Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

ARTICLE IN PRESS

NDT&E International 39 (2006) 525–541


www.elsevier.com/locate/ndteint

Ultrasonic arrays for non-destructive evaluation: A review


Bruce W. Drinkwater, Paul D. Wilcox
Department of Mechanical Engineering, University of Bristol, University Walk, Bristol, BS3 1TR, UK
Received 1 September 2005; received in revised form 23 March 2006; accepted 29 March 2006
Available online 22 May 2006

Abstract

An ultrasonic array is a single transducer that contains a number of individually connected elements. Recent years have seen a
dramatic increase in the use of ultrasonic arrays for non-destructive evaluation. Arrays offer great potential to increase inspection quality
and reduce inspection time. Their main advantages are their increased flexibility over traditional single element transducer methods,
meaning that one array can be used to perform a number of different inspections, and their ability to produce immediate images of the
test structure. These advantages have led to the rapid uptake of arrays by the engineering industry. These industrial applications are
underpinned by a wide range of published research which describes new piezoelectric materials, array geometries, modelling methods and
inspection modalities. The aim of this paper is to bring together the most relevant published work on arrays for non-destructive
evaluation applications, comment on the state-of the art and discuss future directions. There is also a significant body of published
literature referring to use of arrays in the medical and sonar fields and the most relevant papers from these related areas are also
reviewed. However, although there is much common ground, the use of arrays in non-destructive evaluation offers some distinctly
different challenges to these other disciplines.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Arrays; Ultrasonic; Modelling; Signal processing

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
2. Array design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
2.1. Transducer geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
2.2. Array transducer materials and manufacture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528
2.3. Inspection modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
3. Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
3.1. Modelling array elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
3.2. Beam profile modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
3.3. Point spread function modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
3.3.1. Array modelling example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
3.3.2. Apodisation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
3.3.3. Element spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
4. Signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536
4.1. Full matrix capture and post-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536
4.2. Signal encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
5. Future directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
5.1. Adaptive arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538

Corresponding author. Tel.: +44 117 9289749; fax: +44 117 9294423.
E-mail address: b.drinkwater@bristol.ac.uk (B.W. Drinkwater).

0963-8695/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ndteint.2006.03.006
ARTICLE IN PRESS
526 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

5.2. Data fusion . . . . . ............... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539


5.3. Harmonic imaging and elastic imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
6. Conclusions. . . . . . . . . . ............... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
References . . . . . . . . . . . . . . ............... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540

1. Introduction centre frequency of 2.25 MHz. They demonstrate the use of


an array to steer the beam through a range of angles thus
This paper reviews recent developments in ultrasonic ensuring good sensitivity to a range of crack angles in
array technology. Ultrasonic arrays are used for a wide geometrically complex components. Smith et al. [5] and
range of applications including sonar and medical imaging. Brotherhood et al. [6] show applications of arrays in the
However, the focus of this paper is on their application to aerospace industry including the inspection of both
non-destructive evaluation (NDE) using bulk waves. aluminium and carbon fibre reinforced composite wing
Ultrasonic arrays offer two key advantages over standard sections. Both use 64 element, 10–12 MHz arrays and
monolithic transducers. Firstly, a particular array is able to describe the principal advantage as being the ability to
undertake a range of different inspections from a single rapidly image large areas of uniform geometry.
location and so is more flexible than a single element Section 2 of this paper gives an overview of the different
transducer. In fact, an array can generate ultrasonic fields array transducer geometries and inspection modalities used
of almost infinite variety. However, they are most in NDE applications, with a focus on the array transducers
commonly used to produce fields similar to those from rather than the instrumentation. It should also be noted
traditional single element transducers, i.e. plane, focused that this paper describes exclusively bulk wave based
and steered beams. Secondly, most types of array (with the inspection modalities and so the use of array technology
exception of annular arrays) can be used to produce images for Guided/Lamb wave inspections is not reviewed. This
at each test location. This allows rapid visualisation of the section also covers transducer materials, although this is a
internal structure of a component. large research field in its own right and not the principal
An ideal array would have an infinite number of focus of this review. Section 3 reviews techniques for array
infinitesimally spaced elements, each able to independently modelling, and represents the most detailed part of this
generate an arbitrary waveform and receive reflected paper. Due to the wide variety of different array geometries
signals. Although such an array is far from achievable at and inspection modalities, it is the authors’ belief that
the time of writing, the increased performance and reduced modelling is a key step in the selection and optimisation of
size of electronics, as well as the yearly increase in an array for a given inspection. Alongside array modelling,
computing power means that this aim is getting closer. signal processing is the other focus of this review and is
The best achievable performance is limited by both the discussed in Section 4. Techniques employed to produce
transmit–receive instrumentation and the array transducer improved images of test structures using signal processing
design. Currently the transmit–receive instruments used are highlighted. Finally, Section 5 describes a number of
with NDE arrays are limited to below 256 independent possible future directions of array technology. An extensive
transmit–receive channels and so this represents the list of references is then given. Whilst the research papers in
maximum number of elements which can be used the field of ultrasonic arrays for NDE have been
simultaneously. Multiplexing then allows the available thoroughly searched, those referenced have been selected
transmit–receive channels to be switched to excite and to be the most useful in terms of relevance and novelty.
receive using a subset (termed the aperture) of the elements
in a larger array. For example, in the medical field Yen and 2. Array design
Smith [1] used a multiplexing system to enable a 256
channel transmit–receive instrument to be connected to a 2.1. Transducer geometries
64,000 element array.
These advantages have led to a significant take-up of Ultrasonic array transducers can be loosely classified as
ultrasonic array technology by the engineering industry. At one-dimensional (1-D), two-dimensional (2-D) or annular.
the time of writing there are at least five companies 1-D linear arrays are by far the most widespread in
manufacturing array systems world-wide and many hun- industry. These consist of a row of strip shaped elements as
dreds of array systems in use in industry. For example, shown in Fig. 1(a). Throughout this paper, the Cartesian
Song et al. [2], Chatillon et al. [3] and Mahaut et al. [4] all coordinate system shown in this figure will be used. Here,
describe the application of ultrasonic arrays in the nuclear the array lies in the x–y plane and the normal or axial
power industry. Song et al. describes the inspection of direction is parallel to the z-axis.
turbine blade roots using a 64 element array with a centre In the case of 1-D arrays, the array elements are
frequency of 7.5 MHz and Chatillon et al. and Mahaut et distributed in the x direction. Typically the elements in
al. describe the inspection of welds in thick (30–50 mm) 1-D arrays are relatively long in the y direction and hence
steel plates using a 24 element conformable array with a approximately behave as infinitely long strip sources. The
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 527

Imaging plane

Array element

a
a

z y z y
L

x a x
(a) Array element (b)

(c) Array element

Fig. 1. Definitions of most common array transducer geometries: (a) 1-D linear array, (b) annular array, (c) 1.5-D array, and (d) 2-D array with elements
arranged in a grid pattern.

image produced is therefore 2-D and lies in the x–z plane as but the addition of the extra rows enables a limited amount
shown in Fig. 1(a). of steering of the imaging plane in the y–z plane.
Although full 2-D arrays are gaining in importance in To allow the full 3-D imaging potential of 2-D arrays to
the medical sector they have yet to see widespread take-up be realised, various alternative array element distributions
by the NDE industry. Fig. 1(b) shows a 2-D array which have been suggested. For example, Martinez et al. [10]
uses square elements arranged in the x–y plane in a grid compared the imaging performance of a fully populated
pattern and which is able to image the three dimensional array with an array in which the elements were arranged in
(3-D) volume in front of it. a series of three concentric circles, as shown in Fig. 2(c).
Annular arrays of the type shown in Fig. 1(c) are They showed both experimentally and through simulation
somewhat different from other arrays in that they are that this array had similar imaging performance to that of
designed to provide variable focal depth only and do not a fully populated array of the same overall size. Norton [11]
allow beam steering. In this context they are used as a modelled the use of the configuration shown in Fig. 2(d) in
direct replacement for one or more conventional focused which elements were distributed around a single circle for
probes to perform, for example, the inspection of titanium medical imaging and showed that its effective aperture is
billets [7]. Such probes are typically convex rather than the same as a fully populated array when correct
planar as this provides some initial focussing and reduces transmit–receive weightings are used to compensate for
the number of elements required [8]. the element directivity. The Mills cross configuration
The array configurations shown in Fig. 1 are the simplest shown in Fig. 2(e), is commonly used in radar and was
but there are many other possible element patterns. also investigated by Mondal et al. [12]. In that paper, the
Needless to say, the 2-D array offers the most room for Mills cross configuration was shown to be inferior to the
innovative element patterns. This is particularly important circular configuration for the same number of elements.
as the number of elements required to fully populate a grid Recently in the medical imaging field a number of authors
pattern as shown in Fig. 2(a) quickly becomes large. 1.5-D have suggested the use of arrays with very large numbers of
arrays have been used to reduce the numbers of elements elements and investigated methods of selecting the optimal
relative to a fully populated 2-D array [9]. Typically 1.5-D numbers and distribution of elements for the transmit and
arrays contain a small number of parallel rows of elements receive apertures. For example, Yen and Smith [1] selected
as shown in Fig. 2(b). The primary imaging plane still sub-apertures of 256 elements from within a 65,000 element
passes through the x-axis (i.e. along the rows of elements), array. As shown in Fig. 2(f), they used four elements
ARTICLE IN PRESS
528 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

(a) (b)

(c) (d)

(e) (f)

Fig. 2. Schematic diagram of a range of different 2D array geometries: (a) fully populated circular array, (b) elements distributed around three concentric
circles, (c) elements distributed around a single circle, (d) the Mills cross configuration and (e) random element excitation within a fully populated array.

randomly distributed in a 16 element unit cell which was 2.2. Array transducer materials and manufacture
then repeated and demonstrated that this approach
improved the imaging performance when compared to There has been significant development of ultrasonic
regular element distribution patterns even when the array materials over the last 10 years. Much of this
element spacing was large. development has been driven by the medical (see for
The geometry of arrays is physically limited by a number example [15]) and underwater sonar fields (see for example
of factors relating to the physical manufacture, connectiv- [16]), both of which require highly sensitive and linear
ity and the number of channels of array controller transduction systems. Traditionally arrays were manufac-
instrumentation. These limitations have meant that, despite tured from blocks of piezoceramic material, such as lead
the obvious potential for imaging, currently there has been zirconate titanate (PZT-5 H), divided into elements using a
very little industrial take-up of 2-D arrays for NDE and diamond wire saw. The gaps between elements can be filled
such systems are still at the research stage [13]. One with a lossy polymer in an attempt to reduce element-to-
exception is Lasser [14] who developed an ‘‘ultrasonic element cross-talk. As the piezoceramics have high acoustic
camera’’ (marketed though Imperium Inc., USA) which is impedance they are inefficient when coupled to low
a 2-D array receiver with a very high element density impedance materials such as water. Due to the requirement
(128  128 elements over 1 cm2). A plane wave generator is to couple to living tissue in the medical field and water in
used to uniformly illuminate the structure and the array the sonar field, this has driven much of the recent
receives the transmitted or reflected signals. The maximum developments in piezoelectric materials for ultrasonic
signal amplitude of each element is then displayed as an generation and reception. In particular, this has lead to
intensity. This produces a 2-D image of the object with the the development of piezocomposites, in which
array element size being the resultant pixel size. small piezoceramic rods are set in a polymer matrix.
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 529

Piezocomposites have lower acoustic impedance than significant research interest [23]. The advantage of these
traditional piezoceramics and so are better acoustically new materials stems from the fact that their electromecha-
matched to water [17,18]. In addition the cross-talk is nical coupling efficiency is significantly larger than that of
significantly reduced and the bandwidth improved [19]. standard piezoelectric ceramics. This implies better sensi-
Since their initial development, a range of manufacturing tivity and potential for bandwidth enhancement over
techniques including mechanical dicing, injection moulding existing transducers. These new materials are yet to make
and laser machining [20] have been developed. an impression on the NDE marketplace and it will be some
In NDE, the requirement to match to water is less years before they are routinely used in array manufacture.
important than in the medical and sonar fields as the test Capacitive micromachined arrays have also been shown
structures are typically solids. However, the cross-talk to offer excellent acoustic performance, particularly
reduction and bandwidth improvement advantages of matching to water [13] and air [24]. Their key advantage
piezocomposites mean that most arrays in the NDT field over other transducer technologies is the ease with large
are manufactured in this way. Of particular importance are 2-D arrays can be manufactured and integrated with
the 1–3 composites, an example of which is shown in Fig. 3. electronic circuits. This approach offers the potential to
This configuration of piezocomposites has a particularly solve the connectivity issues which become a significant
high piezoelectric efficiency and near unimodal motion in problem when large numbers of elements are used.
the direction of the piezoceramic pillars leading to The common ‘‘dice and fill’’ manufacturing method
extremely low cross-talk. Methods for optimal design for means that the gap between elements is governed by the
1–3 composite piezoceramic transducers and reduction in width of the finest saw blade which is typically of the order
lateral motion are discussed in detail by Hayward and of 50–100 mm [25]. The size of elements governs their
Bennett [19]. A further advantage of these materials is that electrical properties and the range of sizes present in a
they can be adapted to a range of applications. For particular array may have to be limited to suit the electrical
example, the addition of a soft polymer matrix has enabled characteristics of the array controller instrument [26]. This
the development of flexible arrays [21,22] which can can be a particular problem in the case of annular arrays
conform around the test structure. However, despite this where element areas are non-uniform leading to large
promise, flexible arrays for NDT applications are currently capacitance variations between elements.
confined to research laboratories and are not commercially
available. 2.3. Inspection modalities
A new range of single crystal piezoelectric materials,
such as niobate-lead titanate, have recently attracted Arrays can be used for NDE in a wide range of different
inspection modalities. In general, the approach is to use the
array to mimic a range of different ‘standard’ ultrasonic
transducers. An exhaustive description of array inspection
Piezoceramic pillar
modalities would require another paper and so this section
reviews only the most common approaches. The simplest
approach is to position the array parallel to the surface of
the test structure and either couple it directly [3,27] or
through some intervening coupling medium [6]. In this
configuration plane B-scans, focused B-scans or angular
sweeps can be performed as shown in Figs. 4(a–c). In the B-
scan shown in Fig. 4(a) a subset of the array elements
(termed the aperture) is used to inspect the component and
the reflected signals are plotted as intensity. The aperture is
then moved electronically along the array to produce an
image. The elements within the aperture can be either fired
in-phase to produce a plane beam or fired with delays to
produce a focused beam as shown in Fig. 4(b). As each
focal position requires a different set of delays (often
termed a ‘‘focal law’’) each also necessitates a further
excitation-reception cycle. Therefore, as the number of
different focal positions required increases the inspection
speed decreases, which is a major issue in the medical sector
where real-time imaging is a key objective.
Angular sweeps (or sector scans) can be carried out with
Polymer matrix directly coupled arrays as shown in Fig. 4(c), but are more
Fig. 3. Schematic diagram of an element manufactured from 1–3 commonly used in conjunction with an angle wedge. The
connectivity piezocomposite material. angle wedge, typically made from a low loss material, such
ARTICLE IN PRESS
530 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

Aperture can occur over a range of angles, although it can also be


used to increase coverage from a particular location.
Essentially the swept angle approach emulates the more
traditional approach for crack detection where a number of
monolithic wedge transducers with different wedge angles
would be employed. This will be considered further in the
next section. Care must be taken to ignore spurious signals
which can often reverberate in the wedge and contaminate
(a) the resultant image. However, as these remain constant
over time they do not represent a significant problem.

3. Modelling

Because of the wide range of possible array geometries


and inspection modalities, modelling plays a crucial part in
the use of arrays for NDE. Typically the challenge is to
(b) select an array to achieve a required level of performance.
This performance requirement could include a specification
of minimum detectable defect size, defect sizing accuracy,
maximum steering angle, depth of penetration, range of
inspection angles, volume inspected from each position or
speed of inspection. This section reviews the modelling
θ approaches which have been adopted to assess the
performance of a given array and inspection modality.
There are two related, but distinct modelling philoso-
phies described in the literature; beam profile modelling
(c)
(see for example [29–31]) and point spread function (PSF)
modelling (see for example [32,33]). Both types can be
Angle
performed in either the time or frequency domain. Here the
wedge
frequency domain formulation is used as it is straightfor-
ward to relate it directly to the modular systems approach
to ultrasonic NDE modelling used by Schmerr [34] and
others.
Both these modelling techniques are based on directly
emulating the physical operation of an array. A third
approach that will not be covered in detail here is based on
θ a more abstract mathematical formalism in which the array
is regarded as a spatial sampler [35]. In this approach the
operation of an array and subsequent imaging is described
(d) in terms of a spatial Fourier transform. The attraction is
the mathematical insight obtained into array performance
Fig. 4. Array inspection modalities: (a) B-scan, (b) focused B-scan, (c)
angular sweep or sector scan, and (d) angular sweep with angle wedge. and phenomena such as spatial sampling frequency and
aperture size.
As its name suggests, beam profile modelling simulates
the ultrasonic field pattern produced by an array. For a
as polymethyl methacrylate (PMMA, also known by trade given array each focal law must be modelled separately.
names Plexiglas and Perspex), is used to orient the array at The advantage of beam profile modelling is that it is very
a fixed angle to the test piece as shown in Fig. 4(d) so that quick to perform and easy to obtain an impression of the
mode converted shear waves are excited in the test piece by effect of applying a particular focal law. Whilst beam
longitudinal waves in the wedge. In this way the array can profile modelling enables grating lobes and beam width to
be used to steer the shear waves in the test piece over a be visualised, it does not give a quantitative representation
limited range of angles [28]. Typically the range of angles of these effects on the final image produced by the array. A
for longitudinal waves in the wedge is kept beyond the further problem with beam profile modelling is that it does
longitudinal critical angle, so that only shear waves are not take reception into account. For a monolithic
excited in the test piece to simplify the inspection. The transducer the sensitivity in reception is identical to that
reason for sweeping the inspection angle in this configura- in transmission. However, for arrays there is no reason why
tion is usually to increase the detectability of cracks which the same set of element delays or the same signal processing
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 531

approach has to be used in both transmission and (x,y,z)


reception. This means that there is in effect a ‘‘reception’’
beam profile and this should be multiplied with the
‘‘transmission’’ beam profile before any quantitative
performance analysis is undertaken.
In the PSF approach the ability of a given array and z
imaging algorithm to image a point reflector is modelled.
To achieve this, ultrasonic data from an array due to a y
point reflector at a particular point is simulated and the
image of the reflector is then plotted using the appropriate θ
imaging algorithm applied to the simulated data. The
advantage of PSF modelling is that the effect of artefacts
such as grating lobes can be seen exactly as they will appear
in the final image. The disadvantage of PSF modelling is
the computational cost required to produce a single PSF
image and the fact that a single PSF only gives information φ
L
about a reflector at the point modelled. x

3.1. Modelling array elements

Many authors have produced models of transducer and


array transducer performance using either the beam profile
or PSF approach. Both types of models always need to
a
start from a model of an individual element. There are a
variety of techniques for modelling array element output Fig. 5. Definition of parameters for modelling the output of a rectangular
including finite element analysis (see for example [36]) and array element.
Huygens’ principle (see for example [37]). Finite element
analysis has been used to model both the electromechanical
effects within the element and the propagation of ultra- far field, this is:
sound into the surrounding media. For example, Robert-    
pa sin y cos f pL sin y sin f
son et al. [38] used finite elements to relate the voltage Df ðo; y; fÞ ¼ sinc sinc ,
lðoÞ lð o Þ
applied across a piezocomposite array element to the
displacement field in the surrounding media. This approach (1)
is time consuming to set-up but lends itself to modelling where Df is the directivity function due to the finite size of
more complex geometries [39], near field effects and details an element and l is the ultrasonic wavelength. The other
such as inter-element cross-talk [40]. terms in Eq. (1) are shown in Fig. 5 from which it can be
In the Huygens’ principle approach the output of a seen that a and L are the dimensions of the rectangular
transducer is described in terms of the summation of point element and y and f are the elevation and azimuth angles
(for a 3-D model) or line (for a 2-D model) sources respectively.
representing the surface of the transducer. Huygens’ It is common to model 1-D linear arrays using a 2-D
principle only models the mechanical aspects of the element model, in which each element is assumed to behave as a
(or transducer) and so the distribution of pressure across strip source that is infinitely long in the y direction. For
the element must be know a priori. However, for an array it clarity, this approach is adopted for the remainder of this
is common to assume that the elements are small and can paper. In this case, Eq. (1) can be simplified by setting
be modelled by their far field performance, in which case f ¼ 0 to give the directivity of an element in a 2-D plane:
closed form analytic solutions are readily derived for the  
pa sin y
fields from simple element shapes. In 3-D, elements are Df ðo; yÞ ¼ sinc . (2)
modelled as either circular or rectangular emitters and, in lðoÞ
2-D, as infinitely long strip emitters [41]. A directivity Eqs. (1) and (2) are based on integrating the fields from
function is then used to describe the angular distribution of point and line sources that radiate uniformly in three and
the emitted field of each element. Various directivity two dimensions, respectively. This is an acceptable
functions are described in the literature. Consider the approximation for arrays radiating into fluid, but for
array element shown in Fig. 5 and the spherical polar arrays coupled to a solid the fields to be integrated are not
coordinate system shown. The most common approach uniform in all directions. In the case of isotropic solids, the
(see for example [42]) is to integrate a series of point fields to be integrated should be those provided by Miller
sources to obtain the 3-D element directivity function and Pursey [43]. From this paper, the far field directivities,
due to the finite size of a rectangular element. In the DL and Ds, of longitudinal and shear waves generated in
ARTICLE IN PRESS
532 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

the 2-D case by line loading on the surface of a semi-infinite half-space loading are shown in Fig. 6(c). Comparison of
elastic solid can be written as Figs. 6(a) and (c) shows that there are significant
 2  differences between the basic directivity pattern for
cL =cS  2 sin2 y cos y radiation into fluid and into a solid half-space. It is
DL ðyÞ ¼ , (3)
F 0 ðsin yÞ particularly interesting to observe the large amplitude of
shear waves which are predicted in the solid case.
 
 5=2 c =c 2 sin2 y  1 1=2 sin 2y Fortunately, the shear waves travel significantly slower
cL L S
than the longitudinal waves and so their impact on the
D S ð yÞ ¼ , (4)
cS F 0 ðk sin yÞ resultant image is often small. Fig. 6(d) shows the
where cL and cS are the bulk longitudinal and shear wave directivity pattern into a solid half-space of a l width
velocities and element and it can be seen that this exhibits significantly
greater directionality in terms of longitudinal waves. Note
 2 !2
cL that the large, off-axis shear wave lobes due to line loading
F 0 ðzÞ ¼ 2z2  seen in Fig. 6(b) are significantly suppressed by the effect of
cS
the finite element width in Figs. 6(c) and (d).
 2 !1=2

2 2
1=2 2 cL
 4z z  1 z  . ð5Þ
cS 3.2. Beam profile modelling
The far field directivity, D, of a strip source of finite width
on the surface of a semi-infinite solid can therefore be In the frequency domain the radiated pressure field from
obtained by multiplying the line source directivity of the mth element may be written in terms of a combination
Eq. (3) or (4) by Eq. (2): of transfer functions (see for example [34]):

Dðo; yÞ ¼ Df ðo; yÞ  DL;S ðyÞ. (6) pm ðrm ; ym Þ ¼ I ðoÞT ðoÞAðo; rm ÞDðo; ym ÞeikðoÞrm , (7)
Fig. 6 shows the effect of the terms in Eq. (6). In all cases where I is the Fourier transform of an electrical input
cL= cS ¼ 2 and the directivities are normalised by DL(0). In signal, T is the impulse response function of an element, A
Fig. 6(a) the directivity pattern for longitudinal waves is a function to account for energy loss through beam
radiated into fluid by a l/2 width element is shown. The divergence and attenuation, rm and ym are local polar
underlying directivities of longitudinal and shear waves due coordinates with their origin at the centre of the element, k
to line loading of a solid half-space are shown in Fig. 6(b) is the relevant wavenumber and D is the element directivity
and the combined effect of both the element width and the function defined in Eq. (6).
°

°


22.5

22.5
°

°
45

45

° °
67.5 67.5

90° 90°
(a) 0 0.5 1 1.5 2 (b) 0 0.5 1 1.5 2

Longitudinal
Shear
°

°


22.5

22.5
°

°
45

45

° °
67.5 67.5

90° 90°
(c) 0 0.5 1 1.5 2 (d) 0 0.5 1 1.5 2

Fig. 6. Comparison of array element directivity functions: (a) l/2 width element radiating into fluid, (b) longitudinal and shear wave amplitudes due to
line loading of a solid half-space, (c) l/2 width element radiating into solid half-space, and (d) l width element radiating into solid half-space. In all cases
cL =cS ¼ 2 for the solid. All waves are normalised to PL(0).
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 533

In order to model the beam profile from an array 3.3. Point spread function modelling
under a particular focal law, the fields from each individual
element given by Eq. (7) must be converted to a In order to perform PSF modelling it is first necessary to
global coordinate system and superposed with the appro- simulate the complete data set obtained from the array.
priate amplitude weightings and phase delays. The This data set comprises an N  N matrix of signals from
beam profile in global Cartesian coordinates (x,z) is every transmit–receive element pair and is referred to as the
therefore: full matrix. Each signal in the full matrix can be expressed
in either the time- or frequency-domain. Although experi-
X
N mental signals are obviously obtained in the time-domain,
pðx; zÞ ¼ am eikbm pm the frequency-domain representation is again used for
m¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   consistency.
2 2 1 x  xm The signal obtained when the mth element is transmit-
 ðx  xm Þ þ z ; tan , ð8Þ
z ting and the nth element is receiving in the presence of a
single point reflector is given by
where am is the amplitude weighting and kbm is the phase
delay. For clarity, it has been assumed in Eqs. (7) and (8) H m;n ðoÞ ¼ I ðoÞ½T ðoÞ2 Dðo; ym ÞDðo; yn ÞRðo; ym ; yn Þ
that the elements in the array are identical although it is
straightforward to extend the procedure to non-uniform  Aðo; d m ÞAðo; d n ÞeikðoÞðd m þd n Þ , ð9Þ
elements.
The values of am and bm for some common imaging where I is the Fourier transform of the electrical input to
algorithms are shown in Table 1. To model reception an the transmitting element, T is the impulse response
identical construction can be used to create a received function of an element, A is an amplitude factor to account
beam profile. The overall sensitivity of the array is then the for energy loss through beam divergence and attenuation, k
transmit beam profile multiplied by the receive beam is the relevant wavenumber, d is the propagation distance
profile. In the case where the same focal law is used in from an element to the reflector, R is the reflection
transmission and reception these are identical, hence the coefficient of the reflector, D is the element directivity
transmit beam profile can be simply squared to obtain the function defined in Eq. (6) and the subscripts m and n refer
overall sensitivity. to the transmit and receive elements, respectively.

Table 1
Beam and PSF modelling parameters for common imaging algorithms

Beam model parameters PSF model parameters


(for plane and focused B-scans aperture is of width a and (for plane and focused B-scans aperture width is a)
centred on x ¼ x0 ; for beam steering, steering angle is ys)

Algorithm am bm wm,n(x,z) em,n(x,z)


(
Plane B-scan 1 jxm  x0 j  12a em ¼ z
wm ¼
0 jxm  x0 j412a
( (
Contributions from 1 jxm  x0 j  12a 0 1 jxn  x0 j  12a en ¼ z
elements in aperture wn ¼
0 jxm  x0 j412a 0 jxn  x0 j412a
summed without time
delays
( qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Focused B-scan 1 jxm  x0 j  12a em ¼ ðxm  xÞ2 þ z2f
wm ¼
0 jxm  x0 j412a
( qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ( qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Contributions from 1 jxm  x0 j  12a 1 jxn  x0 j  12a ðxn  xÞ2 þ z2f
zf  ðxm  x0 Þ2 þ z2f wn ¼
en ¼
elements in aperture 0 jxm  x0 j412a 0 jxn  x0 j412a
summed with phase delay
to focus at depth zf

Beam steering em ¼ r þ xm xr
Contributions from 1 en ¼ r þ xn xr
1 xm sin ys pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
elements summed with where r ¼ x2 þ z2
phase delay to steer beam
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Total focusing method
em ¼ ðxm  xÞ2 þ z2
Beam model of limited applicability as only one focal spot qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
All elements used to focus 1
can be modelled at a time en ¼ ðxn  xÞ2 þ z2
at every point
ARTICLE IN PRESS
534 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

For PSF predictions, the reflector is assumed to reflect However, because both the transmit and receive apertures
omnidirectionally and material attenuation is neglected. are the same in the cases considered, the overall sensitivity
Hence for a 2-D model the following expressions for R and based on beam profiles is equal to the transmit beam profile
A are used in Eq. (9): squared (i.e. the 20 dB greyscale is effectively doubled to
40 dB). Hence the grey-scales of beam profiles and PSFs are
Rðo; ym ; yn Þ ¼ 1,
directly comparable. The PSF calculations are based on a
1 Gaussian windowed five cycle (length in cycles defined by
Aðo; d Þ ¼ pffiffiffi . ð10Þ
d 40 dB points of window) input signal, a point reflector
The PSF of a given array and imaging algorithm can then located on the central axis at (0, 8l and the impulse
be computed by summing up weighted and delayed response functions of the transducer elements are assumed
contributions from all the transmitter-receiver pairs used. to be unity. The array has 64 elements with l/2 element
The PSF (in the x–z plane), PSF(x,z), of an N element spacing and l/2 element width. The array controller
array can be written as instrumentation is assumed to be such that up to 16
elements can be fired independently.
Z 1 m¼N
X nX ¼N
Fig. 7(a) shows the beam profile and the PSF for a 16
PSFðx; zÞ ¼ wm ðx; zÞwn ðx; zÞ
1 m¼1 n¼1
element aperture plane B-scan, while Fig. 7(b) shows the
beam profile and the PSF for a 16 element aperture B-scan
 H m;n ðoÞeikðoÞðem ðx;zÞþen ðx;zÞÞ do, ð11Þ focused at a depth of 8l. The imaging artefacts seen behind
where w and e are parameters which describe the element the image of the reflector in the PSF in Fig. 7(a) are a result
amplitude weighting and phase for the imaging algorithm. of the cylindrical edge waves generated at the edges of the
The values of w and e for the basic implementation of rectangular aperture interacting with the reflector. Note
various imaging algorithms are summarised in Table 1. that the presence of these artefacts is not readily apparent
from the equivalent beam profile in Fig. 7(a).
3.3.1. Array modelling example
Fig. 7 shows beam profiles and PSFs for a typical NDE 3.3.2. Apodisation
1-D array. In this and subsequent figures, the beam profile In Table 1 the aperture weighting in the plane and
is the transmit beam profile only and is plotted on a 20 dB focused B-scan cases is rectangular. However, in terms of
greyscale, while the PSF is plotted on a 40 dB greyscale. side lobe suppression, performance can be improved by

20 0 20 0
z/λ
z/λ

dB
dB

0 -20 0 -40
-10 0 10 -10 0 10
(a) x/λ x/λ

20 0 20 0
z/λ
z/λ

dB
dB

0 -20 0 -40
-10 0 10 -10 0 10
(b) x/λ x/λ

Fig. 7. Performance of 64 element array with l/2 element width, l/2 element pitch and 16 element active aperture: (a) plane B-scan and (b) B-scan focused
at depth 8l. The PSFs are for a point reflector at (0,8l.
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 535

making the aperture weightings smooth functions and this spacing array (as in Figs. 7 and 8) was used to steer a beam at
is termed apodisation [37]. +451. In Fig. 10(a) the main lobe of the beam can be seen at
The effect of applying a raised cosine apodisation approximately +451 while the other lobes visible are side
function to the aperture is shown in Fig. 8 in the same lobes of diminishing amplitude. These side lobes occur due to
format as Fig. 7. The artefacts seen in the PSF in Fig. 7(a) the rectangular apodisation of the array but no grating lobe
are not visible in Fig. 8(a) due to the apodisation. The is generated as the l/2 element spacing criterion is satisfied.
spatial extent of the PSF is also reduced slightly as, in this Fig 10(b) shows the beam profile from a l element spacing
example, a smaller beam width is produced by the array (as in Fig. 9) steered at +101. In Fig 10(b) it can be
apodisation. In focusing the converse is true and apodisa- seen that, as well as the main lobe and side lobes, an
tion reduces the effective aperture which reduces the additional grating lobe is now present at approximately
focusing performance, although in the case shown in 501, although its amplitude is somewhat smaller than that
Fig. 8(b) this effect is small. of the main lobe. This demonstrates than such an array can
perform limited beam steering with only a small grating lobe
3.3.3. Element spacing occurring. Fig. 10(c) shows the same l array steered at +451
Fig. 9 shows the beam profile and PSFs for the same from which it can be seen that very poor steering in the
configurations as in Fig. 7 except with l element spacing. desired direction is achieved. The grating lobe which occurs
To make this a fair comparison with Figs. 7 and 8, which is at approximately 201 can be seen to be of significantly
used an element spacing of l/2, the number of elements higher amplitude than the main beam.
used to plot Fig. 9 was halved to keep the aperture size In beam steering grating lobes will appear when these time
constant. Extra signals are apparent either side of the delays are such that there are one or more other directions in
expected reflection from the point reflector. These spurious which the waves from the elements are an integer number of
images are the effect of grating lobes and can be thought of wavelengths out of phase. It is straightforward to show that
as resulting from spatial undersampling across the array, grating lobes cannot exist for any steering angle if the
i.e. the spatial equivalent of the well known picket fence spacing between elements is less than half a wavelength–the
effect in the time domain (see for example [44]). so-called l/2 criterion. However there are a number of
Grating lobes are even more problematic for beam steering mechanisms that suppress grating lobes and hence allow this
inspections as one or more extra steered directions can be criterion to be somewhat relaxed in practice.
generated in addition to that intended by the applied delays. Firstly, array excitation is by short pulses a few
Fig. 10(a) shows beam profile generated when a l/2 element wavelengths in duration rather than continuous waves

20 0 20 0
z/λ
z/λ

dB
dB

0 -20 0 -40
-10 0 10 -10 0 10
(a) x/λ x/λ

20 0 20 0
z/λ
z/λ

dB
dB

0 -20 0 -40
-10 0 10 -10 0 10
(b) x/λ x/λ

Fig. 8. Performance of 64 element array with l/2 element width, l/2 element pitch, 16 element active aperture and raised cosine apodisation: (a) plane
B-scan and (b) B-scan focused at depth 8l. The PSFs are for a point reflector at (0,8l).
ARTICLE IN PRESS
536 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

20 0 20 0

z/λ
z/λ

dB
dB
0 -20 0 -40
-10 0 10 -10 0 10
(a) x/λ x/λ

20 0 20 0

z/λ
z/λ

dB
dB

0 -20 0 -40
-10 0 10 -10 0 10
(b) x/λ x/λ

Fig. 9. Performance of 64 element array with l element width, l element pitch and 16 element active aperture: (a) plane B-scan and (b) B-scan focused at
depth 8l. The PSFs are for a point reflector at (0,8l).

and hence the constructive interference produced by grammed delays and amplitudes to reflect the am and bm
superposed pulses separated by one wavelength is less than functions in Eq. (8) for the chosen imaging algorithm. The
in the case of continuous waves. physical wavefront injected into the specimen therefore
Secondly, the element width is normally such that the follows the pattern described by the beam profile. The
gap between elements is very small, hence the element received signals are summed and so contain the linear
width increases with element pitch. Wider elements are combinations of signals from different transmitter ele-
more directional (Fig. 6) and hence grating lobes at angles ments.
greater than the steering angle are suppressed relative to An alternative mode of array operation demonstrated by
the main beam. (However, note that for large steering Holmes et al. [27] is to actually obtain the full matrix of
angles the same effect causes grating lobes at smaller angles data experimentally and this operation is referred to as full
than the steering angle to be amplified relative to the main matrix capture. Any algorithm can then be applied by post-
beam as in the case of Fig. 10(c).) processing the data. The only limiting factor for such post-
Finally, if grating lobes are generated using an array in a processing is computation time. Holmes et al. investigated
wedge configuration it is highly likely that the orientation and quantitatively compared the performance of a number
of the grating lobes will be such that they do not pass of different algorithms in post-processing, such as B-scans
directly through the bottom of the wedge into the test- and focused B-scans, to image a crack in an aluminium
piece. Instead the energy in the grating lobes will block.
reverberate within the wedge itself rather than causing It should be noted that the concept of using full matrix
spurious ghost signals due to features in the test-piece. The capture and digital post-processing is already used
reverberations will be manifested as constant coherent extensively in guided wave arrays. Here, sophisticated
noise at the start of time-domain signals recorded from the algorithms are routinely applied in post-processing for
array. [28]. mode and direction separation [45], dispersion compensa-
tion [46] and frequency dependent beam steering [47,48].
4. Signal processing Holmes et al. also processed the array data such that the
array was focused on every point in the field of view and
4.1. Full matrix capture and post-processing termed this the total focusing method. The relevant w and e
functions for total focusing method are given in Table 1.
Under conventional array operation, a number of Fig. 11 shows the total focusing method applied to data
channels are used as parallel transmitters with pro- from a 64 element linear array with l/2 element width and
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 537

100 20
0 0
z/λ

z/λ
dB 10
dB

-20 -40
0 0
-50 0 50 -10 -5 0 5 10
(a) x/λ x/λ

100 Fig. 11. PSF generated using the Total Focusing Method algorithm for a
0 point reflector at (0,8l using a 64 element array with l/2 element width
and l/2 element pitch.

l/2 element pitch. In the PSF in Fig. 11 the reflector is at


(0,8 l) and so is directly comparable with Fig. 7.
The main drawback of the full matrix capture approach
z/λ

dB is that using a single element in transmission limits the total


acoustic power output and hence degrades the signal to
random noise ratio (SNR) of each transmit-receive pair
relative to traditional multi-element aperture approaches.
This degradation is tempered by an improvement in SNR
due to the averaging effect of the imaging algorithm. The
net effect is that the SNR of data reconstructed from full
0 -20
matrix capture compared to that obtained from an
-50 0 50
(b) x/λ equivalent multi-channel parallel
pffiffiffiffiffi transmitter configuration
100 is reduced by a factor of N where N is the number of
0 elements in the aperture involved [27,49].

4.2. Signal encoding

The signal-to-noise ratio, bandwidth and scanning speed


of arrays can all be improved by the use of various
encoding schemes such as Golay sequences [50] and chirps
z/λ

dB [51]. Typically the code (or sequence) is sent to the


transducer and a decoding filter used on reception. For
example, with Golay codes two binary pulse trains are sent
in two successive transmission cycles. One of these is the
anti-phase version of the other as shown in Fig. 12. The
received signals are then cross-correlated with the corre-
-20 sponding input signals and the pair of cross-correlated
0
-50 0 50 signals summed. The procedure works because the auto-
correlation functions for each input signal have sidelobes in
(c) x/λ
the time-domain that are in anti-phase with each other and
Fig. 10. Beam steering using (a) 16 elements, l/2 element width, l/2 which therefore cancel out on addition. The final time
element pitch, steering angle +451, (b) 8 elements, l element width, domain signal has a signal-to-noise ratio which is propor-
l element pitch, steering angle +101 and (c) 8 elements, l element width,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

tional to 2:sequence length [52]. The net result is that for
l element pitch, steering angle +451.
the same peak-to-peak voltage applied to a given array
element, the resultant signal-to-noise ratio can be improved
ARTICLE IN PRESS
538 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

Cross-correlation Array elements


functions
Decoded
Input Output result
F(t) F(t) F(t) F(t)
Golay Received Transmit 1
code A signal A
F(t) -F(t) F(t) -F(t)
Golay Received Transmit 2
code B signal A
F(t) F(t) -F(t) -F(t)
Transmit 3

F(t) -F(t) -F(t) F(t)


Fig. 12. Schematic diagram of the use of Golay codes (denotes
correlation). Transmit 4

Fig. 13. Sequences of simultaneously fired encoded signals.

above that achievable using simple pulsed excitation. This


approach is therefore particularly useful in ‘lossy’ materials step in array imaging. Two approaches to increasing
where the penetration depth is limited by attenuation. One imaging speed are frequency division and simultaneous
disadvantage of the use of coded sequences is that they transmission of multiple coded waveforms (see for example
result in a ‘dead’ zone in front of the transducer as the [54]). The concept of frequency division is the most simple
elements cannot receive until after they have completed conceptually. It involves splitting the available frequency
the transmission of the sequence. In practice workers in the bandwidth into distinct frequency regions. Different coded
medical field have found that this limits the codes to 40–80 (or uncoded) waveforms are transmitted in each frequency
bits in length [53]. region and processed separately. This yields an imaging
In an array, the signal-to-noise ratio of the full matrix speed increase proportional to the number of frequency
capture approach described in Section 4.1 can be improved regions as the total number of firings can be reduced by this
by simultaneously firing different combinations of binary fraction. The practical limitation of this approach is that
codes on all elements [54]. This idea is illustrated in Fig. 13 the response of the test-piece needs to be frequency
for an array with four elements. The net result is a series of independent over the range of frequencies used. Simulta-
four simultaneous equations, from which the received neous firings of different Golay codes can be used if their
signals from four separate individual element firings can be cross-correlation functions are sufficiently small relative to
reconstructed. The number of firings is equal to the number their auto-correlation functions. This difference defines the
of elements, N, in the array and so the time required to cross-talk from one coded sequence to another. To date
perform capture of the full-matrix remains unchanged these approaches have only been used in the medical sector
relative to that required if individual elements are fired. but there are obvious applications in the NDT field where
This then yields the full matrix of transmit–receive it is required to inspect large structures rapidly.
combinations from which the desired imaging algorithm
can be implemented as before. However, as the signal 5. Future directions
associated with a particular element transmitting is now
obtained by combining the N different signals thepsignal
ffiffiffiffiffi to 5.1. Adaptive arrays
random noise ratio is improved by a factor of N . This
precisely balances the degradation in signal to random The ability of arrays to generate wavefronts of infinite
noise associate with full matrix capture that was discussed variety lends itself to adaptive approaches. The principle is
in the previous subsection. that the element amplitudes and delays (and in principle
The codes are also designed to have a very wide also the waveforms) can be adjusted to adapt to
bandwidth. For example, chirps [55] are defined in terms uncertainties in the test structure. A number of authors
of a start frequency, an end frequency and (typically) a have investigated these approaches for a variety of different
linear variation of frequency with time. However, as the reasons. For example Clark et al. [56] used high power
codes must be passed though the transducer they are lasers to generate an arbitrary distribution of light on a
inevitably affected by the transfer function of the surface which then excited a surface acoustic wave. They
transducer. Due to this effect, although the bandwidth of then adaptively adjusted the distribution of light to
the code may be extremely wide, the transducer still limits compensate for acoustic aberrations due to material
the practically achievable bandwidth. inhomogeneities and hence produce a good focus. With
In addition to the above uses, coded waveforms can also this system they were able to undertake surface acoustic
be used to increase the speed of inspection. This is possible wave imaging at 82 MHz. In the medical imaging field
as encoding strategies have been developed which allow a Karaman et al. [57] used an adaptive approach to
reduced number of firings which is often the time-limiting compensate for tissue motion when using an array to
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 539

perform synthetic aperture focusing. Of more direct results in the distortion of the ultrasonic waveforms, which
relevance to this paper, the group at Commissariat à in turn results in the generation of harmonics in the
l’Energie Atomique, France [3,4,58] have investigated the frequency domain. The harmonics are measured using
effect of surface irregularities on array imaging perfor- higher order resonances of the transducer and their
mance. They have developed a flexible array which can amplitude relates to the extent of the nonlinearity. This
conform to the surface features of the component. The has two advantages; firstly as the harmonics are at higher
element delays are then adaptively adjusted to optimise the frequencies, the imaging resolution is inherently improved
beam profile. Significantly improved sensitivity to defects and secondly, it has been shown to have increased
(side drilled holes) was demonstrated after the adaption. sensitivity to certain medical conditions. In elastic imaging
Very few of these adaptive systems have moved beyond the the test medium is excited using a low frequency pulse or
laboratory stage of development but the potential for step function. The effect of this low frequency loading is
improved performance appears large. As there is a then measured by ultrasonic imaging. For example, in step
significant desire by industry to inspect complex compo- loading an image is taken before and after the application
nents it is anticipated that adaptive array approaches will of the load and the displacement measured. Although it is
become increasingly commonplace and this research area not obvious how these two medical imaging approaches
will expand rapidly. could be transferred to NDE there would be clear
advantages if they could be. For example, nonlinear
5.2. Data fusion NDE is in its own right, a rapidly growing area of research
(see for example [64]) and the possibility of nonlinear
Arrays offer the ability to produce many different images imaging would be very attractive. The measurement of
from each transducer location. It is already standard elastic properties and residual stress is an on-going area of
industrial practice when using arrays to perform a number interest in the NDE community.
of different inspections, for example performing both a
sector scan and a plane B-scan. The challenge, which is as 6. Conclusions
yet unresolved, is how to make best use of the potentially
massive amount of image information. A number of This paper has reviewed current research progress in the
authors have looked at data fusion in NDE. For example use of ultrasonic arrays for NDE. This is an area in which
Horn and Mayo [59] assessed the inspection reliability the industrial uptake is progressing as rapidly as the
gains which could be obtained from combining eddy- research. The benefits of arrays for NDE are in terms of
current and ultrasonic data. They investigated a number of inspection performance, flexibility and rapid imaging. In all
different ways of combining the data, such as logical-OR these areas, arrays offer a step change over traditional
decision making as well as weighted combinations and single element transducers. It is anticipated that the
showed that all these approaches produced measurable performance of commercially available instrumentation
reliability improvements (in terms of probability of will improve dramatically over coming years. This should
detection). Hall et al. [60] used image fusion between lead to systems in which more elements are used and, in
different tomographic reconstruction techniques. They particular, full 3-D imaging is realised. Clearly these
fused reflection, transmission and time-of-flight images systems will require more sophisticated electronics, and
using a fuzzy pixel fusion technique. Essentially this looked faster computational performance. Given the current rate
to see if a given pixel was above a threshold in each image. of development these steps seem inevitable.
If it was above this threshold in any of the images then it As noted, the array imaging algorithms discussed in
was definitely in the final fused image, if it was below then Section 3 are based on the emulation of monolithic
it became part of a weighted sum with the equivalent pixels transducers. For this reason their resolution in all cases is
from the other images. It is clear that there is a huge ultimately diffraction limited. With access to the full data
amount of work to be performed in this area before NDE set from an array, the diffraction limit can potentially be
engineers routinely use data fusion and fuzzy logic overcome using so-called ‘‘super-resolution’’ imaging
approaches. However, it seems undeniable that, given the techniques. For example Prada et al. [65] describes a
wide range of possible inspection modalities, array imaging technique based on decomposition of the time-reversal
data is not being used optimally at present. operator (DORT) to identify the location of a number of
point reflectors and Devaney [66] suggests an enhancement
5.3. Harmonic imaging and elastic imaging termed multiple signal classification (MUSIC). Although
super-resolution imaging can only be used to resolve a
The last 10 years has seen a massive expansion in interest limited number of targets, it is likely that the next few years
in both harmonic imaging [61] and elastic imaging [62] for will see the emergence of this technology in certain
medical applications. Both approaches have now been specialised ultrasonic array applications [67].
demonstrated for various diagnostic purposes (see for This paper has also highlighted the importance of
example [63]). Harmonic imaging relies on the inherent modelling in the selection of arrays to fulfil a given
nonlinearity of various types of tissue. This nonlinearity performance specification. It is the authors’ belief that it is
ARTICLE IN PRESS
540 B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541

essential for such modelling to be performed as part of the made with viscous polymer processing for high frequency ultrasound.
inspection development process. A number of possible Ultrasonics 2004;42:479–84.
[21] Powell DJ, Hayward G. Flexible ultrasonic transducer arrays for
future areas for development including adaptive arrays,
non-destructive evaluation applications—part 1: the theoretical
data fusion and harmonic and elastic imaging have also modelling approach. IEEE Trans Ultrason Ferr Freq Cont 1996;
been discussed. 43(3):385–92.
[22] Powell DJ, Hayward G. Flexible ultrasonic transducer arrays for
non-destructive evaluation applications—part 2: performance assess-
ment of different array configurations. IEEE Trans Ultrason Ferr
References Freq Cont 1996;43(3):393–402.
[23] Marin-Franch P, Cochran S, Kirk K. Progress towards ultrasound
[1] Yen JT, Smith SW. Real-time rectilinear 3D ultrasound using receive applications of new single crystal materials. J Mat Sci Matls in
mode multiplexing. IEEE Trans Ultrason Ferr Freq Cont 2004;51(2): Electroincs 2004;15:715–20.
216–26. [24] Hutchins DA, McIntosh JS, Neild A, Billson DR, Noble RA.
[2] Song SJ, Shin HJ, Jang YH. Development of an ultrasonic phased Radiated fields of capacitive micromachined ultrasonic transducers in
array system for non-destructive tests of nuclear power plant air. J Acoust Soc Am 2003;114(3):1435–49.
components. Nucl Eng Design 2002;214:151–61. [25] Turnbull DH, Foster FS. Fabrication and characterisation of
[3] Chatillon S, Cattiaux G, Serre M, Roy O. Ultrasonic non-destructive transducer elements in two-dimensional arrays for medical imaging.
testing of pieces of complex geometry with a flexible array transducer. IEEE Trans Ultrason Ferr Freq Cont 1992;39(4):464–75.
Ultrasonics 2000;38:131–4. [26] Powell DJ, Hayward G, Ting RY. Unidimensional modelling of
[4] Mahaut S, Roy O, Beroni C, Rotter B. Development of phased array multi-layered piezoelectric transducer structures. IEEE Trans Ultra-
techniques to improve characterisation of defect located in a son Ferr Freq Cont 1998;45(3):667–77.
component of complex geometry. Ultrasonics 2002;40:165–9. [27] Holmes C, Drinkwater BW, Wilcox PD. Post-processing of the full
[5] Smith RA, Bending JM, Jones LD, Jarman TRC, Lines DIA. Rapid matrix of ultrasonic transmit-receive array data for non-destructive
ultrasonic inspection of ageing aircraft. Insight 2003;45(3):174–7. evaluation. NDT&E Int 2005;38(8):701–11.
[6] Brotherhood CJ, Drinkwater BW, Freemantle RJ. An ultrasonic [28] Crowther P. Practical experience of phased array technology for
wheel-array sensor and its application to aerospace structures. Insight power station applications. Insight 2004;46(9):525–8.
2003;45(11):729–34. [29] Wooh SC, Shi Y. Influence of phased array element size on beam
[7] Friedl JH, Gray TA, Khandelwal P, Dunhill T. Ultrasonic phased steering behaviour. Ultrasonics 1998;36:737–49.
array inspection of seeded titanium billet. In: Thompson DO, [30] Lee J-H, Choi S-W. A parametric study of ultrasonic beam profiles
Chimenti DE, editors, Review of Progress in QNDE. vol. 23. 2004. for a linear phased array transducer. IEEE Trans Ultrason Ferr Freq
p. 809–16. Cont 2000;47(3):644–50.
[8] Lupien V. Phased array ultrasonic probe design: from art to science [31] Ahmad R, Kundu T, Placko D. Modelling of phased array
through optimization. Thompson DO, Chimenti DE, Review of transducers. J Acoust Soc Am 2005;117(4):1762–76.
Progress in QNDE, vol. 23. 2004. p. 761–8. [32] Chiao RY, Thomas LJ. Analytic evaluation of sampled aperture
[9] Wildes DG, Chaio RY, Daft CMW, Rigby KW, Smith LS, ultrasonic imaging techniques for NDE. IEEE Trans Ultrason Ferr
Thomenius KE. Elevation performance of 1.25D and 1.5D transdu- Freq Cont 1994;41(4):484–93.
cer arrays. IEEE Trans Ultrason Ferr Freq Cont 1997;44(5):1027–37. [33] Lingvall F, Olofsson T, Stepinski T. Synthetic aperture imaging using
[10] Martinez O, Akhnak M, Ullate LG, de Espinosa FM. A small 2D sources with finite aperture: deconvolution of the impulse response.
ultrasonic array for NDT applications. NDT&E Int 2003;36:57–63. J Acoust Soc Am 2003;114(1):225–34.
[11] Norton SJ. Annular array imaging with full-aperture resolution. J [34] Schmerr Jr LW. Fundamentals of ultrasonic non-destructive evalua-
Acoust Soc Am 1992;92(6):3202–6. tion. New York: Plenum Press; 1998.
[12] Mondal S, Wilcox PD, Drinkwater BW. Design of two-dimensional [35] Goodman J, Moeller M. Introduction to Fourier optics. McGraw-
ultrasonic phased array transducers. Trans ASME—J Pressure Vessel Hill Book Co Ltd; 2004. ISBN:0974707724.
Technol 2005;127(3):336–44. [36] Lerch R. Simulation of piezoelectric devices by two- and three-
[13] Oralkan O, Ergun AS, Cheng C-H, Johnson JA, Karaman M, Lee dimensional finite elements. IEEE Trans Ultrason Ferr Freq Cont
TH, Khuri-Yakub BT. Volumetric ultrasound imaging using 2-D 1990;32(2):233–47.
CMUT arrays. IEEE Trans Ultrason Ferr Freq Cont 2003;50(11): [37] McNab A, Cochran A, Campbell MA. The calculation of acoustic
1581–94. fields in solids for transient normal surface sources of arbitrary
[14] Lasser M, A novel high speed, high resolution, ultrasound imaging geometry and apodisation. J Acoust Soc Am 1990;87(4):1455–65.
system. Thompson DO, Chimenti DE, editors. Review of Progress in [38] Robertson D, Hayward G, Gachagan A, Reynolds P. Minimisation
QNDE, vol. 17. 1998. p. 1713–9. of mechanical cross-talk in periodic piezoelectric composite arrays.
[15] Oralkan O, Ergun AS, Johnson JA, Karaman M, Demirci U, Kaviani Insight 2004;46(11):658–61.
K, Lee TH, Khuri-Yakub BT. Capacitive micromachined ultrasonic [39] Yaralioglu GG, Ergun AS, Khuri-Yakub BT. Finite-element analysis
transducers: next generation arrays for acoustic imaging? IEEE Trans of capacitive micromachined ultrasonic transducers. IEEE Trans
Ultrason Ferr Freq Cont 2002;49(11):1596–610. Ultrason Ferr Freq Cont 2005;52(12):2185–98.
[16] Caiti A, Bergem O, Dybedal J. Parametric sonars for seafloor [40] Wilm M, Armati R, Daniau W, Ballandras S. Cross-talk phenomena
characterisation. Meas Sci Technol 1999;10:1105–15. in a 1-3 connectivity piezocomposite. J Acoust Soc Am 2004;116(5):
[17] Hayward G, Hossack JA. Unidimensional modelling of 1–3 2948–55.
composite transducers. J Acoust Soc Am 1990;88(2):599–608. [41] McNab A, Stumpf I. Monolithic phased array for the transmission of
[18] Smith WA, Auld BA. Modelling 1–3 composite piezoelectrics: ultrasound in NDT ultrasonics. Ultrasonics 1986;24:148–55.
thickness-mode oscillations. IEEE Trans Ultrason Ferr Freq Cont [42] Wooh SC, Shi Y. Three-dimensional beam directivity of phase-
1991;38(1):40–7. steered ultrasound. J Acoust Soc Am 1999;105(6):3275–82.
[19] Hayward G, Bennett J. Assessing the influence of pillar aspect ratio [43] Miller GF, Pursey H. The field and radiation impedance of
on the behaviour of 1–3 connectivity transducers. IEEE Trans mechanical radiators on the free surface of a semi-infinite isotropic
Ultrason Ferr Freq Cont 1996;43(1):98–108. solid. Proc Roy Soc Lond 1954;34:521–41.
[20] Abrar A, Zhang D, Su B, Button TW, Kirk KJ, Cochran S. 1–3 [44] Randall RB. Frequency analysis. 2nd ed. Copenhagen: Bruel &
connectivity piezoelectric ceramic-polymer composite transducers Kjaer; 1987.
ARTICLE IN PRESS
B.W. Drinkwater, P.D. Wilcox / NDT&E International 39 (2006) 525–541 541

[45] Cawley P, Lowe MJS, Alleyne DN, Pavlakovic BN, Wilcox P. [55] Misaridis T, Jensen JA. Use of modulated excitation signals in
Practical long range guided wave testing: applications to pipes and medical ultrasound. Part I: basic concepts and expected benefits.
rails. Mater Eval 2003;61:66–74. IEEE Trans Ultrason Ferr Freq Cont 2005;52(2):177–91.
[46] Wilcox PD. A signal processing technique to remove the effect of [56] Clark M, Sharples SD, Somekh M. Fast, all-optical Rayleigh wave
dispersion from guided wave signals. IEEE Trans Ultrason Ferr Freq microscope: imaging on isotropic and anisotripic materials. IEEE
Cont 2003;50(4):419–27. Trans Ultrason Ferr Freq Cont 2000;47(1):65–74.
[47] Wilcox PD. Omni-directional guided wave transducer arrays for the [57] Karaman M, Bilge HS, O’Donnell M. Adaptive multi-element
rapid inspection of large areas of plate structures. IEEE Trans synthetic aperture imaging with motion and phase aberration
Ultrason Ferr Freq Cont 2003;50(6):699–709. correction. IEEE Trans Ultrason Ferr Freq Cont 1998;45(4):1077–87.
[48] Sicard R, Chahbaz A, Goyette J. Guided lamb waves and L-SAFT [58] Mahaut S, Roy O, Serre M. An adaptive system for advanced NDT
processing technique for enhanced detection and Imaging of applications using phased arrays. Ultrasonics 1998;36:127–31.
corrosion defects in plates with small depth-to-wavelength ratio. [59] Horn D, Mayo WR. NDE reliability gains from combining eddy-
IEEE Trans Ultrason Ferr Freq Cont 2004;51(10):1287–97. current and ultrasonic testing. Ultrasonics 2000;33:351–62.
[49] Sharples SD, Clark M, Somekh MG. All-optical adaptive acoustic [60] Hall ID, McNab A, Hayward G. Improved ultrasonic image
microscope. Ultrasonics 2003;41:295–9. generation through image fusion. Ultrasonics 1999;37:433–43.
[50] Chiao RY, Hao XH. Coded excitation for diagnostic ultrasound: a [61] Duck FA. Nonlinear acoustics in diagnostic ultrasound. Ultrasound
system developer’s perspective. IEEE Trans Ultrason Ferr Freq Cont Med Biol 2002;28(1):1–18.
2005;52(2):160–70. [62] Parker KJ, Taylor LS, Gracewski S. A unified view of imaging the
[51] Gan TH, Hutchins DA, Green RJ, Andrews MK, Harris PD. elastic properties of tissue. J Acoust Soc Am 2005;117(5):2705–12.
Noncontact, high-resolution ultrasonic imaging of wood samples [63] Dresser TS, Jesrzejewicz T, Bradley C. Native tissue harminic
using coded chirp waveforms. IEEE Trans Ultrason Ferr Freq Cont imaging: basic principles and clinical applications. Ultrasound Quart
2005;52(2):280–8. 2000;38:273–7.
[52] Nowicki A, Litniewski J, Secomski W, Lewin PA, Trots I. Estimation [64] Van Den Abeele KE-A, Sutin A, Carmeliet J, Johnson PA. Micro-
of ultrasonic attenuation in a bone using coded excitation. Ultra- damage diagnostics using nonlinear elastic wave spectroscopy.
sonics 2003;41:615–21. NDE&E Int 2001;34:239–48.
[53] Misaridis T, Jensen JA. Use of modulated excitation signals in [65] Prada C, Thomas JL, Fink M. The iterative time reversal process:
medical ultrasound. Part II: Design and performance for medical analysis of the convergence. J Acoust Soc Am 1995;97(1):62–71.
imaging applications. IEEE Trans Ultrason Ferr Freq Cont 2005; [66] Devaney AJ. Time reversal imaging of obscured targets from
52(2):192–207. multistatic data. IEEE Trans Antennas and Propag 2005;53(5):1600–10.
[54] Misaridis T, Jensen JA. Use of modulated excitation signals in [67] Simonetti F. Multiple scattering: the key to unravel the subwave-
medical ultrasound. Part IIII: High frame rate imaging. IEEE Trans length world from the far field pattern of a scattered wave. Phys Rev
Ultrason Ferr Freq Cont 2005;52(2):208–19. E 2006;73(3):036619.

You might also like