhomogeneous dissipative isotropic and anisotropic media, in Review of Progress in
650 14 Quantitative Ultrasonic NDE with Models
Quantitative Nondestructive Evaluation, vol. 14, ed. by D.O. Thompson, D.E. Chimenti
(Plenum Press, New York, 1995), pp. 251–258
67. T.A. Gray, The CAD/NDE interface—designing for inspectability, in Review of Progress in
Quantitative Nondestructive Evaluation, vol. 9A, ed. by D.O. Thompson, D.E. Chimenti
(Plenum Press, New York, 1990), pp. 877–884
68. L.W. Schmerr, D. O. Thompson, NDE and design—a unified life-cycle engineering approach,
in Review of Progress in Quantitative Nondestructive Evaluation, vol. 12B, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1993), pp. 2325–2331
69. L.W. Schmerr, D.O. Thompson, NDE models and design—a unified life-cycle engineering
approach, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 13B, ed. by
D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1994), pp. 2183–2190
70. L.W. Schmerr, T.P. Lerch, A. Sedov, A focused transducer/scatterer model for ultrasonic
reference standards, Review of Progress in Quantitative Nondestructive Evaluation, vol. 12A,
ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1993), pp. 925–931
71. L.W. Schmerr, A. Sedov, The flat-bottom hole: an ultrasonic scattering model. Res.
Nondestruct. Eval. 1, 181–196 (1989)
Chapter 15
Model-Based Flaw Sizing
One important use of models is the development of methods to size unknown flaws
from their measured response. Since natural occurring flaws are typically irregular
in shape, when sizing defects it is necessary to determine the degree to which one
wants to try to recover the details of the flaw geometry.
At one extreme, one may simply size the flaw in terms of some comparable
reference scatterer such as a flat-bottom hole, for example, where “comparable” is
usually taken to mean that both the unknown flaw and known scatterer produce the
same A-scan signal amplitude. This approach suffers from the fact that the signal
amplitude of the unknown flaw is tied closely to its geometry (see the modeling
results in Chap. 10), which may be quite different from that of the reference
scatterer, so that no real size information is recovered in this approach unless the
morphology of the flaw is known a priori.
At the other extreme, one can attempt to collect a large amount of data on a flaw
and use that data to reconstruct an ultrasonic flaw image. This approach, when
successful, has been able to provide qualitative information on flaw properties, but
for small flaws has been less successful to date as a quantitative analysis tool [1].
Equivalent flaw sizing is essentially a middle ground approach between these
two extremes. In this method, one attempts to construct an estimate of the size,
shape, and orientation of an unknown flaw in terms of a best-fit simple equivalent
shape, such as an ellipsoid (for volumetric flaws), or an ellipse (for cracks). These
equivalent shapes, while they ignore many of the detailed irregularities of the
unknown flaw, do provide a quantitative size estimate that can then be used in
modern structural safety and reliability estimates [1].
Electronic Supplementary Material: The online version of this chapter (doi: 10.1007/978-3-
319-30463-2_15) contains supplementary material, which is available to authorized users.
Cracks are often the most dangerous type of defect one can find in a part since they
can lead to rapid, catastrophic failures if they exceed a certain critical size [1]. In fact,
the whole field of fracture mechanics has grown up in the last 50 years to understand
crack growth behavior, crack failure criteria, and to develop means to guarantee the
safety and reliability of cracked structures. NDE methods can play an important part
in these efforts if the methods can provide quantitative crack sizing information.
In this section, we will use an approach, based on the Kirchhoff approximation, to
size isolated cracks in terms of a best fit flat elliptical shape. By isolated, we mean
that the crack is well removed from any boundary as shown in Fig. 15.1. Isolated
cracks may appear in problems such as weld cracking, or in materials where
subsurface inhomogeneities initiate such cracks. It may be possible to use this
method also to size other flaws that are crack-like in nature, such as lack of fusion
or lack of penetration defects in welds, for example, or delaminations in composites.
In the Kirchhoff approximation, the far field scattering amplitude response of an
elliptical shaped flat crack always exhibits a pair of “flash points” in the time
domain as shown in Chap. 10. Those flash points are characterized by an anti-
symmetrical set of square root singularities for an incident P-wave or for an
incident SV-wave that is below the critical angle with respect to the stress-free
crack surface (see Fig. 10.20). For an incident SV-wave above the critical angle, the
flash points are a combination of symmetrical and anti-symmetrical responses
(Fig. 10.33). However, in all cases, these flash points are separated by a time
interval, 2Δt, where this time interval can be written directly in terms of the
y
eαs β
ei
iy a1 u3
iz x u1 a2
ix
z u2
15.2 Kirchhoff-Sizing for Cracks 653
geometry of the crack and the incident and scattered wave directions and mode
type. To see this, recall from Chap. 10 that for an incident plane wave of type β
traveling in the e βi direction, and scattered from a flat elliptical crack in a solid as a
wave of type α traveling in a direction eαs ðα ¼ P, SV Þ, ðβ ¼ P, SV Þ, flash points
appear in the time domain far field scattering amplitude response separated by a
time 2Δt where
Δt ¼ gα;β r eα;β =cα : ð15:1Þ
This time interval, Δt, is a function of the incident and scattered wave speeds of the
solid, cβ and cα, respectively, the incident and scattered wave directions and the
geometry of the ellipse. The “setup parameter”, gα;β, is given by
gα;β ¼ cα =cβ eiβ esα ð15:2Þ
with
eqα;β ¼ gα;β =gα;β ; ð15:4Þ
where recall (a1, a2) are the lengths of the two semi-major axes of the ellipse and
(u1, u2) are unit vectors along those axes. For a given ultrasonic inspection it is
assumed that the modes (and corresponding wave speeds) of the waves being sent
and received in the solid are known, as are the incident and scattered directions, so
that for a measurement of the time between flash points, 2Δt, Eq. (15.1) lets us
calculate the square of the effective radius directly, i.e.
2
α;β 2 cα Δt
re ¼ : ð15:5Þ
jgα;β j
M h
2 i
2
1X 2
Eð a 1 ; a2 ; ϕ1 ; ϕ2 ; ϕ3 Þ ¼ r eα;β m r eα;β ða1 ; a2 ; ϕ1 ; ϕ2 ; ϕ3 Þ ð15:8Þ
2 m¼1
and use a direct search or gradient descent method to find the best set of parameters
that minimize E. It was found, however, that this non-linear least squares minimi-
zation process was sensitive to noise and bandwidth constraints and required
adaptive search strategies when sizing cracks [7]. Fortunately, these difficulties
can be avoided entirely by reformulating the procedure as a two-step linear least
squares/eigenvalue problem.
The first step in this method is to note that the effective radius expression for the
ellipse Eq. (15.3) is identical in form to that of an ellipsoid (see Appendix F) if we
simply set the third dimension, a3, of the ellipsoid equal to zero. Thus, for the
ellipsoid,
2
2
2
2
r eα;β ¼ a21 eqα;β u1 þ a22 eα;β
q u 2 þ a 2
3 eα;β
q u 3 : ð15:9Þ
If we introduce a fixed set of coordinates (x, y, z) (Fig. 15.1) and write both the
known unit vectors eqα;β and the unknown unit vectors un ðn ¼ 1, 2, 3Þ in terms of
their components in this coordinate system, i.e.
eqα;β ¼ ex ix þ ey iy þ ez iz
; ð15:10Þ
un ¼ unx ix þ uny iy þ unz iz
15.2 Kirchhoff-Sizing for Cracks 655
where
The value of introducing these six C coefficients is that, as Eq. (15.11) shows, (r α;β
e )
2
M h
2 i
2
1X 2
EðCÞ ¼ r eα;β m r eα;β ðC; em Þ ; ð15:13Þ
2 m¼1
∂EðCÞ
¼0 ði, j ¼ 1, 2, 3Þ; ð15:14Þ
∂Cij
XM n 2
∂E
¼ r eα;β m C11 e21m þ C22 e22m þ C33 e23m þ C12 e1m e2m
∂Cij m¼1
o
þ C13 e1m e3m þ C23 e2m e3m eim ejm ¼ 0ði ¼ 1, 2, 3Þ ð j ¼ 1, 2, 3Þ: ð15:15Þ
Once Eq. (15.15) is solved for the C coefficients, we can use these coefficients to
determine the size and orientation parameters of the best fit ellipsoid, (a1, a2, a3) and
656 15 Model-Based Flaw Sizing
(u1, u2, u3), in the following manner. First, note that when the (x, y, z) axes are
aligned with the (u1, u2, u3) unit vectors we have
From Eq. (15.12) then it follows that under these conditions the symmetric matrix
C is diagonal, where the diagonal components are the squares of the ellipsoidal
parameters (a1, a2, a3), i.e.
Cxx ¼ a21
Cyy ¼ a22
: ð15:17Þ
Czz ¼ a23
Cxy ¼ Cxz ¼ Cyz ¼ 0
Thus, if once having obtained the C coefficients, we solve the subsequent eigen-
value problem
X
3
Cij λδij lj ¼ 0 ði ¼ 1, 2, 3Þ ð15:18Þ
j¼1
the three eigenvalues (λ1, λ2, λ3) and corresponding eigenvectors (l1, l2, l3) are just
the desired size and orientation parameters:
λ1 ¼ a21 l1 ¼ u 1
λ2 ¼ a22 l2 ¼ u 2 : ð15:19Þ
λ3 ¼ a23 l3 ¼ u 3
For our crack sizing problem, since a3 ¼ 0, solving the linear least squares problem
Eq. (15.15) and subsequent eigenvalue problem Eq. (15.18) should in principle give
one eigenvalue that is identically zero. In practice, however, measurement errors,
noise, and numerical inaccuracies in the solution process will often give a small
positive or negative value for this eigenvalue. This small value can be ignored since
the only pertinent size information is in the other two coordinate directions.
This two-step sizing procedure is fast, robust, and relatively simple to implement
in modern automated immersion scanning systems. Most of the difficulties encoun-
tered with the method center around identifying the flash point signals which are
relatively weak. A set of experimental Δtm measurements taken from such a system
is given in the problems at the end of this Chapter. This data set (together with some
synthetic data also provided in the problems) can be used to demonstrate the entire
equivalent flaw sizing procedure.
15.2 Kirchhoff-Sizing for Cracks 657
This sizing algorithm relies on models in three ways. First, the algorithm itself is
based on an approximate model for the flaw scattering amplitude. Second, the crack
response predicted by the Kirchhoff approximation is for the flaw far field scatter-
ing amplitude, and as we have seen previously, this scattering amplitude is only one
part of an entire ultrasonic measurement process. Thus, it is desirable to extract the
scattering amplitude (as a function of frequency) from the measured ultrasonic
response, using the Thompson-Gray measurement model and the deconvolution
procedures discussed in the previous chapter. This scattering amplitude can then be
inverted on time to obtain a signal from which the Δtm measurements can be made.
Finally, a third way that modeling can help in this sizing method is in the elimina-
tion of errors due to finite bandwidth constraints [8]. These errors arise since even
after deconvolution, the scattering amplitude is obtained as a function of frequency
only over a finite frequency range. Thus, when this finite bandwidth response is
inverted back into time, the time measured between flash points will be changed
systematically from the theoretical value of Δtm. However, one can use models to
also help remove this error. If a simple rectangular window function is used to
represent the effective bandwidth of the system, then one can multiply the theoret-
ical scattering response (see Eq. (10.171) and Eq. (10.172) for example) by this
window function and invert the product back into the time domain to get a
simulated bandlimited signal. Equivalently, one could use the efficiency factor, if
is available, to replace the rectangular window function and give a better represen-
tation of the actual bandwidth of the system. In either case, repeating this process
for many different incident and/or scattered angles and different flaw sizes, one can
then construct a calibration correction curve of the error involved (defined here as
(exact Δt bandlimited Δt)/exact Δt) versus the measured Δt. An example of such
a curve, generated using a rectangular window function, is shown in Fig. 15.2.
Figure 15.3 shows a set of total errors involved when applying the sizing algorithm
to scattering data taken from a 400 μm radius crack in titanium, both with and
without compensation for the systematic bandwidth errors. As can be seen, this
correction does “de-trend” the data and reduce somewhat the overall error level.
The general shape of the calibration curve of Fig. 15.2 can be understood
qualitatively as follows. At relatively large values of Δt, the finite bandwidth effects
increase the rise and fall times of the flash point signals and cause the ideal time
separation between flash points to be reduced somewhat, as shown in Fig. 15.4a.
The percentage error due to this effect tends to grow as Δt becomes smaller. In the
limit Δt ! 0, the flash points ideally go to the infinitely sharp “doublet” shown in
Fig. 15.4b. However, even for an infinitely sharp doublet a finite bandwidth
ultrasonic system can only produce an apparent set of flash points separated by a
minimum time Δtb, where Δtb is characteristic only of the bandwidth of the system.
Thus, for any exact separation times between flash points, 2Δt, that satisfy
2Δt < Δtb , (Fig. 15.4c) the measured time interval in these cases will always be
merely Δtb, eventually producing infinitely large negative percentage errors as the
exact Δt ! 0.
658 15 Model-Based Flaw Sizing
30
10
–10
–30
% Error
–50
–70
–90
–110
15.91 39.79 79.58 127.32 175.08 222.82 270.56 318.31
Δt in Nanoseconds
Fig. 15.2 Look-up curve for the elimination of finite bandwidth effects on time measurements Δt.
Reprinted with permission from: C.P Chiou, and L.W. Schmerr, New approaches to model-based
ultrasonic flaw sizing, Journ. Acoust. Soc. Am., Vol. 92, 435–444, (1992)
30
25
experiment
20 correction
15
% Error
10
–5
–10
30 35 40 45 50 55 60 65 70
Look Angle in Degree
Fig. 15.3 Model-based corrections of Δt for a 400 400 μm “crack” in titanium. (Reprinted with
permission from: Chiou, C.P, and L.W. Schmerr, New approaches to model-based ultrasonic flaw
sizing,” Journ. Acoust. Soc. Am., Vol. 92, pp. 435–444, 1992. Copyright 1992 Acoustical Society
of America)
15.3 Born-Sizing for Volumetric Flaws 659
Δtb
As shown in the previous section, if an experimental method exists for obtaining the
equivalent radius, re, of a flaw in various directions, this information can be used to
put together an equivalent ellipsoidal shape that best matches the scattering data.
Because the previous section was concerned with sizing isolated cracks, we were
only interested in obtaining a best fit degenerate ellipsoid (ellipse). However, for a
volumetric flaw, if a similar determination of re can be made, we can use the same
algorithms to determine the best fit complete 3-D ellipsoid.
Determining re for an isolated crack was made simple by the existence of two
singularities (flash points) in the far field scattering amplitude time domain
response. These singularities are usable from an experimental standpoint because
they remain prominent in a real A-scan response even for a bandlimited system
(other, less singular components will be more severely “washed out” when
bandlimited). For volumetric flaws, our results of Chap. 10 (based on the Kirchhoff
approximation) showed the only singularity in the scattering amplitude was the
leading edge response of the flaw. In contrast, in the weak scattering Born approx-
imation, both singular leading edge and trailing edge responses were predicted for
volumetric flaws. Unfortunately, for the strongly scattering flaws typically found in
practice, only the singular leading edge response, which is predicted by both the
Born and Kirchhoff approximations, is consistently present.
Since the scattering responses of volumetric flaws do not normally contain two
prominent responses whose time separation can be used to find re, other methods of
determining the effective radius must be found. Fortunately, such a method exists
660 15 Model-Based Flaw Sizing
within the Born approximation. To see this, recall from Chap. 10 that the far field
scattering amplitude for a flaw in a solid can be obtained, as a function of frequency,
in terms of a coefficient, fl(eβi ; eαs ), which is a function only of the material
properties of the flaw and the incident and scattered directions, and a shape
function, G(eβi ; eαs , ω), as
ω2 Blnα f l eiβ ; esα
Anα;β eiβ ; esα ¼ G e i
β α
; e s ; ω ; ð15:20Þ
4πc2α 0
where
ð
G eiβ ; esα ; ω ¼ exp ikα 0 gα;β x dV ðxÞ ð15:21Þ
Vf
and
gα;β ¼ cα 0 =cβ 0 eiβ esα : ð15:22Þ
For an ellipsoid, following the same steps as carried out in Sect. 10.3.3 for the fluid
case, we would find explicitly
ð1
4πa1 a2 a3
β α
G ei ; e s ; ω ¼ α;β
sin kα0 gα;β r eα;β r rdr: ð15:23Þ
α;β
kα 0 jg jre
r¼0
From Eq. (15.20) and Eq. (15.23), we can write the far field scattering amplitude of
the ellipsoid as
ð1
α;β α;β α;β
Aα;β
n eiβ ; esα ; kα 0 ¼ Cnα;β sin kα 0 gα;β r α;β r r r dr;
e r kα 0 g e e ð15:24Þ
r¼0
where we have shown the dependency of the scattering amplitude on kα0 explicitly
and where the coefficient, Cα;β
n , is
Blnα f l eiβ ; esα a1 a2 a3
Cnα;β ¼
3 : ð15:25Þ
jgα;β j r eα;β
2
and make the change of variables to u ¼ reα;β r, Eq. (15.24) can be rewritten as
ð
1
Anα;β β α α;β
ei ; es ; kα 0 ¼ Cn Θ u=reα;β sin kα 0 gα;β u kα0 gα;β u du: ð15:27Þ
u¼0
As shown in Chaps. 11 and 12, in the Thompson-Gray measurement model, the flaw
response appears in the form of an explicit component of the scattering amplitude
which we will write here as
h
i
A eiβ ; esα ; kα 0 ¼ Anα;β eiβ ; esα ; kα 0 linα ¼ Aα;β eiβ ; esα ; kα 0 liα
ð15:28Þ
ð
1
β α α;β α
A ei ; es ; kα0 ¼ Cn lin Θ u=r eα;β sin kα0 gα;β u kα0 gα;β u du: ð15:29Þ
u¼0
ð
1
h
i sin k gα;β v
β α α0
Fð v Þ ¼ Re A ei ; es ; kα 0 dkα 0 ; ð15:30Þ
kα 0 jgα;β jv
0
where “Re” indicates the real part of the scattering amplitude. Note that for an
ellipsoid in the Born approximation the scattering amplitude is already purely real
so that taking the real part is superfluous. However, we want to be able to generalize
Eq. (15.23) for other cases (non-ellipsoidal, strong scatterers etc.) where the
scattering amplitude is complex and still recover a real function F(v) that can be
directly related to the flaw geometry. To see the function that is recovered for the
ellipsoid in the Born approximation, placing Eq. (15.27) into Eq. (15.30) gives
ð1
1 ð
FðvÞ ¼ Cnα;β linα Θ u=rα;β sin kα 0 gα;β u sin kα 0 gα;β v u dkα 0 du
e
v
0 0
ð15:31Þ
ð þ1
þ1 ð
Cnα;β linα
u
Fð v Þ ¼ Θ u=reα;β sin kα0 gα;β u sin kα0 gα;β v dkα 0 du:
4 v
1 1
ð15:32Þ
662 15 Model-Based Flaw Sizing
ð
þ1
sin kα0 gα;β u sin kα0 gα;β v dkα 0
1
ð
þ1
1
¼ exp ikα 0 gα;β ðu þ vÞ þ exp ikα 0 gα;β ðu þ vÞ
4
1
exp ikα 0 gα;β ðv uÞ exp ikα 0 gα;β ðu vÞ dkα 0
¼ π δ gα;β ðu þ vÞ δ gα;β ðu vÞ ð15:33Þ
so that using the properties of the delta function the u integration in Eq. (15.30) can
also be done. This shows that the function F(v) is just proportional to the charac-
teristic function, i.e.
1
i sin k gα;β v
2gα;β reα;β α0
Θ v=r eα;β ¼
β α
Re A ei ; es ; kα0 dkα0 :
α α β α
πBmn lin f m ei ; es a1 a2 a3 0 k α0 jgα;β jv
ð15:35Þ
The component of the scattering amplitude in the integrand of Eq. (15.35) is in the
frequency domain. We can also obtain a similar expression for the characteristic
function
using the
same component
of
the time domain scattering amplitude,
i.e. for a eiβ ; esα ; t ¼ linα anα;β eiβ ; esα ; t
h
ð
i þ1
β α
Re A ei ; es ; kα0 ¼ a eiβ ; esα ; t cos ðωtÞdt ð15:36Þ
1
ð
þ1 ð
þ1
1 sin kα0 gα;β v sin kα0 gα;β v
expðiωtÞdkα0 ¼ cos ðωtÞdkα0
2 kα0 jgα;β jv kα0 jgα;β jv
1 0
π
¼ α;β
Θ cα0 t=gα;β v ð15:37Þ
2jg jv
Equations (15.35) and (15.38) show how the characteristic function can be recov-
ered for an ellipsoid (in the Born approximation) if an appropriate integral of the
scattering amplitude is carried out in the frequency or time domains, respectively.
These equations also provide a means for determining the effective radius of an
unknown flaw if the scattering amplitude is obtained experimentally for that flaw.
The procedure is to calculate either of the following integrals:
ð
1
h
i sin k gα;β v
β α α0
I ðvÞ ¼ Re A ei ; es ; kα 0 dkα 0 ð15:39aÞ
kα 0 jgα;β jv
0
or
þjgα;β jv=cα 0
ð
1
I ðvÞ ¼ a eiβ ; esα ; t dt: ð15:39bÞ
v
jgα;β jv=c α0
The value of v where either of these integrals drops to near zero is then taken to be
the effective radius, re, which is the “edge” of the corresponding characteristic
functions in Eq. (15.35) and Eq. (15.38). If this same process is repeated at
M different incident and/or scattered directions, then this data provides the input
needed to extract the best fit equivalent ellipsoid for the unknown flaw via the linear
least squares/eigenvalue approach of the previous section. This method is called the
1-D inverse Born approximation (in the frequency or time domain) [9].
Although the Born weak scattering approximation was used as the basis for this
sizing algorithm, Rose [10] has shown that even a strong scatterer, like a spherical
void, the algorithm produces a correct estimate of the sphere radius. In Fig. 15.5a
the exact pulse-echo far field scattering amplitude of P-waves from a sphere of
radius a is shown, together with the characteristic function recovered via this
method. In this case the effective radius re is the same as the radius of the sphere,
which is estimated accurately, as shown in Fig. 15.5b. Figure 15.6 shows some
664 15 Model-Based Flaw Sizing
a b
1
R (t)
g (r)
0
–2a/c 0 a r
Time
Fig. 15.5 (a) Schematic representation of the true impulse backscattered response function for a
spherical void in a metal. (b) A schematic characteristic function corresponding to the impulse
response of (a). Reprinted with permission from: J.H. Rose, Elastic wave inverse scattering in
Nondestructive Evaluation, Pure Appl. Geophys., 131, 715–738, (1989)
similar results for spherical voids in various materials where the exact scattering
response, as calculated numerically, is bandlimited somewhat (0:1 < kp a < 10:0,
where a is a characteristic dimension for the flaw and kp is the wave number for
P-waves) to more realistically represent actual data. Similarly, Fig. 15.7 gives the
recovered characteristic functions for various bandlimited spherical inclusion
responses. In general, the Born approximation appears to perform well in estimat-
ing the effective radius. However, as Fig. 15.7 shows, for highly resonant scatterers
like Fe in Si3N4, where multiple waves that are internally reflected in the inclusion
are evident in the scattering response, determination of the effective radius may be
difficult. Based on an experimental program which examined sizing different flaws
in titanium and ceramic specimens, Rose [10] has indicated that the Born sizing
method does require some minimum available bandwidth for the ultrasonic
15.3 Born-Sizing for Volumetric Flaws 665
.80
.40
y(r) .00
–.40
–.80
0 1 2 3 4
r/a
measurement system, which he estimated as 0:5 < kp a < 2:5. To apply the 1-D
inverse Born approximation to size unknown flaws, one must be aware of an
important assumption that is implicit in the use of Eq. (15.39a) and Eq. (15.39b)
in either the time domain or frequency domain forms. This assumption is that the
origin of the time coordinates in Eq. (15.39b) is at the “center” of the flaw and,
likewise, that the frequency domain scattering amplitude of Eq. (15.39a) is also
computed from a time domain scattering response with this same time origin.
Placing the zero of time at the correct position in a set of measured data from real
scatterers is difficult because (1) real flaws are often irregular and do not necessarily
possess a definite center and (2) there is nothing in the scattering response of even
regular flaw shapes with definite centers (like spheres) that distinguishes the center
in the time domain scattering response. Fortunately, there is one possible solution to
this zero-of-time problem within the Born approximation itself [4, 11, 12]. To see
this, recall from Chap. 10 that the time domain far field impulse response of a flaw
could be written as (see Eq. (10.229):
Blnα linα f l eiβ ; esα d2 Sα;β c t=gα;β
α0
a eiβ ; esα ; t ¼ ; ð15:40Þ
4πcα0 dt2
8 9
ðt < ðτ
=
sðtÞ ¼ a eiβ ; esα ; u du dτ ð15:41Þ
: ;
1 1
is a maximum. For an unknown, irregular flaw, the integrand of Eq. (15.41) can be
obtained via deconvolution using the measured response of that flaw and the
Thompson-Gray measurement model. Then the location of the peak of the double
integral of Eq. (15.41) can also be taken to be the zero-of-time for that flaw
response. Since a(eβi ; eαs , t) comes from the far field impulse response of the flaw,
s(t) is the corresponding time domain ramp response which has been used exten-
sively in the electromagnetic scattering literature for reconstructing the geometry of
unknown scatterers [13, 14] using the physical optics (Kirchhoff) approximation.
There are also available other methods for finding the zero of time location,
including low frequency methods [10] and the “Born Radius/Zero-of-Time Shift
Domain” (BR/ZOTSD) signature [15]. All of the current methods, however, have
some difficulty in being implemented in practice due to noise, limited transducer
bandwidths, and other system errors. Thus, to date there is still a need to have more
robust ways to solve the zero-of-time problem in general.
The Born and Kirchhoff sizing methods are effective for sizing even very small
flaws provided sufficient information is available to determine the effective radius
of the flaw at various orientations relative to the transducer. In the case of Born
sizing of volumetric flaws, this meant that the location of the center of the flaw
needed to be determined relative to its surface. However, if the size of the flaw is
sufficiently large, both the location of the surface of the flaw and the flaw center can
be determined simultaneously through time-of-flight measurements relative to the
transducer and then used to determine a best-fit ellipsoidal flaw shape, as done in
the Kirchhoff and Born sizing methods. This approach we will refer to here as a
time-of-flight equivalent (TOFE) flaw sizing method [16].
Consider a pulse echo (single mode) immersion set-up as shown in Fig. 15.8,
where measurements of the effective radius, reα;α , and center of the flaw, xc
(as measured in some fixed (x, y, z) coordinate system) are to be obtained by
scanning the transducer at different locations, xT, and orientations, eT. When the
transducer is at normal incidence to the interface, the distance H1 can be obtained
by measuring the time it takes a wave reflected from the interface to travel down
and back to the transducer (assuming the wave speed of the water is known). Even
at non-normal incidence, this interface echo can be used to obtain H1 if the angle of
the transducer relative to the interface is not too high. This is because the transducer
in practice puts out waves with a range of angles instead of just an idealized single
central ray shown in Fig. 15.8 and a real surface is rough and reflects some energy
15.4 Time of Flight Equivalent Flaw Sizing 667
e
xc
H2
re
back to the transducer even at oblique incidence. Also, one only has to use a small
range of angles (about normal incidence to the interface) in the water for the
transducer to generate waves with a wide range of refracted angles in the solid
because of the large velocity difference between the water and the solid. Thus, if H1
is known by the time of flight measurements to the interface and the location, xT,
and orientation, eT, of the transducer are both accurately known, the location at
which a central ray strikes the interface, xI, can be obtained from xI ¼ xT þ H 1 eT
(Fig. 15.8). Having once obtained H1, the distance from the interface to the surface
or edge of the flaw, H2 (Fig. 15.8), can then be found by also measuring the total
time of flight, T, from the transducer to the flaw and back (using the first flash point
response for a crack or the leading edge response for a volumetric flaw) and the
known wave speeds of the fluid and the solid. Finally, the direction of the wave in
the solid, e, can be obtained directly from Snell’s law. Thus, in this setup it is
possible to obtain measured values of the parameters xI, e, H2 which we will denote
here, for the mth measurement ðm ¼ 1, 2, . . . MÞ as xIm, em, H2m, respectively. Then,
from the geometry it follows that we have
ðxIm xC Þ em ¼ H 2m þ r α;α
e m
ð15:42Þ
Combining Eq. (15.42) and Eq. (15.43) we can form a function, Fm, which is
identically zero for perfect measurements of an ellipsoid, i.e.
For M real measurements of an unknown flaw, we can then define an error term
symbolically as
1X M
EðxC ; CÞ ¼ ½Fm ðxC , C ; xIm , em , H 2m Þ2 ð15:45Þ
2 m¼1
where now the error function depends on both the six unknown C coefficients and
the three coordinates of the flaw centroid location, xC. These nine unknowns can be
obtained by again minimizing the error function, i.e. setting its first derivatives
equal to zero and solving the nine equations
∂E
¼0 ðj ¼ 1, 2, 3Þ
∂xCj
; ð15:46Þ
∂E
¼0 ði, j ¼ 1, 2, 3Þ
∂Cij
flaw
15.5 Other Sizing Methods 669
fit equivalent sphere but in this case much of the shape information is lost (see
problem 15.8 at the end of the chapter which discusses this case). Thus, it was
concluded that to adequately reconstruct equivalent ellipsoidal-shaped volumetric
flaws with this method two-sided data is typically needed. This type of data could
be obtained in some cases by using back surface reflections from a part, as has been
verified experimentally [16]. However, when TOFE has been used for sizing crack-
like flaws such as a Teflon insert in a composite plate it was found that even one-sided
scanning over a limited aperture was sufficient to reconstruct the flaw [16]. This result
is likely due to the fact that a best fit ellipsoid which reconstructs the crack surface
over the scanning aperture is very thin, thus effectively constraining the crack center
location as well. A synthetic data set is given in the problems at the end of this chapter
so that the reader can verify for himself/herself these properties of the TOFE sizing
algorithm directly.
Since TOFE is based on absolute time-of-flight measurements from the trans-
ducer to the flaw, there are a number of requirements one must be aware of in order
for this method to work properly. First, the wave speeds of the materials involved
must be known accurately. If there is a measurement error Δc in the wave speed
c over a path length H, then an error in the measurement of the flaw surface location
will be ΔH ¼ ðΔc=cÞH. Thus, for example, a 5 % error in wave speed over a one
inch (25.4 mm) path length in steel would give ΔH ¼ 1:27 mm. Similarly, the
position of the transducer must be known accurately as well. Thus, an automated
scanning system is typically needed for such measurements. Note that the Born and
Kirchhoff sizing methods can achieve much higher accuracy in sizing very small
flaws than TOFE sizing because they rely only on time of flight measurements over
distances on the order of the size of the flaw itself. In practice, TOFE has been
shown to be able to size flaws as small as a few mm in diameter on reasonably well-
characterized materials when using a precision scanning apparatus [16].
Conventional testing practices often only use amplitude information from A-scan
responses for flaw sizing. One example of this approach is the use of distance-gain-
size (DGS) curves to size flaws in terms of “equivalent” flat-bottom holes that produce
the same amplitude responses as the unknown flaw [17–19]. Echo-dynamic patterns
can also be used to size relatively large flaws by mapping the signal amplitude
variation as a function of transducer position while scanning over the flaw [20].
These conventional methods suffer from their strong reliance on amplitude
information. Some methods attempt to avoid this limitation by also incorporating
time-of-flight information into the sizing scheme. The time-of-flight diffraction
(TOFD) and satellite-pulse techniques, in particular, are examples of this type of
approach [53, 21–24].
The TOFD method, like the Kirchhoff method, uses the flash point (edge-
diffracted waves) from a crack as the basis for sizing. In a typical application of
this method, a transmitter and receiver are located in the neighborhood of the crack
670 15 Model-Based Flaw Sizing
and the edge waves diffracted from the two extremities of the crack are identified.
By measuring the time-of-flight of these two signals and assuming a simple linear
(slit-like) two dimensional crack shape, the length of the crack can then be
calculated. A detailed description of TOFD sizing and its advantages over
amplitude-only based methods can be found in [24]. The crack mapping approach
by Achenbach and his co-workers is a more detailed time-of-flight algorithm that
also tries to take advantage of similar information for sizing purposes [25–27].
Today, the TOFD has become widely used, often in conjunction with imaging
approaches to help identify the relatively weak flashpoint signals.
The satellite-pulse technique is very similar to TOFD sizing for the case of
cracks, but uses a single transducer (pulse-echo) setup instead. This method also has
been used for sizing volumetric flaws in terms of a best-fit equivalent sphere by
measuring the time difference between the leading edge response of the flaw and a
following creeping wave. Unfortunately, this creeping wave response is often very
small in amplitude, particularly in pulse-echo setups, and may be difficult to
identify.
Another common conventional approach to flaw sizing is to generate a flaw
“image” using a combination of time (or position) and amplitude data. The B- and
C-scans mentioned in chapter one are simple examples of this method. More
advanced techniques of this type are the amplitude and transit time locus curves
(ALOK) [28], the synthetic aperture focusing technique (SAFT) [29, 30], the
physical optics far field inverse scattering (POFFIS) method [31], and the total
focusing method.(TFM) [32]. Other imaging methods that provide more flaw shape
information (but at the expense of detailed scanning and/or processing) include
acoustical holography [33] and acoustic microscopy [34]. Implementation of imag-
ing methods has been made easier with the use of ultrasonic phased arrays that can
perform electronic scanning. For more details see, for example, Schmerr [35].
When the first edition of this book was published, it was hoped that equivalent flaw
sizing would be adopted as a practical sizing tool by some of the ultrasonic NDE
community. Unfortunately, this has not happened. Flaw sizing today is still done
primarily with methods such as the time-of-flight diffraction (TOFD) method,
amplitude-based dB-drop methods, and some phased array imaging approaches.
Equivalent flaw sizing with the Born approximation, in particular, has remained
elusive because of the difficulty of experimentally determining the equivalent
radius from the boundaries of a characteristic function and the lack of an adequate
solution of the zero of time problem. One of the reasons for the lack of interest in
methods such as the Kirchhoff sizing method or TOFE flaw sizing, even though the
Kirchhoff method is closely related to the popularly used TOFD method, likely
comes from the need to have precise linear and angular positioning of the trans-
ducer(s), a requirement that leads to an expensive scanning system. In fact, all of the
15.5 Other Sizing Methods 671
successful Kirchhoff and TOFE sizing experiments performed to date were done
with a multi-viewing transducer system consisting of a collection of single-element
10 MHz wide band (~2 to 16 MHz) transducers (Fig. 15.10) which could be
precisely positioned to examine a flaw from multiple directions [5]. This one-of-
a-kind system was built at the Center for NDE, Iowa State University, specifically
to implement equivalent flaw sizing methods. With the growing popularity and
improved economics of phased array systems, however, equivalent flaw sizing may
now be more viable. This is possible since with a phased array one can replace
expensive setups such as the multi-viewing transducer system with a single phased
array and take advantage of the electronic scanning capabilities of the array to
generate much of the needed data. An example of this approach was given by Engle
et al. [36]. In this case a linear phased array was used in an immersion setup as
shown in Fig. 15.11 to generate a varying incident wave direction e(θ, ϕ), where
(θ, ϕ) are spherical coordinate angles. Changes in the angle θ were obtained by
phasing the linear array appropriately and changes in the angle ϕ were obtained by
rotating the specimens. Two specimens were tested—an aluminum specimen with a
#5 flat-bottom hole (to simulate a plane 1.984 mm diameter circular crack) and a
diffusion-bonded titanium specimen containing an isolated plane elliptical crack
(actually a very thin “pillbox”) with a semi-major axis of 2.5 mm and a semi-minor
axis of 0.6 mm. These flaws were interrogated with a 32 element, 10 MHz linear
672 15 Model-Based Flaw Sizing
array. The array was phased to produce steering at combinations of four θ angles
and three ϕ angles for a total of 12 “look angles”. Kirchhoff sizing of these cracks
was done using the observed flashpoint signals and the two step linear least squares/
eigenvalue sizing method, and these measured data were compensated for finite
band width effects, as discussed in Sect. 15.2.2. Note: in both of these cases the
measured voltage data was used directly rather than extracting the far field scatter-
ing amplitude by deconvolution and then applying the sizing method to that
deconvolved data, so that there remain some modeling errors here. This was done
to show that the Kirchhoff sizing can be applied without the use of the additional
deconvolution step and to show that the bandwidth effects, which were compen-
sated for, are likely the major contributors to errors in the time separation of the
flashpoints. The unit normals of both the flat-bottom hole and elliptical crack were
along the z-axis. The flat-bottom hole size and orientation parameters were:
8 9 8 9 8 9 8 9
< a1 = < 0:992 = < u1x u2x u3x = < 1 0 0 =
a ¼ 0:992 u u2y u3y ¼ 0 1 0 ð15:47Þ
: 2; : ; : 1y ; : ;
a3 0 u1z u2z u3z 0 0 1
where (a1, a2, a3) are the semi-major axis values (in mm) and (u1, u2, u3) are the
directions of those axes. Experimentally, the values found by Engle et al. [36] with
the phased array and the Kirchhoff sizing method were:
8 9 8 98 9 8 9
< a1 >
> = > < 1:05 >=> < u1x u2x u3x > = > < 0:097 0:990 0:106 > =
a2 ¼ 1:00 u1y u2y u3y ¼ 0:995 0:096 0:013 :
: >
> ; > : >
;> : ; >
> : >
;
a3 0:48i u1z u2z u3z 0:002 0:107 0:994
ð15:48Þ
Note that the unit vectors (u1, u2) are arbitrary in this case so that u3 is the only
unit vector of interest, where there is in fact very good agreement with the flat-
bottom hole orientation. Also note that the value of a3 was an imaginary number.
This occurred because a23 can come out to be a small negative number in the
Kirchhoff sizing method, a result that is of no consequence since both measured
non-zero values were close to their actual values.
For the elliptical crack in titanium the actual size and orientation parameters
were:
8 9 8 98 9 8 9
< a1 = < 2:5 = < u1x u2x u3x = < 1 0 0 =
a ¼ 0:6 u u2y u3y ¼ 0 1 0 ; ð15:49Þ
: 2; : ; : 1y ; : ;
a3 0 u1z u2z u3z 0 0 1
15.5 Other Sizing Methods 673
which again were in good agreement with the non-zero sizes. Note that in this case
the measured value of u2 was approximately in the negative y-direction, but this is
acceptable since both the positive and negative y-axis of the ellipse are the same
principal axis. To show that these sizing examples can be easily implemented in
MATLAB® the elliptical crack data from [36] was placed in a MATLAB® function
ellipse_data( ) whose calling sequence is:
This function simply returns three vectors containing the angles theta and phi
(in deg.) and the band-width-corrected time difference, dt (in μs) between the flash
point signals. To use this data a MATLAB® function K-sizing(c, theta, phi, dt) was
also written that implements the linear least squares /eigenvalue Kirchhoff sizing of
the crack. Its calling sequence is
where c is the wave speed (in mm/μs), and (theta, phi, dt ) are the data output from
ellipse_data. This function returns the ellipse axis orientations in the columns of the
matrix pds and semi-major axes (in mm) in the diagonal matrix a. Combining these
two functions we can size the elliptical crack in titanium (c ¼ 6.1 mm/μs) with
MATLAB® yielding:
The results are the same as given in Eq. (15.50) except that direction cosines of
the ellipse axes (columns of the pds matrix) and sizes are in a different order and
the real part was taken of the size matrix to eliminate any imaginary values.
674 15 Model-Based Flaw Sizing
[Note: the function K_sizing can also be used for a number of the Kirchhoff and
Born sizing examples in the Problems section of this chapter, but in those examples
the measured or simulated time data, Δt, is the time difference for waves traveling
from the flaw center and the flaw edge, not from edge to edge (i.e. between flash
points) as used here. Thus, in those problems the input argument to K_sizing, dt,
must be given as dt ¼ 2Δt]. A linear array was used in these cases but it could be
replaced by a 2-D array which would allow one to also use beam steering to obtain
at least some of the ϕ -angles needed. It is important that in these experiments only
12 single-sided pulse-echo responses were used. In principal one could use as few
as six responses but having more helps to average out experimental errors. This is
still a very modest amount of data and certainly considerably less than what is
needed in approaches where flaw images are generated.
Phased arrays could also be used in this same manner for applications such as
angle beam inspections of welds, as shown in Figs. 5.12 and 5.13. Either one-sided
or two-sided pulse echo measurements can be implemented for isolated cracks as
shown in Fig. 15.12a, b. For surface-breaking cracks, as shown in Fig. 15.13, only a
single flash point signal is generally present but a phased array implementation of
TOFE, as described in [37] is a possible sizing method.
Fig. 15.12 Use of a phased array to size an isolated flaw with pulse-echo measurements for
(a) one-sided and (b) two-sided look-angles
The Born sizing method we have described in this chapter is an example of the use
of the 1-D inverse Born approximation. The term “1-D” refers to the fact that in this
algorithm the equivalent radius is the fundamental one-dimensional quantity that is
extracted from the scattering data. A more detailed 3-D inverse Born method has
also been developed for reconstructing the 3-D characteristic function of arbitrarily
shaped convex flaws [10, 38]. Similarly, a Kirchhoff-based algorithm exists for
reconstructing the 2-D characteristic function of a planar crack [7, 39].
Estimates of the equivalent radius for volumetric flaws have be obtained exper-
imentally in several other ways than the approach described here for the Born sizing
method. Zeros in the frequency domain response, for example, have been used
instead to find equivalent radii [40], as have “first-moment” and amplitude-based
methods [41]. In addition to the non-linear least squares and linear least squares/
eigenvalue approaches for obtaining the best-fit ellipsoidal shapes from measured re
data, methods based on neural nets have also been used [42]. There is an extensive
literature on flaw sizing with the Born method, imaging, and other techniques, a
sampling of which can be found in [43–57].
15.7 Problems
z
em
θm
iz
iy
ix y
φm
x
Fig. P15.1 Spherical coordinates (θm, ϕm) of the look angle vector, em, which is the mth
measurement of the vector eqα;β . For a pulse-echo setup where the scattered waves are of the
same mode type as the incident waves ðα ¼ βÞ, then eqα;β ¼ eiβ where eβi is a unit vector in the
direction of the wave incident on the flaw
676 15 Model-Based Flaw Sizing
and the axes of the spheroid (in Cartesian and spherical coordinates) are:
Using the MATLAB® function K-sizing with dt ¼ 2Δt, determine the shape
and orientation of the best-fit ellipsoid that matches this scattering data. Take
the P-wave speed of titanium to be cp ¼ 6:1 mm=μs.
15.2. As shown in the Table P15.1, two data sets (numbers 2 and 12) were
considerably more in error than the rest. Repeat the sizing of problem 15.1
where these two “outliers” are not used, leaving a total of ten data points
altogether. What can you say about the sensitivity of the sizing method to
such errors?
15.3. A 400 μm radius circular crack in titanium was interrogated using a pulse-
echo immersion setup and P-waves incident on the flaw. The Kirchhoff
approximation, as described in this chapter, was used to obtain the time, Δt,
between the flaw center and the leading edge of the flaw (one half the time
between flash points). The results are shown in Table P15.2 (in spherical
coordinates—see Fig. P15.1).
The unit vector normal to the circular crack was aligned along the z-axis.
Using the MATLAB® function K_sizing, determine the shape and orientation
of the best-fit ellipsoid that matches this scattering data. Take the P-wave
speed of titanium to again be cp ¼ 6:1 mm=μs.
15.4. As shown in Table P15.2, two data sets (numbers 1 and 3) had a considerably
larger error than the rest. Repeat the sizing of problem 15.3 where these two
transducer
xI
scanning plane 1
z e
iz
iy
ix y H2
x flaw
xC
scanning plane 2
Fig. P15.2 Simulated planar scan of an ellipsoidal flaw in a single medium where the center of the
transducer, xI, is either in scanning plane 1 or scanning plane 2
“outliers” are not used, leaving a total of ten data points altogether. What can
you say about the sensitivity of the sizing method to such errors?
15.5. A crack of unknown size and orientation in a steel part cp ¼ 5:0 mm=μs
was interrogated using P-waves in an immersion setup. Ten simulated pulse-
echo measurements of the time, Δt, between the flaw center and the leading
edge of the flaw (one half the time between flash points) are listed in
Table P15.3 below for a variety of incident angles (given in spherical
coordinates—see Fig. P15.1). Using the Kirchhoff sizing algorithm, deter-
mine the best fit ellipsoidal shape and orientation that matches this
simulated data.
15.6. Consider a simulated planar scan (from two sides) of an ellipsoidal scatterer
in water where ða1 ; a2 ; a3 Þ ¼ ð20; 25; 30Þ mm, ðu1 ; u2 ; u3 Þ ¼ ix ; iy ; iz , and
the center of the scatterer is at ð3, 5, 6Þ mm (see Fig. P15.2). To apply the
TOFE sizing algorithm to this single medium problem the location of the
interface, xI, is taken as the location of the transducer itself, the unit vector
along the refracted ray, e, is just the orientation of the transducer, and H2 is
the distance from the transducer to where a planar wave front first touches the
scatterer. In Table P15.4 are given a set of transducer locations (in mm) and
orientations. The first 17 of these simulated values are taken when the
transducer is scanned over a plane on one side of the flaw, while the
remaining eight values are for planar scanning over a plane on the other
side (Fig. P15.2).We will use this simple test setup to illustrate some of the
properties of TOFE sizing.
Table P15.4 Simulated transducer locations and orientations for TOFE sizing
m xIm yIm zIm exm eym ezm
1 0.00 0.00 200.00 0.0000 0.0000 1.0000
2 119.00 0.00 200.00 0.5000 0.0000 0.8660
3 84.15 84.15 200.00 0.3536 0.3536 0.8660
4 0.00 119.00 200.00 0.0000 0.5000 0.8660
5 84.15 84.15 200.00 0.3536 0.3536 0.8660
6 119.00 0.00 200.00 0.5000 0.0000 0.8660
7 84.15 84.15 200.00 0.3536 0.3536 0.8660
8 0.00 119.00 200.00 0.0000 0.5000 0.8660
(continued)
15.7 Problems 679
xI
scanning plane 1
z e
iz
iy
ix y H2
circular
x
“crack”
xC = (3,4,–6) mm
Fig. P15.3 Simulated one-sided planar scanning of a 5 mm radius circular “crack” whose center
is at xc ¼ ð3, 5, 6Þ mm and whose normal is in the z-direction
(a) First, consider the case where time-of-flight measurements from the
transducer to the front-surface of the flaw and back are used to obtain
exact simulated distancemeasurements for H2 as shown in Table P15.5.
Using this full data set and the TOFE sizing algorithm, determine the
size and orientation of the best-fit ellipsoid. As an initial guess, take the
location of the center of the flaw at (0, 0, 0) and the C coefficients as Cxx
¼ Cyy ¼ Czz ¼ 10:0 mm2 , Cxy ¼ Cxz ¼ Cyz ¼ 0:0.
680 15 Model-Based Flaw Sizing
b) Using the same exact data as in part (a), again apply TOFE sizing but
now using only the first 17 “measurements” to simulate one-sided
scanning. Do you still get back the exact results?
c) Simulate the effects of unknown material variations by modifying the
exact Hm values as if there was an unknown 2 % random variation in
wave speed over the scanning aperture. Repeat the TOFE sizing using
only one-sided scanning (the first 17 values). Now do you get back the
exact results? Add a single data point from the other side (with the same
type of simulated error) and apply the sizing method again. Now, what
happens? Finally, repeat the sizing with the full data set (as modified by
wave speed variations).
Table P15.5 Simulated data for TOFE sizing
m H2m(mm)
1 176.00
2 208.56
3 206.74
4 206.57
5 208.86
6 211.56
7 212.39
8 211.57
9 210.27
10 263.71
11 260.75
12 260.18
13 263.75
14 267.95
15 268.75
16 267.25
17 265.75
18 380.66
19 376.59
20 375.50
21 380.26
22 385.86
23 386.39
24 384.16
25 382.71
Investigate the sensitivity of the TOFE sizing method to both systematic and
random errors.
15.8. As mentioned previously, another way to reduce the error sensitivity of the
TOFE sizing algorithm when trying to size volumetric flaws with one-sided
scanning is to reduce the “degrees of freedom” in the algorithm by limiting
the reconstruction to a more restricted shape. One can, for example, recon-
struct instead the best fit sphere of radius a that matches the scattering data.
This can be done by setting Cxx ¼ Cyy ¼ Czz ¼ C (where C ¼ a2 ) and Cxy
¼ Cxz ¼ Cyz ¼ 0 in the sizing algorithm and solving, via non-linear least
squares, for the four unknown parameters (xc, yc, zc, C). Using the same
one-sided scanning data (with simulated wave speed errors) as in problem
15.6, repeat the TOFE sizing algorithm to reconstruct the best-fit sphere. Are
your results acceptable?
References
1. M.G. Silk, A.M. Stoneham, J.A.G. Temple, The Reliability of Non-destructive Inspection
(Adam Hilger, Bristol, 1987)
2. A. Sedov, L.W. Schmerr, The time-domain elastodynamic Kirchhoff approximation for
cracks: the inverse problem. Wave Motion 8, 15–26 (1986)
3. D.K. Hsu, J.H. Rose, D.O. Thompson, Reconstruction of inclusions in solids using ultrasonic
Born inversion. J. Appl. Phys. 55, 162–168 (1984)
4. D.K. Hsu, J.H. Rose, R.B. Thompson, D.O. Thompson, Quantitative characterization of flaws
near a surface using inverse Born approximation. J. Nondestruct. Eval. 3, 183 (1982)
5. D.O. Thompson, S.J. Wormley, D.K. Hsu, An apparatus and technique for reconstruction of
flaws using model-based elastic wave inverse ultrasonic scattering. Rev. Sci. Instrum. 57,
3089–3098 (1986)
6. D.K. Hsu, D.O. Thompson, S.J. Wormley, Reliability of reconstruction of arbitrarily oriented
flaws using multiview transducers. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 34,
508–514 (1987)
7. L.W. Schmerr, A. Sedov, C.P. Chiou, A unified constrained inversion model for ultrasonic
flaw sizing. Res. Nondestruct. Eval. 1, 77–97 (1989)
8. C.P. Chiou, L.W. Schmerr, New approaches to model-based ultrasonic flaw sizing. J. Acoust.
Soc. Am. 92, 435–444 (1992)
9. J.H. Rose, R.K. Elsley, B.R. Tittman, V.V. Varadan, V.K. Varadan, Inversion of ultrasonic
scattering data, in Acoustic, Electromagnetic, and Elastic Wave Scattering—Focus on the
T-Matrix Approach, ed. by V.K. Varadan, V.V. Varadan (Pergamon Press, New York, 1980),
pp. 605–614
10. J.H. Rose, Elastic wave inverse scattering in nondestructive evaluation. Pure Appl. Geophys.
131, 715–739 (1989)
11. J. Yang, L.J. Bond, Ultrasonic technique for sizing voids by using area functions. IEEE Proc. A
Sci. Meas. Technol. 139, 45–50 (1992)
12. J. Yang, L.J. Bond, Errors in determining the flaw centroid by using area functions, in Review
of Progress in Quantitative Nondestructive Evaluation, vol. 11A, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1992), pp. 849–856
13. J.D. Young, Approximate image reconstruction from transient signature, in Acoustic, Electro-
magnetic, and Elastic Wave Scattering—Focus on the T-Matrix Approach, ed. by
V.K. Varadan, V.V. Varadan (Pergamon Press, New York, 1980), pp. 631–653
682 15 Model-Based Flaw Sizing
14. E.M. Kennaugh, D.L. Moffatt, Transient and impulse response approximations. Proc. IEEE 53,
898–901 (1965)
15. L.J. Bond, Born inversion: application to spherical and spheroidal voids, in Ultrasonics
International Conference Proceedings, 1989, pp. 768–773
16. S.J. Song, L.W. Schmerr, An ultrasonic time-of-flight equivalent flaw sizing method. Res.
Nondestruct. Eval. 4, 1–18 (1992)
17. J. Krautkramer, Determination of the size of defects by the ultrasonic impulse method.
Br. J. Appl. Phys. 19, 240–245 (1959)
18. Handbook on the Ultrasonic Examination of Welds (The Welding Institute, Cambridge, 1977)
19. H.J. Meyer, The international state of the art in nondestructive inspection of welds with special
emphasis on flaw characterization. Mater. Eval. 42, 793–802 (1984)
20. The Evaluation of Ultrasonic Signals (The International Institute of Welding, Cambridge,
1987)
21. M.G. Silk, Sizing crack-like defects by ultrasonic means, in Research Techniques in Nonde-
structive Testing, vol. 2, ed. by R.S. Sharpe (Academic Press, London, Chapter 2, 1977)
22. M.G. Silk, The use of diffraction based time-of-flight measurements to locate and size defects.
Br. J. NDT 26, 208–213 (1984)
23. G.J. Gruber, G.J. Hendrix, W.R. Schick, Characterization of flaws in piping welds using
satellite pulses. Mater. Eval. 42, 426–432 (1984)
24. J.P. Charlesworth, J.A.G. Temple, Engineering Applications of Ultrasonic Time-of-Flight
Diffraction (Research Studies Press, Somerset, 1989)
25. J.D. Achenbach, A. Norris, K. Viswanathan, Inversion of crack scattering data in the high
frequency domain, in Acoustic, Electromagnetic, and Elastic Wave Scattering—Focus on the
T-Matrix Approach, ed. by V.K. Varadan, V.V. Varadan (Pergamon Press, New York, 1980),
pp. 591–604
26. J.D. Achenbach, K. Viswanathan, A. Norris, An inversion integral for crack-scattering data.
Wave Motion 1, 299–316 (1979)
27. A. Norris, J.D. Achenbach, Inversion of first arrival crack scattering data, in Elastic Wave
Scattering and Propagation, ed. by V.K. Varadan, V.V. Varadan (Ann Arbor Science Pub-
lishers, Ann Arbor, 1982), pp. 61–75
28. B. Grohs, O.A. Barbian, W. Kappes, H. Paul, R. Licht, F.W. Hoh, Characterization of flaw
location, shape, and dimensions with the ALOK system. Mater. Eval. 40, 84–89 (1982)
29. J. Seydel, Ultrasonic synthetic-aperture focusing techniques, in Research Techniques in
Nondestructive Testing, vol. 6, ed. by R.S. Sharpe (Academic Press, London, 1982),
pp. 49–106
30. S.O. Harrold, Ultrasonic focusing techniques, in Research Techniques in Nondestructive
Testing, vol. 6, ed. by R.S. Sharpe (Academic Press¸ London, 1982), pp. 49–106
31. J.K. Cohen, N. Bleistein, The singular function of a surface and physical optics inverse
scattering. Wave Motion 1, 153–161 (1979)
32. C. Holmes, B.W. Drinkwater, P.D. Wilcox, Post-processing of the full matrix of ultrasonic
transmit–receive array data for non-destructive evaluation. NDT & E Int. 38, 701–711 (2005)
33. B.P. Hildebrand, B.B. Brenden, An Introduction to Acoustical Holography (Plenum Press,
New York, 1972)
34. C.F. Quate, A. Ataar, H.K. Wickramasinghe, Acoustic microscopy with mechanical scan-
ning—a review. Proc. IEEE 67, 1092–1114 (1979)
35. L.W. Schmerr, Fundamentals of Ultrasonic Phased Arrays (Springer, New York, 2014)
36. B.J. Engle, L.W. Schmerr, A. Sedov, Equivalent flaw time-of-flight diffraction sizing with
ultrasonic phased arrays, in Review of Progress in Quantitative Nondestructive Evaluation,
vol. 32B, ed. by D.O. Thompson, D.E. Chimenti (American Institute of Physics, Melville,
2013), pp. 1817–1824
37. L.W. Schmerr, B.J. Engle A. Sedov, X. Li, Ultrasonic flaw sizing—an overview, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 32A, ed. by D.O. Thompson,
D.E. Chimenti (American Institute of Physics, Melville, 2013), pp. 895–901
References 683
38. J.H. Rose, Exterior reconstruction of a three dimensional scatterer. Wave Motion 6, 149–154
(1984)
39. A. Sedov, L.W. Schmerr, Pulse distortion and the elastodynamic Kirchhoff approximation for
cracks. The direct and inverse problems. SIAM J. Appl. Math. 47, 1201–1215 (1987)
40. V.G. Kogan, D.K. Hsu, J.H. Rose, Characterization of flaws using the zeros of the real and
imaginary parts of the ultrasonic scattering amplitude. J. Nondestruct. Eval. 5, 57–67 (1985)
41. S.J. Song, L.W. Schmerr, New approaches to ultrasonic equivalent sizing for small flaws.
Nondestruct. Test. Eval. 10, 97–125 (1992)
42. C.P. Chiou, L.W. Schmerr, A neural network model for ultrasonic flaw sizing. Nondestruct.
Test. Eval. 10, 167–182 (1993)
43. H. Zhang, L.J. Bond, Ultrasonic scattering by spherical voids. Ultrasonics 27, 116–119 (1989)
44. C. Chaloner, L.J. Bond, Investigation of the 1-D inverse Born technique. IEE Proc. 134,
257–265 (1987)
45. K.C. Tam, Two-dimensional inverse Born approximation in ultrasonic flaw characterization.
J. Nondestruct. Eval. 5, 95–106 (1985)
46. J.H. Ros, J.L. Opsal, The inverse Born approximation. Exact determination of the shape of
convex voids, in Review of Progress in Quantitative Nondestructive Evaluation, vol. 2B,
ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1983), pp. 949–959
47. R.B. Thompson, Status of implementation of the inverse Born algorithm, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 4A, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1985), pp. 611–621
48. R.B. Thompson, K.M. Lakin, J.H. Rose, A comparison of the inverse Born and imaging
techniques for flaw reconstruction, in 1981 Ultrasonics Symposium Proceedings (IEEE
Press, Piscataway, 1981), pp. 930–935
49. L.J. Bond, C.A. Chaloner, S.J. Wormley, S.P. Neal, J.H. Rose, Recent advances in Born
inversion (weak scatterers), in Review of Progress in Quantitative Nondestructive Evaluation,
vol. 7A, ed. by D.O. Thompson, D.E. Chimenti (Plenum Press, New York, 1988), pp. 437–444
50. J.H. Rose, J.A. Krumhansl, Determination of flaw characteristics from ultrasonic scattering
data. J. Appl. Phys. 50, 2951–2952 (1979)
51. R.B. Thompson, T.A. Gray, Range of applicability of inversion algorithms, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 1, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1982), pp. 233–249
52. F. Cohen-Tenoudji, G. Quentin, B.R. Tittman, Elastic wave transformation, in Review of
Progress in Quantitative Nondestructive Evaluation, vol. 2B, ed. by D.O. Thompson,
D.E. Chimenti (Plenum Press, New York, 1982), pp. 961–974
53. G.J. Gruber, Defect identification and sizing by the ultrasonic satellite-pulse technique.
J. Nondestruct. Eval. 1, 263–275 (1980)
54. L. Adler, J.D. Achenbach, Elastic wave diffraction by elliptical cracks: theory and experiment.
J. Nondestruct. Eval. 1, 87–99 (1980)
55. S. Golan, Optimization of the ultrasonic crack tip ultrasonic diffraction technique for sizing of
cracks. Mater. Eval. 39, 166–169 (1981)
56. D. Heinrich, K.H. Mayer, G. Muller, W. Prestel, Manual and mechanized ultrasonic inspection
of large components in respect of flaw estimation by fracture mechanics. Nucl. Eng. Des. 112,
127–137 (1989)
57. C.F. Schueler, H. Lee, G. Wade, Fundamentals of digital ultrasonic imaging. IEEE Trans. Son.
Ultrason. SU-31, 195–217 (1984)
Chapter 16
Probability of Detection and Reliability
where V0 is the noise-free envelope of the received voltage signal, a quantity that
can be determined from an ultrasonic measurement model, and σ is the standard
deviation of the noise. For a low signal-to-noise ratio (SNR) this model reduces to a
Rayleigh noise distribution and in the case of high SNR it becomes a normal
(Gaussian) distribution centered around the noise-free signal, V0. If one established
a voltage threshold, T, such that when the signal amplitude goes above T it is
considered a flaw detection event, then the probability of detection is
ð
1
Similarly, the probability of a false indication, PFI, i.e. the probability that we
mistake the noise for a flaw, is
ð
1
ð
1
However, in most cases there are also design parameters, x, that are uncertain. Let
X be a random vector that characterizes those uncertain parameters and let pX(x) be
the joint probability density function for those uncertain parameters. Then, assum-
ing that V0 depends on both parameters, y, that have negligible uncertainty, and the
uncertain parameters, x, we have, in place of Eq. (16.4)
16.1 Probability of Detection (POD) Models 687
ð
ð 1
PODðyÞ ¼ f ðr; V 0 ðy; xÞÞpX ðxÞdr dx; ð16:5Þ
DðxÞ T
where D(x) is the domain that defines the range of the uncertain parameters. Gray and
Thompson [2] have used this type of approach. In particular, they examined the effect
of scan plan spacing (distance the transducer was moved between measurements) and
the radius of curvature of a fluid-solid interface in an immersion inspection on
the POD. They simulated the influence of uncertainty of the orientation of a crack
that was being interrogated by assuming a range of possible crack misalignments.
Figure 16.2 shows in diagram form how noise models, ultrasonic measurement
models, and variability models can be combined to form an estimate of POD.
As can be seen from that diagram, one of the issues that must be addressed is how to
model the noise and signal-plus noise distributions to obtain the distribution of
gated peak-to-peak time domain signals that are typically used to determine detect-
ability in ultrasonic inspections [3]. Margetan and others [4–6] have examined this
issue and developed some methods for modeling these peak-to-peak distributions
when studying grain distribution noise.
The noise in a POD model can arise from a number of sources. Electronic noise, in
particular, can be present in the electrical and electromechanical components. If
one has ideal noise-free models of those components, as obtained, for example, by
Schmerr and Song [7], then this type of noise can be added to those components by
following the procedures outlined by Oakley [8], who considered the specific case
of electronic noise in the receiver.
Roughness of the surfaces involved in an ultrasonic inspection can also generate
noise in the received signals. Ogilvy [9, 10] has described in detail methods that can
model such noise. The Kirchhoff approximation, discussed in Chap. 10 for model-
ing flaw signals, can also be used to model these roughness effects [9]. Ogilvy
obtained one simple result with the Kirchhoff approximation [10] when modeling
the effects of flaw surface roughness in a pulse echo setup, where a specular
response is being measured (e.g. the near-normal response of a crack). In such
cases the far field scattering amplitude for a rough flaw, Arough(ω), is related to that
of the smooth flaw, Asmooth(ω), by the relationship
Arough ðωÞ ¼ Asmooth ðωÞexp 2k2 σ 2R cos 2 θ0 ; ð16:6Þ
where k is the wave number, σ R is the RMS surface roughness, and θ0 is the angle of
incidence relative to the defect normal.
R.B. Thompson and his co-workers have also done extensive work on modeling
the noise due material microstructure (grain scattering) [11–13]. These studies have
led to the definition of a quantity called the noise figure of merit, FOM(ω), which is
a measure of a material’s capacity to generate noise [14–17]. This FOM(ω) can
either be measured with experiments or related analytically to the properties of the
material microstructure. For example, for a P-wave traveling in the x3-direction the
FOM is given by [3]
ð
k4 0
½FOMðωÞ2 ¼
2 hδC33 ðxÞδC33 ðx Þiexpð2iks3 ÞdV ðsÞ; ð16:7Þ
4πρc2p
where k is the wave number, ρ is the density of the solid (assumed here to be
independent of position), cp is the compressional wave speed, and the braces h . . . i
16.2 Reliability Modeling 689
denote an ensemble average that only depends on s ¼ x x0 [3]. The quantity in the
braces in Eq. (16.7) represents a two point correlation of the elastic constants that
has been studied in depth, in particular, for titanium alloys [17]. If the FOM of a
specimen is known, either by experimental determination or by evaluating
Eq. (16.7), then one can estimate the grain noise distribution and place that noise
estimate into the procedure diagram of Fig. 16.2 to ultimately obtain the gated peak-
to-peak noise and signal-plus-noise distributions shown as outputs in that figure.
Models are versatile tools that can examine issues such as the effect of scan plans,
choice of instrumentation, etc. on POD values, but there are some sources of
variability, such as material variations and flaw morphology, for example, that
may be impossible to estimate and other uncontrollable (and difficult to character-
ize) parameters in NDE inspections. Meeker and his co-workers [18, 19] have
defined a method for combining model-based simulations of system variability
with statistical methods that can quantify and adjust for systematic biases and
sources of variability that are not accounted for by the model. Having such a
methodology is important to make POD modeling activities more practical and
for helping to demonstrate to the NDE community that models can play an
important role in the determination of POD. In fact, a model-assisted POD
(MAPOD) working group was formed [20] specifically to leverage the use of
physics-based models in POD studies.
Since nondestructive evaluation plays a key role in guaranteeing safety and reli-
ability, modeling of inspections and connecting those NDE models to associated
reliability models is also an important task. This section will briefly outline how
NDE models and reliability models can be combined and the effects of NDE
inspections on probability of failure can be rationally estimated.
ð
þ1 ð
þ1
where
0 : g ð xÞ > 0
I ð gð xÞ Þ ¼ ð16:10Þ
1 : g ð xÞ 0
and fX(x) is the joint probability density function of the random variables. Nor-
mally, g(x) is described implicitly through various algorithmic procedures. For
example, for fatigue failure problems this may include finite element stress analyses
and flaw growth analyses. Because such procedures may be computationally
intensive, a key issue when determining reliability is how one determines the
probability of failure expressed by Eq. (16.9). Generally there are three types of
methods that are used: Monte Carlo simulation (MCS) methods, response surface
methods (RSM), and first or second order reliability methods (FORM), (SORM) [21].
The relative accuracy and speed of these three approaches follows the trend line
shown in Fig. 16.3. The Monte Carlo approach numerically evaluates the integrals of
Eq. (16.9) and so has a high computational cost but in principle has the most accuracy
of these methods. The response surface method approximates the limit state function
in terms of relatively simple functions (polynomials, polyhedrals, etc.), an approach
that is most effective when the number of random variables is small. The FORM and
SORM methods assume that one can transform the given problem into one involving
a set of variables, y, whose distributions are just standardized normal distributions. In
this y-space, then one determines, using linear programming techniques, the point on
the limit state surface gðyÞ ¼ 0 that is closest to the origin (usually called the design
point). The distance to that point is called the safety index (or reliability index), β, i.e.
!1=2
X
n
1=2
β ¼ min y2i ¼ minðy yT Þ
i¼1
subject to gðyÞ ¼ 0: ð16:11Þ
The limit state surface is then approximated locally near the design point as either its
tangent hyperplane (FORM method) or hyper-paraboloid (SORM method). In the
case of the FORM method, the integrations in Eq. (16.9) yield the simple result [22]
Pf ¼ ΦðβÞ; ð16:12Þ
where Φ is the cumulative probability derived from the standard normal distribu-
tion function.
When inspections are used to help guarantee safety and reliability one must
characterize the effects of those inspections on the probability of failure. This can
be done by the use of Bayesian updating [22–25]. As an example, consider the case
of fatigue failure where we follow the growth of a crack to fracture. In this example,
we take the crack growth law as the Paris-Erdogan law, i.e.
da
¼ CΔK m ; ð16:13Þ
dN
where a is the crack length, N is the number of fatigue cycles, and C and m are
material parameters. The variable ΔK is the stress intensity range, where ΔK is
normally given in fracture mechanics theory in the form
pffiffiffiffiffi
ΔK ¼ Fða; GÞΔσ πa; ð16:14Þ
where Δσ is the range of tensile stresses, F is a function that accounts for the effects
of the geometry on the stress intensity factor, and G is a vector of parameters that
define those geometric effects. Integrating Eq. (16.13) from crack size a1 to crack
size a2 corresponding to the number of cycles, N1 and N2, respectively, gives
að2
da
pffiffiffiffiffi m ¼ CΔσ m ðN 2 N 1 Þ; ð16:15Þ
½Fða; GÞ πa
a1
where Δσ m is a mean stress range. Equation (16.15) can also be written in terms of a
damage accumulation function, Ψ(a1, a2), given by
að2
da
Ψ ð a1 ; a 2 Þ ¼ pffiffiffiffiffi m ð16:16Þ
½Fða; GÞ πa
a1
692 16 Probability of Detection and Reliability
as
This damage accumulation function can be used together with a Bayesian updating
procedure to examine the effects of an inspection on fatigue failure. We will
consider here the case where a crack is not detected during a single inspection
after Nd cycles with a measurement having the capability of detecting a flaw of size
ad. This event of not detecting the crack can be described in terms of the damage
accumulation function as
where a0 is some initial flaw size at N ¼ N 0 cycles. Similarly, if the critical flaw size
at failure is ac, we can define the event of a fatigue failure as
In terms of our previous discussion of reliability we can take our limit state function
here as
However, with Bayes theorem we can estimate the updated probability of failure
given that there was an inspection at N ¼ N d and no flaw was detected as
P fL 0 \ I 0 g
Pupd
f ð N Þ ¼ P a ð N Þ > a c a ð N d Þ < ad ¼ , N > Nd :
PfI 0g
ð16:22Þ
The terms on the right side of Eq. (16.22) can then be estimated according to the
various reliability schemes discussed previously. A similar procedure can be
followed for the case where a flaw is detected or for multiple inspections at different
times, but we will not show those extensions here. They are discussed in depth
in [22].
Although Eq. (16.22) shows the manner in which one can in principle calculate
the influence of inspections on reliability, it does not show explicitly how the
capability of the NDE inspection affects this updated probability of failure. To
see this connection one needs to examine the effects of an inspection on the
underlying probability distributions [23, 24]. Since the details are lengthy we will
only quote the end result here for the case we have just considered, namely
16.2 Reliability Modeling 693
ð c x¼1
a¼a ð
1
Pupd
f ðN Þ ¼1 f A a; N x; N d f1 PODðxÞgf A ðx; N d Þdx da , N > N d ;
Kd
a¼0 x¼0
ð16:23Þ
where
ð
1
and
ð
1
f A ða; N Þ ¼ f A a; N x; N 0 f A0 ðxÞdx: ð16:25Þ
0
Here f A a; N x : N τ is the probability of flaw sizes at N > N τ conditional on a
ðN τ Þ ¼ x and f A0 ðxÞ is the initial distribution of flaw sizes at N ¼ N 0 . Equations
(16.23)–(16.25) show that the probability of detection versus flaw size curve, POD
(a), is the key inspection quantity that affects the updated reliability. The other
key
quantities are the initial flaw distribution, f A0 ðxÞ, and the distributions f a; N x; N τ
ðτ ¼ N 0 , N d Þ which can be obtained with the aid of the flaw growth law. Thus, POD
is indeed a fundamental quantity that can define the capabilities of an NDE
inspection, both in terms of the ability to detect flaws and to estimate the effects
of that inspection on reliability.
There are two software packages that have been developed commercially to
combine reliability and NDE inspection capabilities. The Profast system was
developed in Norway by the company Det Norske Veritas (DNV) for simulating
the probabilistic inspections of off-shore structures (http://www.dnv.jp/services/
software/products/sesam/sesamgenie/profast.asp) such as oil platforms that are
subject to fatigue failure. It is based on the reliability program Proban which
takes the FORM/SORM approach to evaluating reliability [25]. It uses 1-D and
2_D crack growth models, using the Paris-Erdogan law and allows for the definition
of inspection plans at various “hot-spots” in the structure. It can estimate the
influence on those inspections on reliability and examine various inspection sched-
ules. The software package DARWIN (Design Assessment of Reliability With
INspections) was developed at Southwest Research Institute (http://www.swri.
org/4org/d18/mateng/probfrac/default.htm) under funding from the Federal Avia-
tion Administration. This package is a probabilistic damage tolerance design code
that can estimate the risk of fracture of turbine engine rotor disks containing
material flaws. DARWIN integrates finite element stress analysis, fracture mechan-
ics analysis, nondestructive inspection capabilities (through a POD(a) curve) and
probabilistic reliability analyses to determine the effects of inspection on the
probability of failure. The probabilistic reliability analysis is done with a Monte
Carlo methods coupled with fast integration routines.
694 16 Probability of Detection and Reliability
There are a few overview papers and specific application papers on the role that
modeling plays in POD evaluation [26–30]. See also the MAPOD working group
papers [20] for discussions of many of the issues involved in bringing models into
the determination of POD. For books that describe the major aspects of structural
safety and reliability methods see [31, 32].
References
1. S.O. Rice, Mathematical analysis of random noise. Bell Syst. Tech. J. 24, 46–156 (1945)
2. T.A. Gray, R.B. Thompson, Use of models to predict ultrasonic NDT reliability, in Review of
Progress in Quantitative Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti,
vol. 5A (Plenum Press, New York, 1986), pp. 911–918
3. R.B. Thompson, F.J. Margetan, Use of elastodynamic theories in the stochastic description of
the effects of microstructure on ultrasonic flaw and noise signals. Wave Motion 36, 347–365
(2002)
4. F.J. Margetan, I. Yalda, R.B. Thompson, Predicting gated peak-to-peak grain noise distribu-
tions for ultrasonic inspection of materials, in Review of Progress in Quantitative Nondestruc-
tive Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 15B (Plenum Press, New York,
1996), pp. 1509–1516
5. I. Yalda-Mooshabad, F.J. Margetan, R.B. Thompson, Applying the 2D random walk formal-
ism to predict ultrasonic grain noise distributions, in Review of Progress in Quantitative
Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 13B (Plenum Press,
New York, 1994), pp. 1713–1720
6. I. Yalda, F.J. Margetan, R.B. Thompson, Survey of ultrasonic grain noise characteristics in jet
engine titanium, in Review of Progress in Quantitative Nondestructive Evaluation, ed. by
D.O. Thompson, D.E. Chimenti, vol. 15B (Plenum Press, New York, 1996), pp. 1487–1494
7. L.W. Schmerr, S.-J. Song, Ultrasonic Nondestructive Evaluation Systems—Models and Mea-
surements (Springer, New York, 2007)
8. C.G. Oakley, Calculation of ultrasonic transducer signal-to-noise ratios using the KLM model.
IEEE Trans. Ultrason. Ferroelectr. Freq. Control 44, 1018–1026 (1997)
9. J.A. Ogilvy, Theory of Wave Scattering from Random Rough Surfaces (Institute of Physics,
Bristol, 1991)
10. J.A. Ogilvy, Model for predicting ultrasonic pulse-echo probability of detection. NDT E Int.
26, 19–29 (1993)
11. F.J. Margetan, R.B. Thompson, Microstructural noise in titanium alloys and its influence on
the detectability of hard-alpha inclusions, in Review of Progress in Quantitative Nondestruc-
tive Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 11B (Plenum Press, New York,
1992), pp. 1717–1724
12. F.J. Margetan, R.B. Thompson, I. Yalda-Mooshabad, Modeling ultrasonic microstructural
noise in titanium alloys, in Review of Progress in Quantitative Nondestructive Evaluation,
ed. by D.O. Thompson, D.E. Chimenti, vol. 12B (Plenum Press, New York, 1993),
pp. 1735–1742
13. F.J. Margetan, I. Yalda-Mooshabad, R.B. Thompson, Estimating ultrasonic signal-to-noise
ratios for inspection of cylindrical billets, in Review of Progress in Quantitative Nondestruc-
tive Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 13B (Plenum Press, New York,
1994), pp. 1737–1744
References 695
14. J.H. Rose, Ultrasonic backscattering from polycrystalline aggregates using time-domain linear
response theory, in Review of Progress in Quantitative Nondestructive Evaluation, ed. by
D.O. Thompson, D.E. Chimenti, vol. 10B (Plenum Press, New York, 1991), pp. 1715–1720
15. J.H. Rose, Ultrasonic backscatter from microstructure, in Review of Progress in Quantitative
Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 11B (Plenum Press,
New York, 1992), pp. 1677–1684
16. J.H. Rose, Theory of ultrasonic backscatter for multiphase polycrystalline solids, in Review of
Progress in Quantitative Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti,
vol. 12B (Plenum Press, New York, 1993), pp. 1719–1726
17. K.Y. Han, R.B. Thompson, Ultrasonic backscattering in duplex microstructures: theory and
application to titanium alloys. Metall. Trans. 28A, 91 (1997)
18. W.Q. Meeker, V. Chan, R.B. Thompson, C.P. Chiou, A methodology for predicting probabil-
ity of detection for ultrasonic testing, in Review of Progress in Quantitative Nondestructive
Evaluation, ed. by D.O. Thompson, D.E. Chimenti, vol. 20B (American Institute of Physics,
Melville, 2001), pp. 1972–1978
19. W.Q. Meeker, S.L. Jeng, C.P. Chiou, R.B. Thompson, Improved methodology for inspection
reliability assessment for detecting synthetic hard alpha inclusions in titanium, in Review of
Progress in Quantitative Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti,
vol. 17B (Plenum Press, New York, 1998), pp. 2061–2068
20. http://www.cnde.iastate.edu/mapod/
21. R.E. Melchers, Structural Reliability Analysis and Prediction, 2nd edn. (Wiley, New York,
1999)
22. Z. Zhao, A. Haldar, Reliability-based structural fatigue damage evaluations and maintenance
using non-destructive inspections, in Uncertainty Modeling in Finite Element, Fatigue, and
Stability Analysis of Systems, ed. by A. Haldar, A. Guran, B.M. Ayyub (World Scientific,
Singapore, 1997), pp. 159–214
23. K. Sobczyk, B.F. Spencer Jr., Random Fatigue—From Data to Theory (Academic, San Diego,
1992)
24. B.F. Spencer, Stochastic diffusion models for fatigue crack growth and reliability estimation,
in Stochastic Approach to Fatigue: Experiments, Modeling, and Reliability Estimation, ed. by
K. Sobczyk (Springer Verlag, New York, 1993), pp. 185–241
25. H.O. Madsen, Fatigue reliability of marine structures, in Stochastic Approach to Fatigue:
Experiments, Modeling, and Reliability Estimation, ed. by K. Sobczyk (Springer Verlag,
New York, 1993), pp. 243–301
26. R.B. Thompson, T.A. Gray, Use of ultrasonic models in the design and validation of new NDE
techniques. Philos. Trans. R. Soc. A320, 329–340 (1986)
27. T.A. Gray, F. Amin, R.B. Thompson, Application of ultrasonic POD models, in Review of
Progress in Quantitative Nondestructive Evaluation, ed. by D.O. Thompson, D.E. Chimenti,
vol. 7B (Plenum Press, New York, 1988), pp. 1737–1744
28. F. Amin, T.A. Gray, F.J. Margetan, Ultrasonic POD model validation and development for
focused probes, in Review of Progress in Quantitative Nondestructive Evaluation, ed. by
D.O. Thompson, D.E. Chimenti, vol. 9A (Plenum Press, New York, 1990), pp. 869–879
29. P. Wagner, V. Durbec, B. Nouailhas, Probability of detection of artificial flaws by a statistical
analysis of the ultrasonic testing parameters, in Proceedings of the 14th World Conference on
Nondestructive Testing (Oxford and IBH Publishing, New Delhi, 1996) pp. 953–956
30. R.B. Thompson, Using physical models of the testing process in the determination of proba-
bility of detection. Mater. Eval. 59, 861–865 (2001)
31. H.O. Madsen, S. Krenik, N.C. Lind, Methods of Structural Safety (Prentice-Hall, Englewood
Cliffs, 1986)
32. O. Ditlevsen, H.O. Madsen, Structural Reliability Methods (Wiley, New York, 1996)
Appendix A: The Fourier Transform
In Chap. 2 it was shown that the Fourier transform plays an important role in the
analysis of LTI systems. This appendix summarizes some of the important proper-
ties of Fourier transforms and their numerical calculation.
A function f(t) and its Fourier transform, F(ω), are related by the Fourier
transform pair [1, 3]:
ð
þ1
F ð ωÞ ¼ f ðtÞexpðiωtÞdt; ðA:1Þ
1
ð
þ1
1
f ðt Þ ¼ FðωÞexpðiωtÞ dω: ðA:2Þ
2π
1
2. Shifting
f ðt t0 Þ $ FðωÞexpðiωt0 Þ
3. Scaling
4. Integration
ðt
F ð ωÞ
f ðtÞdt $ πFð0ÞδðωÞ
iω
1
ðω
f ðtÞ
πf ð0ÞδðtÞ þ $ FðωÞdω
it
1
5. Conjugation
f * ðtÞ $ F* ðωÞ
f * ðtÞ $ F* ðωÞ where ða þ ibÞ* ¼ ða ibÞ
6. Convolution
ð
þ1
7. Correlation
ð
þ1
ð
þ1
8. Parsevals’ theorem
ð
þ1 ð
þ1
and when f 1 ¼ f 2 ¼ f
ð
þ1 ð
þ1
f ðtÞ F ð ωÞ
δðtÞ 1
3. Sign function
(
1 t>0 2i
sgnðtÞ ¼
1 t<0 ω
2
4. jtj
ω2
1
5. iπ sgnðωÞ
t
a þ iω
6. expðatÞHðtÞ
a2 þ ω 2
7. Laplace function
2a
expðajtjÞ
a 2 þ ω2
700 Appendix A: The Fourier Transform
8. Gauss function
pffiffiffi
π
exp at2 exp ω2 =4a
a
9. Rectangular function
(
1 jtj < 1=2 sin ðω=2Þ
ΠðtÞ ¼ ¼ sincðω=2π Þ
0 jtj > 1=2 ω=2
10.
sin ðatÞ
Π ðω=2aÞ
πt
The typical output of an ultrasonic system is an analog voltage versus time wave
form. Most modern NDE measurement systems digitize such wave forms so that the
sampled values can be manipulated by computer. In this section we will examine
the consequences of dealing with a sampled signal and consider the way in which
the Fourier transform can be computed from these samples in an efficient manner.
To begin, we rewrite the Fourier transform (and its inverse) in terms of the
frequency f (in Hertz) instead of the circular frequency ω:
ð
þ1
X ðf Þ ¼ xðtÞexpð2πi f tÞ dt
1
ðA:4Þ
ð
þ1
Now we consider the discrete (sampled) values of x(t) at the times tj ¼ jΔt where
j ¼ 0, 1, 2, . . . and we let f s ¼ 1=Δt. Then these sampled values are given by
ð
þ1
x tj ¼ Xðf Þexpð2πi j f =f s Þ df ; ðA:5Þ
1
Appendix A: The Fourier Transform 701
which can be written equivalently as an infinite sum over a set of finite intervals as
ðkþ1
ð Þf s
X
1
x tj ¼ Xðf Þexpð2πi j f =f s Þ df : ðA:6Þ
k¼1
k fs
Since the exponential function in Eq. (A.6) is periodic in f with period fs,
the expression in Eq. (A.6) is unchanged if we replace expð2πi j f =f s Þ by
expð2πi jðf kf s Þ=f s Þ. However,
ðkþ1
ð Þf s
Xðf Þexpð2πi jðf kf s Þ=f s Þ df
kf s
ðA:7Þ
ðf s
¼ Xðu þ kf s Þexpð2πi j u=f s Þ du;
0
1 ðs
f
X
x tj ¼ Xðu þ kf s Þexpð2πi u=f s Þ du: ðA:8Þ
k¼1
0
X
k¼þ1
Xp ð f Þ ¼ Xðf þ kf s Þ: ðA:9Þ
k¼1
Then
ðf s
x tj ¼ Xp ð f Þexpð2πij f =f s Þ df : ðA:10Þ
0
Since Xp( f ) is a periodic function in f with period fs, it can be expanded in a Fourier
series as
X
1
Xp ð f Þ ¼ Cj expð2πij f =f s Þ; ðA:11Þ
j¼1
ðf s
Cj ¼ 1=f s Xp ð f Þexpð2πij f =f s Þ df ðA:12Þ
0
702 Appendix A: The Fourier Transform
Recognizing these Fourier coefficients as just x(tj)/fs from Eq. (A.11), we have
X
1
Xp ð f Þ ¼ 1=f s x tj expð2πij f =f s Þ: ðA:13Þ
j¼1
Thus, knowing all the sampled values of x(t), Eq. (A.13) shows that we recover
Xp( f ) rather than X( f ) itself. The function Xp( f ) differs from our original X( f ) by
the sum of X( f )’s displaced at multiples of fs (Fig. A.1). This difference is referred
to as aliasing. If X( f ) is band limited, i.e. X( f ) ¼ 0 for j f j > f max where fmax is the
maximum frequency in the signal, then we see that Xp ð f Þ ¼ Xð f Þ for j f j < f max if
f s
2 f max (Fig. A.2). Even when X( f ) is not band limited, if fmax represents the
b
Xp ( f )
f
– fs fs
f
– fmax fmax
b
Xp ( f )
f
– fs – fmax fmax fs
Appendix A: The Fourier Transform 703
f s 2f max ; ðA:14Þ
which is the well-known Nyquist sampling criterion. This criterion says that in
order to recover a true replica of the frequency spectra of our original signal in the
frequency range j f j < f max , we should sample our signal at least twice the value of
the maximum significant frequency in that signal. Since the Nyquist criterion only
gives a lower bound for the sampling frequency, in practice we usually sample a
signal at a much higher rate such as 5f max ! 10f max to be conservative.
Now, suppose that we sample Xp( f ) in Eq. (A.13) at f ¼ nΔf , n ¼ 0, 1,
2, . . . where Δf ¼ 1=T and T is the total time interval sampled. Then
X
1
Xp ð f n Þ ¼ 1=f s x tj expð2πi jn=Tf s Þ: ðA:15Þ
j¼1
However, if we take Tf s ¼ N where N is an integer and use the fact that exp(2πijn/
N ) is periodic in j with period N, we can rewrite Xp( fn) as
X
N1
Xp ð f n Þ ¼ T=N xp tj expð2πijn=N Þ; ðA:16Þ
j¼0
where
X
1
xp t j ¼ x tj þ kT ðA:17Þ
k¼1
X
1
xp ðt Þ ¼ xðt þ kT Þ; ðA:18Þ
k¼1
which is formed from x(t) in the same manner as Xp( f ) is formed from X( f ).
Eq. (A.16) is in the form of an inverse discrete Fourier transform [2] that relates the
sampled values of Xp( f ) to the sampled values of xp(t). If aliasing is negligible, then
Eq. (A.16) is also a relation between the sampled values of our original wave form
x(t) and its frequency components X( f ).
Given the sampled values Xp( f ), we can also recover xp(t) via an inverse
relationship similar to Eq. (A.16). To see this consider the sum
X
N 1 X
N 1 X
N 1
Xp ð f n Þexpð2πi kn=N Þ ¼ T=N xp tj expð2πi ðk jÞn=N Þ: ðA:19Þ
n¼0 n¼0 j¼0
704 Appendix A: The Fourier Transform
However,
X
N 1
1=N expð2πi ðk jÞn=N Þ ¼ Δð j kÞ; ðA:20Þ
n¼0
where Δðj kÞ is the discrete delta function, having a sampling property similar to
the Dirac delta function (see problem A.1), i.e.
X
N 1
cj Δ ð j k Þ ¼ ck k ¼ 0, 1, 2, . . . N 1 ðA:21Þ
j¼0
Thus, we find
X
N 1
T xp ðt k Þ ¼ Xp ð f n Þexpð2πi kn=N Þ; ðA:22Þ
n¼0
which is the desired inversion formula. In the absence of aliasing, the discrete
forward Fourier transform pair (Eq. (A.22) and Eq. (A.16)) give us an explicit way
to calculate sampled values of the Fourier transform X( f ) directly from the sampled
values of the wave form x(t) and vice versa. However, for large N, this may still
involve a considerable amount of computation. Fortunately, there is a way to reduce
the computation required as discussed in the next section.
A direct evaluation of the discrete Fourier transforms in Eqs. (A.16) and (A.22)
requires on the order of ðN 1Þ2 multiplications, a value which grows rapidly as
N increases (see Table A.1). In 1965, however, Cooley and Tukey [5] published an
algorithm that can reduce the number of multiplications required to an order of
N/2 log2N, a value that grows much slower (Table A.1). This algorithm, now known
as the fast Fourier transform (FFT) [2], has made the numerical calculation of
Fourier integrals and their inverses practical.
Many software packages, such as MATLAB, MathCad, and Mathematica,
for example, compute the discrete Fourier transforms with an FFT algorithm.
One must be aware of some subtle differences between those various
implementations. In this book we have defined the forward and inverse Fourier
transforms by the relations in Eqs. (A.1) and (A.2). However, we can also define
these pair of equations in the forms
ð
þ1
~ ð ωÞ ¼ n 1
F f ðtÞexpðiωtÞ
1
ðA:23Þ
ð
þ1
f ð t Þ ¼ n2 ~ ðωÞexpðiωtÞ
F
1
as long as the constants (n1, n2) satisfy n1 n2 ¼ 1=2π. Different choices will obvi-
ously lead to different values for the Fourier transform which is why we have placed
a tilde over the transform in Eq. (A.23). In analytical discussions of wave
propagation, one must be especially aware of the signs chosen in Eq. (A.23),
since this can alter the physical meaning of the result. In Chap. 4, for example,
we expressed a 1-D plane wave traveling in the + x-direction by its Fourier
transform (see Eq. (4.5)) as
ð
þ1
1
f ðt x=cÞ ¼ FðωÞexp½iωðx=c tÞdω; ðA:24aÞ
2π
1
where
ð
þ1
F ð ωÞ ¼ f ðtÞexpðiωtÞdt: ðA:24bÞ
1
where the Fourier transform is now the complex conjugate of the Fourier transform
appearing in Eq. (A.24a) since
ð
þ1
~ ð ωÞ ¼
F f ðtÞexpðiωtÞdt ¼ F* ðωÞ: ðA:26Þ
1
706 Appendix A: The Fourier Transform
In discussing harmonic waves some authors will implicitly assume one of these sign
choices without stating them explicitly. For example, a plane wave traveling in the
+x-direction may be written as AðωÞexpðikxÞ. This means that the author is
implicitly using the form of Eq. (A.25) and the complex amplitude, A(ω), will be
the complex conjugate of the same amplitude obtained when using our definitions
of Eqs. (A.1) and (A.2). Differences due to different definitions can also arise in the
case of the discrete Fourier transform and its implementation. In our case, we
derived the discrete Fourier transform and its inverse directly from Eqs. (A.1) and
(A.2), obtaining Eqs. (A.16) and (A.22), which we rewrite here as:
X
N 1
Xp ð f n Þ ¼ Δt x tj expð2πi jn=N Þ
j¼0
ðA:27Þ
1 X
N 1
xp ðtk Þ ¼ Xp ð f n Þexpð2πi kn=N Þ:
NΔt n¼0
X
N 1
~ p ð f n Þ ¼ m1
X x tj expð2πi jn=N Þ
j¼0
ðA:28Þ
X
N 1
xp ðt k Þ ¼ m 2 ~ p ð f n Þexpð2πi kn=N Þ
X
n¼0
as long as the constants (m1, m2) satisfy m1 m2 ¼ 1=N. In this book, we will
exclusively use Eqs. (A.1) and (A.2) for the Fourier transform pair and
Eq. (A.27) for the discrete Fourier transform pair. Numerical evaluation of the
discrete Fourier transform pair will be performed in MATLAB, but the built-in
MATLAB functions fft and ifft do not implement Eq. (A.27). Instead these func-
tions are defined as
X
N
fftðxÞ ! X ð nÞ ¼ xð jÞexp½2πi ðj 1Þðn 1Þ=N
j¼1
ðA:29Þ
1X N
ifftðXÞ ! x ðk Þ ¼ XðnÞexp½2πi ðn 1Þðk 1Þ=N :
N n¼1
We see that the signs have been changed on the exponentials from our definitions
and that the sampling interval, Δt, is missing in the coefficients. Also, the indexing
is shifted by one so the sum ranges from one to N rather than zero to N-1.
To perform the Fourier transform pair of relations in MATLAB using our defini-
tions (but keeping the same indexing as MATLAB), we have therefore defined two
new MATLAB functions, FourierT(x, Δt) and IFourierT(X, Δt) where x is a vector
Appendix A: The Fourier Transform 707
Using s_space in conjunction with FourierT and IFourierT gives you the tools
necessary to perform time and frequency domain evaluations consistent with the
Fourier transform pair definitions used throughout this book. For additional infor-
mation on Fourier analysis see [4].
Appendix A: The Fourier Transform 709
A.5 Problems
From this definition, the sampling property of the discrete delta function
(Eq. (A.21)) follows directly. Show that
X
N 1
Δðk nÞ ¼ 1=N expð2πi mðk nÞ=N Þ
m¼0
i.e. prove that Eq. (A.20) is valid. (Hint: show that the series can be summed
explicitly to the value
wNðknÞ 1 = wðknÞ 1 ;
where w ¼ expð2πi=N Þ.
A.2. Consider the wave form given by
(
et t>0
x ðt Þ ¼ ;
0 t<0
References
In Chap. 2 we saw that the delta function also plays an important role in determin-
ing the response of LTI systems since it acts as an ideal input function whose
Fourier transform contains all frequencies equally weighted. Strictly speaking the
delta function is not an ordinary function but is a distribution [1, 2]. However, we
will continue to speak of δ(t) as an ordinary function, remembering that δ only has a
meaning in terms of its sifting property, i.e.
ðb
δðt xÞf ðxÞdx ¼ f ðtÞ a<t<b
a ðB:1Þ
¼0 t<a or t>b
¼ f ðtÞ=2 t¼a or t ¼ b:
The delta function also plays an important role in dealing with integral equations
and boundary value problems, as discussed in Chap. 5 [3].
Several of the more useful properties of the delta function are summarized below,
given the real constants a, b, c, and A and the function f(t):
1. Scaling properties
t c
δ ¼ jbjδðt cÞ
b
δðtÞ ¼ δðtÞ
X δðt tk Þ 0
δ½f ðtÞ ¼ f ðtk Þ ¼ 0, f ¼ df =dt
k
j f 0 ðtk Þj
f ðtÞδðt cÞ ¼ f ðcÞδðt cÞ
δðtÞδðt cÞ ¼ 0 c 6¼ 0
δðt cÞδðt cÞ is not defined
3. Integrals
ð
þ1
Aδðt cÞdt ¼ A
1
ð
þ1
δ ðt c Þ ¼ 1
1
ðt
δðu cÞdu ¼ H ðt cÞ where HðtÞ is the unit step function:
1
8
>
> 1 t>0
<
H ðtÞ ¼ 1=2 t¼o
>
>
:
0 t<0
4. Derivatives
ðb
f ðtÞδðnÞ ðt cÞ dt ¼ ð1Þn f ðnÞ ðcÞ a < c < b, f ðnÞ ¼ dn f =dtn
a
ð
þ1
1¼ δðtÞexpðiωtÞdt
1
ð
þ1
1
δðtÞ ¼ expðiωtÞdω
2π
1
Appendix B: The Dirac Delta Function 713
References
This Appendix will give a brief discussion of indicial and vector notation and
review some of the basic concepts of motion, mass, and stress/strain needed for our
consideration of the equations governing elastic media. Continuum mechanics texts
[1–6] are good sources for a more complete description of these topics.
The equations involved in elasticity problems are often algebraically quite com-
plex. The use of Cartesian tensor (index) notation significantly simplifies the
expression and manipulation of such equations.
In index notation, the Cartesian coordinates (x, y, z) will be denoted by the
subscripted variables (x1, x2, x3) (Fig. C.1) and the corresponding unit base vectors
by the vectors (e1, e2, e3). An arbitrary three-dimensional vector u with Cartesian
components (u1, u2, u3) can thus be written as
X
3
u¼ ui e i : ðC:1Þ
i¼1
If we adopt the Einstein summation convention, where the appearance twice of the
same index symbol implies summation over the range of that index, Eq. (C.1) can
be written simply as
u ¼ ui ei : ðC:2Þ
In Eq. (C.2), the repeated index i is merely a “dummy” index since it is “summed
out”. We can replace any such dummy index symbol by any other symbol without
e2
e3 x (x1)
e1
z (x3)
changing the meaning of the equation. Thus, for example, the following expressions
are all equally valid:
f ¼ ap b p ¼ a q b q ðC:3aÞ
u ¼ ui e i ¼ uk e k ðC:3bÞ
In Eq. (C.3c), the j and k indices are dummy indices that assume all values in the
range (1, 2, 3) while the i index is a “free” index which assumes only any one of
these values at a time. In some cases, as in Eq. (C.3c), we will show the range of the
free indices explicitly. Otherwise, the range will be assumed to be implicitly given
by (1, 2, 3). In Eq. (C.3d), for example, both free indices (i, j) can take on any one of
the three values (1, 2, 3).
Quantities with no indices (see f in Eq. (C.3a)) are scalars, or tensors of order
zero, while quantities with only one free index are normally components of vectors,
or tensors of order one (see ap, bp, wi in Eqs. (C.3b, c)). Similarly, quantities with
n free indices are often components of tensors of order n. For example, σ ij, and ekl
are components of tensors of order two, while Cijkl is a tensor of order four. Whether
or not a quantity is a tensor of a particular order, of course, depends on its behavior
under coordinate transformation [1].
Several special quantities appear frequently in indicial notation. One of these is
the Kronecker delta tensor, δij, defined as
(
1 i¼j
δij ¼ : ðC:4Þ
0 i 6¼ j
Appendix C: Basic Notations and Concepts 717
The Kronecker delta tensor and the alternating tensor are related through a useful
expression called the ε δ identity given by
Since the variable name is dropped by this convention, where derivatives of more
than one variable occur in the same expression, it is sometimes convenient to
indicate which variable is being differentiated by using distinct index symbols.
To illustrate, consider the function f ðx; yÞ ¼ f ðx1 ; x2 ; x3 ; y1 ; y2 ; y3 Þ. We write
f , k ¼ ∂f =∂xk f , K ¼ ∂f =∂yK
so
2
f , iK ¼ ∂ f =∂xi ∂yK
2
f , KK ¼ ∂ f =∂yK ∂yK
2
f , kk ¼ ∂ f =∂xk ∂xk
etc.
Using indicial notation, many of the standard vector operations are easily
written:
1. Dot product of two vectors, u and v
u v ¼ ui v i
if c ¼ a b then ci ¼ εijk aj bk
718 Appendix C: Basic Notations and Concepts
∇f ¼ f , p ep
∇ f ¼ f i, i
∇2 f ¼ ∇ ð∇f Þ ¼ f , kk
∇2 u ¼ uk, jj ek
where nk are components of the unit normal vector of the surface S which points
outward from V. Some important special cases of Gauss’ theorem arise from the
following specific choices for the general tensor a:
Appendix C: Basic Notations and Concepts 719
1. a ¼ ϕ, a scalar
ð ð
ϕ, k dV ¼ ϕ nk dS ðC:8Þ
V S
2. a ¼ vi , a vector
ð ð
vi, k dV ¼ vi nk dS ðC:9Þ
V S
3. a ¼ vi , a vector
ð ð
vi, i dV ¼ vi ni dS ðC:10Þ
V S
Another fundamental integral theorem that we will make use of is Stokes’ theorem.
This theorem relates the integral over an open surface S to a corresponding closed
contour (line) integral over the edge C of S where, if n is the outward normal to S,
the contour is obtained from n according to the right hand rule. There are also
various forms of this theorem. The form which we will use in this book for
transducer modeling with boundary diffraction waves (Chap. 8) is
720 Appendix C: Basic Notations and Concepts
ð I
ð∇ f Þ ndS ¼ f dc: ðC:12Þ
S C
Several other forms of Stokes’ theorem that are particularly useful in regularizing
the hypersingular integral equations appearing in crack problems (see Chap. 5) are
1. For a vector fk
ð I
½ð∂f k =∂xr Þnk ð∂f k =∂xk Þnr dS ¼ εkrq f k dxq : ðC:13Þ
S C
u ¼ x X: ðC:15Þ
Now, consider another two points Q and Q0 which are located in the neighborhood
of points P and P0 , respectively, as shown. The differential length between these
two pairs of neighboring points is
x1
x3
Appendix C: Basic Notations and Concepts 721
in volume V, and
in volume V0 . Any distortion that occurs between these points during the deforma-
tion is given by
Thus, this local distortion, or strain, is described by the second order tensor, εij,
given by
1
εij ¼ δ X k , i X k, j ; ðC:19Þ
2 ij
where the factor of 1/2 is introduced merely for convenience. The quantity εij is
known as the Almanski strain tensor. From Eq. (C.15), the Almanski strain tensor
can be rewritten in terms of the displacement components, ui, as
1
εij ¼ u þ uj , i u k , i uk , j : ðC:20Þ
2 i, j
In all the applications considered in this book, the displacement gradients will be
everywhere small, so to first order we can write εij ¼ eij where
1
eij ¼ u þ uj , i ðC:21Þ
2 i, j
is the Cauchy strain tensor. Since to first order we also have a change in displace-
ment, dui, given by
1
ωij ¼ εijk ωk ðC:23Þ
2
722 Appendix C: Basic Notations and Concepts
ω¼∇u ðC:24Þ
so that the total change in displacement du can be written in vector notation, from
Eq. (C.22) as
1
du ¼ e dx þ ω dx: ðC:25Þ
2
Since we will assume the displacement (and velocity) gradients of the motion are
everywhere small, these gradients will also be neglected when computing the
velocity, v, and the acceleration, a. For the velocity v, for example, we have
where
is a total material time derivative. Thus for small displacement gradients we have
If we let ρ0(X) be the mass density of the material in some original volume V at a
fixed time t ¼ 0 (Fig. C.2) and ρ(x, t) the corresponding density in volume V0 at
time t, then by conservation of mass we have
ð ð
ρðx; tÞdx1 dx2 dx3 ¼ ρ0 ðXÞdX1 dX2 dX3
0 V
V
or
ð
½ρðxðX; tÞ, tÞJ ρ0 ðXÞdX1 dX2 dX3 ; ðC:31Þ
0
V
Appendix C: Basic Notations and Concepts 723
ρJ ¼ ρ0 : ðC:32Þ
For small displacement gradients, however, it can be shown that the Jacobian is
approximately
J ¼ ð 1 þ uk , k Þ ðC:33Þ
ρ ¼ ρ0 ð1 uk, k Þ ffi ρ0 : ðC:34Þ
In most of our discussions we will take the original density, ρ0, to be independent of
X, so that Eq. (C.34) then shows that ρ is also, to first order, merely a constant.
C.5 Stress
ΔF −n − ΔF
n
V
F2
P ΔA
F3
V–
F2
F3
724 Appendix C: Basic Notations and Concepts
internal forces exerted across the plane by V þ . We define the traction vector, t(n),
acting at P as
Note that since an equal and opposite force, ΔF, acts on an identical area ΔA for
V þ with unit normal -n, it follows that
x3
t3
Appendix C: Basic Notations and Concepts 725
t12 t1
t13 x1
t11
x3
n
t(n)
– t1 P x1
– t2
x3
Dividing by dA and noting that the components, nk of the unit vector normal to dA,
n ¼ nk ek , are given by nk ¼ dAk =dA we find
tðnÞ ¼ tk nk ðC:38Þ
or
Equation (C.39) shows that a knowledge of the stress tensor, τkl, at any point
P determines the traction vector at P for any cutting plane through P. For this
reason, we say that the stress components τkl completely determine the state of
stress at point P.
References
1. G.E. Mase, G.T. Mase, Continuum Mechanics for Engineers (CRC, Boca Raton, 1992)
2. A.C. Eringen, Mechanics of Continua (Wiley, New York, 1967)
3. D. Frederick, T.S. Chang, Continuum Mechanics (Allyn and Bacon, Boston, 1965)
4. Y.C. Fung, A First Course in Continuum Mechanics (Prentice-Hall, Englewood Cliffs, 1977)
5. W. Jaunzemis, Continuum Mechanics (Macmillan, New York, 1967)
6. L.E. Malvern, Introduction to the Mechanics of a Continuous Medium (Prentice-Hall, Engle-
wood Cliffs, 1969)
Appendix D: The Hilbert Transform
ð
þ1
f ðtÞdt
f H ðuÞ ¼ H½f ðtÞ; u ¼ 1=π ; ðD:1Þ
tu
1
As with the Fourier transform, there is an inverse Hilbert transform that can recover
f(t) given by
ð
þ1
1 f H ðuÞdu
f ðtÞ ¼ : ðD:3Þ
π ut
1
1
δðtÞ
πu
t a
Rea > 0
t2 þ a2 u2 þ a2
a u
Rea > 0
t2 þ a2 u2 þ a2
References
1. I.H. Sneddon, The Use of Integral Transforms (McGraw-Hill Book Co., New York, 1972)
Appendix E: The Method of Stationary Phase
ðb
I ¼ f ðxÞexp½ikϕðxÞdx; ðE:1Þ
a
where the wave number, k, is assumed to be large and f(x) is a smoothly varying
function. Since the phase of the complex exponential will vary rapidly in this case,
with each half period oscillation nearly canceling the adjacent half period oscilla-
tion of opposite sign, we expect that I ! 0 as k ! 1 . Under such circumstances
the major contribution of the integral will come from points where the phase term
0
ϕ(x) is stationary, i.e. ϕ ðxÞ ¼ dϕðxÞ=dx ¼ 0. To see this, consider the following
integral:
h i
Fig. E.1 The function xð4 xÞ cos kðx 2Þ2 plotted a for k ¼ 20, and b for k ¼ 40.
ð4 h i
I p ¼ xð4 xÞexp ikðx 2Þ2 dx:
0
Figures E.1(a),(b) show the behavior of the real part of this integrand for k ¼ 20 and
k ¼ 40, respectively. One can see from the behavior of the integrand that most
oscillations are canceling except near x ¼ 2 which is a stationary point for the phase
term ϕðxÞ ¼ ðx 2Þ2 .
Assume now, that the integral in Eq. (E.1) has one stationary point within
0
the open interval (a, b) at the point xs where ϕ ðxs Þ ¼ 0. Then, expanding the
phase ϕ(x) in a Taylor series about this point gives
00
ϕðxÞ ffi ϕðxs Þ þ ϕ ðxs Þs2 =2; ðE:2Þ
00 00
where ϕ ðxs Þ ¼ d2 ϕ=dx2 and s ¼ x xs and where we have assumed ϕ ðxs Þ 6¼ 0.
Since f is assumed to be slowly varying, it is nearly a constant and can be taken out
of the integrand, together with the constant phase part of the exponential, and
the major contribution of the integrand remaining comes from the region
ðxs ε, xs þ εÞ about the stationary phase point. Thus, we have approximately
ð
s¼þε
h 00 i
I ¼ f ðxs Þexp½i k ϕðxs Þ exp ikϕ ðxs Þs2 =2 ds: ðE:3Þ
s¼ε
ð
þ1
h 00 i
I ¼ f ðxs Þexp½i k ϕðxs Þ exp ikϕ ðxs Þs2 =2 ds: ðE:4Þ
1
Appendix E: The Method of Stationary Phase 731
00 00 1=2
Now, if we let kϕ ðxs Þs2 =2 ¼ t2 , it follows that ds ¼ 2=kϕ ðxs Þ dt and the
stationary phase approximation of the integral, Is, is
h ð
þ1
00 i1=2 h n 00 oi
I s ¼ f ðxs Þ 2=kϕ ðxs Þ exp½ikϕðxs Þ exp i t2 sgn ϕ ðxs Þ dt: ðE:5Þ
1
The integral in Eq. (E.5) can be performed exactly since we have [1]
ð
þ1
pffiffiffiffiffiffiffiffi
cos t2 dt ¼ π=2
1
ðE:6Þ
ð
þ1
pffiffiffiffiffiffiffiffi
sin t2 dt ¼ π=2;
1
so that
ð
þ1
h n 00 oi 00
pffiffiffiffiffiffiffiffi
exp it2 sgn ϕ ðxs Þ dt ¼ 1 þ isgn ϕ ðxs Þ π=2
ðE:7Þ
1
pffiffiffi 00
¼ π exp iπsgn ϕ ðxs Þ =4
Equation (E.8) gives an explicit expression for the stationary phase evaluation of
the original integral when a single isolated stationary phase point exists in the open
interval (a, b). Special cases when multiple stationary phase points exist, or when
00
the stationary phase point(s) are at the interval end points, and cases where ϕ ¼ 0
at the stationary point can also be considered (see [1] for further details). However,
for the specific problems covered in this book, Eq. (E.8) will generally suffice.
The only special case not covered by Eq. (E.8) we would like to consider is when
there is no stationary phase point at all in the closed interval [a, b]. Then we can
write I in the form
ðb
f ðxÞ 0
I¼ 0 expfikϕðxÞgikϕ ðxÞdx; ðE:9Þ
ikϕ ðxÞ
a
where the second term in square brackets in Eq. (E.9) is a perfect differential so that
an integration by parts gives
732 Appendix E: The Method of Stationary Phase
ðb
f ðxÞexp½ikϕðxÞ b 1 h 0
i0
I¼ 0 f ðxÞ=ϕ ðxÞ exp½ikϕðxÞdx: ðE:10Þ
ikϕ ðxÞ a
ik
a
0
Since ϕ ðxÞ 6¼ 0 in [a, b] we can again write the integral in Eq. (E.10) as the product
of two terms, one being a perfect differential, and perform the integration by parts
again. Thus, keeping only the first terms, for k large we have
f ðbÞexpðikbÞ f ðaÞexpðikaÞ 1
I¼ 0 0 þ O : ðE:11Þ
ikϕ ðbÞ ikϕ ðaÞ k2
bð1 bð2
where again k is large and the function f is assumed to be smoothly varying. We will
assume a stationary phase point exists at the point xs ¼ ðxs ; ys Þ inside the limits of
the integrals. Then ∂ϕðxs Þ=∂x ¼ ∂ϕðxs Þ=∂y ¼ 0 and we can expand ϕ in a Taylor
series to second order as
1n s o
ϕðx; yÞ ¼ ϕðxs ; ys Þ þ ϕ, xx ðx xs Þ2 þ 2ϕ,sxy ðx xs Þðy ys Þ þ ϕ,syy ðy ys Þ2 ;
2
ðE:13Þ
2
where ϕ,sxy ¼ ∂ ϕðxs Þ=∂x∂y,etc. If we change to new coordinates (u, v), where
u ¼ ð x xs Þ
ðE:14Þ
v ¼ ðy ys Þ þ ϕ,sxy ðx xs Þ=ϕ,syy
and
ϕ,svv ¼ ϕ,syy
ϕ,suv ¼ 0
ðE:17Þ
2
ϕ,suu ¼ ϕ,sxx ϕ,sxy =ϕ,syy
ð þ1
þ1 ð
Is ¼ f ðxs ; ys Þexp½ikϕðxs ; ys Þ exp ik ϕ,suu u2 þ ϕ,svv v2 dudv: ðE:18Þ
1 1
Since the u and v integrals in Eq. (E.18) are identical to the forms we evaluated
previously in the one-dimensional case (see Eq. (E.4)), we can obtain immediately
" #1=2 " #1=2
2π 2π
I s ¼ s f ðx s ; y s Þ
k ϕ, uu kϕ,svv ðE:19Þ
exp ikϕðxs ; ys Þ þ iπ sgn ϕ,suu þ sgn ϕ,svv =4 ;
which can be written in terms of the original x and y coordinate values by defining
2
H ¼ ϕ,suu ϕ,svv ¼ ϕ,sxx ϕ,syy ϕ,sxy
ðE:20Þ
σ ¼ sgn ϕ,suu þ sgn ϕ,svv
2π f ðx s ; y s Þ
Is ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi exp½ikϕðxs ; ys Þ þ iπσ=4: ðE:22Þ
k s s
2
ϕ ϕ ϕ s
, xx , yy , xy
734 Appendix E: The Method of Stationary Phase
The last case that we will consider by the method of stationary phase is the
evaluation of a particular integral over a curved surface given by
ð
I ¼ f ðxÞexp½ikða xÞdSðxÞ; ðE:23Þ
S
where the surface S will be assumed to be smooth, but otherwise arbitrary, and the
vector a, is a constant. Again, we wish to evaluate this integral at high frequencies,
i.e. when k is large. In this case, we have taken the phase term to be of the particular
form ϕðxÞ ¼ a x which will occur in a number of applications.
As in the previous cases, we will assume the major contribution to this integral
comes from the evaluation near a stationary phase point xs on the surface. We can
parameterize S in terms of two surface coordinates (t1, t2) so that on S we can write
x ¼ xðt1 ; t2 Þ. Then in this case the stationary phase conditions become
∂ϕ=∂t1 ¼ ∂ϕ=∂t2 ¼ 0
or, explicitly
a ∂x=∂t1 ¼ 0
ðE:24Þ
a ∂x=∂t2 ¼ 0:
Since the vectors ∂x=∂tα ðα ¼ 1, 2Þ are tangent to the surface, it follows from
stationary phase that at the stationary phase point xs ¼ ðb1 ; b2 Þ, a is parallel to the
unit normal, nI, which is taken here to be the inward normal. Then, expanding ϕ in a
Taylor series about the stationary phase point as before, we obtain in this case
1n o
ϕ ¼ a xð t 1 ; t 2 Þ ¼ a xs þ ðtα bα Þ tβ bβ a x,sαβ ; ðE:25Þ
2
2
where x,sαβ ¼ ∂ xðb1 ; b2 Þ=∂tα ∂tβ Then, the integral can be written approximately as
þε2 b1ð
b2 ð þε1 o
1 n
I ¼ f ðxs Þexpðik a xs ÞJ ðxs Þ exp k ðtα bα Þ tβ bβ a x, αβ dt1 dt2;
s
2
b2 ε2 b1 ε1
ðE:26Þ
where dS ¼ Jdt1 dt2 is the area element in the (t1, t2) coordinates and J is a Jacobian.
From differential geometry [3] we have
γ
x,sαβ ¼ Γ αβ ðxs Þx,sγ þ hαβ ðxs ÞnI ðxs Þ: ðE:27Þ
Appendix E: The Method of Stationary Phase 735
where Γ γαβ are Christoffel’s symbols of the second kind, hαβ is the curvature tensor,
and x,sγ ¼ ∂xðb1 ; b2 Þ=∂tγ . The dot product of a with these second order partial
derivatives of the position vector can be expressed in general as
γ
a x,sαβ ¼ Γ αβ ðxs Þ a x,sγ þ hαβ ðxs Þ a nI ðxs Þ : ðE:28Þ
However, from the stationary phase conditions the first term on the right side of
Eq. (E.27) vanishes, leaving
a x,sαβ ¼ hαβ ðxs Þ a nI ðxs Þ : ðE:29Þ
If t1 and t2 are chosen in particular to be the arc length parameters taken along the
principal directions of the surface at xs, then
and J ðxs Þ ¼ 1 where κ 1 and κ 2 are the principal curvatures of the surface at xs and
the stationary phase contribution to the integral becomes (expanding the limits of
integration, as before to infinity):
I s ¼ f ðxs Þexp½i k a xs
ð þ1
þ1 ð h n oi
1 ðE:31Þ
exp k a nI ðxs Þ κ1 ðt1 b1 Þ2 þ κ2 ðt2 b2 Þ2 dt1 dt2 :
2
1 1
With this choice of coordinates the integrals appearing in Eq. (E.31) can be
independently evaluated since they are in the same general form as found in the
one- and two-dimensional integral cases considered previously. Thus, we find
1=2 1=2
2π 2π
Is ¼ f ðxs Þexp½ika xs þ iπσ=4; ðE:32Þ
kjκ1 a nI ðxs Þj kjκ 2 a nI ðxs Þj
where
σ ¼ sgn κ 1 a nI ðxs Þ þ sgn κ 2 a nI ðxs Þ : ðE:33Þ
If we let
jκ1 j ¼ 1=R1
ðE:34Þ
jκ2 j ¼ 1=R2 ;
736 Appendix E: The Method of Stationary Phase
where R1 and R2 are the magnitudes of the principal radii of curvature of S at xs,
we have
pffiffiffiffiffiffiffiffiffiffi
2π R1 R2
Is ¼ f ðxs Þexp½i k a xs þ iπσ=4: ðE:35Þ
kja nI ðxs Þj
References
1. I.S. Gradshteyn and I.M. Ryzhik, Table of Integrals, Series, and Products (Academic,
New York, 1980)
2. L.B. Felsen, N. Marcuvitz, Radiation and Scattering of Waves (Prentice-Hall, Inglewood
Cliffs, 1973)
3. I.S. Sokolnikoff, Tensor Analysis, 2nd edn. (Wiley, New York, 1964)
4. N. Bleistein, R.A. Handelsman, Asymptotic Expansion of Integrals (Dover Publications,
New York, 1986)
5. N. Bleistein, Mathematical Methods for Wave Phenomena (Academic, New York, 1984)
Appendix F: Properties of Ellipsoids
In Chap. 10, it was shown that the far field scattering response of ellipsoidal
inclusions and elliptical cracks can be obtained explicitly through the use of the
Born and Kirchhoff approximations. These same approximations were used in
Chap. 15 to develop a number of equivalent flaw sizing algorithms. In this Appen-
dix, we will derive some of the geometrical properties of ellipsoids that are useful in
such applications. The details discussed here can also be typically found in texts on
tensor analysis and differential geometry [1–3].
One geometrical quantity that appears frequently in Chaps. 10 and 15 is the
effective radius of an ellipsoid, re, which was defined to be the perpendicular
distance from the center of an ellipsoid to a plane P which is tangent to its surface
at some point, x0, as shown in Fig. F.1. If we write the equation of the ellipsoid as
then since the unit outward normal, n, is given by n ¼ ∇g=j∇gj its components are
x1 x2 x3
n1 ¼ , n2 ¼ , n3 ¼ ðF:2Þ
Ha21 Ha22 Ha23
with
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x21 x22 x23
H¼ þ þ : ðF:3Þ
a41 a42 a43
n0 X ¼ r e ; ðF:4Þ
which is the general equation for a plane whose unit normal is n0 ¼ n01 ; n02 ; n03 and
whose perpendicular
distance from the origin is the “effective radius”, re. Since
point x0 ¼ x01 ; x02 ; x03 also lies in the plane P, for this point Eq. (F.4) gives
so that re is given by
1 1
re ¼ ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ðF:7Þ
H0 ð x1 Þ
0 2
ðx02 Þ
2
ðx03 Þ
2
a4
þ a4 þ a4
1 2 3
2 2 2
To obtain a more useful form for re, consider a21 n01 þ a22 n02 þ a23 n03 . From
Eqs. (F.1), (F.2), and (F.7)
2 2 2
a21 n01 þ a22 n02 þ a23 n03
0 2 2 0 2 2 0 2 2
x =a1 þ x2 =a2 þ x3 =a3
¼ 1 0 2 ðF:8Þ
H
¼ r 2e ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðF:9Þ
¼ a21 ðei u1 Þ2 þ a22 ðei u2 Þ2 þ a23 ðei u3 Þ2 ;
Appendix F: Properties of Ellipsoids 739
where (u1, u2, u3) are unit vectors along the axes of the ellipsoid (Fig. F.1). Thus, if
we are given the normal to the plane, ei, and the size and orientation of the ellipsoid
through the parameters (a1, a2, a3) and (u1, u2, u3), respectively, Eq. (F.9) deter-
mines the effective radius. Also, the point where the plane touches the ellipsoid can
then be obtained, since
Finally, it is interesting to note that Eqs. (F.2), (F.4), and (F.6) also imply that an
equivalent form for the plane P is given by
x1 ¼ a1 cos t1 cos t2
x2 ¼ a2 cos t1 sin t2 ðF:12Þ
x3 ¼ a3 sin t1 :
From differential geometry the mean and Gaussian curvatures, M and K, respec-
tively, of a surface are given by [1]
where gαβ and hαβ are the first and second fundamental forms
gαβ ¼ x, α x, β
ðF:15Þ
hαβ ¼ nI x, αβ
and the comma denotes partial differentiation with respect to the t parameters, i.e.
x, α ¼ ∂x=∂tα , etc. The vector nI is the inward unit normal to the ellipsoid which is
given parametrically from Eqs. (F.2), (F.7), and (F.12) as
re re re
nI ¼ cos t1 cos t2 u1 cos t1 sin t2 u2 sin t1 u3 : ðF:16Þ
a1 a2 a3
from which the mean and Gaussian curvatures can be computed at x0 from
Eq. (F.14) as
κ1 þ κ2 re n
M¼ ¼ 2 2 2 a21 a22 þ a23 ðei u1 Þ2 þ a22 a21 þ a23 ðei u1 Þ2
2 2 a 1 a2 a3
o ðF:18Þ
þ a23 a21 þ a21 ðei u3 Þ2
and
1 r4
K ¼ κ1 κ2 ¼ ¼ 2 e2 2 ; ðF:19Þ
R1 R 2 a 1 a 2 a 3
where κ 1 and κ2 are the principal curvatures of the ellipsoid and R1 and R2 are the
corresponding principal radii of curvature. Just as the Gaussian curvature appears in
the first order leading edge response of volumetric flaws, Chen [4] has shown that
the mean curvature also appears in a higher order expansion of this leading edge
expression.
Appendix F: Properties of Ellipsoids 741
References
In this second edition, MATLAB® functions and scripts have been included to
implement some of the ultrasonic models we have considered. This Appendix
briefly describes those functions and scripts. Additional MATLAB models for a
variety of ultrasonic NDE problems are also available in two related books:
L.W. Schmerr and S.J. Song, Ultrasonic Nondestructive Evaluation Systems –
Models and Measurements, Springer, 2007, and L.W. Schmerr, Fundamentals of
Ultrasonic Phased Arrays, Springer, Springer, 2015. The MATLAB code listings
for this book are available on the web from the publisher or they can be obtained
directly by sending an email with the subject title “Ultrasonic NDE Codes” to the
author at lschmerr@iastate.edu.
fluid_fluid
>> [R, T] ¼ fluid_fluid(iangd, d1, d2, c1, c2). A function which returns
(R,T) the reflection and transmission coefficients (based on velocity
ratios) for a plane wave incident on a plane interface between two fluids
at an angle iangd (in degrees).The parameters (d1,c1) and (d2, c2) are
the densities and wave speeds in the first and second fluids. The units of
these quantities are arbitrary but must be consistent.
fluid_solid
>> [tpp, tps] ¼ fluid_solid(iangd, d1, d2, cp1, cp2, cs2). A function
which returns (tpp, tps) the plane wave transmission coefficients for a
P-wave and S-wave, respectively, produced in a solid at a fluid-solid
interface due to a P-wave incident in the fluid at an angle iangd
(in degrees). The coefficients are both based on velocity ratios. The
parameters (d1, cp1) are the density and wave speed of the fluid and (d2,
cp2, cs2) are the density, P-wave speed, and S-wave speed, respec-
tively. The units of these quantities are arbitrary but must be
consistent.
solid_f_solid
>> [tpp, tps] ¼ solid_f_solid(iangd, d1, d2, cp1, cs1, cp2, cs2). A
function which returns the P-P(tpp)and P-S(tps) transmission coeffi-
cients, based on velocity ratios, for a P-wave incident on a plane
interface between two solids that are in smooth contact through an
intermediate fluid layer of zero thickness. The parameters (d1, cp1,
cs1) are the density, P-wave speed, and S-wave speed for the first solid
while (d2, cp2, cs2) are the corresponding parameters for the second
solid. The units of these parameters are arbitrary but must be
consistent.
solid_solid
>> [tp, ts, rp, rs] ¼ solid_solid(iangd, d1, d2, cp1, cs1, cp2, cs2,
type). A function which returns the transmitted P-wave (tp), transmit-
ted SV-wave (ts), reflected P-wave (rp), and reflected SV-wave
(rs) transmission/reflection coefficients (based on velocity ratios)
for two solids in welded contact. The inputs are the incident angle
(s),iangd,(in degrees), (d1, d2), the densities of the two media,
(cp1, cs1), the compressional and shear wave speeds of the first medium,
and (cp2, cs2) the compressional and shear wave speeds of the second
medium. The parameter type is a string,(’P’ or ’S’), which indicates the
type of incident wave in medium one. If cs1 ¼0 and type ¼ ’P’ the function
returns the coefficients for a fluid-solid interface with rs ¼ 0. The wave
speed cs2 cannot be set equal to zero.
stress_freeP
>> [Rp, Rs] ¼ stress_freeP(ang, cp, cs). A function which returns (Rp,
Rs),the reflection coefficients for the reflected P-waves and S-waves,
respectively,(based on velocity ratios)for a plane P-wave incident on
a plane stress-free surface of an elastic solid. The parameter ang is
the incident angle (in degrees), cp, cs are the P- and S-wave speeds,
respectively, and. The units of the wave speeds are arbitrary but must
be consistent. Note that there are no critical angles for this case.
snells_law
>> [ang_in, ang_out]¼ snells_law(ang, c1, c2, type). A function which
returns (ang_in, ang_out) the incident and refracted angle at a plane
interface, respectively,(in degrees)that satisfy Snell’s law. The
first input, ang, is a given incident or refracted angle, ang,
(in degrees) that a travelling wave makes with respect to the normal of
a plane interface. The wave speed in the first medium is c1 and the wave
speed in the second medium is c2. The input parameter, type, is either
the string ’f’ or the string ’r’, indicating a forward or reverse,
Appendix G: Matlab Functions and Scripts 745
Rayleigh_speed
>> cr ¼ Rayleigh_speed(cp, cs). A function which returns the Rayleigh
wave speed, cr, at a free surface of a solid where (cp, cs) are the P-wave
and S-wave speeds, respectively. The units of the wave speeds are arbi-
trary but must be consistent.
dispersion_plots
>> dispersion_plots. A script which creates a 2-D set of normalized
frequency and wave speed values at which the Rayleigh-Lamb dispersion
function for symmetrical or anti-symmetrical plate waves is evaluated.
The MATLAB function contour is then used to plot the dispersion curves
for this 2-D region. Note that the fundamental anti-symmetrical mode
plot is unreliable at very small frequencies and must be replaced by the
analytical results for this mode at those values.
Rayleigh_LambM
>> y ¼ Rayleigh_LambM(c, cp, cs, fh, type). A function which returns the
Rayleigh-Lamb function values for symmetrical or unsymmetrical plate
waves. The input parameter, c, is the wave speed of the plate wave, while
(cp, cs) are the compressional and shear wave speeds of the plate,
respectively. The wave speed values are arbitrary but must be consis-
tent. The parameter fh is the frequency times the half width of the
plate. The input parameter, type, is a string,(’s’ or ’a’)for symmet-
rical or ant-symmetrical modes, respectively.
dispersion_curves
>> dispersion_curves. A script, which like the script
dispersion_plots, generates a 2-D set of non-dimensional frequency
and wave speed values at which the plate wave Rayleigh-Lamb dispersion
function is evaluated for symmetrical or anti-symmetrical plate waves.
The dispersion curves are then plotted with the MATLAB function con-
tour. Individual dispersion curves are extracted and ordered in fre-
quency. The fundamental flexural mode values for small frequencies where
the contour values are unreliable are modified with analytical results.
The specific dispersion curve specified by the user is then plotted.
746 Appendix G: Matlab Functions and Scripts
bdw_fluid
>> p ¼ bdw_fluid(x, y, z, a, c, f, N). A function which returns the
normalized pressure, p, for a planar circular piston transducer of
radius, a,(in mm), radiating into a fluid of wave speed, c, (in m/sec),
at a frequency, f,(in MHz). The pressure is calculated at the points (x,
y, z) (in mm), where (x, y, z) can be scalars, vectors, or matrices. N, is
an optional input parameter that specifies the number of line segments
used to approximate the edge integral in a boundary diffraction wave
model. If N is not specified, N is determined automatically so that the
line segment length is no larger than one tenth of a wavelength. This
function calls the supporting function bdw_model which implements the
boundary diffraction wave model of the transducer.
bdw_model
>> p ¼ bdw_model(rho, z, a, c, f, N). A function which returns the nor-
malized pressure, p, for a circular transducer of radius, a,(in mm),
radiating into a fluid of wave speed, c, (in m/sec), at a frequency, f,
(in MHz). The pressure is calculated at the radial distance rho(in mm)
and axial distance, z,(in mm) in the fluid. N is the number of line seg-
ments used to approximate the edge integral in a boundary diffraction
wave model. All input parameters must be scalars. This is a supporting
function which called multiple times by the function bdw_fluid.
onaxis_foc
>> p ¼ onaxis_foc(z, f, a, R, c). A function which calculates the
on-axis normalized pressure,p, at the locations z(in mm)of a spheri-
cally focused transducer of radius a (in mm) and focal length R(in mm)
radiating into a fluid with wave speed c(in m/sec), using the O’Neil
model. If R ¼ inf, the function returns the on-axis pressure of a planar
transducer of radius a.
parameters2
>> parameters2. A script which is used to define the input parameters
needed to model a planar or focused circular piston transducer radiat-
ing through a curved interface. These parameters are then placed in a
MATLAB structure called setup.
gauss_c15
>> [A, B] ¼ gauss_c15(). A function which returns the 15 "optimized"
coefficients obtained by Wen and Breazeale to simulate the wave field of a
circular planar piston transducer radiating into a fluid by a superpo-
sition of Gaussian beams: Wen, J.J. and M. A. Breazeale," Computer
optimization of the Gaussian beam description of an ultrasonic field,"
Computational Acoustics, Vol.2, D. Lee, A. Cakmak, R. Vichnevetsky,
Eds. Elsevier Science Publishers, Amsterdam, 1990, pp. 181-196.
Appendix G: Matlab Functions and Scripts 747
MG_beam2
>> v ¼ MG_beam2(setup). A function which returns the normalized veloc-
ity amplitude, v, due to a planar or spherically focused piston trans-
ducer radiating obliquely through a curved fluid-solid or a smooth
solid-solid interface where the plane of incidence must be aligned
with one of the principal axes of the curved surface. The input parame-
ter setup is a MATLAB structure that is generated by the MATLAB script
parameters2 and contains all of the input parameters needed to define a
given inspection. This is a multi-Gaussian beam model that uses the
15 optimized coefficients of Wen and Breazeale to calculate the wave
field. Those coefficients are returned by the MATLAB function gauss_c15.
onaxis_interface
>> [v, vp] ¼ onaxis_interface(z1, z2, f, a, d1, d2, cp1, cp2, cs2). A
function which computes the on-axis velocity of a circular piston
transducer of radius a (in mm)radiating at a frequency f (in MHz)
through a planar fluid-solid interface at normal incidence. The dis-
tances z1, z2 (in mm) are in the fluid and solid, respectively, whose
densities are d1, d2. The compressional wave speed of the fluid is cp1
and the compressional and shear wave speeds of the solid are cp2, cs2.
All the wave speeds are measured in m/sec. The units of the densities are
arbitrary but must be consistent. The function returns both the normal-
ized on-axis velocity, v, and the corresponding paraxial approxima-
tion, vp, for the on-axis velocity of the transmitted P-wave. The
distance z1 in the fluid must be a scalar.
angle_beam2
>> angle_beam2. A script which generates an image of the wave field of a
45 degree angle beam shear wave transducer in the plane of incidence.
The script uses the multi-Gaussian beam model MG_beam2.
pulse2
>> pulse2. A script which generates a wave form at an on-axis point in
water produced by a circular planar transducer. It uses the multi-Gauss-
ian beam model MG_beam2 and the function spectrum to generate the pulse.
A_crack_pe
>> A ¼ A_crack_pe(f, b, c, angd). A function which returns the pulse-echo
scattering amplitude,A, (in mm)for a circular crack of radius b (in mm)
at a frequency,f, (in MHz) in a material with a wave speed, c, (in m/sec)
and at an angle angd(in degrees) using the Kirchhoff approximation.
748 Appendix G: Matlab Functions and Scripts
A_void_pe
>> A ¼ A_void_pe(f, b, c). A function which returns the pulse-echo
scattering amplitude, A,(in mm) of a spherical void of radius b (in mm)
at frequency f (in MHz) in a solid or fluid with wave speed c (in m/sec)
using the Kirchhoff approximation.
A_incl_pe
>> A ¼ A_incl_pe(f,b,c1,d1,c0,d0). A function which calculates the
pulse-echo scattering amplitude, A,(in mm) for a spherical inclusion
of radius b (in mm) where (c1, d1) and (c0, d0) are the wave speeds and
densities of the flaw and host materials, respectively. Wave speeds are
in m/sec and densities are in arbitrary but consistent units. The func-
tion uses the modified Born approximation where the wave travels in the
flaw at the flaw wave speed and the amplitude of the front and back
responses are in terms of the reflection coefficient without making the
weak scattering approximation. A phase correction puts the front and
back responses at the correct times.
sphere_rigid_pe
>> A ¼ sphere_rigid_pe(f, b, c). A function which calculates the pulse-
echo scattering amplitude, A, (in mm) for a rigid sphere of radius b
(in mm) in a fluid with wave speed, c, (in m/sec). This function uses the
method of separation of variables.
eff_parameter
>> [aeff, Reff] ¼ eff_parameters(c, f, zmin, zmax). A function which
computes the effective radius (aeff) and effective focal length (Reff)
of a spherically focused transducer from measurements of zmin and zmax
(both in mm),the on-axis min and max locations in the frequency domain
response of the transducer, at a frequency f (in MHz) where the wave
speed c is in meters/sec. If zmax is not specified, then Reff ¼ inf (pla-
nar transducer) is returned and only aeff is calculated. This function
calls a supporting function, transeq.
transeq
>> y ¼ transeq(x, zmax, zmin, k). A function which returns the values y
of the equation y ¼ f(x, zmin, zmax, k), where (zmin, zmax) are measured
values (in mm)of the on-axis min and max response of a focused trans-
ducer and k is the wave number (in rad/mm). The function f is a supporting
function needed to determine the effective parameters of a spherically
focused transducer.
Appendix G: Matlab Functions and Scripts 749
ellipse_data
>> [theta, phi, dt] ¼ ellipse_data(). A function which returns exper-
imental sizing data for a 2.5 x 0.6 mm elliptical crack in titanium(c ¼
6100 m/sec) as measured with a phased array where (theta, phi) are the
spherical angles (in degrees) at which dt is measured, where dt is the
time between flash points (in microseconds). These measured values are
corrected for finite bandwidth errors.
K_sizing
>> [pds, a] ¼ K_sizing(c, theta, phi, dt). A function which returns
matrices containing the direction cosines, pds, and lengths of the
semi-major axes,a,(in mm)of an equivalent flat elliptical crack in a
solid with wave speed,c, (in mm/microsec), based on measured times
between flashpoints, dt, (in microsec) at the spherical angles angles
(theta, phi), measured in degrees.
FourierT
>> y ¼ FourierT(x, dt). A function which returns y, the forward FFT of
the sampled time domain values,x, where dt is the sampling time interval
between points. FourierT approximates the Fourier transform where the
integrand of the transform is x*exp(2*pi*i*f*t). This is consistent
with the definition of the forward Fourier transform used in this book.
For NDE applications the frequency components, f, are normally in MHz,
and the sampling time, dt, in microseconds. The values, y, are complex
while the values of x are real.
IFourierT
>> y ¼ IFourierT(x, dt). A function which computes the inverse FFT of
the sampled frequency domain values, x, for a sampling time interval
dt. IFourierT assumes the integrand of the inverse transform is given by
x*exp(-2*pi*i*f*t). This is consistent with the definition of the
inverse Fourier transform used in this book. The first half of the sam-
pled values of x are the spectral components for positive frequencies
ranging from 0 to the Nyquist frequency 1/(2*dt). The second half of the
sampled values are the spectral components for the corresponding neg-
ative frequencies. The values, y, should be real but may contain small
imaginary parts from numerical round off, which can be eliminated by
taking the real part, i.e. real(y).If the negative frequency values are
set equal to zero then to recover the inverse FFT of x we must replace x
(1) by x(1)/2 and then compute 2*real(IFourierT(x,dt)).
750 Appendix G: Matlab Functions and Scripts
s_space
>> y ¼ s_space(xstart, xend, num). A function which generates num
evenly spaced sampled values from xstart to (xend - dx), where dx is the
sample spacing. This is useful in FFT analysis where we generate sampled
periodic functions. Example: generate 1000 sampled frequencies from
0 to 100MHz via f ¼ s_space(0,100,1000); in this case the last value of
f will be 99.9 MHz and the sampling interval will be 100/1000 ¼0.1 MHz.
c_shift
>> y ¼ c_shift(x, n). A function which moves the last n components of the
vector x into the first n component places and shifts the remaining com-
ponents of x to follow those n components, i.e. this is a circular shift.
It can be used to prevent a periodic function, as found in FFT analyses,
from being split into the first and last parts of the window in which it is
displayed. Note: x must be a row vector.
t_shift
>> y ¼ t_shift(x, n). A function which is used in conjunction with the
c_shift function to change the time axis values appropriately so that
the time axis is shifted along with the function. Example use: plot
(t_shift(t, 100), c_shift(fun, 100))
spectrum
>> V ¼ spectrum(f, B, fc, bw). A function which returns a Gaussian-
shaped spectrum, V, for a pulse whose time domain waveform is given by
v ¼ B*cos(2*pi*fc*t)*exp(-t^2/4a^2), where fc is the center frequency
(in MHz when t is in microseconds), and bw is the -6dB bandwidth (in MHz).
Note that this is a function defined only for positive frequencies so to
recover the real time domain waveform, v, we must let V(1) ¼ V(1)/2, and
take twice the real part of the inverse FFT.
lp_filter
>> y ¼ lp_filter(f, fstart, fend). A function which returns low-pass
filter, y, at the frequencies contained in the vector f. The filter has
values of 1.0 below the frequency value fstart and tapers to zero at
frequencies above the value fend with a cosine function. An error is
returned if fend is outside the range of values contained in the vector f
or if fend is less than fstart.
display_struct
>> display_struct(setup). A function which returns all the parameters
contained in the structure, setup, where setup is generated by the
script parameters2.
sphH
>> y ¼ sphH(n,x). A function which returns the spherical Hankel func-
tion of order n at x, where x is a non-dimensional argument. It is used in
separation of variables solutions.
Appendix G: Matlab Functions and Scripts 751
sphJ
>> y ¼ sphJ(n,x). A function which returns the spherical Bessel func-
tion of order n at x, where x is a non-dimensional argument. It is used in
separation of variables solutions.
Index
Helmholtz equation, 66, 89, 90, 92, 101, 487, Linear least squares/eigenvalue sizing,
488, 492, 502 654–657, 663, 672, 675
Hilbert transform, 129, 131, 473, 727–728 Love waves, 215
Hypersingular integrals, 104, 720 Low frequency scattering, 448, 499, 500,
515, 521
LTI model
I fluid medium, 540–544
Immersion testing, 6, 10, 22, 113, 122, 132, general, 390–398
133, 146, 219, 272, 279, 375, 398, 539, immersion testing, 545–552, 579
540, 545, 552–563, 577–579, 611, 624,
634, 635, 667
Impedance. See Acoustic impedance M
Impulse response function, 20–23, 29, Mass, conservation of, 722–723
444–445, 456–457, 462, 471–475, 482, Mean curvature, 740
606, 664 Measurement models
Indicial notation, 715–718 angle beam shear wave (contact), 563–570
Inhomogeneous waves, 113, 122–125, direct derivation, 539–552
127–132, 145, 155, 187, 197, 200, near field (see Near field models)
201, 250 reciprocity-based derivation, 552–570
Integral equations Mode conversion, 8, 141, 146, 194, 352, 606
fluid, 89, 96–98, 108 Model-assisted POD (MAPOD), 689, 694
solid, 89, 103–104, 108 Modified Born approximation (MBA),
Integral representation theorem 482–486, 510, 514, 748
fluid, 91, 92, 111, 220 Multi-Gaussian beam model, 219, 265, 325,
solid, 99, 100 327, 361–363, 365–368, 370, 371, 374,
Interface/boundary conditions 576, 640, 641, 747
fluid, 36–38
solid, 42–44
N
Navier’s equations, 41–42, 44–47, 49, 50,
J 59–63, 69, 78, 99, 202
Jacobian, 723, 732, 734 Near field distances, 225, 226, 372, 374, 381,
600, 601
Near field models
K pressure-pressure, 597–598
Kirchhoff approximation pressure-velocity, 598
curved interface, 167, 168, 301, 339 velocity-pressure, 583–596
fluid (cracks), 438–448 velocity-velocity, 596
fluid (volumetric flaws), 427–438 Noise figure of merit (FOM), 688, 689
solid (cracks), 466–477 Nonlinear least squares sizing, 654
solid (volumetric flaws), 457–466 Numerical beam modeling
Kirchhoff flaw sizing, 169, 449, 475, 652, 666, boundary diffraction wave model,
670, 737 238–243
Kronecker delta function, 716, 717 focused transducer (on-axis), 258
multi-Gaussian beam model, 361–374
planar transducer (on-axis) radiating
L through a plane interface, 367–368
Lamb waves. See Plate waves Numerical flaw modeling
Leading edge response crack, 510, 513
fluid, 436–438 inclusion, 510
solid, 463–466, 605 rigid sphere, 510, 514
Legrendre polynomials, 590 void, 510–512
Lens. See Acoustic lens Nyquist criterion, 703
756 Index