Vous êtes sur la page 1sur 191

Plane Wave Expansion Analysis of Lossy

Composite Transducers Incorporating


Anisotropic Polymers

Leigh-Ann Scott
Department of Mathematics
University of Strathclyde
Glasgow, U.K.
July 2007

This thesis is submitted to the University of Strathclyde for the


degree of Doctor of Philosophy in the Faculty of Science.
The copyright of this thesis belongs to the author under the terms of the United
Kingdom Copyright Acts as qualified by University of Strathclyde Regulation
3.50. Due acknowledgement must always be made of the use of any material in,
or derived from, this thesis.
Abstract
In this thesis a one-dimensional model of a 1-3 composite piezoelectric transducer
which includes elastic loss is described. This extends the Linear Systems Mod-
elling (LSM) approach by incorporating frequency dependent elastic loss into
the polymer phase. The operating characteristics of a device are presented for a
range of passive phase materials including an anisotropic material. A compari-
son with finite element simulations is also reported. Elastic loss is incorporated
into a three dimensional plane wave expansion model (PWE) of these transduc-
ers. A comparison with experimental and finite element data is conducted and
a design to damp out these lateral modes is investigated. Scaling and regulari-
sation techniques are introduced to the PWE method to reduce ill-conditioning
in the large matrices which can arise. The identification of the modes of vibra-
tion is aided by examining profiles of the displacements, electrical potential and
Poynting vector. The dispersive behaviour of a 2-2 composite transducer with
high shear attenuation in the passive phase is examined. The model shows that
the use of a high shear attenuation filler material improves the frequency band
gap surrounding the fundamental thickness mode. The plane wave expansion
model is utilised to investigate the behaviour of a transducer with a self-similar
architecture. The Cantor set is utilised to design a 2-2 configuration, and a 1-3
configuration is investigated with a Sierpinski Carpet geometry. It was found
that by increasing the fractal generation level, the bandwidth surrounding the
main thickness mode will increase, but there will be a corresponding reduction
in the sensitivity amplitude.

i
Contents

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Outline of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Thickness Mode Response of a 1-3 Composite Transducer 13


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Linear Systems Modelling . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Transmission Mode . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 Receiving Mode . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 1-3 Composite Transducers . . . . . . . . . . . . . . . . . . . . . 27
2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Thickness Mode Response of a lossy 1-3 Composite Transducer


with an SBS Passive Phase 41
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3 SBS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Comparison to Finite Element Modelling . . . . . . . . . . . . . 58
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 Modelling of Composite Transducers using the Plane Wave Ex-


pansion Method 63

ii
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Formulation of the method . . . . . . . . . . . . . . . . . . . . . 64
4.2.1 The Geometry . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 The Admittance . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.5.1 Modal Analysis of Device A . . . . . . . . . . . . . . . . 79
4.5.2 Modal Analysis of Device B . . . . . . . . . . . . . . . . 97
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5 Incorporation of Elastic Loss into the Plane Wave Expansion


Method 102
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.2 Frequency Dependent, Elastic Loss Model . . . . . . . . . . . . 103
5.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.4.1 Modal Analysis of Device C . . . . . . . . . . . . . . . . 109
5.4.1.1 Thickness Mode . . . . . . . . . . . . . . . . . 112
5.4.1.2 The first antisymmetrical Mode . . . . . . . . . 116
5.4.1.3 Rayleigh Mode . . . . . . . . . . . . . . . . . . 120
5.4.1.4 Intra-Pillar Mode . . . . . . . . . . . . . . . . . 124
5.4.2 Damping of Unwanted Lateral Modes . . . . . . . . . . . 128
5.5 Comparison with Finite Element Modelling and Experimental Data135
5.6 Inclusion of an Anisotropic Passive Phase . . . . . . . . . . . . . 141
5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

6 Fractal Transducers 145


6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

iii
6.2 The Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.3.1 The Cantor Set Transducer . . . . . . . . . . . . . . . . 151
6.3.2 The Sierpinski Carpet Transducer . . . . . . . . . . . . . 166
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

7 Conclusions 174

iv
Chapter 1

Introduction

1.1 Background
In 1880, Jacques and Pierre Curie discovered that within certain crystalline
minerals, when subjected to a mechanical force, the crystals became electrically
polarized. Conversely, by applying an electric field to these materials, a me-
chanical deformation would occur. This behaviour was named the piezoelectric
effect from the Greek word piezein, meaning to press or squeeze. Piezoelectric
materials have been tailored to suit an impressive range of applications, such
as sensing (force or displacement sensors), actuation (motors and devices that
precisely control positioning) and sonic and ultrasonic signal generation. The
20th century introduced more powerful metal oxide-based piezoelectric ceramics
and man-made materials which enabled designers to develop more applications
for piezoelectricity. Manufacturing of these materials was relatively inexpensive
and ceramics could be easily created for a specific requirement. The most widely
used piezoelectric ceramics are ”PZT” ceramics, which are manufactured from
formulations of lead zirconate / lead titanate and exhibit great sensitivity and
can withstand high operating temperatures.
Ultrasonic transducers have found widespread use, particularly in ultrasonic
medical diagnosis and non-destructive testing. Ultrasound imaging is a stan-
dard procedure that catches images inside the body using high-frequency sound

1
2

waves. Since the images are captured in real time, ultrasound examinations
can show blood flowing through blood vessels, as well as the structure and
movement of the body’s internal organs. These devices are used frequently to
diagnose and treat medical conditions due to their affordability, portability and
that they don’t produce ionizing radiation (x-rays). Conventional ultrasound
devices display images in thin, flat sections of the body, though advancements
in technology include four-dimensional (4-D) ultrasound (3-D ultrasound image
in motion). Non-destructive testing is used to detect defects, such as crack or
corrosion, without destroying the object. This is vital in applications such as
rail and bridge inspections. The properties of these transducers dictate the abil-
ity of the transducer to work effectively and have been extensively investigated
over the last 30 years [24, 23, 10, 71, 25, 31, 26, 27, 46, 57, 14, 44, 54, 17].
There are 4 types of transducers which are widely used: monolithic piezoelectric
plate, composite, polyvinyliden difluoride (PVDF) and electromagnetic acoustic
(EMATS) transducers [70]. Piezoelectric transducers are the most common type
of transducer and can be used at higher frequencies than EMAT’s (1-50 MHz)
[34]. By applying an electric field to a transducer a mechanical deformation
occurs and conversely, a mechanical strain leads to an electric field being pro-
duced. Transducers can therefore work in both receiving and transmitting mode.
When the transducer is transmitting a voltage is applied which is converted from
electrical energy to mechanical energy (much like a loudspeaker). In reception
mode a stress wave is sensed by the transducer which is converted into electrical
energy to produce an electrical signal (much like the human ear). Composite
transducers are increasingly becoming the design of choice in biomedical, sonar
and nondestructive testing applications. This is due to the constituent mate-
rials combining to realise better operational characteristics and the availability
of new materials. The most frequently used designs are 1-3 and 2-2 composites
manufactured from piezoelectric ceramic materials. A 1-3 composite transducer
3

(a) (b)

Figure 1.1: Schematic of a periodic composite transducer where the piezoelectric


ceramic pillars are in black and the polymer filler is in white, (a) 1-3 topology,
(b) 2-2 topology

is made by dicing the ceramic block into a series of pillars and then filling the
void with a passive phase. Topologically the connectivity is in only one direc-
tion for the ceramic (the black vertical pillars) but in all three directions for the
polymer (the white filler) (see Figure 1.1 (a)). For a 2-2 composite the ceramic
is cut longitudinally in one direction so that there is connectivity in two direc-
tions for both the ceramic and polymer (see Figure 1.1 (b)). The advantage of
these composites is that the combined materials give better electromechanical
coupling and acoustic impedance characteristics [29].
Due to the ability of molecules to support vibrations in solids, a number of
different types of wave modes can be found within a transducer. These wave
modes can be characterized by their oscillatory patterns. A longitudinal mode
is characterised by the motion of the material particles being in the same direc-
tion as the wave is travelling in. A shear wave is characterised by the motion
of the material particles being perpendicular to the direction the wave is trav-
elling in. Elliptical and complex vibrations at the surfaces and interfaces make
other waves possible such as surface and plate waves. A surface (or Rayleigh)
wave travels along the surface of a semi-infinite plate with the material particles
undergoing an elliptical motion. As the wave propagates deeper into the plate,
the particles will move in smaller ellipses until a no-movement depth is reached.
Plate waves only propagate in very thin metals, hence they are very common
4

within ultrasonic transducers. The most frequently found plate waves are Lamb
waves which are complex vibrational waves that travel through the entire thick-
ness of the transducer. Their propagation depends on the material properties
and the transducer thickness. One of the problems with the composite pillar
architecture is the presence of surface waves, which are generated between the
adjacent pillars (inter-pillar modes) or within the pillars (intra-pillar modes), in-
terfering with the piston like behaviour of the main thickness mode. Extensive
experimental observations have highlighted the intricate dependency between
the geometry of the design, the material properties and the operational charac-
teristics of the device. One key to an efficient composite transducer is to arrange
for the two phases to move in concert at the driving frequency. This is aided by
ensuring that the transverse wavelength (the wavelength in the direction per-
pendicular to the thickness direction) in the passive phase is much larger than
the lattice periodicity.
A transducer typically has a backing layer and matching layer and a dis-
crete set of electrodes on the upper face (see Figure 1.2). A backing layer will
damp out unwanted oscillations but reduces the sensitivity of the device and a
matching layer will optimise the amount of energy transferred into the mechan-
ical load. The geometry of a composite transducer is described by the pitch
(period of the polymer and ceramic pillar), the kerf width (the slot that the
blade cuts in the material) and the thickness. Since a piezoelectric ceramic is
anisotropic (the properties are directionally dependent), the physical constants
relate to both the direction of the applied mechanical or electric force and the
directions perpendicular to the applied force (transverse direction). A trans-
ducer can be described by the volume fraction of ceramic and the mechanical
quality factor Q (the ratio of the reactance to the resistance in the series equiv-
alent circuit representing the piezoelectric transducer). Beamforming is a signal
processing technique used with arrays of transmitting or receiving transducers
5

that controls the directionality of, or sensitivity to, a radiation pattern. Elastic
loss (mechanical energy lost in the medium due to heat, friction and deforma-
tion) must be taken into account when modelling these devices in order that
results can be compared to experimental investigations. There are numerous
processing techniques for fabrication of piezoelectric composite transducers such
as dice and fill, injection moulding and lost mold techniques [?]. The dice and
fill method involves performing a series of parallel cuts into a piezoelectric plate,
filling these cuts with an epoxy and slicing the base of the device to reveal a
composite transducer. Once the transducer design is complete, the device must
be electroded with silver paint on the main faces, poled and characterised by
measuring techniques.
As the frequency is increased, the element’s oscillations approach a frequency
at which the impedance is a minimum (maximum admittance). This minimum
impedance frequency approximates the series resonance frequency, the frequency
at which impedance in an electrical circuit describing the element is zero, if re-
sistance caused by mechanical losses is ignored. This is the electrical resonant
frequency (fe ), the frequency at which the ceramic most effectively converts
the electrical energy into mechanical energy and vibrates most readily. As the
frequency is further increased, the impedance increases to a maximum (mini-
mum admittance). The maximum impedance frequency, fm , approximates the
parallel resonance frequency, the frequency at which parallel resistance in the
equivalent electrical circuit is infinite if resistance caused by mechanical losses
is ignored. This is the mechanical resonant frequency (fm ), the frequency at
which the ceramic most effectively converts the mechanical energy into electri-
cal energy.
The composition of the ceramic material and the shape and volume of the
piezoelectric pillars determine the electrical resonant frequency. Generally, a
thicker transducer has a lower fe than a thin transducer.
6

Top electrodes

Piezoelectric Ceramic

Backing Layer Matching Layer


V

Figure 1.2: Illustration of a piezoelectric transducer with a backing and matching


layer and an applied voltage V .

An ideal transducer design would have a constant response in the frequency


domain in order that a precise signal can be achieved in the time domain (which
provides a high resolution ultrasound image). This goal is hard to accomplish, so
the aim is to manufacture a transducer which will perform over a wide frequency
range and will have large band gaps (a region where there is no unwanted wave
propagation) around the main thickness mode. The bandwidth is defined as
the range of frequencies where the signal’s Fourier transform has a power above
half the maximum value of the amplitude. The maximum response from the
transducer will be at a point between fe and fm . Early models of these periodic
structures neglected the piezoelectric effect in order to make their dynamic study
7

possible [8]. Various periodic arrangements were considered using a Floquet


expansion to describe the solution to the governing elastodynamic equations.
By restricting these expansions to a finite number of terms a modal analysis was
performed. However, due to the computational complexity of these models the
evolution of their transducer design has been chiefly guided by experiments and
empirical models.
Oakley illustrated the advantages of composites by comparing an annular
array composite to a ceramic only transducer using a 1-dimensional equivalent
circuit model [42]. He also investigated how to achieve an optimum design com-
posite by identifying the main parameters which, when varied, have the most
effect. These are the dielectric constant and electromechanical coupling coeffi-
cient of the piezoelectric material, the Lamé coefficients for the isotropic passive
phase and two parameters to describe the geometry such as the volume fraction
and kerf width. Note that the benefits of using an anisotropic filler and the ef-
fects of elastic loss were not considered. Smith and Auld [67] developed a model
which examined thickness mode oscillations for a wide thin 1-3 composite plate
in a homogeneous medium with no loss. The piezoelectric constitutive equations
were reduced to an effective one-dimensional model using a set of simplifying as-
sumptions. The simplicity of the final model makes computation very quick and
permits the use of linear systems modelling for 1-3 transducers [31]. However,
such one-dimensional models cannot comment on unwanted lateral modes.
There has also been some theoretical investigations of composite transduc-
ers using finite element modelling [28, 9]. Such models can incorporate realistic
operating conditions, such as loss, backing and matching layers, mechanical and
electrical loads and electrode patterning. The major drawback is the high com-
putational cost. The method is mainly used in the time domain and so the
modal analysis requires a Fourier analysis of the predicted displacements. This
can be problematic if the mode amplitudes have large variance. Due to the
8

computational cost, a quarter symmetry of a unit cell with appropriate peri-


odic boundary conditions is often employed. For regular geometries this does
not present a problem but essentially excludes the study of irregular designs
in three dimensions. However the results do compare well with experimental
results and one-dimensional, effective medium models. Analysis has shown that
a wide bandwidth transducer is associated with high loss, or low Q value, ma-
terials. It has also been shown that Lamb wave propagation is responsible for
the inter element crosstalk and has a detrimental effect on beam forming and
the transducer sensitivity [29]. It has been suggested that a passive filler with
high shear loss may aid the damping of these unwanted Lamb waves.
Vasseur et al. [73] used a plane wave expansion method to study the modal
behaviour of an anisotropic passive material consisting of a regular array of uni-
directional cylindrical fibres embedded in an epoxy matrix. They found several
large frequency band gaps and that a marked contrast between the two con-
stituent material parameters provides the best opportunities for realising such
gaps. Shui et al. [66] also studied striated non-piezoelectric composite mate-
rials although they restricted their attention to 2-2 geometries. The dynamic
equations were formulated in each phase and a set of interface conditions were
applied. This allowed them to study the effects of varying the volume fraction
of each constituent on the lateral resonant frequencies. These typically provide
the upper and lower limits of any stop band gap which encompasses the main
thickness mode. The boundary effects, as well as loss, were neglected in their
modelling. Geng and Zhang [21] used partial wave expansion methods to expand
the material properties in a finite thickness plate with the periodic structure of
a 2-2 composite. The dynamic equations were solved in each material domain
and then a set of interface conditions were imposed. For a plate consisting of
a half-space the dispersion curves in the complex wave vector space were ob-
tained and loss was included in the thickness direction. The physical properties
9

of the passive phase were varied and the effect on these dispersion curves dis-
cussed. For the finite thickness plate they found that if the thickness resonant
frequency is lower than the lateral mode, then there will be a frequency near the
ceramic thickness mode where both constituents vibrate in phase. Their analy-
sis also predicted the modes which occur due to the periodicity of the structure.
They also studied the role that the pillar aspect ratio has on the surface profile.
One deficiency in their model is that no loss mechanisms in the materials were
incorporated into the finite thickness case.
Certon et al. [13] studied the lateral modes using an infinitesimally thin
(membrane model) 1-3 composite with no piezoelectric effect. Again the dy-
namical equations were constructed in each material and a set of periodic in-
terface conditions imposed. This does simplify the computations but restricts
the model outcomes to stop band gaps in the transverse modes and, by its very
nature, does not permit variations in material parameters in the thickness mode
direction. It was found that a large ceramic width to pitch ratio (ratio of the
pillar width and period) resulted in the largest frequency band gaps. The mem-
brane model was also studied using Bloch wave theory and a truncated Fourier
series to describe the dependent variables [12]. This led to slightly longer com-
putational times but the need for interface conditions was avoided. Again no
loss mechanisms were invoked but one of the benefits of this approach is that
more irregular geometries could be studied.
Geng and Zhang [22] used an ingenious annular unit cell to study a three di-
mensional piezoelectric transducer with a hexagonal symmetry pattern. Within
each material a partial wave expansion solution of the governing dynamical
equations was derived and interface conditions imposed. A cylindrical coor-
dinate system reduced the dimensionality of the problem by one. Mechanical
and dielectric losses were included in both the ceramic and passive phases, and
piezoelectric loss was included in the ceramic. This permitted a study of the
10

mechanical Q value, that is the ratio of the mechanical energy stored to the me-
chanical energy lost in one cycle. They found that the Q value depended on the
two material properties in a non-trivial fashion and that it is not a simple matter
of interpolating between two sets of values. The predicted lateral modes were
identified from the resonant frequencies associated with the kerf widths and the
diagonal lengths in the passive phase. However, the model has a very restric-
tive geometry and the annular unit cell and hexagonal periodic arrangement is
unrealistic.

1.2 Outline of Thesis


The overall aim of this thesis is to examine the benefits of using anisotropic, high
shear loss polymer filler materials in composite transducers. To this end all of the
modelling used had to be extended to incorporate elastic loss. For comparison
to the three dimensional modelling Chapter 2 investigates the one-dimensional
linear systems (LSM) model. The piezoelectric equations are defined and the
equations which describe the transmission and reception of a thickness mode
transducer are derived. A model for a 1-3 composite transducer is configured by
using the methods of Smith and Auld [67]. These derivations are also extended
to incorporate anisotropic polymer fillers. The LSM model is then extended
in Chapter 3 to include elastic losses. The effect that this has on the recep-
tion/transmission sensitivity profiles and the effective properties of a composite
transducer with an anisotropic polymer phase is reported. The impedance char-
acteristics are also compared to finite element modelling (FEM) results.
The 3-dimensional plane wave expansion (PWE) model is derived in Chapter
4. The geometry of the transducer is described in terms of a two-dimensional,
spatial Fourier series and then the PWE method and associated boundary con-
ditions are outlined. A comparison of the PWE method with the Rayleigh-Lamb
equation and LSM is made and a mode identification methodology for both 2-
11

2 and 1-3 composite transducers using profiles of the displacements, Poynting


vectors and electrical potential are outlined. In Chapter 5 frequency dependent
elastic loss is incorporated into the PWE model. The dispersive behaviour of a
2-2 composite transducer is examined and the effects of introducing high shear
attenuation into the passive phase is investigated. Scaling and regularisation
techniques are introduced and modes are identified by examining profiles of the
displacements, electrical potential and Poynting vector. A new geometrical con-
struction for a transducer using Cantor set and Sierpinski carpet structures is
detailed in Chapter 6. The effects of introducing up to three fractal genera-
tion levels is investigated and a modal analysis is performed to help explain the
characteristics of each fractal device.
The original work in the thesis is stated below:

1. In Chapter 2, the LSM method is extended to show the impact of varying


the volume fraction of the ceramic and the derivation of Smith and Auld
[67] is extended to incorporate anisotropic polymers into their formulation.

2. The LSM method is extended in Chapter 3 to include expressions for fre-


quency dependent mechanical loss, a comparison of the reception/transmission
sensitivity profiles with and without loss, the effective properties of a com-
posite transducer with an anisotropic polymer phase and a comparison to
finite element modelling (FEM) using a range of polymer composites.

3. Chapter 4 extends the derivation of Wilm et al [74] to include the Poynt-


ing vector, a parameter scaling to reduce ill-conditioning, an admittance
that incorporates a dependency on the electrode spacing, a comparison of
the PWE method with the Rayleigh-Lamb equation and linear systems
modelling, the dependency of the determinant of the harmonic analysis
matrix on the driving frequency and wavenumber and a mode identifica-
tion methodology for both 2-2 and 1-3 composite transducers using profiles
12

of the displacements, Poynting vectors and electrical potential.

4. In Chapter 5, the plane wave expansion method is extended to include


mechanical loss in both the piezoelectric and polymer material, associated
scaling and regularisation techniques, a comparison of the PWE results
incorporating loss and a comparison with the LSM model, mode identifi-
cation for a 2-2 composite using profiles of the displacements and Poynting
vectors, the effects of using a high shear loss passive phase and a compar-
ison to FEM modeling and experimental results.

5. The plane wave expansion method is utilised in Chapter 6 to investigate


the electrical impedance and admittance characteristics of a self-similar
composite transducer. Electrical impedance characteristics of a Sierpinski
carpet (1-3) and a Cantor set (2-2) composite transducer along with surface
profiles are used to discuss the behaviour of these devices.
Chapter 2

Thickness Mode Response of a


1-3 Composite Transducer

2.1 Introduction
In this chapter a one-dimensional model of a piezoelectric transducer is inves-
tigated. This consists of a plane parallel plate of piezoelectric material, with a
backing material and a front matching layer. A 1-3 composite transducer with a
polymer filler material will also be considered. Polymer composite transducers
combine two or more materials to gain better characteristics than those of a
monolithic piezoelectric crystal. The thickness mode of the device; a piston like
oscillation, is the predominant dynamical behaviour of such devices [70]. The
LSM model is extended here to show the impact of varying the volume fraction
of the ceramic. In later chapters results will be compared to an LSM model
incorporating elastic loss and a plane wave expansion method.

2.2 Linear Systems Modelling


For thickness mode transducers the dynamics can be approximately described
by a one-dimensional model. By coupling the piezoelectric constitutive equa-
tions with the one-dimensional wave equation for the mechanical displacement

13
14

a Linear Systems Model (LSM) can be derived [25]. In a series of papers, the
dependency of both the reception and transmission characteristics of the trans-
ducer on its physical parameters was investigated using a systems block diagram
approach [31, 26, 27]. Previous authors have used this approach to study some
special cases such as, the open-circuit response of a lossless transducer, with a
rigid backing material [59], the short or open-circuit dynamics of a mechanically
free transducer [69] or the form of the transmitted pressure wave when a step-like
charge is applied [20]. In this section the derivation of the equations describing
the transmission and reception of the thickness mode transducer is detailed.
The two unknowns are the mechanical displacement u(x, t) and the electrical
displacement E(x, t). The one-dimensional wave equation for u(x, t) is
∂ 2 u(x, t) ∂ 2 u(x, t)
ρ = Y (2.1)
∂t2 ∂x2
where ρ is the density and u is the particle displacement, subject to the boundary
conditions at x = 0 and x = L of continuity of displacement and force into the
adjacent media. From Gauss’s Law

∂D/∂x = 0, (2.2)

as there is no free charge inside the transducer.


The one dimensional piezoelectric equations are [25]

T = Y S − hD (2.3)

D
E = −hS + (2.4)


where T is the stress, Y is the Young’s modulus, S is the strain, h is the piezo-
electric constant, D is the electrical displacement, E is the electric field and  is
the permittivity. The strain is defined as
∂u
S= , (2.5)
∂x
15

and voltage V is given by Z L


V = Edx (2.6)
0

where E is the electric field and L is the thickness of the transducer. The
force at each face of the transducer is F , where F = Ar T and Ar is the cross-
sectional area of the transducer. The transducer typically has a backing material
(subscript 2) and a matching layer, but for simplicity, there is no matching layer
here; the device will simply transmit into a load medium (subscript 1) at x = 0
(see Figure 2.1). At the two interfaces there is continuity and differentiability of
the displacement u; the latter of these corresponding to the continuity of force.
Applying these boundary conditions implies that u(0) = u1 (0), u(L) = u2 (L),
F (0) = F1 (0) and F (L) = F2 (L) where

F = (Y S − hD)Ar . (2.7)

Taking Laplace transforms of equation (2.1) with respect to t (assuming zero

A2 A A1

PSfrag replacements
B B1

x=h
x=0

Figure 2.1: Illustration of the transmitted and received waves within a trans-
ducer.

intitial displacement and velocity) gives


∂ 2 ū(x, p)
p2 ū(x, p) = v 2 (2.8)
∂x2
16

where the wave speed v is given by v 2 = Y /ρ. Solving equation (2.8) inside the
transducer gives,
p p
ū(x, p) = Ae− v x + Be v x . (2.9)

Using the acoustic impedance of the transducer Zc , defined as

Zc = ρvAr , (2.10)

with equation (2.9), it is found that

∂ ū Zc ∂ ū x x
Y S̄Ar = Y = Zc v = pZc (−Ae−p( v ) + Bep( v ) ). (2.11)
∂x ρv ∂x

So from equation (2.3)

x x
F̄ + hD̄Ar = pZc (−Ae−p( v ) + Bep( v ) ).

At the surface of the transducer, D̄ = Q̄/Ar where Q̄ is the electrical charge.


That is
x x
F̄ = pZc (−Ae−p( v ) + Bep( v ) ) − hQ̄. (2.12)

Using equation (2.4), equation (2.6) can be rewritten as


Z L 

V̄ = −hS̄ + dx
0 
Z L 
∂ ū Q̄
= −h + dx
0 ∂x Ar 

= −h(ū(L) − ū(0)) + L.
Ar 

That is
 Q̄L
V̄ = −h A(e−pξ − 1) + B(epξ − 1) +

, (2.13)
Ar 
where ξ = L/v is the transit time of a plane wave through the transducer. The
transducer is placed in parallel with a load impedance Z̄E and the combination
17

PSfrag replacements
Z0
I
V
ZE ZT
Vs IE IT

Figure 2.2: Illustration of the transducer circuit.

is placed in series with a load Z̄0 as shown in Figure 2.2. If the input voltage,
V̄s is put into the transducer and the total impedance is Z̄0 + Z̄E Z̄T /(Z̄E + Z̄T ),
then
V̄s
I¯ = Z̄E Z̄T
, (2.14)
Z̄0 + Z̄E +Z̄T

where I is the current and ZT is the electrical impedance of the transducer. The
voltage across the transducer is defined as

V̄s Z̄E Z̄T


V̄ = Z̄E Z̄T
× , (2.15)
Z̄0 + Z̄E + Z̄T
Z̄E +Z̄T

so that the current across the transducer can be expressed as


 
¯ 1 V̄s Z̄E Z̄T
IT = Z̄ Z̄
×
Z̄T Z̄0 + E T Z̄E + Z̄T
Z̄E +Z̄T

V̄s Z̄E
=
(Z̄0 + Z̄E )Z̄T + Z̄0 Z̄E

= āV̄s /(Z̄T + b̄), (2.16)

where ā = Z̄E /(Z̄0 + Z̄E ) and b̄ = Z̄0 Z̄E /(Z̄0 + Z̄E ). Now since I¯T = V̄s /Z̄0 − V̄ /b̄,
the electrical charge Z
Q= IT dt (2.17)
t
18

can be written as
I¯T V̄s V̄
Q̄ = = − . (2.18)
p pZ̄0 pb̄
Ar 
Since C0 = L
, where C0 is the (clamped) capacitance of the transducer, equa-
tion (2.13) becomes
V̄s V̄
V̄ = −h A(e−pξ − 1) + B(epξ − 1) +

− ,
pC0 Z̄0 pC0 b̄
which can be written as
V̄s
V̄ = (−h A(e−pξ − 1) + B(epξ − 1) +

)Ū , (2.19)
pC0 Z̄0
where Ū = pC0 b̄/(1 + pC0 b̄). Within each medium, there is a transmitted wave
to the right (amplitude Bi ) and an incoming wave to the left (amplitude Ai )
and it is assumed that the reflected wave within the backing layer is damped
out (B2 = 0) (see Figure 2.1). Applying the boundary conditions for continuity
of displacement implies that ū1 (0) = ū(0) which gives

A1 + B1 = A + B, (2.20)

and ū(L) = ū2 (L) which gives

−p vL
Ae−pξ + Bepξ = A2 e 2 . (2.21)

For continuity of force from (2.12),

pZ1 (−A1 + B1 ) = pZc (−A + B) − hQ̄ (2.22)

and
−p vL
pZc (−Ae−pξ + Bepξ ) − hQ̄ = pZ2 (−A2 e 2 ), (2.23)

where Z1 is the mechanical impedance of the front load and Z2 of the backing
material. Hence there are four equations and five unknowns (A1 , B1 , A, B, A2 ).
Substituting equation (2.20) in equation (2.22) gives

pZ1 (A + B − 2A1 ) = pZc (−A + B) − hQ̄, (2.24)


19

which can be rearranged to give

hQ̄
A(Zc + Z1 ) − B(Zc − Z1 ) = 2A1 Z1 − , (2.25)
p

and then
   
Zc − Z 1 Zc + Z 1 + Z 1 − Z c hQ̄
A−B = A1 − , (2.26)
Zc + Z 1 Zc + Z 1 (Zc + Z1 )p

so that
hQ̄
A − RF B = A1 (1 − RF ) − , (2.27)
(Zc + Z1 )p
where RF = (Zc − Z1 )/(Zc + Z1 ) is the reflection coefficient at the front face of
the transducer. Multiplying equation (2.21) by pZ2 , and adding it to equation
(2.23) gives

0 = pZ2 (Ae−pξ + Bepξ ) + pZc (−Ae−pξ + Bepξ ) − hQ̄, (2.28)

which can be rearranged to give

hQ̄
Bepξ (Zc + Z2 ) − Ae−pξ (Zc − Z2 ) = , (2.29)
p

so that  
pξ hQ̄ Zc − Z 2
Be = + Ae−pξ . (2.30)
(Zc + Z2 )p Zc + Z 2
Hence B can be expressed as

hQ̄e−pξ
B= + Ae−2pξ RB , (2.31)
(Zc + Z2 )p

where RB = (Zc − Z2 )/(Zc + Z2 ) is the reflection coefficient at the back face of


the transducer. Substituting equation (2.31) into (2.27) gives

RF hQ̄e−pξ hQ̄
A − Ae−2pξ RF RB = + A1 (1 − RF ) − , (2.32)
(Zc + Z2 )p (Zc + Z1 )p

which gives the expression

(RF hQ̄e−pξ )/((Zc + Z2 )p) + A1 (1 − RF ) − (hQ̄)/((Zc + Z1 )p)


A= . (2.33)
1 − e−2pξ RF RB
20

Substituting equation (2.33) into equation (2.27) then gives


RF hQ̄e−pξ hQ̄
(Zc +Z2 )p
+ A1 (1 − RF ) − (Zc +Z1 )p hQ̄
− RF B = A1 (1 − RF ) − , (2.34)
1− e−2pξ R F RB (Zc + Z1 )p
and then
RF hQ̄e−pξ RF RB hQ̄e−2pξ
(Zc +Z2 )p
+ A1 (1 − RF )(RF RB e−2pξ ) + (Zc +Z1 )p
RF B = , (2.35)
1 − e−2pξ RF RB
so that
 
(−hQ̄)/((Zc + Z2 )p) − A1 (1 − RF )(RB e−pξ ) − (RB hQ̄e−pξ )/((Zc + Z1 )p) −pξ

B= −e .
1 − e−2pξ RF RB
(2.36)
Hence, substituting equations (2.33) and (2.36) into equation (2.19) gives
 
h(1 − e−pξ ) hQ̄RF e−pξ hQ̄
V̄ = + A 1 (1 − R F ) −
1 − RF RB e−2pξ (Zc + Z2 )p (Zc + Z1 )p
  
pξ hQ̄e−pξ −2pξ hQ̄RB e−2pξ V̄s
− e + A1 (1 − RF )(RB e )− + Ū ,
(Zc + Z2 )p (Zc + Z1 )p pC0 Z0
(2.37)

and therefore
 
h(1 − e−pξ ) −pξ hQ̄(1 − RB e−pξ )
V̄ = A 1 (1 − R F )(1 − R B e ) −
1 − RF RB e−2pξ (Zc + Z1 )p
 
hQ̄(1 − RF e−pξ ) V̄s
− + Ū , (2.38)
(Zc + Z2 )p pC0 Z0 h
so that
   
hQ̄ K̄B hQ̄ V̄s
V̄ = h K̄F Ā1 (1 − RF ) − − + Ū , (2.39)
p(Zc + Z1 ) p(Zc + Z2 ) pC0 Z0
where K̄F = ((1−e−pξ )(1−RB e−pξ ))/(1−RF RB e−2pξ ) and K̄B = ((1−e−pξ )(1−
RF e−pξ ))/(1 − RF RB e−2pξ ).

2.2.1 Transmission Mode

When the transducer is transmitting there is no force incident on the front face
of the transducer from the load medium and so A1 = 0. Hence
∂u
F̄F = pZc (0) − hQ̄,
∂x
= pZc (−A + B) − hQ̄. (2.40)
21

By substituting equations (2.33) and (2.36), equation (2.40) can be rewritten as


 
pZc hQ̄(1 − RF )e−pξ hQ̄(1 + RB e−2pξ ) hQ̄(1 − RF RB e−2pξ )
F̄F = + − ,
1 − RF RB e−2pξ (Zc + Z2 )p (Zc + Z1 )p pZc


hQ̄Z̄c 2Z̄1 Z̄c e−pξ Zc + Zc RB e−2pξ
= +
1 − RF RB e−2pξ (Zc + Z2 )(Zc + Z1 )Zc (Zc + Z1 )Zc


Zc + Z1 − (Zc − Z1 )RB e−2pξ
− ,
(Zc + Z1 )Zc

hQ̄Z̄1 −pξ −2pξ



= (1 + R B )e − 1 − R B e ,
(1 − RF RB e−2pξ )(Zc + Z1 )

hQ̄K̄F Z1
= − .
Zc + Z 1
(2.41)

Similarly at the interface with the backing material

∂u
F̄B = pZc (L) − hQ̄,
∂x
= pZc (−Ae−pT + BepT ) − hQ̄, (2.42)
22

which can be expressed as


 
pZc hQ̄(1 − RF e−2pξ ) hQ̄(1 + RB )e−pξ hQ̄(1 − RF RB e−2pξ )
F̄B = + − ,
1 − RF RB e−2pξ (Zc + Z2 )p (Zc + Z1 )p pZc


hQ̄Z̄c Z̄c − Z̄c RF e−2pξ 2Z̄c e−pξ
= +
1 − RF RB e−2pξ (Zc + Z2 )Zc (Zc + Z2 )(Zc + Z1 )Zc


Zc + Z2 − (ZC − Z2 )RF e−2pξ

(Zc + Z2 )Zc

hQ̄Z̄2 
= (1 + RF )e−pξ − 1 − RF e−2pξ ,
(1 − RF RB e −2pξ )(Zc + Z2 )

hQ̄K̄B Z2
= − .
Zc + Z 2
(2.43)

Since A1 = 0,

h(1 − e−pξ ) hQ̄RF e−pξ hQ̄
V̄ = −
1 − RF RB e−2pξ (Zc + Z2 )p (Zc + Z1 )p

 
pξ hQ̄RF e−pξ hQ̄RB e−2pξ Q̄
− e − + ,
(Zc + Z2 )p (Zc + Z1 )p C0

 
−h(1 − e−pξ ) hQ̄(1 − RB e−pξ ) hQ̄(1 − RF e−pξ ) Q̄
= + + .
1 − RF RB e−2pξ (Zc + Z1 )p (Zc + Z2 )p C0
(2.44)

Hence the voltage response of the transducer given by equation (2.39) can be
expressed as
 
K̄F hQ̄ K̄B hQ̄ Q̄
V̄ = −h + + ,
p(Zc + Z1 ) p(Zc + Z2 ) Co

 
Q̄ 2 (K̄F TF + K̄B TB )
= 1 − h Co ,
Co 2pZc
(2.45)
23

2Zc 2Zc
where TF = Zc +Z1
and TB = Zc +Z2
.
The electrical impedance of the device ZT , defined as Z̄T = V̄ /I¯T , provides
useful information on its operating characteristics. The electrical impedance can
be obtained by substituting equation (2.18) into equation (2.45) to get
 
1 2 (K̄F TF + K̄B TB )
Z̄T = 1 − h Co , (2.46)
pCo 2pZc
where p = iω and ω = 2πf . A dimensional analysis of equation (2.46) shows that
h2 C0 /pZc , defined as the square of the electromechanical coupling coefficient k̄t ,
is dimensionless [25]. Since K̄F , K̄B , TF and TB are O(1) then it is required that
k̄t is also O(1) so that the two additive terms in equation (2.46) are of similar
order and resonant behaviour can then occur. In transmitting mode a voltage
V̄s is put into the transducer to produce a force at the front face (F̄F ). Equation
(2.16) can be substituted into equation (2.41) along with Q̄ = I¯T /p to obtain
F̄F hāAF K̄F
=− (2.47)
V̄s 2p(Z̄T + b̄)
where AF = 2Z1 /(Zc + Z1 ). To maximize F̄F it is best that Z̄T is as small as
possible and this occurs at the electrical resonant frequency (fe ). The general
transfer function relating the stress wave generated into the load medium to
the input voltage can be obtained by substituting equation (2.46) into equation
(2.47) to get
F̄F hāĀF K̄F
= − ,
V̄s 2p((1/pC0 ) (1 − h2 C 0 (K̄F TF + K̄B TB )/2pZc ) + b̄)

hā(ĀF /2)K̄F
= − (2.48)
(1/C0 )(1 + C0 pb̄ − h2 C0 (K̄F TF + K̄B TB )/2pZc)
Equation (2.48) is then rearranged to give the transmission sensitivity;
F̄F −1
= −hā(AF /2)Ȳ K̄F 1 − h2 Ȳ (K̄F TF + K̄B TB )/2pZc

(2.49)
V̄s
where Ȳ = C0 /(1 + pC0 b̄). If the transducer was in vacuum with no backing or
matching layer, true resonance would occur. In this model energy is lost since the
24

wave transmitted into the backing layer is not reflected back to the piezoelectric
layer, therefore the amplitude of the transmission sensitivity is finite (this also
applies to the reception sensitivity). The aim is to have the transducer perform
efficiently over a range of frequencies. This range of frequency can be quantified
by the bandwidth. For a unimodal function A(ω), with a local maximum, Amax ,
at ωmax and A(ω) = Amax /2 at ωL and ωR , the bandwidth is defined as ωR − ωL .
The percentage bandwidth is then given by (ωR − ωL )/ωmax × 100. The band-
width can also be defined in terms of decibels (dB). Since decibels are defined
as the logarithm of a ratio of power, and since power is proportional to voltage
squared, then a voltage change from V1 to V2 is calculated as 20 log10 (V2 /V1 )
dB. In the bandwidth calculation V2 = Vmax ,V1 = 21 Vmax and so the equivalent
 
change in decibels is 20 log10 1VVmax
max
= 20 log10 2 ≈ 6 dB.
2

2.2.2 Receiving Mode

In receiving mode the displacement amplitude at the front face is given by


F̄F
Ā1 = − , (2.50)
pZ̄1
and, since there is no initial load (Z0 ), equation (2.39) becomes
   
hQ̄ K̄B hQ̄
V̄ = h K̄F Ā1 (1 − RF ) − − Ū . (2.51)
p(Zc + Z1 ) p(Zc + Z2 )
Substituting equation (2.50) and Q̄ = −V̄ /pZ̄E into equation (2.51) gives
−hF̄F K̄F (1 − RF )Ū /pZ1
V̄ = ,
1 − h2 ( ZcK̄+Z
F
1
+ K̄B
Zc +Z2
)Ū /p2 Z̄E

−2hF̄F K̄F Ū Zc /((Zc + Z1 )p)


= . (2.52)
Zc − h2 ( ZK̄cF+Z
Zc
1
+ K̄B Zc
Zc +Z2
)Ū /p2 Z̄E
Equation (2.52) is then rearranged to give the reception sensitivity; the ratio of
the received voltage to the magnitude of the incident force
V̄ −hTF K̄F Ū /pZc
= . (2.53)
F̄F 1 − h (K̄F TF /2 + K̄B TB /2)Ū /p2 Zc Z̄E
2
25

When in response mode the aim is to maximize the voltage Vmax resulting from a
force at the front face (F̄F ), and this occurs at the mechanical resonant frequency
fm .

- Constant Units Value


elastic constant c11 Nm−2 12.72 × 1010
elastic constant c12 Nm−2 8.02 × 1010
elastic constant c13 Nm−2 8.47 × 1010
elastic constant c33 Nm−2 11.74 × 1010
dielectric constant 33 - 1.70 × 103
dielectric constant 11 - 1.47 × 103
Piezoelectric constant h V m−1 2.60 × 109
Specific Mechanical impedance Zc Rayls 7 × 106
density ρb kg m−3 7.50 × 103
Piezoelectric stress coefficient e33 C m−2 23.30
Piezoelectric stress coefficient e31 C m−2 −6.50
Cross sectional area Ar m2 10−4
load impedance ZE Ω 106
load impedance Z0 Ω 50
Young’s modulus Y kg m−1 s−2 1.57 × 1011

Table 2.1: Physical properties of the ceramic phase PZT5H [19].

- Constant Units Value


Transducer thickness L m 2 × 10−4
Backing material impedance Z1 Rayls 2 × 106
Front material impedance Z2 Rayls 1.5 × 106

Table 2.2: Physical properties of the transducer.

In Figure 2.3 the reception sensitivity of the device, given by equation (2.53),
is plotted as a function of the driving frequency, with reflection and transmis-
sion coefficients RF = 0.65, RB = 0.52, TF = 1.65 and TB = 1.52. The backing
material is rubber and the front material is water. Looking at Figure 2.3 the
first peak has a maximum reception sensitivity at 11.3 MHz with a 6dB per-
26


F̄F
(Zc /(h33 Co b))
1

0.75

0.5

0.25
PSfrag replacements

f
1 2 3 4

Figure 2.3: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 for a piezoelectric ce-
ramic transducer (see Tables 2.1 and 2.2 for material properties).

ZT
20

15

10

5
PSfrag replacements
f
1 2 3 4

Figure 2.4: Electrical impedance ZT (kΩ) against frequency f (Hertz) ×107 for
a piezoelectric ceramic transducer.

centage bandwidth of 14%. Figure 2.4 shows a graph of equation (2.46) for the
same transducer. The location of the first minimum in this plot is the electrical
resonant frequency fe and the location of the first maximum is the mechanical
resonant frequency fm . For a transducer to be most effective it will transmit
at the electrical resonant frequency and receive at the mechanical resonant fre-
quency. In this case the mechanical resonant frequency is fm = 11.4 MHz with
27

F̄F

/(h33 Co )

0.6
0.5
0.4
0.3
0.2
PSfrag replacements 0.1

f
1 2 3 4

Figure 2.5: Nondimensionalised transmission sensitivity (absolute value)


((F̄F )/V̄ )/(h33 Co ) against frequency f (Hertz) ×107 for a piezoelectric ceramic
transducer.

an electrical impedance of 6.4kΩ and the electrical resonant frequency is fe = 10


MHz with an electrical impedance of 0.95kΩ. The graph of equation (2.49) (the
transmission sensitivity) for the same PZT5H ceramic transducer is shown in
Figure 2.5. The maximum peak is at 9.8 MHz, with a percentage bandwidth of
18%.

2.3 1-3 Composite Transducers

Figure 2.6: Illustration of a 1-3 composite transducer where the ceramic is black
and the polymer is white.

Composite transducers composed of a piezoelectric ceramic and a passive


28

polymer phase give better electromechanical coupling and acoustic impedance


characteristics than conventional single phase transducers [28]. They have there-
fore found widespread use particularly in ultrasound imaging within medicine
and nondestructive evaluation [28]. 1-3 transducers are typically manufactured
by slicing the piezoelectric ceramic into a bristle block of vertical pillars and
then filling the inter-pillar space with a passive polymer. The ceramic has a
connectivity in only one direction whilst the polymer has connectivity in all
three directions. This topology is therefore described as 1-3 (see Figure 2.6).
The linear systems model in the previous section is based on a transducer whose
active layer is a single material. In order to utilise this model here the effective
properties of the 1-3 composite transducer must be derived [67]. This starts
with the general form for equations (2.3) and (2.4). Hookes’ law for an elastic
material states
Ti = cij Sj (2.54)

where Ti is the stress tensor, Si is the strain tensor and cij is the elastic modulus
tensor [11]. For a cubic crystal polymer the elastic modulus tensor is
 
c11 c12 c12 0 0 0
 c12 c11 c12 0 0 0 
 
p
 c12 c12 c11 0 0 0 
cij = 
 
 0 0 0 c 44 0 0 

 0 0 0 0 c44 0 
0 0 0 0 0 c44
and for a tetragonal ceramic
cc11 cc12 cc13 0
 
0 0

 cc12 cc11 cc13 0 0 0 

c
 cc13 cc13 cc33 0 0 0 
cij =  c
.

 0 0 0 c44 0 0 

 0 0 0 0 cc44 0 
0 0 0 0 0 cc66
From Beyer and Letcher [11]

Di = eij Sj + ij Ej for i, j = 1, 2, 3 (2.55)


29

where Di is the electric displacement, eij = him mj are the piezoelectric stress
coefficients, ij is the permittivity and Ej is the electric field. In an isotropic,
passive polymer epij = 0 and
 
11 0 0
pij =  0 11 0  ,
0 0 11

where 11 is the isotropic permittivity of the polymer. In a typical ceramic such
as PZT (lead zirconate titanate) [11]
 
0 0 0 0 e15 0
ecij =  0 0 0 e15 0 0  ,
e31 e31 e33 0 0 0

and
c11 0 0
 

cij =  0 c11 0 
0 0 c33
since the ceramic is poled in the direction of the pillars. So the piezoelectric
equations for the polymer phase are

Tip = cpij Sjp ,

(2.56)

Dip = pij Ejp ,

and for the ceramic phase,

Tic = ccij Sjc − ecji Ejc , (2.57)

and
Dic = ecij Sjc + cij Ejc . (2.58)

These equations are then reduced to a one-dimensional caricature of the com-


posite. In order to do this six assumptions are utilised as listed below:
30

1. The electric field and the strain are assumed to be functions of z only,
where z is directed along the pillar length (the vertical direction in Figure
2.6).

2. It is assumed that the transducer is a large thin plate (E1 = E2 = 0) and


there is symmetry in the x-y plane, i.e. T1 = T2 and S1 = S2 .

3. The ceramic and polymer move together in the z direction so that S3p =
S3c = S̄3 .

4. The electric field is the same in both phases, hence E3p = E3c = Ē3 .

5. Lateral stresses are equal in both phases so that T1p = T1c = T̄1 . In order
that the composite as a whole is laterally clamped (since the composite
is normally placed in a rigid container that prevents any lateral motion
at its extremities) it is assumed that the lateral strain in the ceramic is
compensated by a complementary strain in the polymer to give

S̄1 = (1 − ψ)S1p + ψS1c = 0 (2.59)

where ψ is the volume fraction of the ceramic. This assumption is intended


to capture the local clamping of the ceramic pillars by the polymer in
the average sense [67]. Equations (2.56), (2.57) and 2.58 have now been
reduced to

T̄1 = (c11 + c12 )S1p + c12 S̄3 , (2.60)

T3p = 2c12 S1p + c11 S̄3 , (2.61)

D3 = 11 Ē3 , (2.62)

T̄1 = (cc11 + cc12 )S1c + cc13 S̄3 − e31 Ē3 , (2.63)

T3c = 2cc13 S1c + c33 S̄3 − e33 Ē3 , (2.64)

and D3c = 2e31 S1c + e33 S̄3 + c33 Ē3 . (2.65)


31

Combining equations (2.60) and (2.63) gives

1
S1p = (cc11 + cc12 )S1c + (cc13 − c12 )S̄3 − e31 Ē3 .

(2.66)
(c11 + c12 )

By using equations (2.59) and (2.66) it is found that

−(cc13 − c12 )S̄3 + e31 Ē3


 
c
S1 = ψ̃ (2.67)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )

where ψ̃ = 1 − ψ and

(cc13 − c12 )S̄3 − e31 Ē3


 
S1p =ψ (2.68)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )
for the ceramic and the polymer, respectively. From equations (2.60) and
(2.68)
T̄1 = c̄13 S̄3 − ē31 Ē3 (2.69)

where
ψcc13 (c11 + c12 ) + ψ̃c12 (cc11 + cc12 )
c̄13 = , (2.70)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )
and
ψe31 (c11 + c12 )
ē31 = . (2.71)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )
In a similar way equations (2.67) and (2.68) can be substituted into the
expressions for T3p , T3c and D3c .

6. If the lateral periodicity is sufficiently fine then the effective total stress
T̄3 and electric displacement D3 are given by volume averaging

T̄3 = ψT3c + ψ̃T3p , (2.72)

and
D̄3 = ψD3c + ψ̃D3p . (2.73)

This then leads directly to

T̄3 = c̄33 S̄3 − ē33 Ē3 (2.74)


32

and
D̄3 = ē33 S̄3 + ¯33 Ē3 (2.75)

where
!
2ψ̃(cc13 − c12 )2
c̄33 = ψ cc33 − + ψ̃c11 , (2.76)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )
!
2ψ̃e31 (cc13 − c12 )
ē33 = ψ e33 − , (2.77)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )
and
!
2ψ̃(e31 )2
¯33 = ψ c33 + + ψ̃11 . (2.78)
ψ(c11 + c12 ) + ψ̃(cc11 + cc12 )

Equation (2.74) can be rewritten using equation (2.75) to get

T̄3 = c̄D
33 S̄3 − h33 D̄3 (2.79)

and equation (2.75) is therefore

Ē3 = −h33 S̄3 + β33 D̄3 (2.80)

where cD 2
33 = c̄33 + (ē33 ) /¯
33 , h33 = ē33 /¯
33 and β33 = 1/¯
33 . The density is
defined
ρ̄ = ψρc + ψ̃ρp (2.81)

where ρc is the density of the ceramic and ρp is the density of the polymer.
For the case when the polymer phase is anisotropic (transversely isotropic) the
expressions for the effective properties must be altered slightly from those in
Equations (2.76), (2.77) and (2.78) to

c̄33 = ψ(cc33 − 2(1 − ψ)(cc13 − c13 )2 /(ψ(c11 + c12 ) + (1 − ψ)(cc11 + cc12 ))) + (1 − ψ)c11 ,
(2.82)
ē33 = ψ(e33 − 2(1 − ψ)e31 (cc13 − c13 )/(ψ(c11 + c12 ) + (1 − ψ)(cc11 + cc12 ))), (2.83)
33

and

¯33 = ψ(c33 + 2(1 − ψ)e231 /(ψ(c11 + c12 ) + (1 − ψ)(cc11 + cc12 ))) + (1 − ψ). (2.84)

A measure of the efficiency with which the transducer converts electrical energy
to mechanical energy is given by the electromechanical coupling coefficient k̄t .
Rewriting the nondimensional term discussed with regard to equation (2.46)
p
(Co = Ar ¯33 /L, p = 1/ξ = v/L, v = cD
33 /ρ̄ ) an expression is obtained for the

electromechanical coupling coefficient

h̄33 ē33
k̄t = p =p . (2.85)
cD33 β 33 c D
33 
¯ 33

This value can be compared with the thickness electromechanical coupling con-
stant for a freely vibrating piezoelectric ceramic rod given by

dc33
k33 = √ (2.86)
εc33 sc33

where scij = (ccij )−1 is the elastic compliance constant, dcij = scij ecij and εcij =
ecij scij ecji + cij [1]. The specific acoustic impedance Z̄c (equation (2.10)) is now
given by
q
Z̄c = cD
33 ρ̄ (2.87)

where the longitudinal velocity is


s
cD
33
v̄ = . (2.88)
ρ̄

For the transducer to operate efficiently it is required that the acoustic


impedance Z̄c matches, as well as is possible, the acoustic impedance of the
load medium Z2 . A closer acoustic impedance matching with the load medium
reduces internal wave reflection at the front face of the transducer. Typically the
load medium has a far lower impedance than the piezoelectric ceramic. How-
ever, the inclusion of the polymer in the 1-3 composite serves to reduce the
value of Z̄c . A plot of Z̄c versus ceramic volume fraction ψ, as given by equation
34

Zc

3
2.5
2
1.5
1
PSfrag replacements 0.5
ψ
0.2 0.4 0.6 0.8 1

Figure 2.7: Mechanical impedance Zc (MRayls) against ceramic volume frac-


tion ψ for a 1-3 composite transducer (see Tables 2.1, 2.2 and 2.3 for material
properties).

(2.87) is shown in Figure 2.7, where the polymer filler is the hardset material
HY1300/CY1301 [48] and the piezoelectric ceramic is the same as in the previous
section.

Data Values Constant Hardset


Shear modulus (real part) G0 (kg m−1 s−2 ) 1.57 × 109
Young’s modulus (real part) E 0 (kg m−1 s−2 ) 4.28 × 109
Shear Velocity Vs (m s−1 ) 1.17 × 103
Longitudinal Velocity Vl (m s−1 ) 2.51 × 103
Density ρ ( kg m−3 ) 1.15 × 103
Dielectric constant  4
Elastic constant c11 7.19 × 109
Elastic constant c44 1.57 × 109
Frequency of measurement f0 (Hz) 5.00 × 105

Table 2.3: Physical properties of the polymer phase HY1300/CY1301 Hardset


[48].

A plot of k̄t versus the volume fraction of ceramic ψ, given by equation


(2.85), is shown in Figure 2.8. Note that ψ = 1 corresponds to pure ceramic
and there is an optimal value for the volume fraction (of around ψ = 0.6) that
35

k̄t

0.6

0.5

0.4
PSfrag replacements
ψ
0.2 0.4 0.6 0.8 1

Figure 2.8: Electromechanical coupling coefficient k̄t against ceramic volume


fraction ψ for a 1-3 composite transducer.

produces an electromechanical coupling coefficient which is almost that of a free


ceramic rod (k33 = 0.752) [45]. It is also important to have a reasonably large

0

1500
1250
1000
750
500
PSfrag replacements 250
ψ
0.2 0.4 0.6 0.8 1

Figure 2.9: Dielectric constant ¯/0 against ceramic volume fraction ψ for a 1-3
composite transducer.

dielectric constant so that the clamped capacitance Co (and hence Ū and Ȳ ) is


large enough to produce a measurable reception sensitivity (see equation (2.53))
and a reasonable transmission sensitivity (see equation (2.49)). Again the 1-3
composite transducer can achieve the desired magnitude for ¯ (see Figure 2.9).
36


F̄F
(Zc /(h33 Co b))
1

0.75

0.5

0.25
PSfrag replacements

f
1 2 3 4

Figure 2.10: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 for a piezoelectric ce-
ramic transducer (dashed line) and a 1-3 composite transducer (full line).

ZT
20

15

10

5
PSfrag replacements
f
1 2 3 4

Figure 2.11: Impedance ZT (kΩ) against frequency f (Hertz) ×107 for a piezo-
electric ceramic transducer (dashed line) and a 1-3 composite transducer (full
line).

Figure 2.10 shows the reception sensitivity against frequency for a monolithic
transducer and a 1-3 composite transducer with the hardset material as the
passive phase and ceramic volume fraction ψ = 0.7. Comparison of the peaks
show that the 1-3 composite will resonate around 2 MHz lower than the mono-
lithic transducer and that the peak amplitudes have increased by around 15%.
37

F̄F
V¯s
/(h33 Co )

0.7
0.6
0.5
0.4
0.3
0.2
PSfrag replacements
0.1
f
1 2 3 4

Figure 2.12: Nondimensionalised transmission sensitivity (absolute value)


(F̄F /V¯s )/(h33 Co ) against frequency f (Hertz) ×107 for a piezoelectric ceramic
transducer (dashed line) and a 1-3 composite transducer (full line).

0
0.25
0.5 ψ

0.75

1
PSfrag replacements

1 2 3 4 F̄V̄F (Zc /(h33 Co b))


f /

Figure 2.13: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F (Zc )/(h33 Co b)) against frequency f (Hertz) ×107 and volume fraction
ψ for a 1-3 composite transducer.

The inclusion of a polymer phase within the transducer will give a reduction in
velocity which denotes an increase in impedance. This also indicates that the
transit time ξ has increased, hence the transducer will resonate at a lower fre-
quency since ω = 2πf = 1/ξ. In addition, the percentage bandwidth has nearly
doubled to 24%. In Figure 2.11 the electrical impedance of each device is plot-
38

0
0.25
ψ 0.5

0.75

1
PSfrag replacements ZT
1 2 3 4
f

Figure 2.14: Electrical impedance log ZT (kΩ) against frequency f (Hertz) ×107
and volume fraction ψ for a 1-3 composite transducer.

kt

0.5

0.45

0.4

0.35
PSfrag replacements
ψ
0.2 0.4 0.6 0.8 1

Figure 2.15: Electromechanical coupling coefficient kt against ceramic volume


fraction ψ for a 1-3 composite transducer using equation (2.89).

ted. For the 1-3 composite, the mechanical resonant frequency is fm = 9.8 MHz
with an electrical impedance of 12.2 kΩ and the electrical resonant frequency is
fe = 7.2 MHz with associated electrical impedance of 1.3 kΩ. It is clear that by
using a 1-3 composite transducer the thickness mode response has significantly
improved.
Looking at Figure 2.12, which shows the graph of the transmission sensitivity
39

against frequency, it can be seen that the 1-3 composite has a maximum peak
at fm = 7.3 MHz with a percentage bandwidth of 33%. By comparing this
bandwidth to the monolithic plate, it is found that there is a twofold increase in
the transmission sensitivity bandwidth. Figure 2.13 shows the variation in the
reception sensitivity profile as the volume fraction of the ceramic (ψ) is varied
and Figure 2.14 shows the variation in the electrical impedance profile as ψ is
varied. As the ceramic volume fraction decreases the mechanical impedance
decreases, hence the wave velocities decrease along with the associated resonant
frequencies. It can also be seen that as the polymer phase is slowly introduced
to the pure ceramic (ψ = 1) the magnitude of the reception sensitivity decreases
and the resonant frequencies increase. In the electrical impedance frequency
profile, it can be seen that an ideal profile (low amplitude at the electrical
resonant frequency, high value at the mechanical resonant frequency) is achieved
at an intermediate ceramic volume fraction. As the volume fraction diminishes
the resonant behaviour vanishes and the device acts like a simple capacitor.
Figure 2.15 illustrates how the impedance graph (Figure 2.14) can also be
used to define an alternative electromechanical coupling coefficient using the
expression [1] v  
π fe
u
u
2 fm
kt = t . (2.89)
u

π fe
tan 2 fm

Comparing the results to Figure 2.8 both curves are of similar shape and mag-
nitude. However k̄t is generally of larger magnitude with a long plateau at
intermediate values of the ceramic volume factor and a less well pronounced
maximum.
40

2.4 Conclusions
A linear systems model has been derived and used to model a PZT5H ceramic
transducer. The methods of Smith and Auld (1991) were then introduced to
develop a model for a 1-3 transducer with polymer filler HY1300/CY1301. Re-
ception sensitivity, transmission sensitivity and impedance curves were produced
to compare a piezoelectric plate transducer and a 1-3 composite transducer. The
composite transducer was found to have better operational characteristics than
the piezoelectric plate. Importantly the derivation and subsequent use of the
linear systems model does not incorporate any loss mechanisms. The model will
therefore be extended in the next chapter to include elastic losses.
Chapter 3

Thickness Mode Response of a


lossy 1-3 Composite Transducer
with an SBS Passive Phase

3.1 Introduction
In this chapter the LSM model will be developed to include elastic losses. The
method is extended here to include expressions for frequency dependent mechan-
ical loss, and a comparison of the reception/transmission sensitivity profiles for a
transducer with and without loss is made. The effective properties of a composite
transducer with an anisotropic polymer phase are investigated and a comparison
to a finite element model (FEM), using a range of polymer composites conducted.
A disadvantage of composite transducers is that there is crosstalk between the
active ceramic pillars which reduces the electromechanical coupling efficiency
[29]. To address this disadvantage the use of an anisotropic polymer filler is in-
vestigated here. Obviously a one-dimensional model such as the linear systems
model cannot predict these lateral modes. This investigation will be done in
Chapter 5 using the three dimensional plane wave expansion (PWE) method.
However, it will prove useful to compare the output from the LSM with the PWE
and FEM. The range of polymer fillers that will be compared are a standard

41
42

hardset, a softset, material HP20 and an anisotropic styrene-butadiene-styrene


(SBS) polymer. The SBS polymer is anisotropic with effective styrene-like prop-
erties in the thickness direction and polybutadiene characteristics in the lateral
directions. These characteristics of the material should maintain the piston-like
behavior of the transducer whilst reducing crosstalk between the ceramic pil-
lars. For the hardset material the performance of the transducer with no loss
mechanisms and that with loss will be compared [52].

3.2 Loss
Smith and Auld [67] do not explicitly have loss within their effective medium
model but in any real transducer there will always be wave attenuation. Elastic
damping depends on the temperature, frequency of vibration and the physi-
cal properties of the materials. Acoustic losses can be described by a viscous
damping term [7] and in this section it is shown how mechanical losses can be
introduced into the formulation. The mechanical loss for the shear modulus G
is calculated using the relaxation equation [40]

G = G0 + iG00 (3.1)

where

(Gu − Gr )ω 2 τ 2
G0 = G r + , (3.2)
1 + ω2τ 2
(Gu − Gr )ωτ
G00 = , (3.3)
1 + ω2τ 2

Gr is the relaxed shear modulus, Gu is the unrelaxed shear modulus, ω is the


angular frequency and τ is the shear relaxation time. Loss can be included in
the Bulk and Young’s moduli in a similar fashion. Although dielectric loss is
not considered here, it can also be included in the dielectric constant , via

 = 0 − i00 (3.4)
43

where

(r − u )ω 2 τE2
0 =  u + ,
1 + ω 2 τE2
(r − u )ωτE
00 = ,
1 + ω 2 τE2

r is the relaxed dielectric constant, u is the unrelaxed dielectric constant and


τE is the dielectric relaxation time. The degree of loss is usually expressed in
terms of a dimensionless loss tangent tan δ [40], or attenuation coefficient α [35].
For example,
G00
tan δ = (3.5)
G0
for the mechanical shear loss. Hence equation (3.1) can be written as

G = G0 (1 + i tan δ). (3.6)

The shear wave phase velocity is defined as


s
G
vs = . (3.7)
ρ

However
ω ω ω/k 0 c
vs = = = = , (3.8)
k k − iα
0 1 − iα/k 0 1 − iα/k 0
where c is the measured shear wave velocity and the imaginary part of the
wavenumber k is the attenuation coefficient. Combining equations (3.7) and
(3.8) with equation (3.6) gives
s
G0 c
(1 + i tan δ) = , (3.9)
ρ 1 − iα/k 0
p
which, by using c = G0 /ρ, can be rearranged to give

1
c(1 + i tan δ) 2 = c(1 − iαc/ω)−1 . (3.10)
44

Now, by using a Taylor series expansion, the attenuation coefficient (Nepers/m)


can be expressed as
ω
α= tan δ, (3.11)
2c
To find τ , Gu and Gr , the attenuation coefficient α and the (measured/real
part) wave speed c must be measured at a certain frequency ωI , and then the
associated G0I is calculated. The frequency where tan δ achieves its maximum
ωmax [40] is also required. Substituting equations (3.2) and (3.3) into equation
(3.5) then differentiating with respect to ω gives
2
(Gu − Gr )τ (Gr + Gu ωmax τ 2 ) − 2Gu ωmax τ 2 (Gu − Gr )ωmax τ
2
= 0. (3.12)
(Gr + Gu ωmax τ 2 )2

This expression simplifies to

2
Gr + Gu ωmax τ 2 − 2Gu ωmax
2
τ 2 = 0, (3.13)

which can then be rearranged to give


r
Gr 1
ωmax = . (3.14)
Gu τ

Substituting equations (3.2) and (3.3) into equations (3.5) and (3.11) yields the
expression
(Gu − Gr )ωI τ 2cα
2 2
= , (3.15)
G r + G u ωI τ ωI
so that
Gu (ωI2 τ − 2cαωI2 τ 2 ) = Gr (2cα + ωI2 τ ), (3.16)

which can be rearranged to obtain

Gr ω 2 τ (1 − 2cατ )
= I . (3.17)
Gu 2cα + ωI2 τ

Substituting equation (3.17) equation (3.14) gives

2 ωI2 τ (1 − 2cατ )
ωmax τ2 = (3.18)
2cα + ωI2 τ
45

which can be rearranged to give the quadratic equation

ωI2 ωmax
2
τ 2 + 2cα(ωI2 + ωmax
2
)τ − ωI2 = 0. (3.19)

Solving equation (3.19) and taking the positive root gives


p
−cα(ωI2 + ωmax
2
) + c2 α2 (ωI2 + ωmax
2 )2 + ωI4 ωmax
2
τ= . (3.20)
ωI2 ωmax
2

Gr can be expressed in equation (3.2) as

Gr = G0I + (G0I − Gu )ωI2 τ 2 , (3.21)

which is then substituted into equation (3.17) to get

Gu ωI2 τ (1 − 2cατ ) = (G0I + (G0I − Gu )ωI2 τ 2 )(2cα + ωI2 τ ), (3.22)

which simplifies to

Gu (ωI2 τ + ωI4 τ 3 ) = G0I (1 + ωI2 τ 2 )(2cα + ωI2 τ ). (3.23)

This is rearranged to give the expression

G0I (2cα + ωI2 τ )


Gu = . (3.24)
ωI2 τ

Substituting equation (3.24) into equation (3.17) gives

Gr = G0I (1 − 2cατ ). (3.25)

The attenuation coefficient is often given in dB/m and equation (3.5) requires a
value in Nepers/m. A straightforward conversion factor can be derived however.
Consider the decay in the spatial component of a voltage, expressed in terms of
an attenuation coefficient via V (x) = V0 e−αx . This can be rearranged for α to
give, α = 1/x ln(Vo /V ) Nepers/m. Alternatively logarithms to base 10 can be
taken to give α = 1/(x log10 (e)) log10 (V0 /V ). To state this in decibels per metre
a multiplication of 20 log10 (e) ≈ 8.686 is then needed [7]. The calculated values
46

for the materials in this Chapter are presented in Table 3.1. Due to to the lack
of information on ωmax for these materials in the available literature estimates
were used. A sensitivity analysis was then performed to gauge the effect that
this approach had on the parameters Gu , Gr and τ . This concluded that the
effects were minimal and that these estimates were reasonable. Equations (3.20),
(3.24) and (3.25) use the data in Table 3.1 to calculate a frequency dependent
shear modulus via equation (3.1) (that is, the Lamé coefficient c44 ) [10]. Similar
equations are used for the Young’s modulus (the Lamé coefficient c12 ). Note that
if the imaginary part of c12 turns out to be small and negative, it is set equal to
zero. Also note that it is assumed that the ceramic has frequency independent
viscoelastic losses of the form c11 = c11 (1 + i tan δ).

F̄F
(Zc /(h33 Co b))
1

0.75

0.5

0.25
PSfrag replacements

f
1 2 3 4

Figure 3.1: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 for a 1-3 composite
transducer for the hardset material HY1300/CY1301 with frequency dependent,
viscoelastic loss (full line) and no loss (dashed line) (see Tables 2.1, 2.2 and 3.1
for material properties).

In Figure 3.1 the associated reception sensitivity of the device, given by


equation (2.53), is plotted as a function of the driving frequency; the passive
phase is the hardset material HY1300/CY1301 (see Table 3.1) and the volume
fraction is ψ = 0.7. The effect of incorporating frequency dependent loss into the
47

ZT
20

15

10

5
PSfrag replacements
f
1 2 3 4

Figure 3.2: Electrical impedance ZT (kΩ) against frequency f (Hertz) ×107


for a 1-3 composite transducer with the hardset material HY1300/CY1301 with
frequency dependent, viscoelastic loss (full line) and no loss (dashed line).
F̄F
V¯s
/(h33 Co )

0.6

0.4

0.2
PSfrag replacements

f
1 2 3 4

Figure 3.3: Nondimensionalised transmission sensitivity (absolute value)


(F̄F /V¯s )/(h33 Co ) against frequency f (Hertz) ×107 for a 1-3 composite trans-
ducer with the hardset material HY1300/CY1301 with frequency dependent,
viscoelastic loss (full line) and no loss (dashed line).

model can be seen by comparing the no loss case (dashed line) with the loss case
(full line). The addition of the elastic loss does not affect the profile except at
the main peaks where they are reduced by around 5 per cent. This reduction in
amplitude gives rise to a slight increase in the percentage bandwidth from 24%
48

k̄t

0.6

0.5

0.4
PSfrag replacements
ψ
0.2 0.4 0.6 0.8 1

Figure 3.4: Electromechanical coupling coefficient k̄t against ceramic vol-


ume fraction ψ for a 1-3 composite transducer with the hardset material
HY1300/CY1301 with frequency dependent, viscoelastic loss (full line) and no
loss (dashed line).

kt

0.5

0.45

0.4

PSfrag replacements
0.35
ψ
0.2 0.4 0.6 0.8

Figure 3.5: Electromechanical coupling coefficient kt against ceramic vol-


ume fraction ψ for a 1-3 composite transducer with the hardset material
HY1300/CY1301 with frequency dependent, viscoelastic loss using equation
(2.89).

to 25% MHz. Figure 3.2 shows that the electrical impedance has a mechanical
resonant frequency of fm = 9.7 MHz and an electrical resonant frequency of
fe = 7.3 MHz. It can be seen that introducing loss for these materials has
had very little effect on the profile except at the mechanical resonant frequency.
49

0
0.25
0.5 ψ

0.75

1
PSfrag replacements

1 2 3 4 F̄V̄F (Zc/(h33 Co b))


f

Figure 3.6: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 and volume fraction
ψ for a 1-3 composite transducer for the hardset material HY1300/CY1301 with
frequency dependent, viscoelastic loss.
0
0.25
0.5 ψ

0.75

1
PSfrag replacements ZT
1 2 3 4
f

Figure 3.7: Electrical impedance log ZT (kΩ) against frequency f (Hertz) ×107
and volume fraction ψ for a 1-3 composite transducer with the hardset material
HY1300/CY1301 with frequency dependent, viscoelastic loss.

Here there is a marked decrease in the magnitude of the electrical impedance


although its frequency (the mechanical resonant frequency) remains constant.
In Figure 3.3 the associated transmission sensitivity, given by equation (2.49)
50

is plotted. Again the addition of the viscoelastic loss does not affect the profile
except at the main peaks where they are slightly reduced and the percentage
bandwidth increases by 1% to 34%.
In Figure 3.4 the electromechanical coupling coefficient, given by equation
(2.85), is plotted as a function of the volume fraction of the ceramic phase. It can
be seen that the frequency dependent loss reduces the efficiency across the full
range of volume fractions. From a design perspective the highest efficiency (of
around k̄t = 0.65) occurs at a ceramic volume fraction of around ψ = 0.6. The
diagram also highlights the benefits of using a composite design. The coefficient
has risen by around 30 per cent from k̄t = 0.5 for the pure ceramic (ψ = 1) to
the optimum ceramic volume fraction of around ψ = 0.6. The low permittivity
value in the polymer phase ensures that the piezoelectric constant h33 has a high
value for even low volume fractions of ceramic. This in turn leads to a reasonably
high electromechanical coupling coefficient at low ceramic volume fractions. This
results in the reception and transmission characteristics of the device being near
optimal across a wide range of ceramic volume fractions. Figure 3.5 shows the
electromechanical coupling coefficient calculated using equation (2.89) and the
impedance data in Figure 3.7. Comparing this to the no loss case in Figure 2.15
shows that both curves are very similar.
In Figure 3.6 the nondimensionalised reception sensitivity frequency profile
is plotted against the ceramic volume fraction. As the ceramic volume fraction
decreases the resonant frequencies decrease. Since the mechanical impedance
decreases as the volume fraction of the polymer increases, then the wave ve-
locities decrease along with the associated resonant frequencies. It can also be
seen that the magnitude of the reception sensitivity decreases as the polymer
phase is slowly introduced to the pure ceramic (ψ = 1). This behaviour is main-
tained over a wide range of ceramic volume fractions since the low permittivity
in the polymer keeps the piezoelectric constant at a reasonable level (and the
51

electromechanical coupling coefficient) even for low ceramic volume fractions.


In Figure 3.7 the electrical impedance frequency profile is plotted as a function
of the ceramic volume fraction. Here it can be seen that an optimum profile
(low amplitude at the electrical resonant frequency, high value at the mechani-
cal resonant frequency) is achieved at an intermediate ceramic volume fraction.
As the volume fraction diminishes the resonant behaviour ultimately vanishes
and the device acts like a simple capacitor.

3.3 SBS
In this section the transmission and response characteristics of a 1-3 composite
transducer, with a passive phase consisting of a styrene-butadiene-styrene (SBS)
polymer are investigated. The effective elastic and dielectric properties of the
SBS are derived here. In a series of papers Pethrick and co-workers have studied
the mechanical response of this material to ultrasonic stimulus [5, 4]. They
studied the effects of the volume fraction of styrene and the temperature on
the longitudinal wave speed and loss tangent at a fixed frequency [16]. Scanning
electron microscope images clearly showed the anisotropic nature of the material.
In the direction parallel to the extrusion process the material is striated whereas,
perpendicular to this, the material resembles styrene columns randomly set in
a polybutadiene matrix [56]. Velocity measurements were only presented for
longitudinal propagation on compression moulded samples as the experimental
equipment did not have a large enough angular aperture to cope with the critical
angle for shear wave propagation. This work was extended to examine the
frequency dependence of the attenuation [5] and looked at how phase separation
affects the viscoelastic properties of the composite [4]. In this section a linear
systems model of the SBS 1-3 composite transducer is constructed using the
analysis of the previous sections. The mechanical loss for the Young’s (Y ) and
shear (G) moduli are calculated independently for both the polystyrene and
52

polybutadiene in a similar manner to that used for equation (3.1). An effective


scattering approach is used to calculate the effective bulk (Be ) and shear (Ge )
moduli [10] of the SBS material where [72]

3Bb Bp + 4Bp Gp + 4φBb Gp − 4φBp Gp


Be = , (3.26)
3Bb + 4Gp − 3pBb + 3φBp

Gp (4Gp (2Gp (φ − 1) − Gb (3 + 2φ)) − 3Bp (−3Gp (φ − 1) + Gb (2 + 3φ)))


Ge = ,
Bp (6Gb (φ − 1) − 3Gp (3 + 2φ)) − 4Gp (−3Gb (φ − 1) + Gp (2 + 3φ))
(3.27)
Bp , Bb , Gp , Gb are the Bulk and Shear moduli for the polystyrene and the polybu-
tadiene respectively and φ is the volume fraction of polybutadiene within the
SBS polymer. The effective elastic tensor elements for the polymer are then
given by [11]

c44 = Ge , c12 = Be − 2/3Ge and c11 = Be + 4/3Ge . (3.28)

Note that if there is only experimental data for the shear and Young’s moduli
then
G(2G − Y )
c12 = (3.29)
Y − 3G
can be used to calculate the bulk modulus [11]. The effective dielectric constant
is [72]
p (2p + b + 2φ(b − p ))
e = . (3.30)
2p + b − φ(b − p )
53


F̄F
(Zc /(h33 Co b))
1

0.75

0.5

PSfrag replacements 0.25

f
1 2 3 4

Figure 3.8: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 for an SBS 1-3 com-
posite transducer with frequency dependent, elastic loss (full line) and no loss
(dashed line) (see Tables 2.1, 2.2 and 3.1 for material properties).

ZT
20

15

10

PSfrag replacements 5

f
1 2 3 4

Figure 3.9: Absolute value of the impedance ZT (kΩ) against frequency f


(Hertz)×107 for an SBS 1-3 composite transducer with frequency dependent,
elastic loss (full line) and no loss (dashed line).
54

F̄F
V¯s
/(h33 Co )

0.6

0.4

PSfrag replacements 0.2

f
1 2 3 4

Figure 3.10: Nondimensionalised transmission sensitivity (absolute value)


(F̄F /V¯s )/(h33 Co ) against frequency f (Hertz) ×107 for an SBS 1-3 composite
transducer with frequency dependent, elastic loss (full line) and no loss (dashed
line).

The composite model detailed in Section 2.3 is then used to calculate the
effective properties of the 1-3 transducer for use in the linear systems model.
Figure 3.8 shows the reception sensitivity against frequency of the SBS 1-3 com-
posite transducer and Figure 3.9 shows its electrical impedance.

0
0.25
0.5 ψ

0.75

PSfrag replacements 1

1 2 3 4 F̄V̄F (Zc/(h33 Co b))


f

Figure 3.11: Nondimensionalised reception sensitivity (absolute value)


(V̄ /F̄F )(Zc /(h33 Co b)) against frequency f (Hertz) ×107 and volume fraction
ψ for an SBS 1-3 composite transducer.
55

0
0.25

ψ 0.5

0.75

PSfrag replacements 4
1
3
ZT 2
1
0
2 4 6
f

Figure 3.12: Absolute value of the Electrical impedance log ZT (kΩ) against
frequency f (Hertz) ×107 and volume fraction ψ for an SBS 1-3 composite
transducer.

Zc
3.5
3
2.5
2
1.5
PSfrag replacements 1
0.5
ψ
0.2 0.4 0.6 0.8 1

Figure 3.13: Mechanical impedance Zc (M Rayls) against volume fraction ψ for


an SBS 1-3 composite transducer with frequency dependent, elastic loss (full
line) and no loss (dashed line).
56

kt

0.72
0.7 ψ
0 0.2 0.4 0.6 0.8 1
0.68
0.66
PSfrag replacements 0.64
0.62
0.6

Figure 3.14: Electromechanical coupling coefficient kt against volume fraction ψ


for an SBS 1-3 composite transducer with frequency dependent, elastic loss (full
line) and no loss (dashed line).

The volume fraction of ceramic in Figures 3.8 and 3.9 is the same as that
used in Figures 2.10 and 2.11. The volume fraction (φ) of the polybutadiene
in all the SBS Figures is 0.69. In Figure 3.8 the nondimensionalised reception
sensitivity profile for such a device is shown. Due to the high attenuation in
this polymer there is a marked reduction in amplitude when the loss is included.
Figure 3.9 shows that the electrical impedance has a mechanical resonant fre-
quency of fm = 9.1 MHz, and an electrical resonant frequency of fe = 6.5
MHz. The nondimensionalised transmission sensitivity frequency response in
Figure 3.9 shows a minor decrease in amplitude and a frequency shift. At this
volume fraction of polybutadiene (φ = 0.7) the material has lower values for its
moduli than the hardset material discussed in the previous section. However,
the anisotropy, incorporated into the method through Equations 2.82, 2.83 and
2.84, leads to a similar amplitude for the peak transmission sensitivity.
In Figure 3.11 the nondimensionalised reception sensitivity frequency re-
sponse is shown as a function of the ceramic volume fraction. As before, the
addition of the polymer serves to increase the amplitude of the main lobe. As
57

Data Values Constant Polystyrene Polybutadiene Hardset


Shear modulus (real part) G0 (kg m−1 s−2 ) 1.36 × 109 6.61 × 107 1.57 × 109
Young’s modulus (real part) Y 0 (kg m−1 s−2 ) 3.63 × 109 1.96 × 107 4.28 × 109
tan δ Maximum fmax (Hz) 3.15 × 105 3.15 × 105 3.15 × 105
Frequency of Interest fI (Hz) 5 × 105 5 × 105 5 × 105
0
G Attenuation Coefficient αG (Np/m) 33 1718 41
0
Y Attenuation Coefficient αY (Np/m) 4 13 16
Density ρ (kg m−3 ) 1.04 × 103 0.91 × 103 1.15 × 103
Dielectric constant  2.4 4 4
elastic constant c11 5.58 × 109 1.62 × 109 7.25 × 109
elastic constant c44 1.36 × 109 6.61 × 107 1.57 × 109
Calculated Values
Unrelaxed shear modulus Gu (kg m−1 s−2 ) 1.38 × 109 8.24 × 107 1.60 × 109
Relaxed shear modulus Gr (kg m−1 s−2 ) 1.30 × 109 4.27 × 107 1.50 × 109
Shear relaxation time τG (s) 5.27 × 10−7 3.82 × 10−7 4.88 × 10−7
Unrelaxed Young’s modulus Yu (kg m s ) 3.64 × 109
−1 −2
1.97 × 108 4.35 × 109
Relaxed Young’s modulus Yr (kg m−1 s−2 ) 3.60 × 109 1.92 × 108 4.11 × 109
Young’s relaxation time τY (s) 5.27 × 10−7 5.24 × 10−7 4.91 × 10−7

Table 3.1: Physical properties of the polymer phase polystyrene, polybutadiene


[55] and HY1300/CY1301 [48]

the volume fraction of the ceramic decreases the main peak disappears and the
device no longer displays the desired resonant behaviour. The frequency shift
in the resonant modes is more pronounced than in Figures 3.9 and 3.10 and
this is further highlighted in the electrical impedance frequency profile shown
in Figure 3.12. Figures 3.13 and 3.14 show the mechanical impedance (Zc ) and
the electromechanical coupling coefficient (k̄t ) (calculated using equation (2.85))
against ceramic volume fraction ψ respectively. This efficiency is also exhibited
by the electromechanical coupling coefficient shown in Figure 3.14. The peak
value of around k = 0.73 is achieved at a volume fraction of polymer of ψ = 0.55.
The high attenuation leads to a clearer separation between the loss/no loss pro-
files.
58

3.4 Comparison to Finite Element Modelling

Constant Softset HP20


G0 (kg m−1 s−2 ) 6.5 × 108 3.38 × 108
E 0 (kg m−1 s−2 ) 1.84 × 109 9.68 × 108
Vs (m s−1 ) 747 549
Vl (m s−1 ) 2000 1605
ρ ( kg m−3 ) 1.16 × 103 1.12 × 103
 4 4
c11 1.66 × 109 2.89 × 109
c44 6.50 × 108 3.38 × 108
α0s (db/m) 6063 21281
α0l (db/m) 825 565
f0 (MHz) 0.5 0.5

Table 3.2: Physical properties of the passive phase materials ([32]).

Constant SBS1 SBS2 SBS4


G (kg m−1 s−2 )
0
2.76 × 108 2.30 × 108 4.35 × 108
E 0 (kg m−1 s−2 ) 7.95 × 108 6.66 × 108 1.20 × 109
Vs (m s−1 ) 538.6 498 676
Vl (m s−1 ) 1635 1584 1533
ρ ( kg m−3 ) 0.95 × 103 0.93 × 103 0.95 × 103
 4 4 4
c11 2.55 × 109 2.33 × 109 2.24 × 109
c44 2.7 × 108 2.30 × 108 4.35 × 108
s
α0 (db/m) 10388 10388 7062
α0l (db/m) 87 104 80
f0 (MHz) 0.5 0.5 0.5

Table 3.3: Physical properties of the passive phase materials ([45]).

In Figure 3.15 the LSM model is compared to finite element modelling by


examining the impedance characteristics of a range of 2-2 composite transduc-
ers, all with thickness 3.83 × 10−3 m. A 2-2 composite is made by cutting the
ceramic longitudinally in one direction so that there is connectivity in two direc-
tions for both the ceramic and polymer. The finite element (FE) derived data
59

are produced from a 2-dimensional model of the 1-3 piezoelectric composite us-
ing the PZFlex code [58], the model represents a slice through the transducer
structure and is configured with a boundary of void elements, equivalent me-
chanically and acoustically to vacuum, in order to represent the transducer as
a freely vibrating structure. In order to accurately represent the propagating
wave fronts, the spatial discretisation of the model was 15 elements per wave-
length at the highest frequency of interest, 5MHz in this case. The mechanical
excitation was provided by an impulsive load at one end of the transducer struc-
ture, the mechanical propagation field was allowed to evolve and temporally
and spatially sampled along the width dimension. The data was then processed
using a 2D fast Fourier transform [29, 30] to obtain the behaviour plotted in
Figure 3.15 Plot (a) shows the response when using a standard hardset polymer
which has a low shear attenuation (see Table 2.3). The oscillations in the electri-
cal impedance at the lower frequencies represent the undesirable low frequency
modes which interfere with the piston-like motion of the device. Because of the
high computational cost of FEM, simulations are run in short duration. Due
to the short simulation times, these lower frequency modes are only sampled in
the time domain over relatively few cycles, therefore the fast Fourier transform
of this signal is less accurate and noisy at these lower frequencies. It is neces-
sary to use a higher dimensional model, such as the FEM used here, to display
these modes; the one-dimensional LSM method has a smooth impedance curve
at these frequencies. However, the two methods do agree on the magnitude and
trend of the curve in this region. As the frequency increases the two methods do
differ substantially. The LSM prediction of the electrical impedance magnitude
at the first peak (the mechanical resonant frequency) is much lower than that
given by the FEM, and a larger bandwidth results. The location of this peak
is also slightly shifted to a higher frequency. In plot (b) a softset polymer filler
is used which has high shear and longitudinal attenuation (see Table 3.2). This
60

has damped out the unwanted, low frequency modes but there may well be a
decoupling of the ceramic and polymer phases. This would subsequently inhibit
its ability to transfer energy into the load medium and help explain the drop
in the peak electrical impedance that can be seen. There is better agreement
between the two methods on the magnitude of the first peak impedance, al-
though the frequency discrepancy remains. In plot (c) a material with a very
high shear attenuation and a relatively low longitudinal attenuation is modelled
(see Table 3.2, material HP20). This material should be able to damp out the
unwanted shear modes which propagate through the polymer whilst having suf-
ficient stiffness in the thickness direction to remain in phase with the ceramic
pillars. Both methods are in agreement over the form of the electrical impedance
profile with the FEM approach predicting a far lower value at the electrical res-
onant frequency. Plots (d) and (e) show the electrical impedance curves for
SBS1 and SBS2 in Table 3.3. These materials are similar to HP20, having high
shear attenuation and low longitudinal attenuation, although they are slightly
less stiff and the attenuation coefficients are smaller. Both plots show reasonable
agreement between the two methods. SBS4 has a reduced degree of attenuation
from these materials but it is slightly stiffer (see Table 3.3). Although there
is still good agreement between the two methods around the resonant modes,
this is less true at the lower frequencies. The FEM predicts a lowering of the
impedance in the lower frequency range and this is probably due to the polymer
and ceramic phases oscillating independently.

3.5 Conclusions
Frequency dependent, elastic loss was introduced to the LSM, and an SBS trans-
ducer with high shear loss was compared to an equivalent transducer with a
standard passive filler. By comparing a device with no loss to that with loss in-
corporated, it was shown that the lossy transducer gave very similar results, but
61

Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

1
f (MHz) 1
f (MHz)
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

(a) (b)

Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

1
f (MHz) 1
f (MHz)
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

(c) (d)

Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

0.2 0.4 0.6 0.8 1


f (MHz) 0.2 0.4 0.6 0.8 1
f (MHz)

(e) (f)

Figure 3.15: Electrical impedance characteristics for the frequency dependent,


elastic loss transducers with the passive phase materials described in Tables 3.2
and 3.3 using FE modelling (solid line) and the LSM method (dashed line) with
polymer phase (a) hardset, (b) softset, (c) HP20, (d) SBS1, (e) SBS2 and (f)
SBS4 .

was slightly damped. It was found that by using the anistropic SBS filler, the
transducer will perform well over a larger range of frequencies and dampen out
unwanted lateral modes. There was a comparison made between the LSM and
FEM for a range of polymer composite fillers. This confirmed that the LSM was
robust as the thickness mode response matched well for each comparison. The
drawback with the LSM modelling is that it is only a one-dimensional model,
62

hence additional modes which occur in the lateral directions cannot be investi-
gated here. Plane wave expansion methods [74], that incorporate lamina models
[68, 15] of the SBS to describe its anisotropy, will be investigated in Chapter 5.
Chapter 4

Modelling of Composite
Transducers using the Plane
Wave Expansion Method

4.1 Introduction
In this chapter the three-dimensional plane wave expansion (PWE) model of
the 1-3 composite is derived. The geometry of the transducer is described in
terms of a two-dimensional, spatial Fourier series and then the PWE method
and associated boundary conditions are outlined. The derivation of Wilm et
al [74] is extended here to include the Poynting vector, a parameter scaling
to reduce ill-conditioning, an admittance that incorporates a dependency on
the spatial wavelength of the driving voltage, a comparison of the PWE method
with the Rayleigh-Lamb equation and linear systems modelling, the dependency
of the determinant of the harmonic analysis matrix on the driving frequency
and wavenumber, and a mode identification methodology for both 2-2 and 1-3
composite transducers using profiles of the displacements, Poynting vectors and
electrical potential.
Ideally a single longitudinal mode in the thickness direction will drive the
transducer in a piston like fashion. As discussed in Chapter 1, other modes,
propagating in other directions, can interfere with this behaviour. Hence it

63
64

is of interest to theoretically predict the design criteria, material parameters,


etc. which will provide a large frequency band gap between the desired thick-
ness mode and these other waves. For instance, surface waves will travel along
the boundary between two different media say in a layered design, Rayleigh
waves will propagate near the surface of a semi-infinite solid, Lamb waves will
propagate near both surfaces of a finite thickness plate and bulk waves keep
a constant phase velocity throughout propagation and their energy is evenly
distributed throughout the layer [60]. However these descriptions pertain to
isotropic materials and their use as descriptions of the waves found in heteroge-
neous, piezoelectric and anisotropic layers is to be treated with caution.

4.2 Formulation of the method


4.2.1 The Geometry
PSfrag replacements x2 PSfrag replacements

x1
x2

 


  p2
p2 t 

  x1



 p1
s
s
(a) p1 (b) t

Figure 4.1: (a) Plan view of one period of a 1-3 piezoelectric transducer in the
x1 -x2 plane and (b) three-dimensional geometry (the x3 axis is in the vertical
direction) over a few periods of the 1-3 connectivity piezoelectric transducer.

The model is configured for periodic 2-2 and 1-3 composites with thickness
in the x3 direction (see Figure 4.1). By using the periodicity of the structure in
the x1 − x2 plane, the material constants, M (r), can be expanded as a Fourier
series where M represents either the density ρ, the elasticity tensor cijkl , the
65

piezoelectric stress tensor eijk , or the permittivity tensor ij . For the 1-3 com-
posite structure shown in Figure (4.1) the material constants only depend on x1
and x2 . Denote by p1 the period of the geometry in the x1 direction and by p2
the period of the geometry in the x2 direction. The width of the ceramic pillar
in the x1 and x2 direction is denoted by s and t, respectively. That is

< x1 < − 2s and −p2 2 < x2 < p22


 −p1
 θ, if 2
< x1 < 2s and −p2 2 < x2 < −t
 −s
 θ, if


2 2
M (x1 , x2 ) = θ, if −s
2
< x1 < 2s and t
2
< x2 < p22 (4.1)
θ, if s
< x1 < p21 and −p2 2 < x2 < p22



 2
φ, if −s
< x1 < 2s and −t < x2 < 2t

2 2
where θ and φ are the passive and the ceramic phase respectively. They satisfy
the periodicity relationship M (x1 + p1 a, x2 + p2 b) = M (x1 , x2 ), ∀ a, b  Z. A
Fourier series representation for this geometry can be written,
∞ ∞
−( 2πm x1 + 2πn x2 )
X X
M (x1 , x2 ) = Mmn e p 1 p 2 . (4.2)
m=−∞ n=−∞

Multiplying both sides by e(2πn/p2 )x2 and integrating with respect to x2 from
−p2 /2 to p2 /2 then multiplying by e(2πm/p1 )x1 and integrating with respect to
x1 from −p1 /2 to p1 /2 gives

Z p1 Z p2
2 2
( 2πm x1 + 2πn x2 )
M (x1 , x2 )e p
1 p 2 dx2 dx1 = p1 p2 Mmn , (4.3)
p1 p2
− 2
− 2

since

Z pi 
2
−( 2πb xi ) 0, for b  Z, b 6= 0
e pi dxi = (4.4)

pi pi , for b = 0.
2

Equation (4.3) can be expanded as


Z p1 Z p2
2 2
−( 2πm x1 ) −( 2πn x2 )
p1 p2 Mmn = θ e p 1 dx1 e p 2 dx2
p1 p2
− 2
− 2
Z s Z t
2 2
−( 2πm x1 ) −( 2πn x2 )
+ (φ − θ) e p1 dx1 e p2 dx2 (4.5)
− 2s − 2t
66

which integrates to give the Fourier coefficients,

(φ − θ)t πms
Mm0 = sin( ) m = ±1, ±2, ... (4.6)
p2 πm p1

(φ − θ)s πnt
M0n = sin( ) n = ±1, ±2, ... (4.7)
p1 πn p2
(φ − θ) πms πnt
Mmn = 2
sin( ) sin( ), m, n = ±1, ±2, ... (4.8)
π mn p1 p2
Note that M00 = θ + (φ − θ)st/p1 p2 and this can be expressed as

M00 = φψ + θ(1 − ψ) (4.9)

where ψ = st/p1 p2 is the volume fraction of φ. So, as expected, the first Fourier
coefficient is the simple mixing rule average of the two materials. This double
subscript notation can be simplified by ordering the Fourier coefficients as shown
in Figure 4.2.

•(−N, M ) ... •(0, M ) ... •(N, M )

.. .. .. .. ..
. . . . .

•(−N, 0) ... •(0, 0) ... •(N, 0)

.. .. .. .. ..
. . . . .

•(−N, −M ) . . . •(0, −M ) ... •(N, −M )

Figure 4.2: Illustration of ordered Fourier coefficents


67

Define the ordered set

H = {(−N, −M ), (−N, −M + 1), . . . , (−N, M ), (−N + 1, −M ), . . . , (N, M )}


(4.10)
so that if

 
s 2π s,1 2π s,2
G = H , H ,0 , (4.11)
p1 p2
then (4.2) can be rewritten for a finite number of terms (N in direction x1 and
M in direction x2 ) as
L
s
X
M (x1 , x2 ) = M s e−G .r . (4.12)
s=1
s,i th
where H is the i component of element s of H and L = (2N + 1)(2M + 1).
The 2-2 geometry is described in this framework by setting t = p2 and M = 0.
In a similar way the Floquet theorem states that all fields F (r, t), for in-
stance displacements or stresses, propagating within periodic structures can be
approximated as the series
L
s ·r)
X
F (r, t, k, ω) = F s (k, ω)e(ωt−k·r−G (4.13)
s=1

where r = (x1 , x2 , x3 ), t is time, ω is the angular frequency and k = (k1 , k2 , k3 )


is the wave vector [39].

4.2.2 The Model

The piezoelectric constitutive equations together with Newton’s second law and
Gauss’s law for dielectric media are [1]
68

Tij = cijkl uk,l + elij φ,l (4.14)

Di = eikl uk,l − il φ,l (4.15)

∂ 2 uj
ρ = Tij,i (4.16)
∂t2

Di,i = 0. (4.17)

These constitute 16 equations in the 16 unknowns which are the stresses Tij , the
displacements uk , the electric potential φ and the electrical displacements Di .
Denote the generalized displacement field by u where u = (u1 , u2 , u3 , u4 = φ)
and the generalized stress vectors by ti = (Ti1 , Ti2 , Ti3 , Di ). Substituting the
expansions (4.12) and (4.13) into equation (4.14) and (4.15) yields

L L X
L
p r +Gt )·r
X X
Tijp e−G ·r = −(kl + Grl )(ctijkl urk + etlij ur4 )e−(G (4.18)
p=1 r=1 t=1

and
L L X
L
p r +Gt )·r
X X
Dip e−G ·r = −(kl + Grl )(etikl urk − til ur4 )e−(G . (4.19)
p=1 r=1 t=1

p ·r
Now multiply equations (4.18) and (4.19) by eG and integrate over one
period (from −p1 /2 to p1 /2 in x1 and −p2 /2 to p2 /2 in x2 ); this essentially
p
equates the coefficients of e−G ·r . Obviously only certain combinations of Gr
and Gt on the right hand side of these equations will combine to equate to a
Gp term on the left hand side. This restriction is captured by V p,q which, for
a given Fourier term H p , describes the Fourier terms H q which survive on the
69

right hand side of (4.18) and (4.19) after integration. It transpires that
L+1 L+1
 p,1 q,1
 p + 2 − q, if 1 ≤ p + 2 − q ≤ L, |H − H | ≤ N




p,q
V = and |H p,2 − H q,2 | ≤ M (4.20)




0, otherwise

p,q
where, for example, cVijkl = 0 if V p,q = 0. Using equation (4.20), equations
(4.18) and (4.19) can now be written

L
p,q p,q
X
Tijp = −(kl + Gql )(cVijkl uqk + eVlij uq4 ) (4.21)
q=1

and
L
p,q p,q
X
Dip = −(kl + Gql )(eVikl uqk − ilV uq4 ). (4.22)
q=1

In terms of the generalized stress vectors, (4.21) and (4.22) combine to give

L
X
tpi = (kl + Gql )Ap,q
il u
q
(4.23)
q=1

where

p,q p,q p,q p,q


cVi11l cVi12l cVi13l eVli1
 
 
 V p,q V p,q V p,q p,q 
 ci21l ci22l ci23l
 eVli2 

Ap,q
 
il =
 cV p,q cV p,q cV p,q eV p,q
.
 (4.24)
 i31l i32l i33l li3 
 
 V p,q V p,q V p,q p,q

 ei1l ei2l ei3l −Vil 

The same analysis can be carried out for equations (4.16) and (4.17) to obtain
the expression

L
X
(ki + Gpi )tpi = ω 2 Rp,q uq , (4.25)
q=1

where
70

p,q
ρV
 
0 0 0
p,q
p,q
 0 ρV 0 0 
R =  V p,q
.
0 0 ρ 0 
0 0 0 0
T T
Now let Ti = t1i , . . . , tpi , . . . , tLi , U = u1 , . . . , uq , . . . , uL ,
 

 1,1
Aij A1,2 ij . . . A 1,L 
ij
 A2,1 A2,2 . . . A2,L 
 ij ij ij
Aij =  . , (4.26)

. . .
 .. .. .. ..  
AL,1
ij AL,2
ij . . . AL,L ij

and  
R1,1 R1,2 . . . R1,L
 R2,1 R2,2 . . . R2,L 
R= .. .. .. . (4.27)
 
..
 . . . . 
RL,1 RL,2 . . . RL,L
Equations (4.23) and (4.25) can then be written compactly as

Ti = Aij Γj U (4.28)

and
ω 2 RU = Γi (Ti ) (4.29)

where  
(ki + G1i )I4 0 ... 0
 0 (ki + G2i )I4 . . . 0 
Γi =  .. .. .. . (4.30)
 
..
 . . . . 
0 0 . . . (ki + GLi )I4
Multiplying equation (4.28) by Γi and using equation (4.29) gives

Γi (Ti ) = Γi Aij Γj U = ω 2 RU. (4.31)

Equation (4.31) can now be expanded to obtain the expressions

ω 2 RU = BU + C1 k3 U + k3 (T3 ) (4.32)

and
k3 (T3 ) = k3 (C2 + Dk3 )U (4.33)
71

P P P
where B = i,j=1,2 Γi Aij Γj , C1 = i=1,2 Γi Ai3 , C2 = j=1,2 A3j Γj and D = A33 .
These are combined to give the generalised eigenvalue problem
 2     
ω R−B 0 U C1 I U
= k3 (4.34)
−C2 I T3 D 0 T3
 (r)
(r) U
in the 8L eigenvalues k3
and corresponding eigenvectors . Solving
T3
equation (4.34) and introducing the relative amplitudes A(r) gives

  L 8L  q
(r) !
u(r, t) X q ·r
X (r) u
= e(ωt−k1 x1 −k2 x2 ) e−G A(r) e −k3 x3
. (4.35)
tq3 (r, t) tq3
q=1 r=1

Energy distribution within the transducer can be used to clarify particular types
of modes in conjunction with examining the profiles of the displacements, stresses
and electric potential (see the detailed discussion on waves in Section 1.1). The
energy in the device can be examined using the Poynting vector which is defined
as
Pj = −Tij ui,t + φDj,t . (4.36)

Substituting equations (4.14) and (4.15) into equation (4.36) gives

Pj = −(cijkl uk,l + elij φ,l )ui,t + φ(ejkl uk,l − jl φ,l ),t . (4.37)

Since for m = 1, 2
L 8L
!
(r)
−Gq ·r q,(r)
X X
uk,m = e(ωt−k1 x1 −k2 x2 ) −(km +Gq,m )e A(r) e −k3 x3
uk , (4.38)
q=1 r=1

and for m = 3,
L 8L
!
(r)
−Gq ·r (r) q,(r)
X X
uk,3 = e(ωt−k1 x1 −k2 x2 ) e −k3 A(r) e −k3 x3
uk (4.39)
q=1 r=1

q,(r)
then uk,t = ωuk and (uk,l )t = ωuk,l. The notation uk denotes the entry in
the r th eigenvector pertaining to the q th element of the displacement u in the xk
direction. Hence the Poynting vector can now be written as
72

Pj = −ω(cijkl uk,l + elij φ,l )ui + ωφ(ejkl uk,l − jl φ,l ). (4.40)

4.3 Boundary Conditions


The method is sufficiently general to cope with a wide range of boundary con-
ditions but for simplicity the mechanical boundary conditions of a stress free
plate are considered. From equation (4.35),

L 8L
!
(r)
−Gq ·r q,(r)
X X
(ωt−k1 x1 −k2 x2 ) (r) −k3 x3
T3i = e e A e T3i . (4.41)
q=1 r=1

For a stress free plate T3i = 0 at x3 = h which yields the expression


L 8L
!
(r)
q q,(r)
X X
0= e−G ·r A(r) e−k3 h T3i . (4.42)
q=1 r=1

q ·r
Equation (4.42) is then multiplied by eG and integrated over one period x1 ×
x2  −p2 1 , p21 × −p2 2 , p22 to get
   

8L
(r)
q,(r)
X
0= A(r) e−k3 h T3i , q = 1, . . . , L. (4.43)
r=1

Equation (4.43) consists of 3L scalar equations in the 8L unknowns A(r) . Now


the stress free boundary condition at the lower face (x3 = 0) is applied to obtain

8N
q,(r)
X
0= A(r) T3i , q = 1, . . . , L (4.44)
r=1

which gives a further 3L equations.


For the electrical boundary conditions the electrical potentials at the top and
bottom of the transducer are prescribed. The lower surface is a monolithic plate
with zero electrical potential. The upper plate has a set of electrodes which
follow the periodicity of the underlying composite. If νi is the wavelength of the
73

electrical potential then the surface potential is given by


“ ”
 ωt− 2π x1 − 2π x2
φ(x1 , x2 , t) = V0 e ν 1 ν 2 . (4.45)

The voltage is sinusoidal in time and the device has a discrete set of electrodes on
the upper face (see Figure 4.3). In order to describe this as a boundary condition
in our continuum, the electrode spatial distribution is described as a continu-
ous sinusoid. Hence the wavelength of this sinusoid achieves its minimum value
when the voltage at adjacent modes is precisely 180o out of phase. The electrical

Top electrodes

Voltage

Figure 4.3: The top electrode spacing and an example applied voltage when
γ1 = k1 p1 /(2π) = 1/2.

potential excites waves of wavelength νi = 2π/ki (i = 1, 2). The nondimension-


alised parameter γi = ki pi /2π = pi /νi (i = 1, 2) is used to investigate the effect
that the electrical potential periodicity has on the mode distribution. At one
extreme, γi = 0, this denotes infinitesimally close electrodes or that all the elec-
trodes are in phase. At the other extreme, γi = 1/2, the adjacent electrodes
are precisely 180o out of phase which corresponds to a half wavelength electrode
spacing (see Figure 4.3). From equation (4.35) and (4.45),
L 8L ““ ” “ ” ”
q ·r (r)  k1 − 2π x1 + k2 − 2π x2
X X
e−G A(r) φq,(r) e−k3 h
= V0 e ν
1 2ν
(4.46)
q=1 r=1
74

q
at x3 = h. Equation (4.46) is then multiplied by eG ·r and integrated over the
h i h i
p1 p1 p2 p2
spatial electrode period x1 × x2  − 4γ ,
1 4γ1
× − ,
4γ2 4γ2
. Since
p2 p1
p1 p2
Z Z
4γ2 4γ1 q q
e−G ·r eG ·r dx1 dx2 = , (4.47)
p
− 4γ2
p
− 4γ1 4γ1 γ2
2 1

and Z p2 Z p1
4γ2 4γ1
e((k1 +2π/p1 (H ) dx dx
q,1 −γ q,2 −γ
1 ))x1 +(k2 +2π/p2 (H 2 ))x2
1 2
p p
− 4γ2 − 4γ1
2 1
" # 4γp1 " # 4γp2
(k1 +2π/p1 (H q,1 −γ1 ))x1 1 (k2 +2π/p2 (H q,2 −γ2 ))x2 2
e e
=
(k1 + 2π/p1 (H q,1 − γ1 )) p
(k2 + 2π/p2 (H q,2 − γ2 )) p
− 4γ1 − 4γ2
1 2

p1 p2
! !
−2 sin(k1 + 2π/p1 (H q,1 − γ1 )) 4γ 1
−2 sin(k2 + 2π/p2 (H q,2 − γ2 )) 4γ 2
=
(k1 + 2π/p1 (H q,1 − γ1 )) −(k2 + 2π/p2 (H q,2 − γ2 ))

   
p1 p2 q,1 p1 q,2 p2
= sinc (k1 + 2π/p1 (H − γ1 )) sinc (k2 + 2π/p2 (H − γ2 )) .
4γ1 γ2 4γ1 4γ2
(4.48)

Equation (4.46) can be rewritten as


8L  
X
(r) q,(r) −k3 h
(r) p1 q,1
A φ e = V0 sinc (k1 + 2π/p1 (H − γ1 ))
r=1
4γ1
 
q,2 p2
× sinc (k2 + 2π/p2 (H − γ2 )) q = 1, . . . L.
4γ2
(4.49)

At x3 = 0, V0 = 0 and so
8L
X
A(r) φq,(r) = 0 q = 1, . . . L. (4.50)
r=1

Equations (4.43), (4.44), (4.49) and (4.50) constitute 8L equations in the 8L


unknowns A(r) . Hence this system of linear equations can be solved and the
displacements, stresses etc. can be examined using equation (4.35).
In the first instance the short circuit situation given by V0 = 0 (at x3 = h)
is examined to derive the dispersion curves showing the resonant modes of the
75

transducer. This gives rise to a system of equations in the form XA(r) = 0,where
X is an 8L×8L matrix. The modes are then determined by identifying where the
determinant of X is equal to zero. Unfortunately the matrix X is ill-conditioned
[38]. To help obviate this problem the matrix entries are balanced by scaling the
parameters of the model (see Table 4.1). Each of the parameters is made O(1)
by a judicious choice of the scalings α, β, γ and ϕ so that 5 equations in four
unknowns must be satisfied. This is done by scaling the thickness h by specifying
β, scaling the density ρ by specifying α, scaling the piezoelectric stress tensor
eijk by specifying ϕ, scaling the elasticity tensor cijkl by specifying γ and this
results in an appropriate scaling for the permittivity tensor ij .

Parameter Units Dimensions Scaling


cijkl N m−2 M L−1 T −2 αβ −1 γ −2
ij F m−1 C M −1 T 2 L−3
2 2 −1 2 −3
φα γ β
eijk Cm−2 CL−2 φβ −2
ρ kgm−1 M L−1 αβ −1
h m L β

Table 4.1: Dimensions and scaling parameters for the material properties.

4.4 The Admittance


The admittance expresses the ease with which an alternating current flows
through the transducer and the electrical resonant modes are signified by max-
ima in the real part of the admittance. This is a useful alternative method of
investigating the dispersive characteristics of a device. To derive an expression
for the admittance an alternative electrical boundary condition of continuity in
the electrical potential at the front face of the device is employed,

φsubstrate |x3 =h = φair |x3 =h . (4.51)


76

The electrical potential in air (φair ) satisfies Laplace’s equation

∇2 φair = 0 (4.52)

and, as there is no piezoelectric effect in air,

∂φair
D3air = −0 . (4.53)
∂x3

Equations (4.51) and (4.52) imply that at x3 = h,


substrate substrate substrate
∂2φ ∂2φ ∂2φ
=− 2 − 2 . (4.54)
∂x23 ∂x1 ∂x2

From equation (4.35)


L “ ” 8L
!
(k1 + 2π H q,1 )x1 +(k2 + 2π H q,1 )x2 (r)
X −
X
φsubstrate = eωt e p1 p2
A(r) e −k3 x3
φq,(r)
q=1 r=1
(4.55)
so that, at x3 = h, Equation (4.54) can be rewritten as
L 8L
!
(r)
−Gq ·r (r)
X X
e(ωt−k1 x1 −k2 x2 ) e −(k3 )2 A(r) e−k3 h φq,(r)
q=1 r=1

L
q ·r
X
(ωt−k1 x1 −k2 x2 )
= e e−G
q=1
8L   !
X 2π q,1 2 2π q,2 2 (r)
× (k1 + H ) + (k2 + H ) A(r) e−k3 h φq,(r) .
r=1
p1 p2
(4.56)

(r)
From equation (4.56) it can be seen that k3 = ±|κq | where
 2  2
q 2 2π 2π q,2
(κ ) = k1 + H q,1 + k2 + H . (4.57)
p1 p2
(r)
For bounded solutions in equation (4.35) it is required that −k3 < 0, that is

(r)
k3 = |κq |. (4.58)
77

Now, by using equations (4.51) and (4.58), equation (4.53) can be written as

∂φ substrate
D3air |x3 =h = −0 |x3 =h
∂x3
L 8L
!
X q
X (r)
(ωt−k1 x1 −k2 x2 )
= e e−G ·r 0 |κq |A(r) e −k3 h
φq,(r) .
q=1 r=1
(4.59)

The difference between the normal electrical displacement in the material and in
air is the charge distribution at the interface of the transducer. The admittance
is then given by the ratio of the current I to the voltage V , both averaged
over the transducer surface. Utilising the electrical periodicity of the device
architecture then gives


R ν42 R ν41
∂t −
ν2

ν1 D3substrate − D3air dx1 dx2
4 4
Y = ν2 R ν41 “ ” |x3 =h . (4.60)
R  ωt− 2π x1 − 2π x2
V0 4
ν
− 42 −
ν1 e ν 1 ν 2 dx1 dx2
4

Now
Z ν2 Z ν1 “ ”
4 4 2πγ 2πγ
 ωt− p 1 x1 − p 2 x2
V0 e 1 2 dx1 dx2
ν2 ν1
− 4
− 4
" 2πγ1 # 4γp1 " 2πγ2 # 4γp2
− x1 1 − x2 2
ωt e p1
e p2
= V0 e
− 2πγ
p1
1
p − 2πγ
p2
2
p
− 4γ1 − 4γ2
1 2
! !
− π2  π2 − π2  π2
e −e e −e
= V0 eωt
− 2πγ
p1
1
− 2πγ
p2
2

! !
ωt −2 −2
= V0 e
− 2πγ
p1
1
− 2πγ
p2
2

p1 p2
= V0 eωt . (4.61)
π 2 γ1 γ2
78

Using equations (4.59) and (4.35) the current can be written as,
L Z p2 Z p1 “ ”
4γ2 4γ1
− (k1 + 2π H q,1 )x1 +(k2 + 2π H q,1 )x2
X
ωt
I = ωe e p 1 p 2 dx1 dx2
p p
q=1 − 4γ2 − 4γ1
2 1
8L
!
(r)
 
q,(r)
X
× A(r) e −k3 h
D3 − 0 |κq |φq,(r) . (4.62)
r=1

Following similar lines to those in the derivation of equation (4.48) gives


L 
π2
    
X 2π q,1 p1 2π q,2 p2
Y = ω sinc (k1 + H ) sinc (k2 + H )
q=1
16V 0 p 1 4γ 1 p 2 4γ2
8L  
(r)
q,(r)
X
× A(r) e−k3 h
D3 − 0 |κq |φq,(r) . (4.63)
r=1

Hence the admittance is extended from the derivation of Wilm et al [74], to


include a dependency on the electrode spacing. To find the amplitudes A(r) the
system of equations (4.43), (4.44), (4.49) and (4.50) is solved but with V0 set to
a non-zero input voltage.

4.5 Results
The methodology presented in the previous section is utilised here to investi-
gate the modal behaviour of a 2-2 composite transducer (Section 4.5.1) and a
1-3 composite transducer (Section 4.5.2) both composed of PZT5H ceramic and
HY1300/CY1301 polymer. For piezoelectric transducers the resonant frequency
is dictated by the thickness of the transducer, therefore in NDT applications
(1 − 10 MHz) the thickness required are in hundreds of microns. The thick-
ness of the transducer is also central to the choice of scaling parameters and a
thickness of 200µm is chosen here to illustrate the operational characteristics of
the transducer. This scaling allows essentially any thickness to be adequately
simulated. The 2-2 topology is described by setting t = p2 and the wave number
k2 to zero.
79

4.5.1 Modal Analysis of Device A

Device A B
Composite type 2-2 1-3
Ceramic PZT5H PZT5H
Polymer HY1300/CY1301 Hardset HY1300/CY1301 Hardset
Thickness (mm) (h) 0.2 0.2
Kerf (mm)(p1 − s) 0.06 0.06
Saw pitch (mm) (p1 ) 0.2 0.2

Table 4.2: Geometry of the two composite piezoelectric transducers.

- Constant Units Value


elastic constant c11 Nm−2 12.72 × 1010
elastic constant c12 Nm−2 8.02 × 1010
elastic constant c13 Nm−2 8.47 × 1010
elastic constant c33 Nm−2 11.74 × 1010
dielectric constant 33 - 1.70 × 103
dielectric constant 11 - 1.47 × 103
Piezoelectric constant h V m−1 2.60 × 109
density ρb kg m−3 7.50 × 103
Piezoelectric stress coefficient e33 C m−2 23.30
Piezoelectric stress coefficient e31 C m−2 −6.50

Table 4.3: Physical properties of the ceramic phase PZT5H [19].

To obtain the dispersion curves the short circuit boundary conditions are
used (V0 = 0 in equation (4.49)) to give a system of equations of type XA(r) = 0.
The absolute value of the determinant of the matrix X is then calculated over
a selected range of frequencies and k1 values. Taking the logarithm of this
determinant, as shown in Figure 4.4, highlights the minima which correspond
to the modes where these boundary conditions are satisfied. It is the relative
magnitude of the determinant that is significant and it is important that this
surface is sufficiently smooth so that the ridges (modes) in Figure 4.4 can be
easily identified in an automated search algorithm. To identify the dispersive
80

Data Values Constant Hardset


Shear modulus (real part) G0 (kg m−1 s−2 ) 1.57 × 109
Young’s modulus (real part) E 0 (kg m−1 s−2 ) 4.28 × 109
Shear Velocity Vs (m s−1 ) 1.17 × 103
Longitudinal Velocity Vl (m s−1 ) 2.51 × 103
Density ρ ( kg m−3 ) 1.15 × 103
Dielectric constant  4
Elastic constant c11 7.19 × 109
Elastic constant c44 1.57 × 109
Frequency of measurement f0 (Hz) 5.00 × 105

Table 4.4: Physical properties of the polymer phase HY1300/CY1301 Hardset


[48].

behaviour, a loop is created to calculate the logarithm of the determinant of the


matrix X over a selected frequency range, for a selected range of k1 values. A
search is then performed and the k1 and ω that correspond to each minima are
stored. This procedure gives rise to the dispersion curves presented in Figure
4.5. The type of vibration represented by each dispersion curve in Figure 4.5
needs to be identified. Ideally the Rayleigh-Lamb modes, the lateral modes and
any bulk modes should be as far as possible from the thickness mode to prevent
any interference which could reduce the efficiency of the transducer and cause
poor beam formation.
81

ω
0 5 10
0

0.2
k 1 p1

0.4

PSfrag replacements

-70
-75
log |Det X| -80

Figure 4.4: The determinant of the matrix of coefficients arising from the short
circuit boundary conditions versus the driving frequency ω (MHz) and the nondi-
mensionalised parameter k1 p1 /2π for device A as described in Table 4.2 (see
Tables 4.3 and 4.4 for material properties).

To help identify the thickness mode, linear systems modelling is used to


calculate the frequency at which the device will resonate (see Chapter 2 ). From
the impedance characteristics in Figure 2.11, the device is expected to resonate
around 7 MHz. The bulk waves (see Figures 4.7, 4.8 and 4.9) and the first
anti-symmetrical Lamb wave (a0 ) (see Figures 4.10, 4.11 and 4.12) are at the
lower frequencies of Figure 4.5 and the isolated mode at around f = 6 MHz
is the thickness mode (see Figures 4.13, 4.14 and 4.15). The surface skimming
(SS) bulk mode, which is a transverse surface acoustic wave that propagates on
the surface of a piezoelectric crystal (see Figures 4.16, 4.17 and 4.18), normal
to the vertical plane [60], is also present in Figure 4.5. Due to the architecture
of piezoelectric composites, surface waves are generated between the adjacent
82

ω

10

Mode b
8

6 Thickness Mode

First Anti-Symmetrical
Lamb Mode
PSfrag replacements 4

Mode a
2

k 1 p1
0.1 0.2 0.3 0.4 0.5 2π

Figure 4.5: Dispersion characteristics of device A (Frequency (MHz) versus


nondimensionalised wavenumber).

pillars (intra-pillar modes) and within the pillars (inter-pillar modes). These
unwanted modes normally occur at a higher frequency than the thickness mode.
To help identify the symmetric and anti-symmetric Lamb modes the dispersion
relationship for a homogeneous, isotropic, non-piezoelectric and lossless material,
the Rayleigh-Lamb equation, can be used. It is defined as [29]

p p !
tan((ωh/2) (c2 − c22 )/c22 c2 ) 4 ((c/c1 )2 − 1)((c/c2 )2 − 1)
± = 0, (4.64)
(2 − (c/c2 )2 )2
p
tan((ωh/2) (c2 − c21 )/c21 c2 )

where c is the (unknown) phase velocity, ω is the frequency, c1 is the longitudinal


velocity, c2 is the shear velocity and h is the thickness. By solving equation (4.64)
it is found that the first anti-symmetrical Lamb mode is at a lower frequency
83

vp (kms−1 )
10

8 a1

s1
6 s0
PSfrag replacements
4

a0
0.5 1 1.5 2 2.5
f × h (kHzm)

Figure 4.6: Phase velocity (vp ) versus frequency (f ) × thickness (h) product for
device A using the PWE method (dotted lines) and the Rayleigh-Lamb equation
(solid lines).

than the thickness mode. These results are presented in Figure 4.6 where the
Rayleigh-Lamb equation is compared to the homogenised form of the PWE
method (N, M = 0). It is clear from Figure 4.6 that both methods predict the
first anti-symmetrical Lamb mode (a0 ) and the first symmetrical mode (s0 ) to
have the same form but the results of each method do not match quantitatively.
This is to be expected since the PWE method is for a piezoelectric, heterogeneous
and anisotropic material and the Rayleigh-Lamb equation is for an isotropic
material.
Since the lateral dimension of the ceramic pillar (s) is smaller than the thick-
ness (h) the intra-pillar mode will be at a higher frequency than the thickness
mode [29], but it can be confused with inter-pillar modes and other high fre-
quency modes. Hence a modal analysis involving the variations in the displace-
ment, electrical potential and Poynting vector is used to properly identify each
mode.
For example, the results for k1 = 500m−1 and f = 0.41 MHz are shown
in Figure 4.7. To classify the modes the relative magnitudes of the Poynting
84

P1 P3
x3 -6
50 100 150 200 8·10
0.95
-6
6·10
0.9
PSfrag replacements 4·10-6
PSfrag replacements 0.85
-6
0.8 P1 2·10
P3
x3
P3 0.75 50 100 150 200

(a) PSfrag replacements (b)


PSfrag replacements x3
x3 P1 u3
u1
P1 P3
x3
P3 50 100 150 200 P3 0.06
P3 0.8
0.05
0.6 u1 0.04
0.8 0.03
0.4
0.6 0.02
0.2 0.4 0.01
0.2
x3
50 100 150 200
PSfrag replacements
(c) x3 (d)
P1
P3 φ
P3 1

0.8
u1
0.8 0.6
0.6 0.4
0.4
0.2 0.2
u3
x3
50 100 150 200

(e)

Figure 4.7: Characteristics of mode a within device A (k1 = 500 m−1 and
f = 0.41 MHz). (a) Normalised Poynting vector P1 (W/m2 ), (b) normalised
Poynting vector P3 (W/m2 ), (c) displacement in the x1 direction u1 (µm), (d)
displacement in the x3 direction u3 (µm) and normalised electrical potential φ
(V) against x3 (µm).

vector and displacement in each direction is important, therefore each mode has
been normalised independently. The plots are all nearly symmetric about the
midplane in the thickness direction x3 . In plot (a), the Poynting vector P1 has
maxima at x3 = 24.5µm and 175µm with thickness h = 200µm. By investigating
the displacements it is found that u1 in plot (c) has its maximum displacement
in the midplane of the transducer and is significantly larger than u3 , which
85

x3

2
1.5
PSfrag replacements 1
0.5

-4 4
x1
P3 -2 2

(a)

x3

2
1.5
PSfrag replacements 1
0.5

-4 4
x1
P3 -2 2

(b)

Figure 4.8: The real part of the in-plane displacement of mode a within device
A (k1 = 500 m−1 and f = 0.41 MHz) in the x3 -x1 plane x2 = 0 where (a) is at
time t0 and (b) is at time t0 plus half the period. The thick regions bounded by
the black lines represent the ceramic pillars and the thin regions represent the
polymer.

has a maximum displacement of 0.0567µm at both ends of the transducer (plot


(d)). The plots of P3 and P1 indicate that the wave energy is mainly at the
top and bottom face with the component parallel to these transducer surfaces
dominant. By investigating the displacement in the x3 -x1 plane x2 = 0 in
Figure 4.8 the displacement is dominant in the x1 direction and there is very
little movement in the thickness direction. The x1 axis is in arbitrary units
and the amplitude of the oscillations has been exaggerated in both directions so
that the ’particle’ motion can be seen. Figure 4.9 illustrates a material particle
charting its evolution over time. Each dot is plotted at regular time intervals.
The displacements have been exaggerated by equal scaling in both directions
86

x3

1.5

0.5

PSfrag replacements 0
x1
-2 -1 1 2

(a)
x3

1.5

0.5

PSfrag replacements 0
x1
0.25 0.5 0.75 1 1.25 1.5 1.75

(b)

Figure 4.9: Displacement (real part) (u) of a bulk wave within device A (k1 = 500
m−1 and f = 0.41 MHz) during excitation. Plot (a) illustrates the movement
of specific points within the ceramic as time is varied and (b) illustrates the
movement of specific points over time within the polymer (x1 is in arbitrary
units).
87

in order to clarify the particle’s motion. The general motion of the particle is
elliptical, with the particle periodically returning to its original position. The
curved arrows indicate whether the motion is clockwise or anti-clockwise. It is
also clear in Figure 4.9 that the displacement is dominant in the x1 direction,
which is indicative of this being a bulk longitudinal wave or the first symmetrical
Lamb wave.
P1 P3
1
0.2
0.8
0.15
0.6
0.1
PSfrag replacements 0.4 PSfrag replacements
PSfrag replacements 0.05
0.2
x3
P 1
x3 P1 x3
P3 50 100 150 200 50 100 150 200
P3
(a) (b)
u1
u1 u3
1 50
0.005
PSfrag replacements 200
0.8
x3 0.004
150
P1 0.6 100
PSfrag replacements 0.003
P3 50
0.4 x3
P1 0.002
0.2
u3 P3
x3 x3
50 100 150 200 50 100 150 200
u1
(c) u3 (d)
50
200 φ
150 1
100 0.8
50
0.005 0.6
0.004 0.4
0.003
0.002 0.2
x3
50 100 150 200

(e)

Figure 4.10: The first anti-symmetrical Lamb mode for device A (k1 = 5500 m−1
and f = 1.93 MHz). (a) Normalised Poynting vector P1 (W/m2 ), (b) normalised
Poynting vector P3 (W/m2 ), (c) displacement u1 (µm), (d) displacement u3 (µm)
and normalised electrical potential φ (V) against x3 (µm).

To find the first anti-symmetrical Lamb wave, equation (4.64) is used to give
88

x3
2

1.5

PSfrag replacements 1

0.5
x1
P3 -4 -2 2 4
(a)

x3
2
1.5
PSfrag replacements 1
0.5
x1
P3 -4 -2 2 4
(b)

Figure 4.11: The real part of the in-plane displacement of the first anti-
symmetrical Lamb mode for device A (k1 = 5500 m−1 and f = 1.93 MHz)
in the x3 -x1 plane x2 = 0 where (a) at time t0 and (b) at time t0 plus half
the period. The thick regions bounded by the black lines represent the ceramic
pillars and the thin regions represent the polymer.

a qualitative form for the mode and the PWE method is then used to pinpoint
it more precisely. The point k1 = 5500m−1 and f = 1.93 MHz is investigated.
Looking at Figure 4.10 the Poynting vector in the x1 direction dominates. Most
of the energy is concentrated at the transducer faces, although this persists
for at least a couple of wavelengths into the interior of the transducer. The
displacement is also predominantly in the x1 direction with any movement re-
stricted to two wavelengths from the transducer faces. Figure 4.11 illustrates
that the wave is anti-symmetric and Figure 4.12 shows elliptical motion in the
anti-clockwise direction at the bottom face, elliptical motion in the clockwise
direction at the top face but only motion in the x3 direction in the middle of
89

x3

1.5

0.5

PSfrag replacements 0
x1
-0.2 -0.1
5 -0.1 -0.0
5 0.05 0.1 0.15

(a)
x3

1.5

0.5

PSfrag replacements
0
x1
0.9 0.95 1 1.05 1.1

(b)

Figure 4.12: Displacement (real part) (u) of the first anti-symmetrical Lamb
mode for device A (k1 = 5500 m−1 and f = 1.93 MHz) during excitation. Plot
(a) illustrates the movement of specific points within the ceramic as time is
varied and (b) illustrates the movement of specific points over time within the
polymer (x1 is in arbitrary units).
90

the transducer for both materials. This mode clearly has most of its energy and
displacement at the faces of the transducer therefore it can be categorised as the
first anti-symmetrical Lamb mode.
P1 P3
1
0.00012
0.0001 0.8
0.00008 0.6
0.00006
PSfrag replacements 0.00004 PSfrag replacements 0.4
0.00002 0.2
x3 P1
50 100 150 200 x3
P3 50 100 150 200

(a) (b)

u1 u3
0.0006 1

0.0005
PSfrag replacements PSfrag replacements 0.8
0.0004
x3 x3 0.6
P1 0.0003 P1
0.4
P3 0.0002 P3
0.0001 0.2
x3 u1
50 100 150 200 x3
u3 500 100 150 200

(c) (d)

φ
1
PSfrag replacements
0.8
x3
P1 0.6
P3
0.4

u1 0.2
u3
x3
50 100 150 200

(e)

Figure 4.13: The thickness mode for device A (k1 = 500 (m−1 ) and f = 6.84
MHz). (a) Normalised Poynting vector P1 (W/m2 ), (b) normalised Poynting
vector P3 (W/m2 ), (c) displacement u1 (µm), (d) displacement u3 (µm) and
normalised electrical potential φ (V) against x3 (µm).

The thickness mode at k1 = 500m−1 and f = 6.84 MHz shown in Figure 4.13
shows that all the displacements, energy and electrical potential are symmetric
in the thickness direction. The Poynting vector in the x1 direction has very little
energy compared to P3 which has energy distributed throughout the transducer,
91

x3

2
1.5
PSfrag replacements 1
0.5
x1
P3 -4 -3 -2 -1 1 2 3 4

(a)

x3

2
1.5
PSfrag replacements 1
0.5
x1
P3 -4 -3 -2 -1 1 2 3 4

(b)

Figure 4.14: The real part of the in-plane displacement of the thickness mode
for device A (k1 = 500 (m−1 ) and f = 6.84 MHz) in the x3 -x1 plane x2 = 0
where (a) is at time t0 and (b) is at time t0 plus half the period. The thick
regions bounded by the black lines represent the ceramic pillars and the thin
regions represent the polymer.

and u1 is negligible compared to u3 . By investigating the displacement in the


x3 -x1 plane x2 = 0 in Figure 4.14 it can be seen that the transducer is moving
up and down with very little motion in the x1 direction. Figure 4.15 verifies this
and also shows that both the ceramic and polymer are moving in a clockwise
motion at the top and middle of the transducer and are moving anti-clockwise
at the bottom.
At the point k1 = 1.4 × 104 m−1 and f = 11.81 MHz (see Figure 4.16) the
Poynting vector in the x3 direction dominates. Most of the energy is concen-
trated at the transducer faces, although this only persists for a small distance
into the interior of the transducer. The displacement is predominantly in the x1
92

x3

1.5

0.5

PSfrag replacements 0
x1
-0.0
5 -0.0
4 -0.0
3 -0.0
2 -0.0
1 0.01

(a)
x3

1.5

0.5

PSfrag replacements
0
x1
0.94 0.96 0.98 1 1.02

(b)

Figure 4.15: Displacement (real part) (u) of the thickness mode for device A
(k1 = 500 (m−1 ) and f = 6.84 MHz) during excitation. Plot (a) illustrates
the movement of specific points within the ceramic as time is varied and (b)
illustrates the movement of specific points over time within the polymer (x1 is
in arbitrary units).
93

P1 P3
0.6 1
0.5 0.8
0.4
0.6
0.3
0.2 PSfrag replacements 0.4
PSfrag replacements 0.1 0.2
P1
x3 x3
50 100 150 200 50 100 150 200

(a) (b)

u1 u3
30
1.2
25
PSfrag replacements PSfrag replacements 1
20
x3 x3 0.8
P1 15 P1 0.6
P3 10 P3 0.4
5 0.2
u1
x3 x3
u3 50 100 150 200 50 100 150 200

(c) (d)

φ
1
PSfrag replacements
0.8
x3
P1 0.6
P3
0.4

u1 0.2
u3
x3
50 100 150 200

(e)

Figure 4.16: Characteristics of mode b for device A (k1 = 1.4 × 104 m−1 and
f = 11.81 MHz). (a) Normalised Poynting vector P1 (W/m2 ), (b) normalised
Poynting vector P3 (W/m2 ), (c) displacement u1 (µm), (d) displacement u3 (µm)
and normalised electrical potential φ (V) against x3 (µm).

direction and distributed evenly throughout the transducer. The in-plane com-
ponents of the displacement in Figure 4.17 show the ceramic being stretched
and squashed at opposite sides of the transducer and being mainly stationary
within the transducer. Figure 4.18 shows motion in the anti-clockwise direction
at the top face, motion in the clockwise direction at the bottom face and less
motion in the anti-clockwise direction near the middle of the transducer for both
materials. This mode clearly has most of its energy at the faces of the trans-
94

ducer and therefore it could be identified as a surface skimming bulk wave (or
Bleustein-Gulyaev wave) or a Rayleigh wave.

x3
2
1.5
PSfrag replacements 1
0.5
x1
P3 -4 -2 2 4
(a)

x3
2

1.5

PSfrag replacements 1

0.5
x1
P3 -4 -2 2 4
(b)

Figure 4.17: The real part of the in-plane displacement of mode b for device A
(k1 = 1.4 × 104 m−1 and f = 11.81 MHz) in the x3 -x1 plane x2 = 0 where (a) is
at time t0 and (b) is at time t0 plus half the period. The thick regions bounded
by the black lines represent the ceramic pillars and the thin regions represent
the polymer.
95

Resonant frequencies are also given by the maxima of the electrical con-
ductance. By examining the conductance plot in Figure 4.19, the thickness
mode can be identified as the central ridge of the plot at around 6.5 MHz. The
lower frequency maxima correspond to the first symmetrical Lamb wave passing
through the path of the bulk waves whilst the intra-pillar mode is the first set
of peaks to the right of the thickness mode at around 10 MHz. Figure 4.19
is in good agreement with the modes identified in Figure 4.5. The advantages
of plotting the conductance is that the relative ampiltude of each mode can
clearly be seen and any spurious points found in the dispersion diagram can be
eradicated. There is a reasonable band gap around the thickness mode which
indicates that the transducer should perform efficiently. Figure 4.20 shows the
surface displacements for the thickness mode over three periods of the structure.
The displacement in the x3 direction dominates with the ceramic and passive
phase moving in phase, although the profile is not ideally flat.

0.1

0.2 k 1 p1

0.3

0.4
2
1.5
PSfrag replacements
1 Y
0.5
0
0 2.5 5 7.5 10
f

Figure 4.19: The conductance of the transducer Y (normalised) plotted against


the nondimensionalised wavenumber k1 p1 /2π (expressed in fractions of the wave-
length) and the driving frequency ω (MHz) for device A.
96

x3

1.5

0.5

PSfrag replacements 0
x1
-0.0
6 -0.0
4 -0.0
2 0.02

(a)
x3

1.5

0.5

PSfrag replacements
0
x1
0.975 0.98 0.985 0.99

(b)

Figure 4.18: Displacement (real part) (u) of mode b for device A (k1 = 1.4 × 104
m−1 and f = 11.81 MHz) during excitation. Plot (a) illustrates the movement
of specific points within the ceramic as time is varied and (b) illustrates the
movement of specific points over time within the polymer (x1 is in arbitrary
units).
97

u1 u3
0.4 9

0.3 7
PSfrag replacements
0.2 5
PSfrag replacements
0.1
3
u1
x1 x1
-100 100 200 300 400 500 -100 100 200 300 400 500

(a) (b)

Figure 4.20: The displacements at the surface x3 = h for the thickness mode
for device A (k1 = 500 m−1 and f = 6.84 MHz). (a) in-plane displacement u1
versus x1 at x2 = 0, (b) out-of-plane displacement u3 versus x1 at x2 = 0. The
thick regions bounded by the dashed lines represent the ceramic pillars and the
thin regions represent the polymer.

4.5.2 Modal Analysis of Device B

The modal behaviour for a 1-3 composite transducer using PZT5H ceramic and
HY1300/CY1301 polymer can also be examined. The corresponding determi-
nant plot is shown in Figure 4.21. Additional modes will arise in the 1-3 compos-
ite since there is now cross-coupling in both the x1 and x2 direction (see Figure
4.22).
98

ω
0 5 10
0

0.2
k 1 p1

0.4

PSfrag replacements

20
10
0 log|Det X|

Figure 4.21: The determinant of the matrix of coefficients arising from the
short circuit boundary conditions versus the driving frequency ω (MHz) and
thenondimensionalised wavenumber k1 p1 /2π for device B as described in Table
4.2 (see Tables 4.3 and 4.4 for material properties).

By examining the conductance plot in Figure 4.23, the thickness mode can be
identified as the central ridge of the plot at around 6.5 MHz. The lower frequency
maxima (the low amplitude peaks to the left of the main ridge) correspond to
a bulk wave and the first symmetrical Lamb wave passing through the path of
the thickness mode whilst the inter-pillar mode is the first set of peaks to the
right of the thickness mode at around 9 MHz. Figure 4.23 is in good agreement
with the modes identified in Figure 4.22.
99

ω

10

PSfrag replacements

k 1 p1
γ1 = 2π
0.1 0.2 0.3 0.4 0.5

Figure 4.22: Dispersion characteristics of device B.

Figure 4.24 shows the surface displacements for the thickness mode over three
periods of the structure of a 1-3 composite transducer. The displacement in the
x3 direction dominates with the ceramic and passive phase moving in phase,
although again the surface is somewhat uneven.
100

0.1

0.2 k 1 p1

0.3

0.4
2
1.5
1 Y
0.5
PSfrag replacements
0
0 2.5 5 7.5 10
ω

Figure 4.23: The conductance of the transducer Y (normalised) plotted against


the nondimensionalised wavenumber k1 p1 /2π (expressed in fractions of the wave-
length) and the driving frequency ω (MHz) for device B.

u1 u2

0.06 0.02

0.015
0.04 PSfrag replacements
0.01
PSfrag replacements
0.02
0.005
u1
x1 x1
-100 100 200 300 400 500 -100 100 200 300 400 500

(a) (b)

u3
1

PSfrag replacements 0.8

0.6
u1
u2 0.4
x1
-100 100 200 300 400 500

(c)

Figure 4.24: The displacements of the surface x3 = h for the thickness mode
for device B (k1 = 500 m−1 and f = 6.99 MHz). (a) in-plane displacement u1
versus x1 at x2 = 0, (b) in-plane displacement u2 versus x1 at x2 = 0,(c) out-
of-plane displacement u3 versus x1 . The thick regions bounded by the dashed
lines represent the ceramic pillars and the thin regions represent the polymer.
101

4.6 Conclusions
The plane wave expansion method has been derived in this Chapter and extended
to provide a more complete analysis of the supported waves. Results have been
presented using this model for both 2-2 and 1-3 composite transducers made from
PZT5H ceramic and the standard hardset material HY1300/CY1301. Disper-
sion curves were produced and particular modes on these curves were identified
using the Poynting vector, electrical potential and the displacement. Admit-
tance curves were also produced which helped to clarify the important modes.
Ill-conditioning is present in the matrix inversion procedure used to find the
amplitudes of each mode, particularly for 1-3 configurations. Regularisation
techniques such as Tikhonov and truncated singular value decomposition can be
used to address this issue and will be discussed in the next chapter. Validation
of the model will be shown by comparison to finite element modelling in the
next chapter and the model will also be extended to incorporate elastic losses.
Chapter 5

Incorporation of Elastic Loss


into the Plane Wave Expansion
Method

5.1 Introduction
In this chapter, frequency dependent viscoelastic loss is incorporated into the
three-dimensional plane wave expansion model (PWE). The dispersive behaviour
of a 2-2 composite transducer will be examined and the effects of introducing
high shear attenuation into the passive phase will be investigated. The aim
is to increase the frequency band gap between the thickness mode and other
parasitic waves to obtain a higher transmission bandwidth. Scaling and regular-
isation techniques are introduced to reduce ill-conditioning in the large matrices
which arise in the PWE method. The identification of the modes is aided by
examining profiles of the displacements, electrical potential and Poynting vector
[49, 50, 51].
In Chapter 4 the PWE method was used to investigate dispersion relation-
ships for both 1-3 and 2-2 composite transducers but this analysis did not in-
clude any loss mechanisms. Loss is vital for reducing crosstalk between adjacent
ceramic pillars in the transducer. In addition the PWE method cannot be com-
pared to experimental results until loss is incorporated into the model. An

102
103

alternative approach is to use finite element modelling (FEM)[29]. Such models


can incorporate realistic operating conditions, such as viscoelastic loss, backing
and matching layers, mechanical and electrical loads and electrode patterning.
The major drawback is the high computational cost. FE modelling can be used
in the frequency domain [6] but for the modelling of periodic composite trans-
ducers the method is often used in the time domain [58]. In the time domain
approach the harmonic analysis requires a numerical Fourier transform of the
predicted displacements, which can be problematic if the mode amplitudes have
a large variance. Due to the computational cost, a quarter symmetry of a unit
cell with appropriate periodic boundary conditions is often employed. For reg-
ular geometries this does not present a problem but essentially excludes the
study of irregular designs in three dimensions. FE modelling has recently shown
that Lamb wave propagation is responsible for inter-element crosstalk and has a
detrimental effect on beam forming and transducer sensitivity [29]. It has been
suggested that a passive filler with high shear loss may aid the damping of these
unwanted Lamb waves [47]. The method is extended here to include mechani-
cal loss in both the piezoelectric and polymer material, associated scaling and
regularisation techniques, a comparison of the PWE method with elastic loss to
that with no loss mechanisms and the LSM model, mode identification for a 2-2
composite using profiles of the displacements and Poynting vectors, the effects
of using a high shear loss passive phase and a comparison to FEM modeling and
experimental results. Note that dielectric and other loss mechanisms are not
included here.

5.2 Frequency Dependent, Elastic Loss Model


As discussed in Chapter 3, the degree of mechanical loss is usually expressed
in terms of a dimensionless loss tangent tan δ [40], or an attenuation coefficient
α [35]. There are lots of different mechanisms and models [11] and this is the
104

model we have chosen. Frequency dependent loss can be introduced into the
ceramic via  
ω
tan δ = tan δ(ωI ), (5.1)
ωI
where ω is the angular frequency, tan δ(ωI ) is the measured loss tangent and ωI
is the angular frequency at which tan δ is measured. Equation (5.1) is used to
obtain the imaginary parts of the elastic constants within the ceramic at a given
frequency using
cijkl = cijkl (1 +  tan δ). (5.2)

If the polymer phase is assumed to be isotropic then the elastic constants are
expressed in terms of the complex Lamé coefficients µ and λ, defined as

µ = µ0 + µ00 and λ = λ0 + λ00 (5.3)

so that the complex velocities are given by [38]


s s
µ + µ
0 00 (λ0 + 2µ0 ) + (λ00 + 2µ00 )
vs = , vl = , (5.4)
ρ ρ

where vs is the shear velocity, vl is the longitudinal velocity and ρ is the density.
The loss is however usually expressed for these materials in terms of a longitu-
dinal attenuation coefficient (αl ) and a shear attenuation coefficient (αs ). These
can be related to tan δ via equation (3.11),

2αs |vs | 2αl |vl |


tan δs = and tan δl = . (5.5)
ω ω

The frequency dependency can be included via the attenuation coefficients using
0
αl/s = αl,s (ω/ω0 )2 , where αl/s
0
is the experimentally measured attenuation coef-
ficient at frequency ω0 . Using Equations (5.4) and (5.5), the imaginary parts of
the Lamé coefficients are then given by
v s !
u
u 8α4 µ04 ρ 2ω4
µ00 = t 4s 4 1 + 1 + 4 04 (5.6)
ω ρ 4αs µ
105

and v
u s !
u 8α4 (λ0 + 2µ0 )4 ρ2 ω 4
λ00 = t l 4 4 1+ 1+ 4 0 − 2µ00 . (5.7)
ω ρ 4αl (λ + 2µ0 )4
0
So given αl,s , ω0 , λ0 , µ0 , ρ and ω, equations (5.6) and (5.7) can be used to give
the complex valued elasticity tensor in equation (4.14) and associated Fourier
coefficient in equation (4.18). The degree of shear loss can be varied by mul-
tiplying the shear attenuation coefficient αs by a parameter ζ. The change in
loss is then incorporated into the model through the imaginary parts of the
Lamé coefficients. The effect that changing from low shear attenuation to high
shear attenuation has on the frequency band gap surrounding the fundamental
thickness mode of the transducer will be investigated in section 5.4.2.
To derive the anisotropic coefficients for an SBS passive phase, the material
properties of polybutadiene (superscript pb) and polystyrene (superscript ps)
are mixed together using ’laminate’ mixing rules [68]. The coefficient c11 is
calculated from the mixing rule

cps pb
11 c11
c11 = , (5.8)
φcps pb
11 + (1 − φ)c11

which is also used to calculate c13 , c44 and 33 . The coefficient c12 is calculated
using
c12 = φcpb ps
12 + (1 − φ)c12 (5.9)

which is also used to calculate 11 , 22 and ρ.

5.3 Implementation
From section 4.3 the vibrational modes of the transducer must satisfy the system
of equations XA(r) = Q, given by equations (4.43), (4.44), (4.49) and (4.50),
where X(k, ω) is an 8L × 8L matrix and Q(k) is a column vector of length 8L
where L = (2N +1)(2M +1). As the number of Fourier coefficients are increased
within the PWE method, the dimensions of X increase and the determinant
106

becomes too large to be stored. This problem occurs due to the exponential
terms which arise when calculating the boundary conditions at x3 = h. This
term affects the rows between 1 and 3(2N + 1) and 6(2N + 1) + 1 and 7(2N + 1)
in the matrix X. To prevent the problem occuring the row entries are multiplied
00 (r) )h
by a scale factor given by eMaxr (k3 , where k300 (r) is the imaginary part of the
(r)
wavenumber k3 . There are also numerical instabilities as the determinant of
X approaches zero due to ill-conditioning, and Tikhonov regularisation is used
here to circumvent this [41]. To implement this the matrix X is converted to a
real, symmetric form by multiplying it by its complex, conjugate X ∗ . The zero
eigenvalues of X are then translated along the real axis, away from the origin,
by adding a small amount, µ, to give

(X ∗ X + µI)A(r) = X ∗ Q. (5.10)

The determinant of X ∗ X + µI is still large and so minima in the cost function


surface log |X ∗ X + µI| are found, parameterised by the angular frequency, ω,
and complex wavenumbers k1 and k2 . This determinant calculation is performed
by finding the eigenvalues λi and using det(X ∗ X + µI) = Li=1 λi , which can be
Q

rewritten as log(det(X ∗ X + µI)) = Li=1 log λi .


P
107

log |X ∗ X + µI|

140
120
100
PSfrag replacements 80
60 (c)

40 (a)

20
(b)
f (MHz)
0.5 1 1.5
Figure 5.1: Determinant of (a) the coefficient matrix X arising from the modal
equations, (b) the real symmetric matrix X ∗ X and (c) the Tikhonov regularised
matrix X ∗ X + µI with µ = 2 × 10−3 versus frequency f (MHz) for device C (See
Tables 2.1, 3.1 and 5.1 for material properties).

The effects of Tikhonov regularisation are shown for a 2-2 composite in Fig-
ure 5.1 for N = 3, M = 0. Ill-conditioning in the determinant calculation is
indicated by the noise in plot (a) which corresponds to no regularisation being
used. A major improvement can already be seen by making the matrix real and
symmetric in plot (b). However the introduction of the small parameter µ re-
sults in the smooth curve in plot (c). Truncated Singular Value Decomposition
[37] was also investigated but proved to be not as robust as Tikhonov regular-
isation in this instance. The algorithm for obtaining a particular mode for the
2-2 composite design sets k1 to a real number (initially) and then searches the
cost function surface in the frequency direction until a number of local minima
are found. These interim minima are used as the initial values for a search in
the direction of the imaginary part of k1 , although here the algorithm stops at
the first local minimum. This orthogonal stepping procedure is then performed
for a range of k1 values.
108

σ6
log |X ∗ X + µI|
σ4
-20

σ1
-25

PSfrag replacements -30

-35

f (MHz)
0.25 0.5 0.75 1 1.25 1.5

Figure 5.2: Determinant of the coefficient matrix arising from the modal equa-
tions versus frequency f (MHz) for device C.

As can be seen in Figure 5.2 the surface given by log |X ∗ X = µI| has many
local minima. Some of these are very localised and do not represent actual
modes but are an artefact of the numerical computations. To help identify such
modes, the distance σi between adjacent maxima which straddle each minima
(see Figure 5.2) is calculated. Each point can then be shaded in the dispersion
curves according to the size of σ, or alternatively the magnitude of the imaginary
part of k1 can also be used (see Section 5.4.2).

5.4 Results
The effects of incorporating loss into the PWE method are investigated here. Pri-
marily, a comparison is made to the LSM model and the PWE method without
loss for a 2-2 composite (Section 5.4.1) Admittance plots and a modal analysis
using displacement and Poynting vector profiles are used to discuss the operat-
109

ing characteristics of this device. In Section 5.4.2, a composite transducer with a


high shear loss passive phase is examined by investigating dispersion curves and
the electrical impedance. In Section 5.5 a comparison between the PWE method,
FE modelling and experimentally measured behaviour is reported. Finally, in
section 5.6, transducers incorporating the SBS filler materials are investigated.

5.4.1 Modal Analysis of Device C

Device A B C
Composite type 2-2 1-3 2-2
Ceramic PZT5H PZT5H PZT5H
Thickness (mm) (h) 3.83 2.4 2
Kerf (mm) 0.23 0.2 0.6
Saw pitch (mm) (p) 0.69 0.49 2

Table 5.1: Geometry of the three composite piezoelectric transducers.

The introduction of loss into the PWE method serves to smooth out the
frequency domain response of the transducer and to lessen the effects of ill-
conditioning. Plot (a) in Figure 5.3 highlights the effect of introducing loss
log |ZT | log |ZT |
10
10
8
8

6 6

4 4

PSfrag replacements 2 PSfrag replacements 2


f f
0.2 0.4 0.6 0.8 1 1.2 1.4 0.2 0.4 0.6 0.8 1 1.2 1.4

(a) (b)

Figure 5.3: The logarithm of the absolute value of the electrical impedance ZT
for device C plotted against the driving frequency f (MHz). In Plot (a), the solid
and dashed lines represent the PWE method with and without loss respectively.
In Plot (b), the solid and dashed lines represent the PWE and LSM methods
with loss respectively.

into the PWE method. Both the loss and no loss cases have similar signatures
110

at low frequencies but, as the loss becomes more pronounced, the higher fre-
quency oscillations are damped. The PWE method is then compared to the
LSM method in plot (b) of Figure 5.3. The LSM method predicts the thick-
ness mode response to be in the same region as the PWE method but, since
the LSM method is one-dimensional, no additional Lamb modes or inter/intra
pillar modes are present. Hence the LSM method is a useful tool which can
be used to quickly find where the thickness mode response will occur within
a new transducer, before beginning an extensive investigation using the PWE
method. Extending this analysis to a range of electrical excitation wavenumbers
0 0

k1 p10.2 k1 p10.2
2π 2π

0.4 0.4

PSfrag replacements PSfrag


5 replacements 5
4 4
3 3
2 2
1 1
0 0.5 1 1.5 0 log |ZT | 0 0.5 1 1.5 0 log |ZT |
f f

(a) (b)

Figure 5.4: The logarithm of the absolute value of the electrical impedance ZT
for device C with (a) no loss and (b) loss, plotted against the nondimensionalised
wavenumber k1 p1 /2π (expressed in fractions of the wavelength) and the driving
frequency f (MHz).

the effects of incorporating loss on the electrical impedance profiles is shown in


Figure 5.4, where it is clear that the loss attenuates the higher frequency modes
and reduces ill-conditioning. Both methods predict the mechanical resonant fre-
quency (fm ) and the electrical resonant frequency fe to be around 0.8 MHz and
0.7 MHz respectively. The lower frequency maxima correspond to Lamb waves
whilst the inter/intra-pillar modes are the first set of peaks to the right of the
thickness mode at around 1MHz.
111

0.1

0.2
k 1 p1

0.3

0.4

0.5
PSfrag replacements 0.4
0.3
0.2
0.1
0 0.5 1 1.5 0 Y
f

Figure 5.5: The conductance Y for device C (normalised) with loss, plotted
against the nondimensionalised wavenumber k1 p1 /2π (expressed in fractions of
the wavelength) and the driving frequency f (MHz).

Equivalently the conductance can be plotted as a function of the non-dimensionalised


electrical excitation wavenumber k1 p1 /2π as shown in Figure 5.5, where the
thickness mode can be identified as the central ridge of the plot at around 0.7
MHz. The advantages of plotting the conductance is that it is clear to see the rel-
ative importance of each mode and in this way it eradicates any spurious modes
that can be found in the dispersion diagrams. Having identified these modes,
their classification can now be performed by utilising spatial and/or temporal
plots of the displacement, the Poynting vector and the electrical potential. The
next few subsections show these plots for the thickness mode (section 5.4.1.1),
the first antisymmetric Lamb mode (section 5.4.1.2), a Rayleigh mode (section
5.4.1.3) and an intrapillar mode (section 5.4.1.4).
112

5.4.1.1 Thickness Mode

|u1 | |u3 |
1
0.05
0.8
0.04
0.6
0.03 PSfrag replacements
0.4
PSfrag replacements 0.02
0.01 0.2
|u1 |
-1 1
x1 x1
-3 -2 2 3 -3 -2 -1 1 2 3

(a) (b)
PSfrag replacements |u3 |
|u1 | 1
PSfrag replacements x1
0.0001 0.8
x1 |u1 |
0.00008
0.6
|u1 | |u3 |
0.00006
0.4
|u3 | 0.00004
0.2
x3 |u1 |
0.5 1 1.5 2
1 1.5
x3
PSfrag replacements 0.5 2

x(c)
1 (d)
|u1 |
x3
|u3 | 2.5
2
x3
1.5
|u1 |
1
|u3 |
PSfrag replacements 0.5
x1 x1
-4 -2 2 4
|u1 | -0.5

|u3 | (e)
x3
x3
|u1 | 2.5

|u3 | 2
1.5
x1
1
x3
0.5
x1
-4 -2 2 4
-0.5

(f)

Figure 5.6: Normalised displacement (ui ) of device C (k1 = 1500 +  m−1 and
f = 0.65 MHz) using the PWE method incorporating loss at the thickness mode.
Plots (a) and (b) show the components of the displacement on the transducer
surface x3 = 0, as x1 is varied , (c) and (d) show the components of the dis-
placement as x3 is varied (at x1 = 0) and (e) and (f) are the real part of the
displacement in the x3 -x1 plane x2 = 0. Plot (e) is at time t0 and (f) is at time
t0 plus half the period (the displacements have been scaled to accentuate the
motion).
113

x3

1.5

0.5

PSfrag replacements
0
x1
-0.05 -0.04 -0.03 -0.02 -0.01

(a)
x3
2

1.5

0.5
PSfrag replacements

0
x1
0.8 0.9 1 1.1 1.2 1.3

(b)

Figure 5.7: Displacement (real part) (u) of device C (k1 = 1500 +  m−1 and
f = 0.6478 MHz) using the PWE method incorporating loss at the thickness
mode during excitation. Plot (a) illustrates the movement of specific points
within the ceramic as time is varied and (b) illustrates the movement of specific
points over time within the polymer (the displacements have been scaled to
accentuate the motion and the scaling in the x1 direction is not identical in each
diagram).
114

By analysing the displacements in Figure 5.6 it is found that u1 has negligible


displacement compared to u3 , which has a very large displacement at the faces
of the transducer and within the ceramic. The thicker lines in plots (e) and
(f) indicate the boundaries between the alternating ceramic and polymer phases
(the ceramic is located at x1 = 0). Figure 5.7 shows that the ceramic is moving
up and down with very little motion in the x1 direction, however the polymer
is being pulled sideways with no motion in the x3 direction. To examine the
real part of the Poynting vector in the x3 -x1 plane x2 = 0, the Poynting vector
is scaled to make it the same order of magnitude as the x1 and x3 components
(see Figure 5.8). It is clear to see that the energy is distributed throughout the
transducer in the thickness direction. This mode is classified as the thickness
mode due to its symmetrical displacement profile in both directions, the large
amplitude of oscillation and the dominant displacement being in the x3 direction.
115

|P1 | |P3 |
0.7
0.6
0.015
0.5

0.01 PSfrag replacements 0.4


0.3
PSfrag replacements
0.005 0.2
|P1 | 0.1
x1 x1
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

(a) (b)

|P1 | PSfrag replacements |P3 |


1
PSfrag replacements x1
0.015 0.8
x1 |P1 |
0.6
|P1 | 0.01 |P3 |
0.4
|P3 |
0.005
0.2
|P1 |
x3 x3
0.5 1 1.5 2 0.5 1 1.5 2
PSfrag replacements
(c) x (d)
1

|P1 | x3
2.5
|P3 |
2
x3 1.5
|P1 | 1
|P3 | 0.5
PSfrag replacements
x1
x1 -4 -2 2 4
-0.5
|P1 |
|P3 | (e)
x3 x3
2.5
|P1 |
2
|P3 |
1.5
x1 1
x3 0.5
x1
-4 -2 2 4
-0.5

(f)

Figure 5.8: Normalised Poynting vector (Pj ) of device C (k1 = 1500 +  m−1 and
f = 0.6478 MHz) using the PWE method incorporating loss at the thickness
mode. Plots (a) and (b) show the components of the Poynting vector on the
transducer surface x3 = 0, as x1 is varied, (c) and (d) show the components of
the Poynting vector as x3 is varied (at x1 = 0) and (e) and (f) are the real part
of the in-plane Poynting vector in the x3 -x1 plane x2 = 0. Plot (e) is at time t0
and (f) is at time t0 plus half the period (the Poynting vector has been scaled
to accentuate the motion).
116

5.4.1.2 The first antisymmetrical Mode

|u1 | |u3 |
x1
-3 -2 -1 1 2 3
0.07
0.8
0.06
0.05 PSfrag replacements 0.6
PSfrag replacements 0.04
0.4
0.03 |u1 |
x1 0.2
-3 -2 -1 1 2 3

(a) (b)

|u1 | PSfrag replacements |u3 |


PSfrag replacements 0.0175 x1 x3
0.5 1 1.5 2
0.015 0.98
x1 |u1 | 0.96
0.0125
|u1 | 0.01 |u3 | 0.94
0.0075 0.92
|u3 |
0.005 0.9
PSfrag
0.0025 replacements |u1 | 0.88
0.5 1 1.5 2
x3 0.86
x1
(c)|u1 | (d)
|u3 |
x3
2.5
x3
2
|u1 |
PSfrag replacements 1.5
|u3 | 1
x1
0.5
|u1 | x1
-4 -2 2 4
-0.5
|u3 |
x3 (e)
|u1 |
x3
|u3 | 2.5
2
x1 1.5
x3 1
0.5

-4 4
x1
-2 2
-0.5

(f)

Figure 5.9: Normalised displacement (ui ) of device C ( k1 = 1500 +  m−1


and f = 0.1178 MHz) using the PWE method incorporating loss at the first
antisymmetrical Lamb mode. Plots (a) and (b) show the components of the
displacement on the transducer surface x3 = 0, as x1 is varied, (c) and (d) show
the components of the displacement as x3 is varied (at x1 = 0) and (e) and (f)
are the real part of the displacement in the x3 -x1 plane x2 = 0. Plot (e) is at
time t0 and (f) is at time t0 plus half the period (the displacements have been
scaled to accentuate the motion).
117

x3
2.5

1.5

0.5

PSfrag replacements 0

-0.5 x1
-0.1 0 0.1 0.2

(a)
x3

1.5

0.5

PSfrag replacements 0

x1
0.6 0.8 1 1.2 1.4

(b)

Figure 5.10: Displacement (real part) (u) of device C (k1 = 1500 +  m−1 and
f = 0.1178 MHz) using the PWE method incorporating loss at the first anti-
symmetrical Lamb mode during excitation. Plot (a) illustrates the movement
of specific points within the ceramic as time is varied and (b) illustrates the
movement of specific points over time within the polymer (the displacements
have been scaled to accentuate the motion and the scaling in the x1 direction is
not identical in each diagram).
118

In Figure 5.9, u3 is once again dominating with high displacement within


the ceramic pillars and fairly uniform displacement in the thickness direction.
Figure 5.10 shows elliptical motion in the anti-clockwise direction at the bottom
face, elliptical motion in the clockwise direction at the top face but only motion
in the x3 direction in the middle of the transducer for both materials, leading
to flexural motion. Looking at Figure 5.11 it is found that the Poynting vector
P3 in the x1 direction dominates and the energy is mainly within the polymer.
In fact most of the energy in the x3 direction is concentrated at the transducer
faces. Figure 5.11 (e) and (f) indicate that the energy is mainly within the
polymer. Figure 5.9 (e) and (f), together with the dynamics of Figure 5.10,
suggest that this is the first anti-symmetrical mode.
119

|P1 | |P3 |
0.07 1
0.06
0.8
0.05
0.04 0.6
PSfrag replacements
0.03 0.4
PSfrag replacements
0.02
0.2
0.01 |P1 |
x1 x1
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

(a) (b)
|P1 | PSfrag replacements |P3 |
PSfrag replacements 0.05 x1 0.12
x1 0.04 |P1 | 0.1
0.08
|P1 | 0.03 |P3 |
0.06
|P3 | 0.02 0.04
0.01 |P1 | 0.02
x3 x3
0.5 1 1.5 2 0.5 1 1.5 2
PSfrag replacements
(c) x (d)
1

|P1 | x3
2.5
|P3 |
2
x3
1.5
|P1 |
1
|P3 | 0.5
PSfrag replacements
x1
x1 -4 -2 2 4
-0.5
|P1 |
|P3 | (e)

x3 x3
2.5
|P1 |
2
|P3 |
1.5
x1 1
x3 0.5

-4 4
x1
-2 2
-0.5

(f)

Figure 5.11: Normalised Poynting vector (Pj ) of device C using the PWE method
incorporating loss (k1 = 1500 +  m−1 and f = 0.1178 MHz) at the first anti-
symmetrical Lamb mode. Plots (a) and (b) show the components of the Poynting
vector on the transducer surface x3 = 0, as x1 is varied, (c) and (d) show the
components of the Poynting vector as x3 is varied (at x1 = 0) and (e) and (f)
are the real part of the Poynting vector in the x3 -x1 plane x2 = 0. Plot (e) is at
time t0 and (f) is at time t0 plus half the period (the Poynting vector has been
scaled to accentuate the motion).
120

5.4.1.3 Rayleigh Mode

|u1 | |u3 |
0.15
0.35
0.125
0.3 0.1
PSfrag replacements
0.075
0.25
PSfrag replacements 0.05
x1
-3 -2 -1 1 2 3 |u1 | 0.025
0.15 -1 1
x1
-3 -2 2 3

(a) (b)

|u1 | PSfrag replacements |u3 |


1
PSfrag replacements 0.7 x1
0.6 0.8
x1 |u1 |
0.6
|u1 | 0.5 |u3 |
0.4
|u3 | 0.4
0.3
|u1 | 0.2
x3 x3
0.5 1 1.5 2 0.5 1 1.5 2
PSfrag replacements
(c) x (d)
1

|u1 | x3
2.5
|u3 |
2
x3 1.5
|u1 | 1
|u3 | 0.5
PSfrag replacements
x1
x1 -4 -2 2 4
-0.5
|u1 |
|u3 | (e)

x3 x3
2.5
|u1 |
2
|u3 |
1.5
x1 1
x3 0.5
x1
-4 -2 2 4
-0.5

(f)

Figure 5.12: Normalised displacement (ui ) of device C (k1 = 1500 +  m−1 and
f = 0.2133 MHz) using the PWE method incorporating loss at a Rayleigh mode.
Plots (a) and (b) show the components of the displacement on the transducer
surface x3 = 0, as x1 is varied, (c) and (d) show the components of the displace-
ment as x3 is varied (at x1 = 0) and (e) and (f) are the real part of the in-plane
displacement in the x3 -x1 plane x2 = 0. Plot (e) is at time t0 and (f) is at time
t0 plus half the period (the displacements have been scaled to accentuate the
motion).
121

x3
2

1.5

0.5
PSfrag replacements

0
x1
-0.1 -0.05 0 0.05 0.1

(a)
x3

1.5

0.5

PSfrag replacements
0
x1
0.85 0.9 0.95 1 1.05 1.1

(b)

Figure 5.13: Displacement (real part) (u) of device C (k1 = 1500 +  m−1 and
f = 0.2133 MHz) using the PWE method incorporating loss at a Rayleigh mode
during excitation. Plot (a) illustrates the movement of specific points within the
ceramic as time is varied and (b) illustrates the movement of specific points over
time within the polymer (the displacements have been scaled to accentuate the
motion and the scaling in the x1 direction is not identical in each diagram).

By investigating the displacements in Figure 5.12 it can be seen that the


122

motion is predominately in the x3 direction near the top face of the transducer.
Figure 5.13 suggests that the ceramic phase moves anti-clockwise, with essen-
tially only lateral motion near the bottom face, with an increasing x3 component
nearer the top face. The polymer moves anti-clockwise, again predominantly
in the lateral direction, with a switch in the direction of rotation at around
x3 = 0.75 mm. Examination of the Poynting vector in Figure 5.14 shows that
all the energy is near the top face of the transducer and this is symptomatic of
a Rayleigh wave.
123

|P1 | |P3 |
0.08 0.05

0.06 0.04

PSfrag replacements 0.03


0.04
PSfrag replacements 0.02
0.02
|P1 | 0.01

x1 x1
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

(a) (b)
|P1 | PSfrag replacements |P3 |
0.08 1
PSfrag replacements x1
0.8
x1 0.06 |P1 |
0.6
|P1 | 0.04 |P3 |
0.4
|P3 |
0.02
0.2
|P1 |
x3 x3
0.5 1 1.5 2 0.5 1 1.5 2
PSfrag replacements
(c) x (d)
1

|P1 | 2.5

|P3 | 2

x3 1.5

|P1 | 1

|P3 | 0.5
PSfrag replacements
x1 -4 -2 2 4
x1
x3 -0.5
|P1 |
|P3 | (e)

x3 2.5

|P1 | 2

|P3 | 1.5

x1 1

x3 0.5

x1 -4 -2 2 4
x3 -0.5

(f)

Figure 5.14: Normalised Poynting vector (Pj ) of device C (k1 = 1500 +  m−1
and f = 0.2133 MHz) using the PWE method incorporating loss at a Rayleigh
mode. Plots (a) and (b) show the components of the Poynting vector on the
transducer surface x3 = 0, as x1 is varied, (c) and (d) show the components of
the Poynting vector as x3 is varied (at x1 = 0) and (e) and (f) are the real part
of the Poynting vector in the x3 -x1 plane x2 = 0. Plot (e) is at time t0 and
(f) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion).
124

5.4.1.4 Intra-Pillar Mode

Investigating the displacements shown in Figure 5.15 it can be seen that the
motion is predominantly in the polymer phase. The overall motion shows the
ceramic phase being stretched and squashed, at alternating faces of the trans-
ducer (i.e. in a flexural motion). Figure 5.16 highlights that the ceramic and
polymer phases move in opposite directions from one another, mainly in the lat-
eral direction. Figure 5.17 shows that the energy is dominant within the ceramic
phases in the lateral direction, away from the surfaces of the transducer. The
evidence all suggests that the motion is being driven by the flexural response of
the ceramic pillar and so this mode is categorised as an intra-pillar mode.
125

|u1 | |u3 |
1
0.4
0.8
0.3
0.6 PSfrag replacements
PSfrag replacements 0.4 0.2

0.2
|u1 | x1
-3 -2 -1 1 2 3
-1 1
x1
-3 -2 2 3

(a) (b)

|u1 | PSfrag replacements |u3 |


PSfrag replacements x1 0.175
0.00012
0.15
x1 0.0001 |u1 |
0.125
0.00008
|u1 | |u3 | 0.1
0.00006
0.075
|u3 | 0.00004
0.05
0.00002 |u1 | 0.025
PSfrag replacements
0.5 1 1.5 2
x3 x3
0.5 1 1.5 2
x1
(c)|u1 | (d)
|u3 | x3
2.5
x3
2
|u1 |
PSfrag replacements 1.5
|u3 | 1
x1 0.5
|u1 | -4 x1
-2 2 4
-0.5
|u3 |
x3 (e)
|u1 | x3
2.5
|u3 |
2
x1 1.5

x3 1
0.5
x1
-4 -2 2 4
-0.5

(f)

Figure 5.15: Normalised displacement (ui ) of device C (k1 = 1500 +  m−1 and
f = 0.1.09 MHz) using the PWE method incorporating loss at the intra-pillar
mode . Plots (a) and (b) show the components of the displacement on the
transducer surface x3 = 0, as x1 is varied, (c) and (d) show the components
of the displacement as x3 is varied (at x1 = 0) and (e) and (f) is the real part
of the in-plane displacement in the x3 -x1 plane x2 = 0. Plot (e) is at time t0
and (f) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).
126

x3

1.5

0.5

PSfrag replacements
0
x1
-0.05 -0.04 -0.03 -0.02 -0.01

(a)
x3
2

1.5

0.5

PSfrag replacements
0
x1
0.85 0.95 1.05 1.15

(b)

Figure 5.16: Displacement (real part) (|u|) of device C (k1 = 1500 +  m−1 and
f = 0.1.09 MHz) using the PWE method incorporating loss at the intra-pillar
mode during excitation. Plot (a) illustrates the movement of specific points
within the ceramic as time is varied and (b) illustrates the movement of specific
points over time within the polymer (the displacements have been scaled to
accentuate the motion and the scaling in the x1 direction is not identical in each
diagram).
127

|P1 | |P3 |
1
0.03
0.8
0.025
0.6 0.02
PSfrag replacements
0.015
0.4
PSfrag replacements 0.01
0.2
|P1 | 0.005
x1 x1
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

(a) (b)
|P1 | PSfrag replacements |P3 |
PSfrag replacements 0.025 x1 0.0175
0.02 0.015
x1 |P1 |
0.0125
|P1 | 0.015 |P3 | 0.01
0.0075
|P3 | 0.01
0.005
0.005
|P1 | 0.0025
x3 x3
0.5 1 1.5 2 0.5 1 1.5 2
PSfrag replacements
(c) x (d)
1

|P1 | 2.5

|P3 | 2

x3 1.5

|P1 | 1

|P3 | 0.5
PSfrag replacements
x1 -4 -2 2 4
x1
x3 -0.5
|P1 |
|P3 | (e)

x3 2.5

|P1 | 2

|P3 | 1.5

x1 1

x3 0.5

x1 -4 -2 2 4
x3 -0.5

(f)

Figure 5.17: Normalised Poynting vector (Pj ) of device C (k1 = 1500 +  m−1
and f = 0.1.09 MHz) using the PWE method incorporating loss at the intra-
pillar mode. Plots (a) and (b) show the components of the Poynting vector on
the transducer surface x3 = 0, as x1 is varied, (c) and (d) show the components
of the Poynting vector as x3 is varied (at x1 = 0) and (e) and (f) are the real
part of the Poynting vector in the x3 -x1 plane x2 = 0. Plot (e) is at time t0
and (f) is at time t0 plus half the period (the Poynting vector has been scaled
to accentuate the motion).
128

5.4.2 Damping of Unwanted Lateral Modes

One of the main problems, which is due to the the architecture of a 1-3 or 2-2
composite transducer, is the presence of parasitic waves. These are generated
between adjacent pillars (inter-pillar modes) or within the pillars (intra-pillar
modes), interfering with the piston like behaviour of the fundamental thickness
mode [24]. Extensive experimental observations have highlighted the intricate
dependency between the geometry of the design, the material properties and
the key operational characteristics of the device. It has been suggested that a
passive material with a low transverse coupling would enhance the transducer’s
efficiency [24, 23]. By introducing high shear loss these parasitic waves will
hopefully be damped and crosstalk within the pillars should be reduced. In this
section the effects of introducing a high shear loss polymer are investigated using
dispersion diagrams and electrical impedance profiles. The model is based on
device C (see Table 5.1), which was discussed in the the previous section, with
a parameter ζ scaling the shear attenuation coefficient αs .
Figures 5.18 and 5.19 contrast the dispersion curves for a passive phase with
low shear loss (ζ = 1) to that with a high shear loss (ζ = 10), for the homoge-
neous case (N=0). Plots (c) and (d) have been coloured according to the size of
the imaginary part of the wavenumber (k1 ) with the red shades corresponding
to the highly attenuated modes. Plots (e) and (f) have been shaded according
to the width (σ) of the minima found in the determinant calculations with the
blue shades being the widest. Plots (c) and (d) suggest that the imaginary part
of k1 for most of the modes are of similar size. Plots (e) and (f) show that σ is
small except for the mode with the lowest phase velocity (highlighted in Figure
5.19). For low shear attenuation in the passive phase the width of the band
gaps around the thickness mode are 0.35 MHz below and 0.42 MHz above at
k1 = 0.106 mm−1 . For high shear attenuation the band gap widens to 0.39 MHz
below and 0.43 MHz above the thickness mode. It can be seen that by intro-
129

f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 1.25
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 thickness mode 0.5 thickness mode
0.25 0.25
1 4 5
k1 (mm−1 ) 1 4 5
k1 (mm−1 )
2 3 2 3

(a) ζ = 1 (b) ζ = 10
f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 1.25
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 thickness mode 0.5 thickness mode
0.25 0.25
k1 (mm−1 ) k1 (mm−1 )
1 2 3 4 5 1 2 3 4 5

(c) ζ = 1 (d) ζ = 10

f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 1.25
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 thickness mode 0.5 thickness mode
0.25 0.25
k1 (mm−1 ) k1 (mm−1 )
1 2 3 4 5 1 2 3 4 5

(e) ζ = 1 (f) ζ = 10

Figure 5.18: Dispersion characteristics for device C (homogenous). Frequency


(f ) versus wavenumber (k1 ) for low shear attenuation in the passive phase (ζ =
1) and high shear attenuation in the passive phase (ζ = 10). Plots (a) and (b)
have no colour, (c) and (d) are coloured with respect to the size of the imaginary
part of the wavenumber k1 and (e) and (f) are coloured with respect to the ’size’
of the minima (σ) of the determinant as defined above.

ducing high shear loss many of the modes have been damped so that a larger
band gap is found around the thickness mode.
Figures 5.20 and 5.21 show a similar analysis for the heterogeneous case
(N=3). As before plots (c) and (d) and have been shaded according to the
130

vp (ms−1 ) vp (ms−1 )
6000 6000
5000 5000
4000 4000
PSfrag replacements s0 PSfrag replacements s0
3000 3000
2000 2000
1000 bulk waves 1000 bulk waves
a0 a0
1 1.5
f × h (kHzm) 1 1.5
f × h (kHzm)
0.5 2 2.5 3 0.5 2 2.5 3

(a) ζ = 1 (b) ζ = 10
vp (ms ) −1
6000
5000 vp (ms−1 )
6000
4000 5000
PSfrag replacements s0 PSfrag replacements
3000 4000
3000 s0
2000
2000
1000 bulk waves 1000 bulk waves
a0 a0
f × h (kHzm) f × h (kHzm)
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3

(c) ζ = 1 (d) ζ = 10

vp (ms−1 ) vp (ms−1 )
6000 6000
5000 5000
4000 4000
PSfrag replacements s0 PSfrag replacements s0
3000 3000
2000 2000
1000 bulk waves 1000 bulk waves
a0 a0
f × h (kHzm) f × h (kHzm)
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3

(e) ζ = 1 (f) ζ = 10

Figure 5.19: Phase velocity (vp ) versus frequency (f ) × thickness (h) product
for device C (homogenous) with low shear attenuation in the passive phases
(ζ = 1) and high shear attenuation in the passive phase (ζ = 10). Plots (a)
and (b) have no colour, (c) and (d) are coloured with respect to the size of the
imaginary part of the wavenumber k1 and (e) and (f) are coloured with respect
to the ’size’ of the minima (σ) of the determinant as defined above.

size of the imaginary part of the wavenumber (k1 ) whilst plots (e) and (f) have
been shaded according to the width of the minima (σ) found in the determinant
calculations. Plots (c) and (d) indicate that the low frequency modes are more
attenuative than the higher frequency modes such as the thickness mode. Plots
131

f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 thickness mode 1.25 thickness mode
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 0.5
0.25 0.25
k1 (mm−1 ) k1 (mm−1 )
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5

(a) ζ = 1 (b) ζ = 10
f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 thickness mode 1.25 thickness mode
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 0.5
0.25 0.25
k1 (mm−1 ) k1 (mm−1 )
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5

(c) ζ = 1 (d) ζ = 10

f (MHz) f (MHz)
2 2
1.75 1.75
1.5 1.5
1.25 thickness mode 1.25 thickness mode
1 1
PSfrag replacements 0.75 PSfrag replacements 0.75
0.5 0.5
0.25 0.25
k1 (mm−1 ) k1 (mm−1 )
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5

(e) ζ = 1 (f) ζ = 10

Figure 5.20: Dispersion characteristics for device C (heterogeneous). Frequency


(f ) versus wavenumber (k1 ) for low shear attenuation in the passive phase (ζ =
1) and high shear attenuation in the passive phase (ζ = 10). Plots (a) and (b)
have no colour, (c) and (d) are coloured with respect to the size of the imaginary
part of the wavenumber k1 and (e) and (f) are coloured with respect to the ’size’
of the minima (σ) of the determinant as defined above.

(e) and (f) show that σ is large for the thickness mode and small for the lower
frequency modes. The thickness mode is more isolated in Figure 5.20 compared
to Figure 5.18 so it is easily identified. By observing Figure 5.21 to it is difficult
to identify the important modes which were seen clearly in Figure 5.19. This is
132

vp (ms−1 ) vp (ms−1 )
2000 2000
1750 1750
1500 1500
PSfrag replacements 1250 PSfrag replacements 1250
1000 1000
750 750
500 500
a0 a0
250 250
s0 s0
f × h (kHzm) f × h (kHzm)
bulk waves 0.5 1 1.5 2 2.5 3 bulk waves 0.5 1 1.5 2 2.5 3

(a) ζ = 1 (b) ζ = 10
vp (ms )
−1
vp (ms )
−1
6000 6000
5000 5000
4000 4000
PSfrag replacements PSfrag replacements
3000 3000
2000 2000
a0 1000 a0 1000
s0 s0
f × h (kHzm) f × h (kHzm)
bulk waves 0.5 1 1.5 2 2.5 3 bulk waves 0.5 1 1.5 2 2.5 3

(c) ζ = 1 (d) ζ = 10

vp (ms−1 ) vp (ms−1 )
6000 6000
5000 5000
4000 4000
PSfrag replacements PSfrag replacements
3000 3000
2000 2000
a0 1000 a0 1000
s0 s0
f × h (kHzm) f × h (kHzm)
bulk waves 0.5 1 1.5 2 2.5 3 bulk waves 0.5 1 1.5 2 2.5 3

(e) ζ = 1 (f) ζ = 10

Figure 5.21: Phase velocity (vp ) versus frequency (f ) × thickness (h) product
for device C (heterogeneous) with low shear attenuation in the passive phases
(ζ = 1) and high shear attenuation in the passive phase (ζ = 10). Plots (a)
and (b) have no colour, (c) and (d) are coloured with respect to the size of the
imaginary part of the wavenumber k1 and (e) and (f) are coloured with respect
to the ’size’ of the minima (σ) of the determinant as defined above.

due in part to the additional modes which arise in the heterogeneous case caused
by the inter-pillar and intra-pillar length scales. For a low shear attenuation in
the passive phase the width of the band gaps around the thickness mode are
0.18 MHz below and 0.58 MHz above at k1 = 0.406 mm−1 . For high shear
133

attenuation the band gap widens to 0.63 MHz below and more than 0.94 MHz
above the thickness mode. This shows that by introducing a high shear loss
passive phase, the thickness mode operates over a wider bandwidth.

G
0.5

0.4 thickness mode

0.3
a0 mode
0.2 interpillar mode
PSfrag replacements

0.1

f (MHz)
0.25 0.5 0.75 1 1.25 1.5

Figure 5.22: Conductance (G) versus frequency (f ) for device C (heterogeneous).


The solid and dash line represent low shear attenuation (ζ = 1) and high shear
attenuation (ζ = 20) respectively (γ = 0.5).

5 Lamb modes Intra/Inter pillar modes





ζ 10 
C
WC
15

20
PSfrag replacements 10
7.5
|ZT | 5
2.5
0
0 0.5 1 1.5
f

Figure 5.23: Absolute value of the electrical impedance of the transducer ZT


(kΩ) versus frequency f (MHz) and passive phase shear loss scaling for device
C

The derivation of the dispersion curves still requires improvement, particu-


134

larly in identifying the guided wave modes of low order. An alternative approach
is to utilise the electrical behaviour of the transducer to plot these curves.
The electrical conductance of the transducer is plotted as a function of fre-
quency in Figure 5.22. Subsequent analysis of the field equations show that the
thickness mode is at f = 0.65 MHz and the first antisymmetric Lamb wave (a0
mode) is at f = 0.12 MHz. When high shear loss is introduced it is apparent
that there is a damping effect on all modes except the thickness mode. In fact
the magnitude of the conductance at the thickness mode has slightly increased
when the high shear attenuation polymer is used. This may be due to unwanted
lateral modes at this frequency being damped out thus allowing the transducer
to more effectively operate in the desired thickness mode. Figure 5.23 shows
the high shear loss gradually taking effect in a plot of the absolute value of the
electrical impedance of the transducer as a function of the driving frequency and
the degree of shear attenuation in the passive phase. The aim is to damp out
any unwanted modes around the electrical resonant frequency fe = 0.65 MHz,
which corresponds to the thickness mode (the mechanical resonant frequency is
fm = 0.85 MHz). From the impedance curve it is clear that as ζ is increased
both the Lamb modes and intra/interpillar modes are being damped.
Note that the thickness mode in Figure 5.20 is predicted to be at a higher
frequency than Figure 5.22. To produce the dispersion diagram in Figure 5.20,
the system of equations XAr = 0 are solved by calculating the determinant of X.
In order to achieve this Tikhonov regularisation (see section 5.3), which shifts the
solution to cope with the numerical instabilities as the determinant approaches
zero, is used. The dispersion diagrams have been shifted and the results give
an approximate representation of where the modes will occur in respect to each
other. However, one advantage of this method is that the imaginary part of
the wavenumber k1 can be calculated. The results in Figure 5.22 are found by
solving a system of equations of the form XAr = Q. This gives an accurate
135

description of where the modes will occur but the imaginary part of k1 has to
be defined.

5.5 Comparison with Finite Element Modelling


and Experimental Data
In this section the PWE method is compared with experimental measurements
and FE results. In Figure 5.24 the PWE method is compared to the FE model
described in section 3.4, by examining the impedance characteristics of device
A (see Table 5.1) for a range of polymer phase materials and a half-wavelength
electrode patterning (γ = 1/2) on the top surface. Plot (a) shows the response
when using a standard hardset polymer HY1300/CY1301 [30] (see Table 3.1).
There is good agreement between both methods in the magnitude of the modes
but the PWE method predicts that the location of the mode will be slightly
shifted to a higher frequency. By performing a modal analysis, it is found that
the low frequency modes comprise a bulk mode and anti-symmetrical Lamb
modes. These unwanted lower frequency Lamb modes are typical of low shear
attenuation materials such as a standard hardset and will interfere with the
transducer performance. Plot (b) shows the response when using a standard
softset polymer. There is agreement between both methods on the magnitude
of the thickness response mode, however, again the PWE method predicts that
the location of the mode will be slightly shifted to a higher frequency. Both
methods have agreement that the high shear and longitudinal attenuation have
damped out any unwanted modes, and the magnitude of the thickness mode
response has reduced dramatically due to the decoupling of the ceramic and
polymer phases. The results in plot (c) for the HP20 passive phase show the
PWE method to predict the same location of the thickness mode response as
the FE modelling, but it is of larger magnitude. The PWE method predicts
an additional mode to the right of the main peak, however modal analysis indi-
136

Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(a) (b)
Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(c) (d)
Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(e) (f)

Figure 5.24: Electrical impedance characteristics for device A using the isotropic
PWE method (solid line in plots (a), (c) and (e)), FE modelling (solid line in
plots (b), (d) and (f)) and the anisotropic PWE method (dashed line) with
polymer phase (a) hardset, (b) softset, (c) HP20, (d) SBS1, (e) SBS2 and (f)
SBS4.

cates that this is spurious. The introduction of the SBS material in plots (d),
(e) and (f) demonstrates that these unwanted low frequency Lamb modes can
be damped down whilst maintaining a reasonable fundamental thickness mode
response magnitude. The PWE method has good agreement with the FE mod-
elling on the location of the modes in plots (d), (e) and (f), however the model
overpredicts the magnitude of these modes. Modal analysis found the additional
137

lower frequency mode in plot (f) to be spurious. The important point to note is
that this enhancement of the device performance is predicted in a similar fash-
ion by both modelling techniques. So although the PWE method has difficulty
here in the quantitative prediction of the impedance magnitude, the qualitative
improvements in performance are still evident.
vp (ms−1 )
4000

3000

PSfrag replacements s0
2000

Loadlines
1000

Bulk waves
a0 f (MHz)
0.2 0.4 0.6 0.8 1

Figure 5.25: Phase velocity (vp ) versus frequency (f ) × thickness (h) product
for device B using the PWE method (triangles), FE modelling (squares) and
experimental data (solid line). The long and short dashed lines represent the
bulk modes and loadlines respectively
138

x3
2.5

1.5

PSfrag replacements 0.5

x1
-1 -0.5 0.5 1

(a)

2.5

1.5

PSfrag replacements
0.5
x1
x3 -1 -0.5 0.5 1

(b)

Figure 5.26: The real part of the in-plane displacement in the x3 -x1 plane at
x2 = 0 for the a0 mode of device B. Plots (a) and (b) show the two extremes of
the periodic motion (f = 0.406 MHz)
139

x3
2.5

1.5

0.5

PSfrag replacements
0

x1
0.2 0.25 0.3 0.35

Figure 5.27: Displacement (real part) of a device B at the a0 mode. This


illustrates the movement of specific points within the transducer as time is varied
(f = 0.406 MHz)

|P |
0.35
0.3
0.25
0.2
0.15
PSfrag replacements
0.1
0.05
x3
0.6 1.2 1.8 2.4

Figure 5.28: Normalised Poynting vector |P | versus x3 of device B at the a0


mode (f = 0.406 MHz)

The PWE method is then compared to experimental and FE modelling data


for device B (see Table 5.1) with a HY1300/CY1301 hardset polymer phase [30]
(see Table 3.1). A dispersion diagram showing the dependency of the phase
velocity on the frequency is shown in Figure 5.25. In addition to predicting
the Lamb modes the PWE method also predicts the presence of bulk waves
and interpillar modes. These latter waves are excited by the spacing between
140

adjacent pillars and diagonally opposite pillars, and are displayed as constant
wavelength loadlines in the dispersion diagram. As such the loadlines should
appear as straight lines in this dispersion diagram. The slight deviations from
linearity which occur in Figure 5.25 are indicative of the accuracy of the method
in locating these modes. The deviation is about 5 per cent and this gives a rough
measure of the accuracy of the method. There is good agreement between all
three sets of results although the FE model predicts a larger frequency range
for the a0 mode than was observed experimentally. It should be borne in mind
however that the PWE method is computationally less intensive than the FE
model and its strength lies in providing a fast, qualitative prediction of the
transducer’s characteristics.
The number of Fourier coefficients used in the PWE method plays an analo-
gous role to that of the spatial discretisation used in the FE analysis. The bal-
ance between computational cost and model accuracy can therefore be controlled
in both methods. The accuracy can be gauged by comparison with experimental
data and by examining the convergence of the model predictions as the compu-
tational complexity is increased. A systematic study of the convergence of the
PWE method and its dependency on the number of Fourier coefficients has not
yet been conducted. Comparing the computational cost of the two methods is
difficult at present for the following reasons. The FE analysis conducted used a
commercial code which has been highly optimised for computational efficiency
[58] whereas the PWE method was implemented using a bespoke, suboptimal
coding. The FE analysis is for a transducer of finite lateral dimensions whereas
the PWE method is for a periodic composite of infinite lateral extent. Thus
reflections from the ends of the plate are included in the FE analysis and this
explains some of the differences. A quantitative comparison of the speed of each
method is therefore problematic at present. However, it is clear from these simu-
lations that the PWE method is at least an order of magnitude faster, although
141

one must bear in mind the aforementioned distinctions in their implementations.


The standard classification of the modes is problematic here as the support-
ing medium is heterogeneous, anisotropic, lossy and piezoelectric. As such the
descriptions of the waves in terms of their symmetry, or as Lamb, Rayleigh, bulk
waves etc. are only psuedo-descriptions and the actual behaviour is far more
complex. Identification of modes is aided by spatial and/or temporal plots of the
displacement, the Poynting vector and the electrical potential. To illustrate the
mode identification process, the a0 mode is investigated. Figure 5.26 highlights
the flexural nature of this mode, Figure 5.27 shows the elliptical motion of the
internal dynamics (with the associated reversals in the direction of rotation) and
demonstrates that the energy is predominantly at the plate boundaries. These
are all characteristics of a flexural Lamb wave.

5.6 Inclusion of an Anisotropic Passive Phase


In this section the electrical impedance and admittance characteristics of a 2-2
composite transducer, with a passive phase of styrene-butadiene-styrene (SBS)
are investigated. As mentioned previously, the SBS polymer is anisotropic
with effective styrene-like properties in the thickness direction and polybuta-
diene characteristics in the lateral directions. These characteristics of the mate-
rial should maintain the piston-like behavior of the transducer whilst reducing
crosstalk between the ceramic pillars. The aim is to damp out any unwanted
modes around the fundamental thickness mode.
Figure 5.29 shows the absolute value of the electrical impedance of the
transducer as a function of the driving frequency and the nondimensionalised
wavenumber k1 p1 /2π for an SBS passive phase with 69% polybutadiene. From
plot (a) the thickness mode response is more prominent than any surround-
ing modes. A lateral mode is present near the mechanical resonant frequency
(fm = 0.9 MHz). As k1 p1 /2π increases this lateral mode will shift around 0.2
142

MHz away from the fm and a lower frequency mode is introduced. In addition,
the magnitude of the fm will decrease. The imaginary part of k1 is 0.3 times the
real part, so as k1 increases, the effects of damping become more apparent.

0
0.1

k1 p10.2

0.3

0.4
PSfrag replacements
0.5
log |ZT |
0 0.5 1 1.5
f

(a)

Figure 5.29: The absolute value of the electrical impedance ZT for device C plot-
ted against the nondimensionalised wavenumber k1 p1 /2π (expressed in fractions
of the wavelength) and the driving frequency f (MHz).

In Figure 5.30, an anisotropic PWE model of 3 SBS passive fillers is compared


to an isotropic PWE model and an FE model by examining the impedance
characteristics of device A (see Table 5.1). Comparison to the isotropic PWE
model in plots (a), (c) and (e) shows that there is good agreement between both
methods in the location of the modes (fm = 0.5 MHz) however the anisotropic
PWE model predicts that the magnitude of thickness mode response will be
smaller. By comparing to the FE models in plots (b), (d) and (f) both methods
have good agreement in location and magnitude of the fundamental thickness
mode, particularly plot (d). The anisotropic PWE model is expected to perform
better since the material properties of the passive phase are more realistic.

5.7 Conclusions
The plane wave expansion (PWE) method is a frequency domain approach for
studying the modal behaviour of periodic piezoelectric composite transducers.
143

Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(a) (b)
Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20


f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(c) (d)
Z (kΩ) Z (kΩ)
80 80

60 60

40 40

PSfrag replacements 20 PSfrag replacements 20

f (MHz) f (MHz)
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(e) (f)

Figure 5.30: Electrical impedance characteristics for device A using the isotropic
PWE model (solid line in plots (a), (c) and (e)), FE modelling (solid line in plots
(b), (d) and (f)) and the anisotropic PWE method (dashed line) with polymer
phase (a) SBS1 , (b) SBS1, (c) SBS2, (d) SBS2, (e) SBS4 and (f) SBS4.

It is shown in this chapter that the method can be extended to incorporate fre-
quency dependent loss in both phases. One advantage of this approach over time
domain methods is that information on low amplitude or high frequency modes
is retained. A strict comparison between the PWE and FE methods in terms
of computational cost and accuracy is difficult at present but it would appear
that the PWE method provides a more qualitative prediction at a fraction of
the computational time. One advantage of investigating piezoelectric compos-
144

ites is that the electrical characteristics provide an additional means of inferring


the mechanical wave dispersion properties. This approach can complement the
harmonic analysis which relies on searching the frequency/complex wavenumber
parameter space for zeros of the determinant of a large, ill-conditioned matrix.
This chapter has also shown however that the use of scaling and Tikhonov reg-
ularisation greatly improves the conditioning of this latter approach. Although
the standard classification of the modes is difficult, as the supporting medium
is heterogeneous, anisotropic, lossy and piezoelectric, pseudo-descriptions of the
main supported modes of vibrations can be obtained by using spatial and/or
temporal plots of the displacement and the Poynting vector. The validation of
the model is proven by the good agreement between the PWE method, FE anal-
ysis and experimental data. Dispersion characteristics for low and high shear
attenuation in the passive phase were also compared and this showed that the
use of a high shear loss polymer as the passive phase in a 2-2 composite trans-
ducer results in an improved stop bad gap around the fundamental thickness
mode.
Chapter 6

Fractal Transducers

6.1 Introduction
In this chapter, the plane wave expansion model (PWE) is utilised to investi-
gate the behaviour of a transducer with a fractal architecture. The traditional
designs used in ultrasound transducers are very regular and have arisen due to
manufacturing constraints rather than performance optimisation. Many of these
restrictions no longer hold due to new manufacturing processes such as computer
controlled, laser cutting machines, and so there is now freedom to investigate
new types of geometry.
If we look to nature for inspiration then devices with irregular and self-
similar constructions may prove beneficial. Fractals for example are infinitely
complex and have a fine structure with detail at all scales. They are usually
self similar, with an irregular geometry which cannot be described in Euclidean
terms. Various types of fractals have provided an excellent representation of
natural objects such as clouds, snowflakes, mountains, lightning, trees and ferns
[61]. However, their use for ultrasonic transducer and array design has never
been investigated before.
In this chapter two systems which exhibit self-similarity, with their design
being built up iteratively from scaled versions of an original component part, are
investigated for their suitability as new transducer array designs [53]. Firstly

145
146

(a) (b)

Figure 6.1: Schematic of a periodic composite transducer where the piezoelectric


ceramic pillars are in black and the polymer filler is in white, (a) 1-3 topology,
(b) 2-2 topology

the Cantor set, which is created by continually deleting the middle third of a set
of line segments in a unit interval, is utilised to design a 2-2 configuration, where
each new fractal generation level will introduce additional ceramic pillars into the
transducer (see Figure ?? (b)). Secondly a 1-3 configuration will be investigated
with a Sierpinski Carpet geometry (see Figure ??(a)). This structure is created
by removing a square of side length one third from the centre of a unit square,
then removing a square of side length one ninth from the remaining eight squares
and so on [18].
Although the analysis of wave propagation in fractal piezoelectric media has
never been attempted, the interest in wave propagation in fractal media has
given rise to various mathematical approaches [33, 36, 75, 63, 62, 64, 65, 18].
For example, previous studies have shown that reaction-diffusion travelling waves
can be supported by Sierpinski gasket pre-fractal lattices [3, 2].

6.2 The Geometry


The model is configured for periodic 2-2 and 1-3 composites with the main
thickness mode of vibration in the x3 direction (see Figure 6.1). By using the
periodicity of the structure in the x1 − x2 plane, the material constants, M (x, %),
can be expressed as 
φ, if xS%
M (x, %) = (6.1)
θ, otherwise
147

where θ and φ are some physical property pertaining to the polymer and the
ceramic phase respectively, and % is the fractal generation level. This satisfies the
periodicity relationship M (x1 + p1 a, x2 + p2 b, %) = M (x1 , x2 , %), ∀ a, b  Z where
p1 is the period of the geometry in the x1 direction and p2 is the period of the
geometry in the x2 direction. A Fourier series representation for the Sierpinski
carpet (1-3 configuration)(see Figure 6.2 can be written as,

X ∞
X
%
M (x1 , x2 , %) = Mmn e−(2πmx1 +2πnx2 ) , (6.2)
m=−∞ n=−∞

where the set S% is given by


% 8 q−1
[ [
S% = [T q (i, 1), T q (i, 1) + (1/3)q ] × [T q (i, 2), T q (i, 2) + (1/3)q ], (6.3)
q=1 i=1

T 1 = {(−1/6, −1/6)}, (6.4)


8
(8q−2    )
[ [ 1 1 1
Tq = γi + Tjq−1 + , (6.5)
i=1 j=1
2 2 3

and γ ={(1/6,1/6),(-1/6,1/6),(-1/2,1/6),(-1/2,-1/6),(-1/2,-1/2),(-1/6,-1/2),(1/6,-
1/2),(1/6,-1/6)}. T q corresponds to the co-ordinates of the bottom left hand
corner of each ceramic pillar and the translations of (1/2, 1/2) is used to facili-
tate the contraction of each pre-fractal in the first quadrant. For the Cantor set
geometry (2-2 configuration)(see Figure 6.2) the Fourier series is expressed as

X
M (x, %) = Mn% e−(2πnx) , (6.6)
n=−∞

and S% simplifies to
% 2 q−1
[ [
S% = [T q (i, 1), T q (i, 1) + (1/3)q ], (6.7)
q=1 i=1

where (2q−2
3   )
[ [ 1 1
Tq = γi + Tjq−1 + , (6.8)
i=1 j=1
2 3
148

(a) % = 1 (b) % = 2

(c) % = 3

Figure 6.2: Plan view of one period of the Cantor set transducer design for the
first three generation levels. (The black material represents ceramic pillars and
the white material is the polymer.)

T 1 = {−1/6} and γ = {−1/2, 1/6}. Multiplying both sides of equation (6.2) by


e2πnx2 and integrating with respect to x2 from −1/2 to 1/2 then multiplying by
e2πmx1 and integrating with respect to x1 from −1/2 to 1/2 gives

Z 1 Z 1
2 2
%
M (x1 , x2 , %)e2πmx1 +2πnx2 dx1 dx2 = Mmn , (6.9)
− 12 − 12

since Z 1 
2
−2πbxi 0, for b  Z, b 6= 0
e dxi = . (6.10)
− 21 1, for b = 0
149

(a) % = 1 (b) % = 2

(c) % = 3

Figure 6.3: Plan view of one period of the Sierpinski carpet transducer design
for the first three generation levels. (The black squares represents ceramic pillars
and the white material is the polymer.)
150

P%
Equation (6.9) is then split into q=1 8q−1 integrals to give for the Sierpinski
carpet
% 8q−1 T q (j,1)+(1/3)q T q (j,2)+(1/3)q
X X Z Z
%
Mmn = (φ − θ)e2π(mx1 +nx2 ) dx
q=1 j=1 T q (j,1) T q (j,2)
% Z 1 Z 1
!
X 2 2
q−1 2π(mx1 +nx2 )
+ 1/ 8 θe dx . (6.11)
q=1 − 12 − 12

Note the last term equals zero unless m = n = 0, in this case it equals
θ 7θ
P% = . (6.12)
q=1 8q−1 8% − 1
Hence the Fourier coefficients at fractal generation level % are given by
% (φ − θ)  πm   πn 
Mmn = sin sin
π 2 mn 3q 3q
q q q q
× e(πm(2T (j,1)+(1/3) )+πn(2T (j,2)+(1/3) )) m, n = ±1, ±2, ..., (6.13)

 
% (φ − θ) 1  πn  q (j,2)+(1/3)q ))
M0n = sin e(πn(2T n = ±1, ±2, ..., (6.14)
πn 3q 3q

 
% (φ − θ) 1  πm  q (j,1)+(1/3)q ))
Mm0 = sin e(πm(2T m = ±1, ±2, ..., (6.15)
πm 3q 3q
and
φ−θ 7θ%
+ M00 =
. (6.16)
32q 8% − 1
For the Cantor set geometry, Equation (6.6) is multiplied on both sides by
e2πnx2 and integrated with respect to x2 from −1/2 to 1/2 to give

Z 1
2
M (x1 , %)e2πnx1 dx1 = Mn% . (6.17)
− 12
P%
Equation (6.17) is then split into q=1 2q−1 integrals to give
% 2q−1 T q (j,1)+(1/3)q
X X Z
Mn% = (φ − θ)e2πnx1 dx1
q=1 j=1 T q (j,1)
% Z 1
!
X 2
+ 1/ 2q−1 θe2πnx1 dx1 . (6.18)
q=1 − 12
151

As before, the last term equals zero unless n = 0, in this case it equals

θ θ
P% = . (6.19)
q=1 2q−1 2% −1

Hence the Fourier coefficients at fractal generation level % are given by

(φ − θ)  πn  q q
Mn% = sin q
e(πn(2T (j,1)+(1/3) )) n = ±1, ±2, ..., (6.20)
πn 3

and
φ−θ θ
M0% = q
+ % . (6.21)
3 2 −1

6.3 Results
In section 6.3.1 the electrical impedance and admittance characteristics of a
Cantor set composite transducer, with a standard hardset passive phase and
PZT5H ceramic (see Tables 3.1 and 2.1 for material properties ) are investi-
gated. The effects of the fractal generation level will be discussed and a modal
analysis will be performed to categorise any additional modes which arise due to
the fractal geometry. In section 6.3.2 the electrical impedance and admittance
characteristics of a Sierpinski carpet composite transducer using these materials
is similarly investigated. The lateral spatial periodicity is set as p1 = p2 = 1
mm and the thickness of the device is also h = 1 mm.

6.3.1 The Cantor Set Transducer

Figure 6.4 illustrates the Fourier approximation of an arbitrary material property


for the first four fractal generation levels of the Cantor set design. Plot (a)
illustrates generation level one which corresponds to one main ceramic pillar of
width 1/3, plot (b) corresponds to generation level two; a generation one pillar
plus two additional pillars of width 1/9, plot (c) then has four additional pillars
of width 1/27 and plot (d) has a further eight pillars of width 1/81. For the
generation level four pillars to be seen, the number of Fourier coefficients must be
152

around 40. As the number of Fourier coefficients increases the implementation


time of the model also increases. It also increases the matrix dimensions and
the effects of ill-conditioning. Hence, in the following results, fifteen Fourier
coefficients are used to approximate the geometry of the transducer.
PSfrag replacements PSfrag replacements

ceramic ceramic

1
9
polymer polymer 1 1 2 7
7 1 2
1 9 3 3 9
1
9 3 3

PSfrag replacements (a) % = 1 PSfrag replacements (b) % = 2

ceramic ceramic

polymer 1 1 2 7
polymer 1 1 2 7
9 3 3 9
1 9 3 3 9
1

(c) % = 3 (d) % = 4

Figure 6.4: Fourier approximation for the first four fractal generation levels of
the Cantor set. Plots (a), (b) and (c) have fifteen Fourier coefficients and plot
(d) has 40.
153

0 0
0.1 0.1

k 1 p1 0.2 k 1 p1 0.2
2π 2π
0.3 0.3
PSfrag replacements
0.4 0.4
PSfrag replacements
0.5
ZT Y
ZT
0 1 2 3 4 0 1 2 3
f f

(a) (b)

Figure 6.5: The absolute value of (a) the electrical impedance ZT (normalised)
and (b) the conductance Y (normalised) plotted against the nondimensionalised
electrode wavenumber k1 p1 /2π (expressed in fractions of the wavelength) and
the driving frequency f (MHz) with fractal generation level % = 1 for the Cantor
set transducer.

By examining the impedance profile in Figure 6.5 (a), the mechanical res-
onant frequency (fm ) is identified as the central ridge at around 2 MHz and
the electrical resonant frequency (fe ) as the minimum at around 1.5 MHz. The
electrical resonant frequency is also equivalent to the thickness mode within the
conductance plot in Figure 6.5 (b). To gauge the effect of the fractal geometry
on the behaviour of the device a comparison with the one dimensional Linear
Systems Model (LSM) [28] is conducted. Fractal generation level % = 1 corre-
sponds to the regular design shown in Figure 6.1(b). Comparison to the LSM
model in Figure 6.7 shows that there is good agreement between both methods
in the location of the modes (fm = 1.9 MHz) however the PWE model predicts
that the magnitude of the thickness mode response will be slightly smaller. As
the top electrode spacing varies from γ1 = 0 (that is a single, infinitely long top
electrode) to γ1 = 1/2 (each alternate ceramic pillar being excited by a voltage
that is shifted 180◦ out of phase (see Figure 6.6)), there is very little change in
the profile.
154

Top electrodes

Voltage

Figure 6.6: The top electrode spacing and an example applied voltage when
γ1 = k1 p1 /(2π) = 1/2.

ZT
40

30

20
PSfrag replacements
10

0.5 1 1.5 2 2.5 3


f

Figure 6.7: Absolute value of the electrical impedance ZT (kΩ) against fre-
quency f (Hertz) ×106 using the LSM method (dashed line) and the fractal
PWE method (solid line) with fractal generation level % = 1 for the Cantor set
transducer.
155

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.8: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
( k1 = 157 + 15.7 m−1 , fe = 1.46 MHz and % = 1). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion, the dark area is the ceramic and the lighter area is the
polymer). The electrical stimulus from the top electrode has a wavenumber of
k1 = 157 m−1 which corresponds to essentially a single electrode covering all the
ceramic pillars.

By investigating the displacement in the x3 -x1 plane x2 = 0 in Figure 6.8


it can be seen that at this frequency the transducer is moving in a piston-like
fashion with very little motion in the x1 direction, and the ceramic pillars move
out of phase with the polymer. This is the thickness mode since the mode is
symmetric, u1 is negligible compared to u3 and the amplitude of the displacement
is large. By examining the Poynting vector in the same plane in Figure 6.9 it can
be seen that the energy is distributed across the transducer in the x3 direction
but mainly in the ceramic phase.
156

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.9: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 157 + 15.7 m−1 , fe = 1.46 MHz and % = 1). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion, the dark area is the ceramic and the lighter area is the
polymer).

0 0
0.1 0.1

k 1 p1 0.2 k 1 p1 0.2
2π 2π
0.3 0.3
PSfrag replacements
0.4 0.4
PSfrag replacements
0.5
ZT Y
ZT
0 1 2 3 4 0 1 2 3
f f

(a) (b)

Figure 6.10: The absolute value of (a) the electrical impedance ZT and (b)
the conductance Y plotted against the nondimensionalised wavenumber k1 p1 /2π
(expressed in fractions of the wavelength of the electrical excitation) and the
driving frequency f (MHz) for fractal generation level % = 2 for the Cantor set
transducer.

Figure 6.10 shows the absolute value of the electrical impedance and con-
ductance of the transducer as a function of the driving frequency and the nondi-
mensionalised wavenumber k1 p1 /2π for fractal generation level two. It is found
that as the spatial wavelength of the electrical excitation decreases (as k1 p1 /2π
increases), an additional mode is introduced. When k1 p1 /2π is small the top
157

electrode acts as a single electrode, essentially treating the device as a homoge-


neous medium. As k1 p1 /2π increases, the heterogeneities in the medium start
to affect its behaviour.
ZT
40

30

20
PSfrag replacements
10

0.5 1 1.5 2 2.5 3


f

Figure 6.11: Absolute value of the electrical impedance ZT (kΩ) against fre-
quency f (Hertz) ×106 using the LSM method (dashed line) and the fractal
PWE method (solid line) for fractal generation level % = 2.

Comparison with the LSM model in Figure 6.11 suggests that the mechanical
resonant frequency is the second peak at 2 MHz.

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.12: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983+298.3 m−1 , fe = 1.08 MHz and % = 2). Plot (a) is at time t0 and (b)
is at time t0 plus half the period (the displacements have been scaled to accentu-
ate the motion). Here the wavenumber for the electrical excitation corresponds
to each alternate generation level 1 ceramic pillar being phase opposed.
158

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.13: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , fe = 1.08 MHz and % = 2). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion).

By investigating the real part of the in-plane displacement in the x3 -x1 plane
x2 = 0 in Figure 6.12 for the lower frequency mode (fe = 1.08 MHz) it is found
that the level alternate % = 1 pillars are out of phase and the level % = 2
pillars, which arise from the new fractal generation level, are in phase with
the neighbouring level % = 1 pillar. The spatial periodicity in the electrical
excitation (k1 = 2983 m−1 ) corresponds to the generation level % = 1 pillars
being spaced at half the wavelength of this excitation. This accounts for the
adjacent level % = 1 pillars being phase opposed. The motion at the top and
bottom of the transducer is predominantly in the x3 direction, with only lateral
motion in the middle of the transducer. As the large pillars become tall and thin,
the smaller pillars are squashed inwards and as they become short, the smaller
pillars are pushed outwards. Figure 6.13 indicates that most of the energy is
spread throughout the pillar and polymer sections which arise from the second
generation level.
159

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.14: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , fe = 1.72 MHz and % = 2). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.15: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , fe = 1.72 MHz and % = 2). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion).

Modal analysis of the higher frequency mode ( the second minimum in Figure
6.11 at fe = 1.72 MHz) also shows a motion that is mainly in the vertical
direction. In contrast to the previous case, as the large pillars become tall, the
smaller pillars are now pushed outwards. Figure 6.15 indicates that the energy
is mainly at the faces of the transducer, particularly at the top face. It can be
seen that as x1 increases, damping will increase since the imaginary part of the
wavenumber is positive. Both of these modes show characteristics of a thickness
mode and the presence of two thickness modes has arisen due to the inclusion
160

of the second generation level of ceramic pillars.

0 0
0.1 0.1

k1 p10.2 k1 p10.2
2π 2π
0.3 0.3

0.4 0.4
PSfrag replacements PSfrag replacements
0.5
ZT ZT
0 1 2 3 4 0 1 2 3
f f

(a) (b)

Figure 6.16: The absolute value of (a) the electrical impedance ZT and (b)
the conductance Y plotted against the nondimensionalised wavenumber k1 p1 /2π
(expressed in fractions of the wavelength) and the driving frequency f (MHz)
for % = 3 for the Cantor set transducer.

The electrical impedance and conductance of a transducer with fractal gener-


ation level % = 3 is plotted as a function of frequency and the nondimensionalised
wavenumber k1 p1 /2π in Figure 6.16. By introducing another fractal generation
level, there are now 4 modes present in each plot. There are two additional
lower frequency modes within the impedance plot and as the wavelength of the
electrical excitation decreases (k1 p1 /2π increases) the main lobe around 2 MHz
is damped out.
ZT
40

30

20
PSfrag replacements
10

0.5 1 1.5 2 2.5 3


f

Figure 6.17: Absolute value of the electrical impedance ZT (kΩ) against fre-
quency f (Hertz) ×106 using the LSM method (dashed line) and the fractal
PWE method (solid line) with % = 3 (γ1 ≈ 0).
161

Comparison with the LSM method when k1 p1 /2π is small shows that there is
good agreement away from the resonant regions where there are now two modes
present in the PWE method.

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.18: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 0.63 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.19: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 0.63 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion).

By analysing the displacement shown in Figure 6.18 of the first mode in


Figure 6.16 (b) (k1 p1 /2π =0.5 and f = 0.63 MHz), it is found that the mode is
antisymmetric. The pillars are stretched and squashed at opposite faces of the
transducer and the energy (see Figure 6.19) is predominantly at the faces of the
162

transducer. This behaviour is symptomatic of an anti-symmetrical Lamb mode,


although it has to be borne in mind that this is a pseudo-description, given that
the medium is heterogeneous, piezoelectric, anisotropic and lossy.

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.20: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 1.14 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.21: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 1.14 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the Poynting vector has been scaled to
accentuate the motion).

Figure 6.20 illustrates the displacement of the second peak in Figure 6.16
(b). It is clear that the large pillars are out of phase from the small pillars and
that the mode is symmetric. As the large pillars become small the additional
pillars are shifted apart in the middle and squashed inwards at the faces of the
163

transducer. By investigating the Poynting vector in Figure 6.21 it is clear that


the energy is distributed throughout the transducer. Since the displacement is
mainly in the x3 direction, the mode is classified as a thickness mode.

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.22: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 1.46 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.23: The real part of the Poynting vector in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 1.46 MHz and % = 3). Plot (a) is at time t0
and (b) is at time t0 plus half the period(the Poynting vector has been scaled to
accentuate the motion).

The displacement, shown in Figure 6.22, of the mode occurring around


1.5 MHz in Figure 6.16(b) is mainly vertical, although the pillars are slightly
stretched and squashed at opposite faces. This mode is anti-symmetric, the
level % = 1, 2 and 3 neighbouring pillars move as one, and each alternating set
164

of these is 180o out of phase. Figure 6.23 shows that, once again, the energy
is mainly at the faces, though there is energy throughout the higher generation
level pillars. By investigating the displacements shown in Figure 6.24 it can
be seen that the overall motion shows the ceramic phase being stretched and
squashed, at alternating faces of the transducer (i.e. in a flexural motion). Here
the largest pillars are out of phase from the two higher fractal generation level
pillars. Figure 6.25 shows that the energy is distributed along the top face of the
transducer.It appears that the motion is being driven by the flexural response
of the large ceramic pillar and so this mode can be categorised as an inter-pillar
mode.

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.24: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 2.43 MHz and % = 3). Plot (a) is at time t0 and
(b) is at time t0 plus half the period (the displacements have been scaled to
accentuate the motion).
165

PSfrag replacements PSfrag replacements

x1 x1
x3 x3

(a) (b)

Figure 6.25: The real part of the in-plane displacement in the x3 -x1 plane x2 = 0
(k1 = 2983 + 298.3 m−1 , f = 2.43 MHz and % = 3) (the Poynting vector has
been scaled to accentuate the motion).

Y
1

0.8

0.6

PSfrag replacements 0.4

0.2

0.5 1 1.5 2 2.5 3


f

Figure 6.26: Absolute value of the Electrical conductance YT (kΩ) against fre-
quency f (Hertz) ×106 using the fractal PWE method. The small dashed, large
dashed and full lines represent fractal generation % = 1, % = 2, and % = 3
respectively of the Cantor set transducer.

The electrical conductance of the transducer is plotted as a function of fre-


quency in Figure 6.26. When using only one fractal generation level, the ampli-
tude of the conductance is found to be 0.6 and the 6 dB percentage bandwidth
is 6%. By introducing another fractal generation level (% = 2), the amplitude
decreases to 0.44 but the bandwidth has increased to 7%. Increasing the fractal
generation level to 3 gives rise to a double lobed thickness mode with amplitude
0.17 and bandwidth 65%. Clearly the inclusion of extra pillars has increased the
frequency range but as a result the amplitude of the thickness mode has been
166

compromised.

6.3.2 The Sierpinski Carpet Transducer

In this section the Sierpinksi carpet (1-3 configuration) device shown in Fig-
ure ??(a) is investigated. Figure 6.27 illustrates the Fourier approximation
(N = M = 4) of an arbitrary material property for the first two fractal gener-
ation levels of the Sierpinski Carpet transducer where the height of the surface
represents the magnitude of the particular property. Plot (a) illustrates gener-
ation level one which corresponds to one central square ceramic pillar of side
length 1/3, plot (b) corresponds to generation level two; a generation one pillar
plus eight additional pillars of side length 1/9.

polymer polymer

ceramic
ceramic
PSfrag replacements PSfrag replacements

(a) % = 1 (b) % = 2

Figure 6.27: Fourier approximation for one spatial period of the first two fractal
generation levels of the Sierpinski Carpet transducer in the x1 − x2 plane (M =
N = 4).
167

0 0
0.1 0.1

k 1 p1 0.2 k 1 p1 0.2
2π 2π
0.3 0.3
PSfrag replacements
0.4 0.4
PSfrag replacements
0.5 0.5
ZT Y
ZT
0 1 2 3 4 0 1 2 3
f f

(a) (b)

Figure 6.28: The absolute value of (a) the electrical impedance ZT and (b)
the conductance Y plotted against the nondimensionalised electrode spacing
k1 p1 /2π (expressed in fractions of the wavelength) and the driving frequency f
(MHz) with fractal generation level % = 1 for the Sierpinski Carpet transducer.

To begin with we examine the generation level % = 1 case which corresponds


to a standard 1-3 design, as there is a single ceramic pillar of side length 1/3 in
each spatial period. Examining the impedance profile in Figure 6.28 (a), the me-
chanical resonant frequency (fm ) can be identified as the central ridge at around
1.8 MHz and the electrical resonant frequency (fe ) as the minima at around 1.4
MHz (fe is also equivalent to the thickness mode within the conductance plot
in Figure 6.28 (b)). It is clear that as k1 p1 /2π increases the main lobes in each
plot are damped.
168

Figure 6.29: Surface displacement for fractal generation level % = 1 for the
Sierpinski Carpet transducer over one period in time (k1 = 3140 + 314 m−1 and
fe = 1.24 MHz )(the displacements have been scaled to accentuate the motion).

By investigating the surface displacement at the thickness mode in Figure


6.29 it can be seen that the transducer is moving in a piston-like fashion with
very little motion in the x1 direction. The ceramic pillars appear to move out of
phase with the polymer but this is due to the ceramic pillars stretching and, due
to continuity at the interface between the phases, this draws the polymer phase
out in the lateral directions thus reducing the height of the polymer phase. The
mode is symmetric and the displacement in the lateral directions (u1 and u2 )
are negligible compared to those in the vertical direction (u3 ).
169

0 0
0.1 0.1

k 1 p1 0.2 k 1 p1 0.2
2π 2π
0.3 0.3
PSfrag replacements
0.4 0.4
PSfrag replacements
0.5 0.5
ZT Y
ZT
0 1 2 3 4 0 1 2 3
f f

(a) (b)

Figure 6.30: The absolute value of (a) the electrical impedance ZT and (b) the
conductance Y for device C plotted against the nondimensionalised wavenumber
k1 p1 /2π (expressed in fractions of the wavelength of the electrical excitation)
and the driving frequency f (MHz) for fractal generation level % = 2 for the
Sierpinski Carpet transducer.

Introducing the next generation of ceramic pillars into the design we now
consider the generation % = 2 transducer. Figure 6.30 shows the absolute value
of the electrical impedance and conductance of the transducer as a function of the
driving frequency and the nondimensionalised wavenumber k1 p1 /2π. Once again
it is found that by introducing an extra fractal generation level an additional
mode is introduced, although the mode appears at a lower frequency (f < 1
MHz) and resembles a Lamb mode.
170

Figure 6.31: Surface displacement for fractal generation level % = 2 for the
Sierpinski Carpet transducer over one period in time (k1 = 3140 + 314 m−1 and
fe = 1.37 MHz )(the displacements have been scaled to accentuate the motion).

By investigating the surface displacement in Figure 6.31 for the thickness


mode it is found that the mode is symmetric and the level % = 1 pillars are in
phase, together with their neighbouring % = 2 pillars. The motion is predomi-
nantly in the x3 direction. As the large pillars become tall and thin, the smaller
pillars are squashed inwards and as they become short, the smaller pillars are
pushed outwards. The introduction of the generation level % = 2 pillars has
created a structure whose surface is more in phase than generation level % = 1
and gives rise to a greater amplitude of displacement in the x3 direction. The
171

addition of the thin ceramic pillars in the heart of the polymer phase helps to
maintain the vertical displacement in these regions.

Figure 6.32: Surface displacement for fractal generation level % = 2 for the
Sierpinski Carpet transducer over one period in time (k1 = 3140 + 314 m−1 and
fe = 0.48 MHz )(the displacements have been scaled to accentuate the motion).

The surface profile at fe = 0.48 MHz, shown in Figure 6.32, demonstrates


that the motion is mainly in the lateral directions and that the displacement
is fairly uniform in the thickness direction. The mode is symmetrical and in
the body of the transducer the displacement of individual points is elliptical
near the top and bottom face, with only lateral motion in the vertical midplane
of the transducer. This behaviour suggests that this is the first symmetrical
172

Lamb mode (s0 ). Higher generation levels of the Sierpinski carpet design could
in theory be investigated using the same methodology. However, in order to
resolve the horizontal dimensions of the ceramic pillars the number of Fourier
coefficients grows exponentially. This in turn leads to very large matrices and
prohibitively long computation times.

6.4 Conclusions
In general, ultrasonic transducers composed of a periodic piezoelectric compos-
ite realise better operational characteristics than single phase designs. The most
frequently used designs are manufactured by dicing the ceramic into a series of
pillars and then filling the void with a passive polymer phase. The architecture of
these devices are very regular and have arisen due to manufacturing constraints
rather than performance optimisation. However, many of these restrictions no
longer hold due to new manufacturing processes such as computer controlled,
laser cutting machines, and so there is now freedom to investigate new types of
geometry. Hence, in this chapter, devices with self-similar constructions have
been investigated. It has been shown that the plane wave expansion model
(PWE) can be utilised to investigate the behaviour of a composite piezoelectric
transducer with a fractal architecture. Of course, from a manufacturing per-
spective, it will only be possible to build fractal devices over a limited number
of generation levels. The effects of introducing up to three fractal generation
levels have been investigated for a Cantor set geometry transducer and a modal
analysis was performed to help explain its characteristics. It was found that
by increasing the fractal generation level, the bandwidth surrounding the main
thickness mode will increase, but there will be a corresponding reduction in the
displacement amplitude. The PWE method was also used to investigate the
effects of using a transducer with a Sierpinski Carpet geometry. It was found
in both fractal architectures (1-3 and 2-2 configurations) that by introducing
173

more fractal generation levels, additional modes will occur although those can
be isolated from the main thickness mode. The investigation of surface profiles
suggests that the inclusion of more fractal generation levels will minimise oscilla-
tion within the passive filler of the transducer. This will optimize the piston like
behaviour of the device. We have developed a model to examine the behaviour
of these devices and the next stage would be to use this model to optimise the
device performance in terms of the aspect ratio, the materials, the side length of
the ceramic pillar at each generation level and indeed other fractal geometries.
Chapter 7

Conclusions

Composite transducers composed of a piezoelectric ceramic and a passive poly-


mer phase provide better electromechanical coupling and acoustic impedance
characteristics than conventional single phase transducers. Ideally a single lon-
gitudinal mode in the thickness direction will drive the transducer in a piston
like fashion, however other parasitic modes, propagating in other directions, can
interfere with this behaviour. Hence it is of interest to theoretically predict the
design criteria that will provide a large frequency band gap between the desired
thickness mode and these other waves. One possibility is to use polymers that
are highly attenuative to shear waves in order to reduce the cross-talk between
the ceramic pillars.
In Chapter 2 a linear systems model (LSM) was derived and used to model
a PZT5H ceramic transducer. A model for a 1-3 transducer with polymer
HY1300/CY1301 using the methods of Smith and Auld [67] was then intro-
duced and reception sensitivity, transmission sensitivity and impedance curves
were produced to compare a piezoelectric plate transducer and a 1-3 composite
transducer. It was found that composite transducers provide better electrome-
chanical coupling and acoustic impedance characteristics than conventional sin-
gle phase transducers.
In Chapter 3 frequency dependent, elastic loss was introduced to the LSM,

174
175

and a composite transducer with a Styrene-Butadiene-Styrene (SBS) polymer


phase was compared to an equivalent transducer with a standard passive filler.
The effective properties of the SBS passive phase were calculated using mixing
rules for parallel and series arrangements. The impedance profiles were affected
by the inclusion of loss via the reduction in the amplitude and the broadening of
the response at the mechanical resonant frequency. The damping of unwanted
lateral modes could not be investigated at this stage since LSM model is only
one-dimensional. The anisotropy in the SBS passive phase does affect the calcu-
lation of the effective material properties, particularly because of the high shear
attenuation. This led to a relatively large reduction in the impedance amplitude
when the loss was included. There was also a slight shift in the mechanical
resonant frequency to a lower value. The anisotropy in the polymer also led to a
higher electromechanical coupling coefficient than the standard isotropic filler.
The LSM model was then compared to a finite element model (FEM) by ex-
amining the impedance characteristics of a range of high shear attenuation, 2-2
composite transducers. The two methods showed reasonable agreement in all
the cases considered. The FEM showed that the use of a high shear attenuation
polymer damps out the unwanted, low frequency modes whilst maintaining a
reasonable impedance magnitude.
The plane wave expansion (PWE) method is a frequency domain approach
for studying the modal behaviour of periodic piezoelectric composite transduc-
ers. It is shown in this thesis that the method can be extended to incorporate
frequency dependent loss in both phases. One advantage of this approach over
time domain methods is that information on low amplitude or high frequency
modes is retained. A strict comparison between the PWE and the FEM methods
in terms of computational cost and accuracy is difficult at present but it would
appear that the PWE method provides a more qualitative prediction at a fraction
of the computational time. One advantage of investigating piezoelectric compos-
176

ites is that the electrical characteristics provide an additional means of inferring


the mechanical wave dispersion properties. This approach can complement the
harmonic analysis which relies on searching the frequency/complex wavenumber
parameter space for zeros of the determinant of a large, ill-conditioned matrix.
In Chapter 4 The plane wave expansion method was derived and extended to
provide a more complete analysis of the supported waves. Results have been pre-
sented for both 2-2 and 1-3 composite transducers made from PZT5H ceramic
and the standard hardset material HY1300/CY1301. Dispersion curves were
produced and particular modes on these curves were identified using the Poynt-
ing vector, electrical potential and the displacement. Admittance curves were
also produced which helped to clarify the important modes. The ill-conditioning
that is present in the matrix inversion procedure used to find the amplitudes of
each mode was circumvented using matrix balancing via parameter scaling and
Tikhonov regularisation.
In Chapter 5 the PWE method was extended to incorporate frequency de-
pendent loss in both phases. One advantage of this approach over time domain
methods is that information on low amplitude or high frequency modes is re-
tained. A strict comparison between the PWE and FE methods in terms of
computational cost and accuracy is difficult at present but it would appear that
the PWE method provides a more qualitative prediction at a fraction of the
computational time. One advantage of investigating piezoelectric composites is
that the electrical characteristics provide an additional means of inferring the
mechanical wave dispersion properties. This approach can complement the har-
monic analysis which relies on searching the frequency/complex wavenumber
parameter space for zeros of the determinant of a large, ill-conditioned matrix.
This chapter also showed however that the use of scaling and Tikhonov regu-
larisation greatly improves the conditioning of this latter approach. Although
the standard classification of the modes is difficult, as the supporting medium
177

is heterogeneous, anisotropic, lossy and piezoelectric, pseudo-descriptions of the


main supported modes of vibrations were obtained by using spatial and/or tem-
poral plots of the displacement and the Poynting vector. It was shown that
there was good agreement between the PWE method, FE analysis and exper-
imental data. Dispersion characteristics for low and high shear attenuation in
the passive phase were also compared and this showed that the use of a high
shear loss polymer as the passive phase in a 2-2 composite transducer results in
an improved stop band gap around the fundamental thickness mode.
In chapter 6 the plane wave expansion model (PWE) was used to investigate
the behaviour of a transducer with a fractal architecture. The effects of intro-
ducing up to four fractal generation levels were investigated for a Cantor set
geometry transducer (2-2) and a modal analysis was performed to help explain
the device characteristics. It was found that by increasing the fractal genera-
tion level, the bandwidth surrounding the main thickness mode increased, but
there was a corresponding reduction in the displacement amplitude. The PWE
method was also used to investigate the effects of using a transducer with a
Sierpinski Carpet geometry (1-3). It was found from both fractal architectures
(1-3 and 2-2 configurations) that by introducing more fractal generation levels,
additional modes will occur. However the thin ceramic pillars embedded in the
polymer phase helps to maintain a more piston-like displacement profile across
the surface of the transducer. A greater aspect ratio in the design of the ceramic
pillars may well give a clearer frequency separation between the thickness mode
and the unwanted lateral modes whilst retaining the benefits of the additional
ceramic pillars in the polymer phase.
Bibliography

[1] ANSI/IEEE standard on piezoelectricity., IEEE Transactions on UFFC,


43 (1996), pp. 717–771.

[2] J. Abdulbake, A. Mulholland, and J. Gomatam, Existence and


stability of reaction-diffusion waves on a fractal lattice, Chaos, Solitons and
Fractals, 20 (2003), pp. 799–814.

[3] , A renormalisation approach to reaction-diffusion processes on fractals,


Fractals, 11 (2003), pp. 315–330.

[4] K. Adachi, D. Hayward, G. Tardajos, and R. Pethrick, High-


frequency ultrasonic studies of solutions of styrene-butadiene-styrene tri-
block copolymers, Polymer, 30 (1989), pp. 1484–1487.

[5] K. Adachi, A. North, R. Pethrick, G. Harrison, and J. Lamb, Ul-


trasonic studies of styrene-butadiene-styrene triblock copolymers, Polymer,
23 (1982), pp. 1451–1456.

[6] ANSYS Europe Ltd, 1st Floor Waterloo House, Riseley Business Park,
Riseley, Berkshire RG7 1NW United Kingdom.

[7] B. Auld, Acoustic Fields and Waves in Solids, John Wiley and Sons,
U.S.A., 1973.

178
179

[8] B. Auld, Y. Shui, and Y. Wang, Elastic wave propagation in three-


dimensional periodic composite materials, Journal de Physique, 45 (1984),
pp. C5–159.

[9] S. Ballandras, M. Wilm, P. Edoa, and A. Soufyane, Finite-


element analysis of periodic piezoelectric transducers, JASA, 93 (2003),
pp. 702–711.

[10] A. Beltzer, Acoustics of Solids, Springer-Verlag, Heidelberg, 1988.

[11] R. Beyer and S. Letcher, Physical Ultrasonics, Academic Press, Lon-


don, 1969.

[12] D. Certon, O. Casula, F. Patat, and D. Royer, Lateral resonances


in 1-3 piezoelectric periodic composites: Modelling and experimental results,
JASA, 101 (1997), pp. 2043–2051.

[13] D. Certon, F. Patat, F. Levassort, F. G., and B. Karlsson,


Lateral resonances in 1-3 piezoelectric periodic composites: Modelling and
experimental results, JASA, 101 (1997), pp. 2043–2051.

[14] A. Cracknell, Ultrasonics, Wykeham Publications, London, 1980.

[15] I. Dainel and O. Ishai, Enigineering Mechanics of Composite Materials,


Oxford University Press, Oxford, U.K., 1994.

[16] P. Datta and R. Pethrick, Acoustic studies of heterophase systems:


styrene-butadiene-styrene, Polymer, 18 (1977), pp. 919–921.

[17] , Ultrasonic studies of glass-filled polymer solids, J. Phys. D: Appl.


Phys., 13 (1980), pp. 153–161.

[18] K. Falconer and J. Hu, Nonlinear diffusion equations on unbounded


fractal domains, J Math Anal App, 256 (2001), pp. 606–624.
180

[19] Ferroperm UK Ltd, Vauxhall Industrial Estate, Ruabon, Wrexham,


United Kingdom, LL14 6HA.

[20] L. Filipczynski, Transients, equivalent circuit and negative capacitance of


a piezoelectric transducer performing thickness vibrations, J. Tech. Phys.,
16 (1975), pp. 121–135.

[21] X. Geng and Q. Zhang, Evaluation of piezocomposites for ultrasonic


transducer applications - influence of the unit cell dimensions and the prop-
erties of constituents on the performance of 2-2 piezocomposites, IEEE
Trans UFFC, 44 (1997), pp. 857–872.

[22] , Resonance modes and losses in 1-3 piezocomposites for ultrasonic


transducer applications, J. Appl. Phys., 85 (1999), pp. 1342–1350.

[23] T. Gururaja, W. Schulze, L. Cross, and R. Newnham, Piezo-


electric composite materials for ultrasonic transducer applications. part 2:
Evaluation of ultrasonic medical applications, IEEE Trans Sonics and Ul-
trasonics, SU-22 (1985), pp. 499–513.

[24] T. Gururaja, W. Schulze, L. Cross, R. Newnham, B. Auld, and


Y. Wang, Piezoelectric composite materials for ultrasonic transducer appli-
cations. part 1: Resonant modes of vibration of pzt rod-polymer composites,
IEEE Trans Sonics and Ultrasonics, SU-22 (1985), pp. 481–498.

[25] G. Hayward, A systems feedback representation of piezoelectric transducer


operational impedance, Ultrasonics, 22 (1984), pp. 153–162.

[26] , The influence of pulser parameters on the transmission response of


piezoelectric transducers., Ultrasonics, 23 (1985), pp. 103–112.

[27] , Using a block diagram approach for the evaluation of electrical loading
effects on piezoelectric reception., Ultrasonics, 24 (1986), pp. 156–164.
181

[28] G. Hayward and J. Hossack, Unidimensional modelling of 1-3 com-


posite transducers, JASA, 88 (1990), pp. 599–607.

[29] G. Hayward and J. Hyslop, Determination of lamb wave dispersion in


data in lossy anisotropic plates using time domain finite element analysis.
part 1: Theory and experimental verification, IEEE Trans UFFC, 53 (2006),
pp. 443–455.

[30] , Determination of lamb wave dispersion in data in lossy anisotropic


plates using time domain finite element analysis. part 2: Application to 2-2
and 1-3 piezoelectric composite transducer arrays, IEEE Trans UFFC, 53
(2006), pp. 449–455.

[31] G. Hayward, C. MacLeod, and T. Durrani, A systems model of the


thickness mode piezoelectric transducer, JASA, 76 (1984), pp. 369–382.

[32] Huntsman Advanced Materials Ltd, Ickleton Road, Duxford, Cam-


bridge, United Kingdom CB2 4QA.

[33] J. Kigami, Harmonic calculus on p.c.f. self-similar sets, Trans. AMS, 335
(1993), pp. 721–755.

[34] G. Kino, Acoustic Waves Devices, Imaging and Analogue Signal Process-
ing, Prentice-Hall, London, 1987.

[35] R. Lakes, Viscoelastic Solids, CRC Press, London, 1999.

[36] M. L. Lapidus, The vibrations of fractal drums and waves in fractal media,
Fractals in the Natural and Applied Sciences, 41 (1994), pp. 255–260.

[37] G. Liu and X. Han, Computational Inverse Techniques in Nondestructive


Evaluation, CRC Press, London, 2003.
182

[38] M. Lowe, Matrix techniques for modeling ultrasonic waves in multilayered


media, IEEE Trans UFFC, 42 (1995), pp. 525–542.

[39] W. Magnus and S. Winkler, Hill’s Equation, New York, Dover, 1979.

[40] N. McCrum, B. Read, and G. Williams, Anelastic and Dielectric


Effects in Polymeric Solids, John Wiley and Sons, 1967.

[41] A. Neumaier, Solving ill-conditioned and singular linear systems; a tuto-


rial on regularization, IEEE Trans UFFC, 44 (1998), pp. 1018–1026.

[42] C. Oakley, Design considerations for 1-3 composites used in transducers


for medical ultrasonic imaging, IEEE, (1997), pp. 233–236.

[43] J. Ogilvy, An approximate analysis of waves in layered piezoelectric plates


from an interdigital source transducer, J. Phys. D: Appl. Phys., 29 (1996),
pp. 876–884.

[44] R. O’Leary, An Investigation into the Passive Materials utilized in the


construction of Piezoelectric Composite Transducers, PhD thesis, University
of Strathclyde, Glasgow, Scotland, 2003.

[45] , Private Communication, Centre for Ultrasonic Engineering, Univer-


sity of Strathclyde, Glasgow, Scotland, 2004.

[46] R. O’Leary, A. Cochran, and G. Hayward, Modification of the pas-


sive phase for frequency agility within 1-3 piezocomposite transducers. Cen-
tre for Ultrasonic Engineering, University of Strathclyde, Glasgow, Scotland
(Unpublished).

[47] R. O’Leary, A. Parr, A. Troge, R. Pethrick, and G. Hayward,


Performance of periodic piezoelectric composite arrays incorporating a pas-
sive phase exhibiting anisotropic properties.
183

[48] R. O’Leary, G. Smillie, G. Hayward, and A. Parr, CUE Materi-


als Database, Centre for Ultrasonic Engineering, University of Strathclyde,
Glasgow, Scotland, June 2002. www.cue.ac.uk.

[49] L.-A. Orr, A. Mulholland, R. O’Leary, and G. Hayward, Har-


monic analysis of lossy piezoelectric composite transducers using the plane
wave expansion method., (submitted to Ultrasonics).

[50] , Incorporation of viscoelastic loss into the plane wave expansion ap-
proach to modelling composite transducers., Proceedings of the 2006 IEEE
International Ultrasonics Symposium, (2006), pp. 472–475.

[51] , Investigation of 2-2 composite transducers with high shear attenua-


tion in the passive phase., Proceedings of GDR 2501: tude de la propaga-
tion ultrasonore en milieux non homognes en vue du contrle non-destructif
Presqu’le de Giens (France), (2006).

[52] L.-A. Orr, A. Mulholland, R. O’Leary, A. Parr, R. Pethrick,


and G. Hayward, Incorporation of viscoelastic losses into the linear sys-
tems modelling of composite ultrasonic transducers., (submitted to Ultra-
sonics).

[53] L.-A. Orr, A. Mulholland, R. O’Leary, R. Pethrick, and


G. Hayward, Theoretical analysis of an ultrasonic transducer with a frac-
tal architecture., (submitted to Chaos, Solitons and Fractals).

[54] E. Papadakis, Use of computer model and experimental methods to design,


analyze, and evaluate ultrasonic nde transducers, Mat. Eval., 41 (1983),
pp. 1378–1388.

[55] A. Parr, Private Communication, Centre for Ultrasonic Engineering, Uni-


versity of Strathclyde, Glasgow, Scotland, 2004.
184

[56] S. Poshyachinda, H. Edwards, and A. Johnson, Preparation and


characterization of poly (1,4-butadiene-b-1, 2-butadiene), Polymer, 32
(1991), pp. 334–337.

[57] D. Powell, G. Hayward, and R. Ting, Unidimensional modelling


of multi-layered piezoelectric transducer structures, IEEE Transactions on
UFFC, 45 (1998), pp. 667–677.

[58] PZFLEX, Weidlinger Associates, 4410 El Camino Real, Los Altos, CA


94022.

[59] M. Redwood, Transient performance of a piezoelectric transducer, JASA,


33 (1961), pp. 527–533.

[60] D. Royer and E. Dieulesaint, Elastic Waves in Solids 1: Free and


Guided Propagation, Springer-Verlag, Berlin, Heidelberg, New York, 1996.

[61] M. Schroeder, Fractals, Chaos, Power Laws, W.H. Freeman and Com-
pany, 1991.

[62] W. Schwalm and M. Schwalm, Electronic properties of fractal-glass


models, Phys. Rev. B, 39 (1988), pp. 12872–12882.

[63] , Extension theory for lattice green functions, Phys. Rev. B, 37 (1988),
pp. 9525–9542.

[64] , Closed formulae for green functions on fractal lattices, Physica A, 47


(1992), pp. 195–201.

[65] , Explicit orbits for renormalisation maps for green functions on fractal
lattices, Phys. Rev. B, 47 (1993), pp. 7847–7858.
185

[66] Y. Shui, X. Geng, and Q. Zhang, Theoretical modelling of resonant


modes of composite ultrasonic transducers, IEEE Transactions on UFFC,
42 (1995), pp. 766–773.

[67] W. Smith and B. Auld, Modelling 1-3 composite piezoelectrics:


Thickness-mode oscillations, IEEE Trans UFFC, 38 (1991), pp. 40–47.

[68] G. Staab, Laminar Composites, Butterworth-Heinemann, U.S.A., 1999.

[69] O. Stuetzer, Multiple reflections in a free piezoelectric plate, JASA, 42


(1967), pp. 502–508.

[70] R. Thurston, A. Pierce, and E. Papadakis, Physical Acoustics: Prin-


ciples and Methods, vol. XXIV, Academic Press, London, 1999.

[71] H. Tiersten, Linear Piezoelectric Plate Vibrations, Plenum Press, New


York, 1969.

[72] E. Tuncer and S. Gubanski, Dielectric relaxation in dielectric mixtures:


Applications of the finite element method and its comparison with dielectric
mixture formulas, J. Appl. Phys., 89 (2001), pp. 8092–8100.

[73] J. Vasseur, B. Djafari-Rouhani, L. Dobrzynski, M. Kushwaha,


and P. Halevi, Complete acoustic band gaps in periodic fibre reinforced
composite materials: The carbon/epoxy composite and some metallic sys-
tems, J. Phys.: Condens. Matter, 6 (1994), pp. 8759–8770.

[74] M. Wilm, S. Ballandras, V. Laude, and T. Pastureaud, A full


3d plane-wave-expansion model for 1-3 piezoelectric composite structures,
JASA, 112 (2002), pp. 943–952.

[75] M. Yamaguti, M. Hata, and J. Kigami, Mathematics of Fractals,


vol. 167, American Mathematical Society, Rhode Island, 1997.