1

Modelling the Generation and Propagation of

Ultrasonic Signals in Cylindrical Waveguides
Fernando Seco and Antonio R. Jimnez
Centro de Automtica y Robtica (CAR)
Ctra. de Campo Real, Madrid
Consejo Superior de Investigaciones Cientficas (CSIC)-UPM
Spain

1. Introduction
Elongated cylindrical structures like rods, pipes, cable strands or bers, support the
propagation of mechanical waves at ultrasonic frequencies along their axes. This waveguide
behaviour is used in a number of scientic and engineering applications: the Non
Destructive Evaluation (NDE) of the structural health of civil engineering elements for safety
purposes (Rose, 2000), in linear displacement sensors (Seco et al., 2009) for high accuracy
absolute linear position estimation, in the evaluation of material properties of metal wires,
optical bers or composites (Nayfeh & Nagy, 1996), and as uid sensors in pipes transporting
liquids (Ma et al., 2007). These applications demand exact quantitative models of the
processes of wave generation, propagation and reception of the ultrasonic signals in the
waveguides.
The mathematical treatment of mechanical wave propagation in cylindrical structures was
provided by J. Pochhammer and C. Chree at the end of the XIX century, but its complexity
prevented researchers from obtaining quantitative results until the advent of computers. D.
Gazis (Gazis, 1959) reported the rst exact solutions of the Pochhammer-Chree frequency
equation, as well as a complete description of propagation modes and displacement and stress
distributions for an isotropic elastic tube, found with an IBM 704 computer. Since then, the
literature on the topic has grown steadily, and references are too numerous for this book
chapter. We will only mention a few landmark developments: the study of multilayered
waveguides beginning with a composite (two-layer) cylinder by H. D. McNiven in 1963;
the extension of Gazis results to anisotropic waveguides, initiated by I. Mirsky in 1965; the
consideration of uids and media with losses surrounding, or contained in the waveguides,
beginning with V. A. Del Grosso in 1968; and nally, the demonstration of ultrasonic guided
waves generated with electromagnetic transducers by W. Mohr and P. Holler in 1976, and
piezoelectrically by M. Silk and K. Bainton in 1979, for the nondestructive testing of pipes.

1.1 Modelling the response of the waveguide to external excitation

Of particular importance for transducer design is the determination of the mechanical
response of a waveguide when subjected to an external excitation. Several approaches exist
to consider this problem.
Integral transform methods (Graff, 1991) convert the differential equations that physically
model the excitation forces and the behaviour of the waveguide into a set of algebraic
2
2 Ultrasonic Waves
Will-be-set-by-IN-TECH

equations, which are more easily solvable. However, in order to nd the actual distribution
of the elastic eld excited in the waveguide, inverse contour integration in the complex
plane has to be performed, which is usually complicated. Due to the complexity of the
Pochhammer-Chree equations, this procedure is only practical with simplied versions of
the wave equation, which in general are not accurate enough for ultrasonic frequencies. See
for example, Folks solution for the transient response of a semi-innite rod to a step pressure
applied to its end (Folk et al., 1958).
The Semi-Analytical Finite Element (SAFE) method is a modication of Finite Element
Methods (FEM) in which the elastic eld is expanded as a superposition of harmonic
waves in the azimuthal-axial (-z) plane, while discretized mechanical equations are used
in the radial (r) direction of the waveguide. This reduction of the number of dimensions
permits a much higher efciency in the computation of the elastic elds (Hayashi et al.,
2003). Waveguides surrounded by innite media (like a pipe submerged in soil) can be
handled by SAFE techniques with proper discretized elements (Jia et al., 2011), as well as
waveguides with arbitrary proles: for example, a railroad rail in (Damljanovic & Weaver,
2004). Although nite element methods are powerful and exible, they have the shortcoming
of great requirements on computer memory and processing time when large structures
or high frequencies of operation are considered, and the difculty encountered in the
parameterization of transducer designs (for example, the determination of the transfer
function of the transducer-waveguide coupling).
Spectral methods are another numerical technique which approximate the differential elastic
equations of the waveguide (Helmholtz equations) by differentiation operators, turning the
problem of nding the wavenumber-frequency roots into a matrix eigenvalues determination
(Doyle, 1997). This numerical method, which is computationally simple and reportedly does
not suffer from the problems associated with large diameter waveguides at high frequencies,
has been recently applied to model multi-layered cylindrical waveguides (Karpnger et al.,
2008).
Modal analysis is an analytical method based on the expansion of the forcing terms acting
in the waveguide into the set of its proper modes (Auld, 1973). In (Ditri & Rose, 1992),
modal analysis is employed to model the loading of a waveguide by a transducer array. This
treatment is extended to more general transducers and antisymmetric modes by (Li & Rose,
2001). Modal analysis is a mathematically exact technique that leads to a closed form integral
equation for the elastic elds in the waveguide, and which incorporates in a natural way the
issue of mode selectivity, offering insight on the physics of waveguide behaviour. For these
reasons, modal analysis will be the approach used in this work.

1.2 Intention and scope of the research

With this book chapter we contribute a numerical simulation treatment of the ultrasonic
behaviour of cylindrical waveguides, based on the Pochhammer-Chree (PC) theory, and
covering the aspects of assembly of the description matrix of the waveguide, tracing of
the frequency-wavenumber curves, computation of mode shapes, use of modal analysis
to determine the response of the waveguide to external excitations, and the dispersive
propagation of signals.
The work described here has resulted in a software package, named PCDISP, written in
the Matlab environment (Matlab, 2004), and freely available1 to be adapted to particular
1 PCDISP webpage: http://www.car.upm-csic.es/lopsi/people/fernando.seco/pcdisp
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 33

circumstances. The main features of the PCDISP software will be introduced in this chapter
alongside with the theoretical concepts upon which it is based.
The purpose of PCDISP is freeing the researchers from the numerically delicate, time
consuming issues arising in the solution of the PC equations, such as the creation of
the waveguide matrix, the numerical instabilities encountered when the thickness of the
waveguide or the operating frequency are high, the determination of proper modes and
the tracing of the dispersive wavenumber-frequency curves. In this way, the researcher can
concentrate in the study of the waveguide/transducer interaction as such.
As far as we are aware of, only two other software suites specically designed for modelling
elastic wave propagation in cylindrical waveguides exist. Disperse (Pavlakovic & Lowe, 1999)
is a commercial package, based on matrix techniques, capable of analyzing cylindrical or
plate waveguides made of perfectly elastic or damped solids, as well as uids. GUIGUW
(Bocchini et al., 2011) is a Matlab-based software which utilizes a SAFE-based approach to
model ultrasonic propagation in cylindrical, plate, and arbitrary cross section waveguides.
However, none of these computer solutions permit to model the waveguide response to
external excitations.
The organization of this chapter is detailed next. Section 2 briey reminds the mathematical
background of the PC theory. Section 3 properly describes the main features of our
methodology and how it is implemented in the PCDISP package. Two common transducer
setups for the generation of ultrasonic waves are studied in section 4 with the help of PCDISP.
Finally, we will offer some conclusions and point to lines in which this research could be
further extended.

2. Background and nomenclature

In this section we present a summarized theoretical background on wave propagation
in cylindrical waveguides, treating such aspects as relevant for our purposes; standard
references can be consulted for further information (Graff, 1991; Meeker & Meitzler, 1972;
Rose, 1999).
A waveguide is a physical structure which supports the propagation of mechanical waves
along its elongated direction z, and modies the behaviour of such waves with respect to free
propagation in the bulk material. There are two fundamental characteristics of waveguide
propagation. The rst is the discretization of waves into propagating modes, of which only
a nite number are permitted for a given frequency, and whose properties are determined
by the shape of the cross section and boundary conditions of the waveguide. The second is
the existence of dispersion, which is the nonlinear relationship between wavenumber and
frequency. As a consequence, signals with a signicant bandwidth are distorted as they travel
along the waveguide, because their spectral components propagate at different phase speeds.
The solutions of the wave equation in a cylindrical material are readily found by the use of
potentials and the technique of separation of variables, arriving at the following general form
for the displacement vector ( ):
u) and stress tensor (

u(r, , z) = u(r, )e jkz = u (r )e jn e jkz (r, , z) = 

 (r, )e jkz = (r )e jn e jkz , (1)

where the cylindrical system is used (with coordinates (r, , z), and unit vectors (er , e , ez )),
harmonic time variation e jt is assumed, and is the angular frequency, k the wavenumber,
4
4 Ultrasonic Waves
Will-be-set-by-IN-TECH

and integer n is a separation constant called the circumferential order, which determines the
symmetry of the solutions in the azimuthal direction.
The radial dependent part of the displacement vector and stress tensor is expressed in matrix
form as (Gazis, 1959):

L+ rr (r ) L+
L (r ) L

u r (r )
SV+ zz (r )
u (r ) = u (r ) = D (r )
u , ( r ) = = D (r ) SV+ .

(2)
SV (r ) SV
z
u z (r )
SH+ rz (r ) SH+
SH r (r ) SH

The amplitude coefcient vector A = [L+ L SV+ SV SH+ SH ] T consists of longitudinal

(L), shear vertical (SV), and horizontal (SH) deformation components, and the + and terms
stand for perturbations propagating in the direction of increasing and decreasing radius,
respectively.
The coefcients of matrices D u and D are of the general form Dij (r; n, k, , c), with c being the
elastic compliance tensor of the solid. These matrices are given explicitly in tables 1 and 2, for
the case of a mechanically isotropic material. They have been obtained from the equations of
motion with a symbolic computation program (Maple, 2007), and match those found in (Gazis,
1959), except for some typographical errors in the paper, also propagated to later works as
(Graff, 1991). Matrices D u and D are implemented in the PCDISP package in function pcmat.
In tables 1 and 2, 2 = 2 /c2vol k2 and 2 = 2 /c2rot k2 , where cvol and crot are respectively
the volumetric and rotational speeds of the solid (Rose, 1999). Functions Zn ( x ) and Wn ( x )
are two independent solutions of Bessels differential equation, with, in general, complex
arguments x = r, r. Of the possible choices for Zn ( x ) and Wn ( x ), the numerical stability
of the frequency equation determinant is increased when Bessels ordinary functions Jn ( x )
and Yn ( x ) are employed for real arguments, and the modied Bessel functions In ( x ) and
Kn ( x ) for purely imaginary arguments. With this choice, the contributions to the elastic eld
from the standing waves propagating towards increasing (+) and decreasing (-) radius are
separated (see section 3.1.4 for more on the stability of the frequency equation determinant).
To cope with the fact that the recurrence relationships between Bessels ordinary functions are
different from those of the modied functions (Abramowitz & Stegun, 1964), signs 1 and 2
are introduced, following the scheme of table 3. For complex wavenumbers, PCDISP uses the
ordinary Bessel functions Jn ( x ), Yn ( x ) with complex values, and 1 = 2 = 1.
The solutions of the wave equation are classied in family modes according to their symmetry
properties, which depend on the circumferential index n of equation 1. Modes for which n = 0
have no dependence on the azimuthal coordinate and are labelled axisymmetric. They are
further divided into torsional modes T(0, m) (which only involve the azimuthal component,
and can be thought of as waves which twist the waveguide), and longitudinal modes L(0, m)
(with both radial and axial components). Antisymmetric modes (n 1) are labelled exural
F(n, m), and involve all three components of the displacement vector. In general, multiple
propagating modes exist for a given circumferential order and frequency, so a second index m
is used to order them. Table 4 summarizes this information.
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 55

u = nW ( r ) rW
D11 n n+1 ( r )
u = nZ ( r ) rZ
D12 n 1 n+1 ( r )
u = krW
D13 n+1 ( r )
D14 = krZn+1 ( r )
u

u = nW ( r )
D15 n
u = nZ ( r )
D16 n
u = jnW ( r )
D21 n
u = jnZ ( r )
D22 n
u = jkrW
D23 n+1 ( r )
u = jkrZ
D24 n+1 ( r )
u = jnW ( r ) jrW
D25 n n+1 ( r )
u = jnZ ( r ) j rZ
D26 n 2 n+1 ( r )
u = jkrW ( r )
D31 n
u = jkrZ ( r )
D32 n
u = j rW ( r )
D33 2 n
u = jrZ ( r )
D34 n
u =0
D35
u =0
D36

Table 1. Coefcients of the displacement matrix D u of equation 2 (all Diju coefcients must be
multiplied by 1/r).

3. Methodology for the simulation of the waveguide behaviour

In this section we describe the methodology used to study waveguide generation and
propagation of ultrasonic signals, discuss the numerical issues encountered, and the approach
used in the PCDISP package. The full process consists in four stages: assembly of the
waveguide description matrix, tracing of the dispersion curves, modal analysis of the excited
modes, and modelling of the signal propagation along the waveguide.
While dealing with these topics, we will introduce the relevant PCDISP routines that should
be used for the computations. Table 5 shows the components of the PCDISP software,
arranged by their functionality. Throughout this chapter, we will use monospace fonts (like
pcmat) to refer to programs of the package.

3.1 Assembling the waveguide description matrix

The waveguide description matrix contains the necessary information to study the mechanical
behaviour of the waveguide. It is built by matching the displacement vector and stress tensor
between adjacent layers, and applying the external boundary conditions. In PCDISP, the
physical data of the waveguide is provided in routine pcwaveguide, and the description
matrix itself is built in pcmatdet.
6
6 Ultrasonic Waves
Will-be-set-by-IN-TECH

= (( k2 2 )r 2 + 2( n 1))W ( r ) + 2rW
D11 n n+1 ( r )
= (( k2 2 )r 2 + 2( n 1)) Z ( r ) + 2 rZ
D12 n 1 n+1 ( r )
= 2 kr 2 W ( r ) 2( n + 1) krW
D13 2 n n+1 ( r )
= 2kr 2 Z ( r ) 2( n + 1) krZ
D14 n n+1 ( r )
= 2n ( n 1)W ( r ) 2nrW
D15 n n+1 ( r )
= 2n ( n 1) Z ( r ) 2n rZ
D16 n 2 n+1 ( r )
= ((22 2 + k2 )r 2 2n ( n 1))W ( r ) 2rW
D21 n n+1 ( r )
= ((22 2 + k2 )r 2 2n ( n 1)) Z ( r ) 2 rZ
D22 n 1 n+1 ( r )
= 2( n + 1) krW
D23 n+1 ( r )
= 2( n + 1) krZ
D24 n+1 ( r )
= 2n ( n 1)W ( r ) + 2nrW
D25 n n+1 ( r )
= 2n ( n 1) Z ( r ) + 2n rZ
D26 n 2 n+1 ( r )
= (22 2 k2 )r 2 W ( r )
D31 n
= (22 2 k2 )r 2 Z ( r )
D32 n
= 2 kr 2 W ( r )
D33 2 n
= 2kr 2 Z ( r )
D34 n
=0
D35
=0
D36
= 2nkrW ( r )
D41 n
= 2nkrZ ( r )
D42 n
= k2 r 2 W
D43 n+1 ( r ) 2 nrWn ( r )
= k2 r 2 Z
D44 n+1 ( r ) nrZn ( r )
= nkrW ( r ) + kr 2 W
D45 n n+1 ( r )
= nkrZ ( r ) + kr 2 Z
D46 n 2 n+1 ( r )
= 2jnkrW ( r ) 2jkr 2 W
D51 n n+1 ( r )
= 2jnkrZ ( r ) 2j kr 2 Z
D52 n 1 n+1 ( r )
= j nrW ( r ) j ( 2 k2 )r 2 W
D53 2 n n+1 ( r )
= jnrZ ( r ) j ( 2 k2 )r 2 Z
D54 n n+1 ( r )
= jnkrW ( r )
D55 n
= jnkrZ ( r )
D56 n
= 2jn ( n 1)W ( r ) 2jnrW
D61 n n+1 ( r )
= 2jn ( n 1) Z ( r ) 2jn rZ
D62 n 1 n+1 ( r )
= j kr 2 W ( r ) + 2jkr ( n + 1)W
D63 2 n n+1 ( r )
= jkr 2 Z ( r ) + 2jkr ( n + 1) Z
D64 n n+1 ( r )
= j (2n ( n 1) 2 r 2 )W ( r ) + 2jrW
D65 n n+1 ( r )
= j (2n ( n 1) 2 r 2 ) Z ( r ) + 2j rZ
D66 n 2 n+1 ( r )

Table 2. Coefcients of the stress matrix D of equation 2 (all Dij coefcients must be
multiplied by G/r2 , where G is the shear modulus of the material).
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 77

Wavenumber Frequency range Coefcients Bessel functions

2 , 2 < 0 Zn (r ) = In (r ) Wn (r ) = Kn (r )
Real /k < cvol , crot
1 , 2 = 1 Zn ( r ) = In ( r ) Wn ( r ) = Kn ( r )
2 < 0, 2 > 0 Zn (r ) = In (r ) Wn (r ) = Kn (r )
Real crot < /k < cvol
1 = 1, 2 = 1 Zn ( r ) = Jn ( r ) Wn ( r ) = Yn ( r )
Real /k > cvol , crot 2 , 2 > 0
Imaginary any 1 , 2 = 1 Zn (r ) = Jn (r ) Wn (r ) = Yn (r )
2 , 2 complex Zn ( r ) = Jn ( r ) Wn ( r ) = Yn ( r )
Complex any
1 , 2 = 1

Table 3. Choice of Bessel functions in the solution of the Pochhammer-Chrees equations.

Modes Coefcients Displacement Stress

Torsional T(0, m) SH u z , r
Longitudinal L(0, m) L , SV ur , u z rr , , zz rz
Flexural F(n, m) L , SV , SH ur , u z , u rr , , zz , rz , z , r

Table 4. Notation and non-null components of the amplitude coefcients, displacement

vector, and stress tensor, for the three family modes of a cylindrical waveguide.

3.1.1 Single layer waveguides

Consider an isotropic tube of inner radius rint and outer radius rext in vacuum or air. The
boundary conditions specify that the traction part of the stress tensor is null in both surfaces
of the tube (Gazis, 1959), so:

t = er = [ rr , r , rz ] T = 0 for r = rint , rext , (3)

which leads to the following matrix determinant equation:


D t (rint )
det D (, k) = det = 0, where D t = Dij with i = 1, 5, 6. (4)
D t (rext )

Equation 4 is called the frequency or characteristic equation of the waveguide, and its roots
(, k) determine the proper modes supported by it. Once these roots are known, the vector
of amplitude coefcients A is determined (up to a multiplicative constant) by solving the
following homogeneous system of equations:

D (, k) A = 0. (5)

Since matrix D (, k) is singular at the modes frequency-wavenumber roots, a robust method

for computing the amplitude, like the singular value decomposition (SVD), is recommended
(Press et al., 1992). With A determined, the distribution of u (r ) and (r ) is computed by
routine pcmatdet.
The original Pochhammer-Chree formulation was developed for the simple case of a one-layer
isotropic waveguide in vacuum; this, however, represents just a fraction of the waveguides of
practical importance. Waveguides may be constituted by several layers, might be built with
8
8 Ultrasonic Waves
Will-be-set-by-IN-TECH

Core routines
pcwaveguide Physical description of the waveguide
pcmat Computes matrices D u (r ) and D (r ) (tables 1 and 2)
pcmatdet Assembles and solves the waveguide description matrix
pcviewmatdet View the entries of the matrix determinant
Plotters and solvers of the frequency equation
pcplotmatdet1D One-dimensional plot of the freq. eq. determinant vs. k, f , or cph
pcplotmatdet2D Two-dimensional plot of the freq. eq. determinant in k- f space
pcsolvebisection Bisection method to nd roots of the freq. eq. vs k, f , or cph
pcdisp Plots the phase and group speeds vs. frequency
pckfcurves Traces k- f curves for real, imaginary, and complex k
pcsolverandom Random solutions of the freq. eq. for complex k
Field computing and wave propagation
pcwaveform Finds the displacement vector u (r ) and the stress tensor (r )
pcorthogonalcheck Checks the orthogonality of modes in the waveguide
pcsignalpropagation Simulates the propagation of a signal along the waveguide
Modal analysis
pcextsurfacestress External traction stresses e acting on the waveguide
pcextvolumforce External volumetric forces f e acting on the waveguide
pcplotexcitation Plots the excitation volumetric force and surface stress
pcmodalanalysis Finds the amplitudes of modes excited in the waveguide

Table 5. Components of the PCDISP software.

anisotropic materials, be embedded in the ground, or transport (or be surrounded by, or both)
uids. We will consider next the extensions of the PC theory which permit to model these
situations.

3.1.2 Multilayered waveguides

Some waveguides are formed by several layers: for example, a metallic rod with external
insulation, or a tube embedded in rock. The Pochhammer-Chree approach was rst used
to analyze a two-layer waveguide in (McNiven et al., 1963; Whittier & Jones, 1967), and later
extended to laminated waveguides (formed by an arbitrary number of layers) in (Nelson et al.,
1971). The modern technique to simulate multilayered waveguides is called the Multiple
Layer Matrix (MLM) (Lowe, 1995), and is adapted from the transfer matrix and global matrix
techniques developed by W.T. Thomson and L. Knopoff in the period 1950-1964 to study wave
propagation in stratied media in seismology.
Following the MLM approach, we assemble a system of linear equations for the complete
waveguide, which includes the equations of the elastic waves for each individual layer
(obtained with pcmat), the equations which match the displacement and traction stresses
at the interface between adjacent layers, and the boundary conditions. Consider the example
of a multilayered waveguide shown in gure 1, where a solid cylinder of radius r1 is enclosed
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 99

r1
l=1
r l=2
2
l=3

Fig. 1. Example of a three-layer cylindrical waveguide.

by a tube of inner radius r1 and outer radius r2 , in turn surrounded by an innite medium.
The corresponding system of equations is:

D1u (r1 ) D2u (r1 )
D t (r ) D t (r ) A1
1 1 2 1
A2 = 0. (6)
D2u (r2 ) D3u+ (r2 )
A3+
D2t (r2 ) D3t+ (r2 )

Note that the radiation conditions are used to simplify the system matrix, leading to discard
the + terms in region 1, since no waves can emanate from r = 0; similarly, the - terms in
region 3 are not considered, as no energy comes from r = ; however, both outgoing and
incoming terms are allowed in the middle region. In each case the unneeded columns of
matrix D t are removed. Next, this equation is solved by the SVD method to nd the modes
amplitude vector, in the same way as the single layer waveguide of section 3.1.1.

3.1.3 Anisotropic and uid-loaded waveguides

Materials with mechanical anisotropy are routinely employed to build waveguides, and
their behaviour is modelled by taking into account the adequate compliance tensor for the
material (Pollard, 1977). The important case of hexagonal symmetry is found in waveguides
with transverse isotropy (with respect to the propagation axis z), like beryllium, or in ber
reinforced composite cylinders. Although in this case there are ve elastic constants (up from
two for an isotropic material), the mechanical elds are still separable with the treatment
of Gazis for isotropic waveguides, with different coefcients for the D u and D matrices,
obviously (Mirsky, 1965). In the case of purely orthotropic symmetry (three orthogonal
planes of symmetry), the solution of the wave equation is not separable; still, closed solutions
can be achieved in the form of a Frobenius power series (obtained in (Mirsky, 1964) for
the axisymmetric case, and extended later in (Markus & Mead, 1995) to the asymmetric
problem). A recent state of the art in the theory of mechanical waves in anisotropic cylindrical
waveguides is found in (Grigorenko, 2005).
Materials with elastic losses (damping) are treated by the Kelvin-Voigt viscoelastic model
(Lowe, 1995), which replaces the elements of the compliance tensor c of the material by
10
10 Ultrasonic Waves
Will-be-set-by-IN-TECH

operators which contain time derivatives, therefore modifying Hookes law:

d
= c = c  + c = (c jc ),
dt
where and  are the stress and strain tensors, and component c of the compliance tensor
models the viscoelastic losses. The solutions of the frequency equation for a waveguide with
viscoelastic losses, are, by default, complex wavenumbers.
A case of particular practical importance is that of waveguides including a uid layer (for
example, a pipe carrying a uid, submerged in a uid, or both). The theoretical treatment
depends on the viscosity of the uid.
The inuence of inviscid uids (which do not support shear stresses) on cylindrical
wave propagation is treated theoretically and experimentally in (Sinha et al., 1992). If the
waveguide is submerged in a liquid, propagating modes with complex wavenumber appear,
which radiate (leak) energy into the surrounding uid. That does not happen for waveguides
containing uids and surrounded by vacuum, although the propagating modes themselves
are modied from the unloaded situation.
A treatment for Newtonian viscous uids which is compatible with the PC based formulation
of wave propagation in cylinders is introduced in (Nagy & Nayfeh, 1996). In a form similar to
the Kelvin-Voigt model, the viscous liquid is modelled as an isotropic solid whose compliance
tensor includes complex elements:
4 2
c11 = + c44 c12 = c44 c44 = j, (7)
3 3
where is the compressibility of the uid, and its viscosity. This simple model has shown a
good accuracy in predicting propagation in waveguides with viscous uids (Aristegui et al.,
2001). Indeed, the changes in the propagation of ultrasonic waves in a pipe caused by the
presence of a uid in its interior can be used to measure the longitudinal wave speed and the
viscosity of the uid (Ma et al., 2007).

3.1.4 The case of large frequency thickness product

Solutions of the frequency equation become numerically unstable when the product frequency
times thickness of the waveguide is high. Physically, this phenomenon arises because
the standing waves established in the radial direction of the waveguide are formed by a
combination of terms which increase exponentially with the radius r and others that decrease
exponentially with it. Since it is the sum of both terms which must match the boundary
conditions at r = rint and r = rext , the dynamic range of positive and negative exponentials
in the frequency equation will eventually overow the numerical capacity of the machine
if the radius or frequency increase. With the 64 bits double precision arithmetic of the
IEEE 754 standard, we have found that a direct implementation of the frequency equation
determinant fails when the product f (rext rint ) is higher than approximately 30 MHzmm.
This threshold is easily reached in the NDE of piping with ultrasonic waves, where frequencies
of a few megahertz are common (Rose, 2000).
From table 3, we can see that the problem arises when a modes phase velocity falls below
the volumetric (cvol ) or rotational (crot ) speeds of the solid, making parameters or ,
respectively, become imaginary. This changes the radial dependence of the mode amplitude
from the bounded Bessel functions J and Y, to the exponentially varying I and K (you can
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 11
11

u(r l ) s(rl )
rL= rext
rL-1
rl
r1
r0 = rint

Fig. 2. Partition of the waveguides cross section for solving the large f d problem.

use pcviewmatdet to see the individual entries of the waveguide description matrix). The
behaviour is different depending on the frequency range:
For cph < crot , both and are imaginary, and there exist two solutions of the frequency
equation, corresponding to two Rayleigh modes propagating close to the outer and inner
surfaces of the waveguide, and with amplitudes decaying exponentially from them. As the
frequency increases, these two modes become decoupled, the opposite waveguide surface
can ultimately be ignored, and the solutions are numerically stable.
For crot < cph < cvol , is imaginary and real. While the SV and SH terms remain
bounded, the terms L + and L decrease and increase, respectively, with the radius,
as described above. The condition number of the waveguide matrix grows with the
frequency, and its solution will eventually become unstable.
For cph > cvol , both and are real, and the solution is stable (this is also the case with
purely imaginary wavenumber).
Since the detection of the large f d problem in 1965, several solutions have been proposed to
increase the numerical stability in plane waveguides (Lowe, 1995). We have not been able
to locate similar studies for cylindrical waveguides, so, for the PCDISP software, we have
developed an algorithm adapted from the transfer matrix and global matrix approaches and
discussed here. We consider this method as a new contribution to the literature.
The cross section of the pipe is divided into L layers of equal thickness, where the l-th layer
is given by rl 1 < r < rl , and r0 = rint and r L = rext are the inner and outer radii of the pipe
(see gure 2). In the l-th layer, the displacement vector and traction part of the stress tensor
are given by:


u (r ) D u (r )
= Al , (8)
(r ) D t (r )
l
where the vector of amplitude coefcients Al = [Ll+ Ll SVl+ SVl SHl+ SHl ] T is permitted
to be different for each layer.
We use the shorthand notation:

D u (r l )
Dl = ,
D t (rl )
and scale this matrix by columns for each layer as:

Dls = Dl Gl , (9)

where Gl is a diagonal matrix whose entries are taken as 1/ max col| Dl |.

12
12 Ultrasonic Waves
Will-be-set-by-IN-TECH

The elastic eld [ u ] T is propagated from the inner to the outer part of a layer by the following
equation:


u (r l ) u (r l 1 )
Pl = 0, (10)
(r l ) (r l 1 )
where Pl is the propagator matrix of layer l, and is given by:

Pl = Dls ( Gl1 Gl 1 ) ( Dls1 )1 . (11)

Applying equation 10 to all the layers of the waveguide, we can assemble a global matrix:

u0


0
P1 I6 u
1

P2 I6 1

.. ..
. = 0, (12)
.

PL1 I6 u L 1

PL I6 L1


uL
L

where I6 is the identity matrix of size 6 6.

The boundary conditions to match the wave elds to the surrounding medium (or to other
layers in multilayered waveguides) are introduced at radii rint and rext . In the case of stress
free boundaries, the terms 0 and L are zero, and their corresponding columns are simply
removed from the global matrix.
Once equation 12 has been solved, and we have determined the displacements and traction
stresses at each layer boundary [u l , l ], the vector of amplitude coefcients for each layer is
found by solving:

u l 1
Dls1
l 1
Asl = , (13)
1 ul
D l ( Gl Gl 1 )
s
l
for l = 1, . . . L. The unscaled amplitude vector is simply: Al = Gl Asl .
The algorithm described increases the f d stability limit by a factor proportional to the number
of layers L, at the expense of larger matrices and longer computational time. In PCIDSP, the
algorithm described is written into routine pcmatdet, and is activated automatically if the
waveguide dened in pcwaveguide consists of NL > 1 layers of the same material.

3.2 Computation of the dispersion curves

The roots of the waveguides frequency equation represent the mechanical modes which
satisfy the boundary conditions. The procedure for computing such solutions is described
in this section.
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 13
13

Material Aluminium Standard DN 25, SCH 80

Inner radius (rint ) 12.15 mm Outer radius (rext ) 16.70 mm
Poissons ratio () 0.35 Bar velocity (c0 ) 5000 m/s
Density () 2700 kg/m3 Shear modulus (G) 25.5 GPa

Table 6. Physical data for the aluminium tube used for demonstration of the PCDISP
software in this book chapter.

3.2.1 Nature and ordering of the solutions of the frequency equation

For a given frequency f , the roots of the frequency equation are in general complex
wavenumbers k = kr + jk i . Wavenumbers with k i = 0 correspond to the propagating
or proper modes of the waveguide; those with k i = 0 represent evanescent modes which
attenuate along the axial distance z. Due to the symmetry of the coefcients of the frequency
equation, purely real or imaginary solutions appear in pairs ( kr and jk i ), while complex
solutions do so in quartets ( kr jk i ). The principle of the conservation of energy dictates
which solutions are valid in a waveguide problem. For example, waves which propagate
towards the z+ axis of an innite waveguide, will necessarily have Im{k} 0. Although
signal propagation does not occur for imaginary or complex wavenumbers, those solutions
are needed to fulll the condition of completeness which will be stated in section 3.3.
Furthermore, stationary waves formed by combination of the two wavenumbers kr + jk i
and kr + jk i (provided that k i > 0), can exist locally, storing but not dissipating energy
(Meeker & Meitzler, 1972). In ultrasonic applications, these waves are important in the region
of generation of ultrasonic waves, at waveguide discontinuities (like defects), and at the
waveguide ends.
Solutions of the frequency equation can be traced like continuous curves in the k- f space,
where k = kr + jk i . As an example, we show the dispersion curves of the longitudinal
modes L(0,m) of a sample waveguide with the data shown in table 6 (this waveguide will
be further used in the examples of section 4). The complete spectrum, up to 3 MHz, has
been computed with the pckfcurves routine of PCDISP, and is shown in gure 3. As it can
be seen, the wavenumber curve of a given mode changes from a real value (shown in blue
colour) to purely imaginary (coloured red, and projected into the negative wavenumber axis),
or complex (plotted in green, with the real part on the positive k axis, and the imaginary part
on the negative k axis), in a complicated fashion.
After nding all possible roots of the frequency equation, branches L(0,m) have to be ordered
in such a way that each mode is assigned a unique, continuous wavenumber between the
maximum frequency f max and zero frequency. Part (b) of gure 3 shows a reduced frequency
range of the spectrum in part (a), and illustrates the convention for labelling modes. PCDISP
uses the following rules about the behaviour of dispersion curves (Meeker & Meitzler, 1972):
Only the rst torsional T(0,1), longitudinal L(0,1), and rst exural F(1,1), modes propagate
down to zero frequency with real wavenumber.
Higher order modes switch from real to imaginary wavenumber at the cutoff frequencies
when the wavenumber becomes null (k = 0, f = f cutoff ), and the phase speed innite.
Miminum points (d/dk = 0, d2 /dk2 > 0) of either real or imaginary wavenumber
branches are also cutoff frequencies (with nite phase speed), from which complex
wavenumber branches start.
14
14 Ultrasonic Waves
Will-be-set-by-IN-TECH

(a) Spectrum of longitudinal modes L(0,m) up to 3 MHz (b) First longitudinal modes L(0,m) up to 850 kHz
3000
800 L(0,5)
L(0,4)

2500 700
L(0,5)
L(0,3)
600
(c)
2000
500 L(0,4) &
(d)

f (kHz)
f (kHz)

L(0,5) L(0,2)
1500 400

300
1000
L(0,1)
(b) 200 L(0,3)
500
100 L(0,2) &
L(0,3)
0 0
6000 4000 2000 0 2000 4000 6000 1000 500 0 500 1000 1500
Im{k} (1/m) Re{k} Im{k} (1/m) Re{k}
(c) Zoom of near cross of real branches
1869.2 (d) Zoom of near cross of imaginary branches

1869.1 1525.15

1869 L(0,10)
L(0,8) 1525.1
1868.9

1868.8 1525.05
f (kHz)

L(0,10)
f (kHz)

1868.7 L(0,7)
1525 L(0,9) &
L(0,10) L(0,9)
1868.6

1868.5 1524.95

1868.4
L(0,9)
1524.9
1868.3

1868.2
1719.5 1720 1720.5 1721 1416.5 1416.481416.461416.441416.42 1416.4 1416.38
Re{k} (1/m) Im{k} (1/m)

Fig. 3. Wavenumber-frequency plot for the L(0,m) modes of the aluminum pipe of table 6, for
frequencies up to 3 MHz (part a). Part b shows the low-frequency spectrum, and the
labelling scheme for the modes. Parts c and d show near crosses of the branches of two
different modes; they have to be largely magnied to be visible.

Complex branches terminate either at zero frequency or at the maximum points (d/dk =
0, d2 /dk2 < 0) of imaginary branches.
The branches are ordered such that k i is positive for modes propagating in the z+ direction,
and that the sign of the group speed does not change along the curve (although it becomes
null at the cutoff frequencies, and at the purely imaginary branches).

3.2.2 Algorithm for curve tracing in PCDISP

In order to generate continuous curves k = k( f ), from f = 0 to f = f max , for each mode,
pckfcurves rst traces the real wavenumber branches, which start at solutions found with
pcsolvebisection at the f = f max and k = kmax axes, and nish at the k = 0 axis. If
one is only interested in propagating modes, this nishes the procedure, and the pcdisp
routine can be used to plot the phase and group speeds of propagating modes in the specied
frequency range. Otherwise, the next step consists in tracing branches with purely imaginary
wavenumber, which start at points (0, f cutoff ) on the vertical axis of zero wavenumber, and
nish when they reach the f = f max , k = jkmax axes, or at another cutoff frequency in the
k = 0 axis.
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 15
15

Search
interval
i+1
dq

dr(i)
i

dr(i-1)
Neighbouring
branch
i-1

dr(i-2)

i-2 Current
branch

Fig. 4. Illustrates the root tracing algorithm used by PCDISP.

Parts (c) and (d) of gure 3 show the reason why a robust algorithm for curve tracing of the
dispersion branches is needed, in order to avoid the apparent crossings between branches,
such as modes L(0,7) and L(0,8), with real wavenumbers, and modes L(0,9) and L(0,10),
with imaginary wavenumbers. The curve tracing method used in pckfcurves is shown
in gure 4. The dispersion curve being traced is extrapolated from the three last computed
points {i 2, i 1, i } to dene an angular interval of width d in a circle of radius dr (i )
centered in the last correctly determined point (i) (dr is the step size of the algorithm, in
normalized coordinates of the k- f space). If a sign change is found in this interval, the
algorithm proceeds with a bisection method to accurately estimate the position of point i + 1
(this is the normal situation). Otherwise, the dispersion curve might have undergone a sudden
change of curvature, or another mode might have come very close to the one being traced,
provoking multiple sign changes. In this case, the step dr (i ) is decreased, or, if needed, points
i 1, i 2, etc, are recomputed with a smaller step dr. Summarizing, the tracing algorithm
of PCDISP keeps track of the curvature of the branch and the proximity of neighbouring
branches, adjusting the interval step between consecutive points accordingly.
It must be pointed out that the frequency equation has spurious solutions at the lines with
slope equal to the volumetric and rotational speeds of the solid (/k = cvol and /k =
crot ), which have to be removed by the root nding algorithm. In the case of multilayered
waveguides, the same phenomenon happens for the speeds of each layer.
The dispersion curves are completed by tracing the complex wavenumber branches k = kr +
jk i , starting from the extrema points of the real/imaginary wavenumber branches. Tracing
the complex wavenumber branches needs more computational effort, since it is required to
solve simultaneously for the real and imaginary parts of the wavenumber; we have obtained
satisfactory results with Mullers method (Press et al., 1992) using also an adaptive step.
Finally, the rules enumerated at the end of section 3.2.1 are used to convert the initially
obtained branches into continuous dispersion curves k = k( f ) for each mode. An example
of this procedure is shown in part (b) of gure 3. First, note that all longitudinal modes except
L(0,1) exhibit cutoff. Mode L(0,2) is cut off at (k = 0, f = 59.7 kHz), where it switches
to a branch with imaginary wavenumber which goes on until (k = 67j m1 , 58.5 kHz),
16
16 Ultrasonic Waves
Will-be-set-by-IN-TECH

eext

eint
r int
D int D ext
rext
D

Fig. 5. Cross section of the waveguide and denition of the regions of integration for modal
analysis.

the minimum point of the imaginary branch. At that point, it is joined by the branch
corresponding to mode L(0,3), and both go down to zero frequency with negative complex
conjugate wavenumbers kr + jk i and kr + jk i .
Similarly, mode L(0,4) is cut off at the minimum point of the real wavenumber branch (k =
345 m1 , f = 621.5 kHz). Mode L(0,5) is cut off at (k = 0, f = 698.6 kHz), changes to
imaginary wavenumber until it reaches point (k = 0, f = 669.4 kHz), and then again to real,
but negative, wavenumber down to (k = 345 m1 , f = 621.5 kHz), where it joins mode L(0,4).
Mode L(0,5) has a negative wavenumber (and consequently, negative phase speed /k), in
order to maintain a positive group velocity d/dk, since this is a propagating mode in the z+
direction of the waveguide, according to the last rule of section 3.2.1. Below 621.5 kHz, the
dispersion curves of modes L(0,4) and L(0,5) descend to zero frequency with negative complex
conjugate wavenumbers, in the same way as modes L(0,2) and L(0,3).

3.3 Modal analysis

Modal analysis is a mathematical technique which permits to compute the dynamic response
of a waveguide subject to arbitrary external forces, as an expansion of the excited wave over
the set of normal modes of the waveguide, as dened in section 2 (Auld, 1973). Modal analysis
is based upon two properties of normal modes: orthogonality, the existence of a scalar product
which is null for any two different modes; and completeness, the capacity of the set of normal
modes to span arbitrary waveforms in the waveguide.
For two different modes of the waveguide (1) and (2) of the form given in equation 1, the
orthogonality relationship (Auld, 1973) establishes that:
 (u1 
2 u2 
1 ) = 0, (14)

which is applicable to linear elastic materials and also to piezoelectric or magnetostrictive

linear materials, assuming no elastic or dielectric losses. Later on this result will be generalized
to include external forces and stresses.
For separable vector elds in the z coordinate, the tridimensional divergence operator can be
written as:
 {} =
 { } + { } ez ,
z
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 17
17

so equation 14 becomes:

 (u1 
2 u2 
1 )e j( k1 k2 ) z + j(k1 k2 )e j( k1 k2 ) z (u1 
2 u2 
1 ) ez = 0. (15)

Discarding the common factor e j( k1 k2 ) z , integrating over the cross section D of the waveguide,
and applying the divergence theorem, we nd that:

2 u2 
(u1  1 ) en dl + j(k1 k2 ) 2 u2 
(u1  1 ) ez dS = 0. (16)
D D

In equation 16, D = Dint Dext represents the inner and outer surfaces of the waveguide,
and the normal unit vector en is taken on each surface pointing out of the waveguides interior,
as shown in gure 5.
Because for proper modes the surface traction stress is null ( t = 
en = 0 in D), the rst
integral of equation 16 is zero. Then a suitable scalar product of modes (1) and (2) is:
rext
j j
P12 = 2 u2 
(u1  1 ) ez dS = (u1 2 u2 1 ) ez r dr. (17)
4 D 2 rint

In the right part of equation 17 we have assumed that the circumferential order n of modes
(1) and (2) is the same; otherwise, P12 is zero automatically due to the integration over the
coordinate. The factor j/4 is introduced so that the quantity P11 equals to the integral of
the acoustic Poynting vector in the cross section of the waveguide, i.e., the power transported
by the mode. For nonpropagating modes with k i = 0, P11 is zero.
With this notation, equation 16 reduces to:

(k1 k2 ) P12 = 0, (18)

which implies that P12 = 0 unless k1 = k2 . In PCDISP, mode orthogonality can be veried
with routine pcorthogonalcheck.
The second condition for modal analysis is completeness, which is based on the premise that
an arbitrary perturbation in the waveguide can be expanded in the set of normal modes:

u1 (r, , z) = a p (z)up (r, ) 1 (r, , z) =

 a p (z)p (r, ). (19)
p p

where the modes are indexed by p, and no distinction has been made between propagating
and evanescent modes. The problem lies in computing the set of coefcients a p (z), when the
waveguide is under an arbitrary excitation composed of:
1. A vector force eld  f e (r, , z) acting on the bulk material of the tube (region D). In PCDISP,
this vector force eld is dened in pcextvolumforce.
e (r, , z) applied to the tube surfaces (region D). In PCDISP, this surface
2. A traction stress 
stress is dened in pcextsurfacestress.
In the case of existence of external elds, the orthogonality relationship, equation 14, must be
generalized to (Auld, 1973):

 (u1 
1 ) = u1 f2 + u2 f1 ,
2 u2  (20)

where 
f 1,2 (r, , z) represent the forcing terms.
18
18 Ultrasonic Waves
Will-be-set-by-IN-TECH

If we take subscript (1) for the wave existing in the waveguide (equation 19) and (2) as the
q-th proper mode of the waveguide:

u2 (r, , z) = uq (r, )e jk q z 2 (r, , z) = 

 q (r, )e jk q z , (21)

we can insert both expressions into equation 20, and, letting f2 = 0 since it corresponds to a
normal mode in the waveguide, we obtain that:


 a p (z)(up 

p )e q = uq f1 e jk q z .
q uq  jk z
(22)
p

Operating with the divergence operator:

 
da p (z)
p
a ( z ) 
( 
u p 
q


u 
q p ) + jk a
q p ( z ) + (up  p ) ez = uq 
q uq  f 1 , (23)
p p dz

which, when integrating across the transversal section of the tube, with the divergence
theorem, reduces to:

a p (z) q uq 
(up  p ) en dl +
p D
  
( A)
 
d
jk q + a p (z) q uq 
(up  p ) ez dS = uq f1 dS . (24)
dz D
p  D  
  
(C)
( B)

q en = 0 in D, and we can combine the sum over the index p to

The rst term simplies as 
nd: 
( A) = (uq 
1 ) en dl.
D
As for the second term, if we assume that p is a propagating mode, we recall the property
of orthogonality of the propagating modes of the waveguide, and the denition of acoustic
power Pp in equation 17, to eliminate all the terms of the sum except the one for which p = q:
 
4Pp d
(B) = jk p + a p ( z ).
j dz

Inserting this result in the preceding equation:

  
4Pp d
jk p + a p (z) = (up 
1 ) en dl + up f1 dS, (25)
j dz D D

where the contributions to the mode amplitude due to the volumetric forces ( f1 = fe ) and the
1 = 
surface tractions ( e ) appear clearly separated. With the following denitions:
 
f ps (z) = j [ up (r, ) 
e (r, , z)] en dl = j e jn p [ u p (r ) 
e (r, , z)] en dl, (26)
D D
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 19
19

and

f pv (z) = j [ up (r, ) fe (r, , z)] dS = j e jn p [ u p (r ) fe (r, , z)] dS, (27)
D D

equation 25 is changed into an ordinary differential equation, solvable by a standard change

of variables, resulting in:

e jk p z 
a p (z) = e jk p z [ f ps (z ) + f pv (z )] dz , (28)
4Pp Rg

where the integration takes place in the region R g where the generating terms f s and f v are not
null, and z is the point where the ultrasonic signal is observed, in the direction of increasing z
from region R g .
If p is a non-propagating mode, our computation method is changed slightly, since Pp = 0.
However, Pp,p = 0, and we can set q = p , k q = kp , and modify equations 26-28 accordingly.
As a summary, we have established the equations that permit to nd the amplitude of the
proper modes excited in the waveguide by an arbitrary set of external driving forces. These
equations are used by routine pcmodalanalysis of PCDISP .

3.4 Propagation of waveforms in the waveguide

The modal analysis equations discussed in section 3.3 permit to obtain the frequency response
of a transducer exciting the waveguide. In many applications we want to predict what
ultrasonic waveforms will be obtained at a certain distance z from the excitation source, when
the transducer is excited by a nite length time signal, i.e., to model the transient behaviour
of the system.
The method to study the propagation of signal waveforms is relatively straightforward
(Doyle, 1997). Let u (0, t) be the input signal in the transducer (placed in region R g ), and
U (0, ) = F [ u (0, t)] its Fourier transform. When this signal excites the waveguide, we
determine the corresponding volumetric forces and surface stresses, and evaluate terms a p (z)
from equation 28 for each signicant frequency component of U and all normal modes
of the waveguide. Note that the terms a p (z) incorporate both the frequency response of
the transducer itself (inside the integral term) and the effect caused by signal propagation
(exp( jk p ( )z)). Thus, the frequency components at a distance z from the generating region
are given by:
U (z, ) = U (0, ) a p (z, )
(29)
U (z, ) = U (z, ),
where the values of the Fourier transform for negative frequencies are taken as complex
conjugate of the positive ones, in order to obtain a real time signal. The waveform at z is
recovered by the inverse Fourier transform:

u (z, t) = F 1 [U (z, )]. (30)

The imaginary and complex wavenumber parts of the spectrum are required in equation 29
if the exciting signal has signicant frequency content below cutoff of the propagating
mode, and the measurement point is not far away from the transducer, as was described in
section 3.2.1.
20
20 Ultrasonic Waves
Will-be-set-by-IN-TECH

period
(B) tail head

Fig. 6. Transducers used for generation of ultrasonic waves in cylindrical waveguides: (a)
Electromagnetic Acoustic Transducer (EMAT) for Lorentz force excitation; (b) Time-Delay
Periodic Ring Array (TDPRA) for piezoelectric excitation.

The dispersive effect characteristic of waveguide propagation is frequently undesired in

applications, since it implies a distortion of the original signal. One practical way of
minimizing dispersion is to employ signals with narrow spectral content (typically by
windowing a sine pulse train) with a central frequency in the region where the curve cph ( ) =
/k( ) is relatively at (Lowe et al., 1998). Where this is not feasible, compensation methods
based on inverting the nonlinear k = k( ) dependence have been developed (Wilcox, 2003).
The propagation of ultrasonic signals in the waveguide is simulated in PCDISP with routine
pcsignalpropagation. When customising this routine, the user must be careful to zero
pad the excitation signal u (0, t) at the end such that the resulting time window has enough
duration to allow propagation of the signal for the distance between transducer and receiving
point. Likewise, a sampling frequency high enough to cover all the spectral content of the
signal should be used; in practical applications, oversampling the signal above the Nyquist
rate is advantageous since it enhances the signal to noise ratio.

4. Demonstration of the methodology

In this section we will illustrate the use of the methodology described in this chapter and the
developed software, to model the performance of a given transducer. The complete procedure
is summarized in table 7, along with the needed routines of the PCDISP package.
Two transducer setups commonly found in ultrasonic guided wave applications will be
analyzed (see gure 6). The rst one is an electromagnetic acoustic transducer (EMAT) used to
generate ultrasound in metallic waveguides without physical contact between the transducer
and the sample; the second, an array of piezoelectric rings which generates ultrasound by
mechanically loading the external tube surface. We begin by considering the mechanical
behaviour of the sample waveguide.

4.1 Dispersive curves of longitudinal modes

We will continue to use the waveguide described in table 6. The complete signal spectrum of
the longitudinal modes was already shown in gure 3; in gure 7 we plot the phase and group
speeds in the range of 0 to 800 kHz. An important requisite for guided waves applications of
ultrasound is the selection and exploitation of a single propagating mode, in a region where
dispersive effects are minimum, since, as a general principle, an external force will excite all
propagating modes existing within its bandwidth (Lowe et al., 1998). We will consider two
possibilities: excitation of mode L(0,2) at frequency f 1 = 250 kHz, and use of mode L(0,3) at
frequency f 2 = 565 kHz (see gure 7 b). The dispersion curves of these modes are relatively
at at these frequencies, and their group speeds are higher than those of other coexisting
modes.
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 21
21

Part A: Input data of the waveguide (pcwaveguide)

Assembling of the waveguide description matrix (pcmatdet)
Compute the dispersion curves k = k( ) (pckfcurves)

Part B: Frequency / transient response of the waveguide (pcmodalanalysis

and pcsignalpropagation)
Fourier transform of excitation signal: U (0, ) = FFT[ u (0, t)]
 Loop over propagating modes p
 Loop over frequencies of the signals spectrum
Compute amplitude vector A p by SVD solution of D (, k p ) A p = 0
Computation of ultrasonic elds of p-th mode: up ,  p (pcmatdet)

 Integration in the region of generation z R g
Compute f ps (z ) by integrating 
e (r, , z) (pcextsurfacestress) in D
Compute f p (z ) by integrating fe (r, , z) (pcextvolumforce) in D
v 

 End of integration in region R g

Computation of mode amplitude a p (z) (equation 28)
Determine gain for frequency : U p (z, ) = U (0, ) a p (z)
 End loop over frequency
Set negative frequencies of the FFT: U p (z, ) = U p (z, )
Inverse Fourier transform: u p (z, t) = IFFT[U p (z, )] (eq. 30)
 End loop over propagating modes
Sum over propagating modes u (z, t) = p u p (z, t)

Table 7. General method used for modal analysis computations, and required PCDISP
routines.

(a) Phase speed of the L(0,m) modes of the waveguide (b) Group speed of the L(0,m) modes of the waveguide
10000 6000

9000 L(0,2) f1
L(0,4) 5000
8000 L(0,3) L(0,4)
TDPRA load
7000 line 1 f
Group speed (m/s)

4000 L(0,3)
Phase speed (m/s)

2
6000 L(0,2) TDPRA load
line 2 f2 L(0,1)
5000 f 3000
1

4000 L(0,1)
2000
3000

2000
1000
1000

0 0
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
Freq (kHz) Freq (kHz)

Fig. 7. (a) Phase and (b) group speeds of the rst axisymmetric modes of the waveguide of
table 6.
22
22 Ultrasonic Waves
Will-be-set-by-IN-TECH

(a) Displacements, mode L(0,2), f = 250 kHz (b) Stresses, mode L(0,2), f = 250 kHz

16.5 16.5
u
16 r 16 rr
uz

15.5 15.5 zz
rz
15 15
Radius (mm)

Radius (mm)
14.5 14.5

14 14

13.5 13.5

13 13

12.5 12.5

1 0.5 0 0.5 1 1 0.5 0 0.5 1

Displacement Stress
(c) Displacements, mode L(0,3), f = 565 kHz (d) Stresses, mode L(0,3), f = 565 kHz

16.5 16.5
ur
16 16
uz
15.5 15.5

15 15
Radius (mm)

Radius (mm)

14.5 14.5

14 14
rr
13.5 13.5
zz
13 13
rz
12.5 12.5

1 0.5 0 0.5 1 1 0.5 0 0.5 1

Displacement Stress

Fig. 8. Displacement and stress proles for mode L(0,2) at f = 250 kHz (upper row), and
mode L(0,3) at f = 565 kHz (bottom row), computed with pcwaveform.

The displacement and stress proles of the selected modes are shown in gure 8. Their
determination is important in NDE applications since the sensitivity of a propagating mode to
a waveguide defect depends on the matching between the defects shape and the mode prole
(Ditri, 1994).
In order to minimize the inuence of dispersion in the propagation of signals, we use as
excitation signal a tone burst consisting of ncyc = 16 cycles of a central frequency modulated
by a raised cosine window. This waveform does a good job in exciting a single frequency of
the waveguide with a nite length signal and minimum sidelobes (Oppenheim et al., 1999).

4.2 Generation of ultrasonic waves with an electromagnetic acoustic transducer

Electromagnetic acoustic transducers (EMATs) are used to excite ultrasonic waves in metallic
waveguides by non contact means (Cawley et al., 2004). Basically, an EMAT consists of a
generating coil, which creates a dynamic eld H (t) at ultrasonic frequencies, and a bias
magnet providing a constant magnetic eld H0 , as shown in gure 6 (a). The physical
phenomenon that couples the magnetic eld with the elastic eld in our non-ferromagnetic
aluminium waveguide is the Lorentz force resulting from the interaction between the eddy
currents J (t) induced in the tube by the dynamic eld and the bias eld H0 , which create a
volumetric force given by:
f em = 0 J H0 . (31)
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 23
23

Previously to computing the Lorentz force, we must determine the distribution of the
electromagnetic eld in the waveguide. Although an exact solution exists, it is complicated
(Dodd & Deeds, 1968), so, for the purposes of this example, we will consider a simplied
model in which the penetration depth of the EM eld in the metal ( = (2/0 e )1/2 , with
0 being the magnetic permeability of vacuum and e = 38 MS/m the electrical conductivity
of aluminum) is small compared with the thickness rext rint of the tube (in our case <
(rext rint )/10 for f > 30 kHz). Then the magnetic eld in the tube is mainly axial and can be
written as:

H (r, z) = Hzext (z) exp[(1 + j)(rext r )/] ez rint r rext , (32)

where Hzext (z) is the axial eld at the outer surface (r = rext ) of the tube. We will further
assume that Hzext (z) can be obtained by the elementary formula:

Ns Is z Ls z ,
Hzext (z) =  + (33)
2L s R2s + z2 R 2 + ( L z )2 s s

where L s and Rs are the solenoids length and radius, Ns the number of turns, and Is the
current through one turn.
Computing the eddy current with Ampres law, J (r, z) = H (r, z), and since the bias eld
is H0 = H0 ez , we obtain:
1+j
f em (r, z) = 0
Hzext (z) H0 exp[(1 + j)(rext r )/] er . (34)

And using equation 27, we can compute the volumetric forcing term for mode p in the
waveguide as : rext
f pv (z) = 2j u pr (r ) f rem (r ) r dr. (35)
rint
If we consider an EMAT with a solenoid length L s = 30 mm and Rs = rext = 16.70 mm
we obtain the dependence of transducer gain with frequency shown in gure 9 (a), where
the radial component of displacement at the surface, ur (rext ), is plotted on a log scale on
the vertical axis. As we can see, the EMAT exhibits high gain in the low frequency region,
which unfortunately coincides with the zone where the dispersive behaviour of mode L(0,1)
is maximum, making it difcult for guided waves applications. In parts (b) and (c) of the
same gure we show the transient response of the waveguide when excited with the pulse
train described in section 4.1, for central frequencies of 250 kHz and 565 kHz, respectively,
when the radial component of surface displacement (ur ) is measured at a point z = 1.5 m from
the EMAT. For 250 kHz, the faster propagating L(0,2) mode is excited with lower amplitude
than the mode L(0,1), while for 565 kHz all modes are excited with approximately equal
amplitudes, appearing also very close in the time domain.
Summarizing, the basic EMAT described in this section shows poor mode selectivity control,
exciting all modes within the bandwidth of the source signal with relatively equal amplitudes,
which makes it a poor choice for this waveguide.

4.3 Generation of ultrasonic wave with piezoelectric surface loading

In this section we consider a Time-Delay Periodic Ring Array (TDPRA), an ultrasonic
transducer with very good mode selectivity and also capable of achieving one directional
emission, emitting much more ultrasonic energy from its enhanced side than in the direction
24
24 Ultrasonic Waves
Will-be-set-by-IN-TECH

(b) Transient response of the EMAT (f = 250 kHz)

(a) EMAT frequency response (Ls = 30 mm) 1
10
0.8
20 L(0,1)
0.6
System gain, |ur(r=b)| (dB)

30 0.4
L(0,2)
40 L(0,1) 0.2

0
50
0.2
60 L(0,4)
L(0,2) 0.4
70
L(0,3) 0.6
80 0.8

90 1
100 200 300 400 500 600 700 800 0 200 400 600 800 1000
Frequency (kHz) Time (s)
(c) Transient response of the EMAT (f = 565 kHz)
1

0.8

0.6 L(0,1) L(0,2)

L(0,3)
0.4

0.2

0.2

0.4

0.6

0.8

1
0 200 400 600 800 1000
Time (s)

Fig. 9. (a) Computed frequency response of the EMAT for solenoid length L s = 30 mm;
transient waveforms at (b) 250 kHz and (c) 565 kHz. The signals in the transient plots have
been normalized to unit amplitude.

of its weakened side (Zhu, 2001). A TDPRA, shown in gure 6 (b), consists in a number of
piezoelectric rings capable of exerting a pressure loading on the outer surface of the tube. The
rings are organized into Np identical periods of Nr rings each, with the length of a period
matched to the wavelength of the mode to be excited. The rings of a period are connected
to the same excitation source, but with a relative delay between them proportional to their
position within the period of the TDPRA; this scheme is repeated throughout the TDPRA.
This reinforces the wavelength matching of a single period and also creates constructive
interference in the enhanced direction (the head of the array) and destructive in the other
(the tail).
To model numerically the TDPRA, we assume that each ring has width zw , with a separation
zs between adjacent rings and vibrates in its thickness mode, exerting a pressure loading,
rr = p over the outer surface of the waveguide, constant for all frequencies. This leads
to an axisymmetric loading (no dependence). In this case, the term corresponding to the
surface loading (equation 26) is:

f ps (z) = 2jrext ur (rext ) p(z), (36)

Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 25
25

while the volumetric term is null. The total pressure p(z) is a sum over the Np periods of Nr
rings each:
Np Nr 1
zci j2mod( i,Nr ) /Nr
p ( z ) = p0 (
zw
)e , (37)
i =0
where zci = (zw + zs )i + zw /2 is the position of the center of each ring, and ( x ) = 1 for
| x | < 1/2, ( x ) = 0 for | x | > 1/2, is the rectangular function.
The parameters of the TDPRA must be tuned to the frequency to be excited. The load line of
the TDPRA, given by cph = Nr (zw + zs ) f , is shown in gure 7 (a), along with the phase speed
curves of the aluminum tube given in section 4.1, for two different designs: the rst one with
Np = 4, Nr = 8, zw = 2.2 mm and zs = 0.4 mm, intended to excite mode L(0,2) at 250 kHz,
and the second with Np = 5, Nr = 6, zw = 1.5 mm, zs = 0.4 mm for excitation of mode L(0,3)
at 565 kHz. The intersection points of these lines with the phase speed curves correspond to
the frequencies for which the TDPRA achieves maximum efciency in mode coupling.
The transducer gain of the rst TDPRA is shown in gure 10 (a). For 250 kHz, the mode L(0,2)
is effectively excited, and the excitation frequency can be ne tuned to make it coincide with
(b) TDPRA transient (f = 250 kHz)
(a) TDPRA frequency response (radial disp ur) 2
0 L(0,2)
L(0,1)
L(0,1) 1.5
5
L(0,2)
head
10 1
System gain, |ur(r=b)| (dB)

15
0.5
20
0
25

30 0.5

35 tail
1
40
1.5
45

50 2
150 200 250 300 350 0 200 400 600 800 1000
Frequency (kHz) Time (s)
(d) TDPRA transient (f = 565 kHz)
(c) TDPRA frequency response (radial disp ur)
2
0
L(0,2)
L(0,3) L(0,3)
5 1.5 L(0,2)
L(0,4) L(0,1)
head
10
1
System gain, |u (r=b)| (dB)

15
0.5
20 L(0,1)
r

25 0

30 0.5
35 tail
1
40
1.5
45

50 2
500 550 600 650 0 200 400 600 800 1000
Frequency (kHz) Time (s)

Fig. 10. Plots of the (a) frequency and (b) transient response of the rst design of the TDPRA
at f = 250 kHz; plots of the (c) frequency and (d) transient response of the second design of
the TDPRA at f = 565 kHz. In the gain plots, the dashed lines correspond to the opposite
(tail) direction. The signals in the transient plots have been normalized to unit amplitude.
26
26 Ultrasonic Waves
Will-be-set-by-IN-TECH

a minimum of the amplitude of the L(0,1) mode, improving the dynamic range between the
two modes. The simulation of the propagated wave (at a distance z = 1.5 m from the TDPRA)
gives the expected results (part (b)). The results obtained with the second TDPRA design are
shown in parts (c) and (d) of gure 10. Mode L(0,3) dominates in this case at 565 kHz, with
a higher relative amplitude over the L(0,1) and L(0,2) modes. In this case, however, as the
group velocities are similar, the received signals appear closer in the time view of part (d). In
all cases, ultrasonic generation in the head side has higher amplitude than in the tail side.
As a conclusion, the TDPRA is an efcient transducer for generating ultrasonic signals
in cylindrical waveguides, and its feature of phase and wavelength matching permits to
excite modes selectively, ne tune the system gain to a desired frequency, and direct the
generated signal in only one direction. Although in this communication we have concerned
ourselves only with axisymmetric transducers, PCDISP can also be used to study excitation of
nonsymmetric modes by piezoelectric arrays (see reference (Li & Rose, 2001) for an example).

5. Conclusions
In this chapter we have presented a methodology to model the dynamic response of a
waveguide of cylindrical symmetry when subject to an arbitrary set of external forces acting
at ultrasonic frequencies, based on the combination of the mechanical Pochhammer-Chree
equations and modal analysis techniques. Furthermore, a software package (named PCDISP),
created in the Matlab environment, is offered freely with the intention of saving other
researchers from the time needed for implementation of the PC theory equations, permitting
them to focus on their particular problems.
Throughout this communication, we have paid special attention to the numerical issues
of stability of the matrix determinant for large frequency thickness products, provided
algorithms for robust root solving and tracing of the dispersion curves, and modelled
the dispersive effect of the waveguide on signal propagation. The methods described in
this chapter are valid for waveguides formed by any number of layers as long as they
have cylindrical symmetry. The PCDISP software can be further extended to consider
materials with anisotropy (transversely isotropic and orthotropic), as well as materials with
elastic damping and waveguides surrounded by, or containing, uids. Guidelines for such
extensions are given in the text.
The performance of modal analysis is illustrated by studying two common transducers
employed in guided wave ultrasonic applications: an electromagnetic-acoustic transducer
and a time-delay piezoelectric ring array. We believe that transducer analysis with
quantitative results is achieved comparatively easier and faster than with other competing
techniques like spectral or nite element methods, obtaining signicant time savings in the
design stage of ultrasonic transducers.

6. Acknowledgments
The nancial support for this work was provided by the Spanish Ministerio de Ciencia e
Innovacin through project Lemur (TIN2009-14114-C04-03).

7. References
Abramowitz, M. & Stegun, I. A. (1964). Handbook of Mathematical Functions with Formulas,
Graphs, and Mathematical Tables, 9 edn, Dover.
Modelling the Generation
Modelling the Generation and
and Propagation of Propagation
Ultrasonic of Ultrasonic
Signals in Cylindrical Waveguides Signals in Cylindrical Waveguides 27
27

Aristegui, C., Lowe, M. J. S. & Cawley, P. (2001). Guided waves in uid-lled pipes
surrounded by different uids, Ultrasonics 39(5): 367375.
Auld, B. A. (1973). Acoustic Fields and Waves in Solids, Wiley Interscience.
Bocchini, P., Marzani, A. & Viola, E. (2011). Graphical user interface for guided acoustic waves,
Journal of Computing in Civil Engineering 25(3): 202210.
Cawley, P., Lowe, M. & Wilcox, P. (2004). An EMAT array for the rapid inspection of large
structures using guided waves, Journal of Nondestructive Testing 9(2): 18.
Damljanovic, V. & Weaver, R. L. (2004). Propagating and evanescent elastic waves in
cylindrical waveguides of arbitrary cross section, Journal of the Acoustical Society of
America 115(4): 15721581.
Ditri, J. J. (1994). Utilization of guided elastic waves for the characterization of circumferential
cracks in hollow cylinders, Journal of the Acoustical Society of America 96(6): 37693775.
Ditri, J. J. & Rose, J. L. (1992). Excitation of guided elastic wave modes in hollow cylinders by
applied surface tractions, Journal of Applied Physics 72(7): 25892597.
Dodd, C. V. & Deeds, W. E. (1968). Analytical solutions to eddy-current probe-coil problems,
Journal of Applied Physics 39(6): 28292838.
Doyle, J. F. (1997). Wave Propagation in Structures: Spectral Analysis Using Fast Discrete Fourier
Transforms, 2nd edn, Springer.
Folk, R., Fox, G., Shook, C. A. & Curtis, C. W. (1958). Elastic strain produced by sudden
application of pressure to one end of a cylindrical bar. I. theory, Journal of the Acoustical
Society of America 30(6): 552558.
Gazis, D. C. (1959). Three-dimensional investigation of the propagation of waves in hollow
circular cylinders. I. Analytical foundation. II. Numerical results., Journal of the
Acoustical Society of America 31(5): 568578.
Graff, K. F. (1991). Wave Motion in Elastic Solids, Dover.
Grigorenko, A. Y. (2005). Numerical analysis of stationary dynamic processes in anisotropic
inhomogeneous cylinders, International Applied Mechanics 41(8): 831866.
Hayashi, T., Kawashima, K., Sun, Z. & Rose, J. L. (2003). Analysis of exural mode focusing
by a semianalytical nite element method, Journal of the Acoustical Society of America
113(3): 12411248.
Jia, H., Jing, M. & Rose, J. (2011). Guided wave propagation in single and double layer hollow
cylinders embedded in innite media, Journal of the Acoustical Society of America
129(2): 691700.
Karpnger, F., Gurevich, B. & Bakulin, A. (2008). Modeling of wave dispersion along
cylindrical structures using the spectral method, Journal of the Acoustical Society of
America 124(2): 859865.
Li, J. & Rose, J. L. (2001). Excitation and propagation of non-axisymmetric guided waves in a
hollow cylinder, Journal of the Acoustical Society of America 109(2): 457464.
Lowe, M. J. S. (1995). Matrix techniques for modeling ultrasonic waves in multilayered media,
IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control 42(4): 525542.
Lowe, M. J. S., Alley, D. N. & Cawley, P. (1998). Defect detection in pipes using guided waves,
Ultrasonics 36(1-5): 147154.
Ma, J., Lowe, M. J. S. & Simonetti, F. (2007). Measurement of the properties of uids inside
pipes using guided longitudinal waves, IEEE Trans. on Ultrasonics, Ferroel. and Freq.
Control 54(3): 647658.
Maple (2007). Release 11 edn, Waterloo Maple, Inc.
Markus, S. & Mead, D. J. (1995). Axisymmetric and asymmetric wave motion in orthotropic
cylinders, Journal of Sound and Vibration 181(1): 127147.
Matlab (2004). Release 14 edn, The Mathworks, Inc.
28
28 Ultrasonic Waves
Will-be-set-by-IN-TECH

McNiven, H., Sackman, J. & Shah, A. (1963). Dispersion of axially symmetric waves in
composite, elastic rods, Journal of the Acoustical Society of America 35(10): 16021609.
Meeker, T. R. & Meitzler, A. H. (1972). Guided wave propagation in elongated cylinders
and plates, in W. P. Mason & R. N. Thurston (eds), Physical Acoustics, Principles and
Methods, Vol. 1A, Academic Press.
Mirsky, I. (1964). Axisymmetric vibrations of orthotropic cylinders, Journal of the Acoustical
Society of America 36(11): 21062112.
Mirsky, I. (1965). Wave propagation in transversely isotropic circular cylinders: Part I: theory;
part II: numerical results, Journal of the Acoustical Society of America 37(6): 10161026.
Nagy, P. B. & Nayfeh, A. H. (1996). Viscosity-induced attenuation of longitudinal guided
waves in uid-loaded rods, J. Acoust. Soc. Am. 100(3): 15011508.
Nayfeh, A. H. & Nagy, P. B. (1996). General study of axisymmetric waves in layered
anisotropic bers and their composites, J. Acoust. Soc. Am. 99(2): 931941.
Nelson, R., Dong, S. & Kalra, R. (1971). Vibrations and waves in laminated orthotropic circular
cylinders, Journal of Sound and Vibration 18(3): 429444.
Oppenheim, A. V., Schafer, R. W. & Buck, J. R. (1999). Discrete-Time Signal Processing, 2 edn,
Prentice Hall.
Pavlakovic, B. & Lowe, M. (1999). A general purpose approach to calculating the longitudinal
and exural modes of multi-layered, embedded, transversely isotropic cylinders,
Annual Review of Progress in Quantitative Nondestructive Evaluation, pp. 239246.
Pollard, H. F. (1977). Sound Waves in Solids, Pion Limited.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. (1992). Numerical Recipes in C.
The Art of Scientic Computing, 4 edn, Cambridge University Press.
Rose, J. L. (1999). Ultrasonic Waves in Solid Media, 1 edn, Cambridge University Press.
Rose, J. L. (2000). Guided wave nuances for ultrasonic nondestructive evaluation, IEEE Trans.
on Ultrasonics, Ferroelectrics and Frequency Control 47(3): 575583.
Seco, F., Martn, J. M. & Jimnez, A. R. (2009). Improving the accuracy of magnetostrictive
linear position sensors, IEEE Trans. on Instrumentation and Measurement 58(3): 722729.
Sinha, B. K., Plona, T. J., Kostek, S. & Chang, S.-K. (1992). Axisymmetric wave propagation in
uid-loaded cylindrical shells. i: Theory, J. Acoust. Soc. Am. 92(2): 11321143.
Whittier, J. & Jones, J. (1967). Axially symmetric wave propagation in a two-layered cylinder,
Int. J. Solids Structures 3(4): 657675.
Wilcox, P. D. (2003). A rapid signal processing technique to remove the effect of dispersion
from guided wave signals, IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency
Control 50(4): 419427.
Zhu, W. (2001). A nite element analysis of the time-delay periodic ring arrays for guided
wave generation and reception in hollow cylinders, IEEE Trans. on Ultrasonics,
Ferroelectrics and Frequency Control 48(5): 14621470.