DOI: 10.1784/insi.2014.56.9.

492 UT SIMULATION

Obtaining geometries of real cracks and using an

efficient finite element method to simulate their
ultrasonic array response
Submitted 25.02.14
M V Felice, A Velichko, P D Wilcox, T Barden and T Dunhill Accepted 29.07.14

Stress corrosion cracks can be a serious issue in many 1. Introduction

industries, so it is imperative to be able to reliably
Stress corrosion cracking (SCC) results from the combined action
inspect for them. However, the branched nature of these
of a tensile stress and a corrosive environment. Furthermore, SCC
cracks causes difficulties when inspecting with a single
occurs at lower stress levels than are required for cracking in a
ultrasonic transducer due to the scattering of ultrasound
non-corrosive environment[1]. SCC can be a serious issue in several
in various directions. An ultrasonic array is more suited
industries, including nuclear power generation and aerospace, and
to this application because it can inspect at many angles.
affected components must be inspected using a reliable technique.
In order to design an array, it is necessary to understand
The particular motivation for this research work is a titanium
the ultrasonic scattering from these complex cracks. In this compressor disc from a turbofan engine, which has the potential
paper, images of real stress corrosion cracks are obtained to suffer from SCC. To save cost and disruption, the inspection
by performing X-ray computed tomography of cracked should be performed in situ or ‘on wing’, without disassembling
parts. Image processing software is used to extract the the engine. Access to the cracked surface is not possible in situ so
crack geometries. Next, the Kirchhoff approximation is ultrasonic inspection is selected, since it offers the best sensitivity
compared with an efficient frequency domain finite element for detecting sub-surface defects. However, stress corrosion
(FE) method, in terms of their ability to correctly simulate cracks have complex morphologies with bridges or connections
the ultrasonic scattering from such cracks. The Kirchhoff between crack surfaces; branches that propagate along different
approximation is dismissed because it does not simulate directions and multiple cracks in the same area[2]. These features
shadowing or multiple reflections between features, neither can scatter ultrasound in various directions, making them more
does it correctly simulate tip diffraction or scattering from unreliable to detect than simple cracks. The surface extents of three
small lengths, whilst the FE method correctly simulates all stress corrosion cracks obtained using fluorescent dye penetrant
these interactions. A hybrid model is implemented, which inspection are shown in Figure 1.
combines the FE method with simple ray tracing to obtain a
simulated ultrasonic array response. In particular, it is the
full matrix capture data that is simulated. The hybrid model
is validated using wire-cut branched shapes and is then
used to simulate array data from the real crack geometries,
which are approximately 4 mm deep. The paper concludes
with a discussion on how the hybrid model and real crack
geometries can be used to optimise an array design. This
can be achieved by running the hybrid model for arrays of Figure 1. Fluorescent dye penetrant images of stress corrosion
different parameters, such as the number of elements, and cracks. A scale for all three images is shown on the left
comparing the crack indications obtained with the different
Ultrasonic arrays offer many advantages over single-element
arrays.
transducers. They can be used to sweep over a range of angles[3]
and when using an array it is possible to obtain an image from a
single location, which can offer more reliable results than a time
trace[4]. These advantages and recent developments, such as probe
miniaturisation, have made ultrasonic arrays a suitable choice
for inspecting for SCC in situ. An efficient and effective way of
optimising an array design, especially for a challenging task, is to
use computer models. This reduces practical trial and error work,
which is time-consuming and expensive, and ensures that the best
Ms Maria V Felice* is with the Department of Mechanical Engineering, possible array inspection is implemented. However, it is common
University of Bristol, Bristol, UK, and the NDE Laboratory, Rolls-Royce practice in ultrasonic simulation models to represent cracks as
plc, Bristol, UK. She is studying for an Engineering Doctorate run through simple planar shapes. This may be justified for certain types of
the Industrial Doctorate Centre in NDE. crack, such as fatigue cracks, but not for stress corrosion cracks,
Dr Alexander Velichko and Professor Paul D Wilcox are with the which are multi-faceted.
Department of Mechanical Engineering, University of Bristol, Bristol, UK. The objective of the research work presented in this paper is
to develop an array design methodology that could be used for
Dr Tim Barden and Professor Tony Dunhill are with the NDE Laboratory, this engine disc inspection and for future tasks where the cracks
Rolls-Royce plc, Bristol, UK.
of concern are not of a simple shape. The research work involves
*Corresponding author. Email: [email protected] three stages, which are described in this paper:

492 Insight Vol 56 No 9 September 2014

q Real branched crack shapes are obtained, in this case from
images of stress corrosion cracks (Section 2).
q A computer simulation method that reliably simulates the
ultrasonic scattering from these cracks is then selected and
implemented (Section 3).
q The ultrasonic array response from these cracks is simulated
(Section 4).

2. Stress corrosion crack (SCC) geometry

2.1 Imaging SCC using X-ray computed tomography
Fluorescent dye penetrant inspection was used to locate a selection
of stress corrosion cracks for three-dimensional (3D) analysis.
The material around these cracks was cut away and the resulting Figure 3. Example images of a stress corrosion crack obtained
using (a) metallography, (b) fluorescent dye penetrant
sample was imaged using X-ray computed tomography (CT). The inspection and (c) X-ray CT. The white indication in (c) is from a
resolution obtained with X-ray CT increases as the size of the sample dense corrosion product
decreases, but obviously care must be taken to not inadvertently
cut away part of the crack. Therefore, a two-step approach was
adopted: X-ray CT was performed on a larger piece of material 3. Simulation of ultrasonic scattering from
to determine the extent of the crack and then the material was cut SCC
down further and scanned again to obtain a higher resolution. The
parameters of both scans and the corresponding voxel dimensions 3.1 Description of different scattering simulation
are shown in Table 1. In both cases, an X-Tek BladeRunner system methods
(Nikon Metrology Inc, Michigan, USA) was used with a 225 kV In this section, the work conducted to select a method to simulate the
source. plane wave ultrasonic scattering from branched cracks is described.
Table 1. X-ray CT parameters for different-sized samples
This was the first stage in the simulation of the ultrasonic array
response from branched cracks. Note that the longitudinal mode was
Parameter Larger sample Smaller sample considered in this paper. Scattering simulation results were stored
Approximate sample dimensions (mm) 40 × 30 × 20 10 × 30 × 20 in a scattering matrix (S-Matrix), which describes the amplitude of
scattered signals as a function of incident and scattered angles[5], as
Source-to-detector distance (mm) 1017 1017 shown in Figure 4 for the example of a slot. The incident amplitude
Source-to-sample distance (mm) 145 94 is equal to unity and the scattered amplitude that is plotted is the
Voxel dimension (μm) 28.5 18.4 amplitude at a distance of 1 wavelength from the scatterer. Note
that the S-Matrix is independent of the array being used and the
The crack geometry was then extracted from the X-ray CT ultrasound propagation paths, therefore the S-Matrix of a scatterer
images using ScanIP software (Simpleware Ltd, Exeter, UK), as can be reused for different inspection simulations (Section 4).
shown in the example in Figure 2. The image processing procedure
involved filtering and thresholding.

Figure 2. (a) X-ray CT image of a stress corrosion crack; (b)

extraction of the crack geometry from the X-ray CT image

2.2 Comparison of X-ray CT and metallography for Figure 4. The scattering matrix at 5 MHz for a slot of length
imaging SCC 1 wavelength, obtained using an analytical solution. The
scattered amplitude at a distance of 1 wavelength from the slot
Following the X-ray CT scans, the small material sample was for unit incident amplitude, for a specified incident angle θin and
mounted and analysed using incremental metallography with scattered angle θsc , can be read off the matrix
fluorescent dye penetrant inspection at each stage, so as to compare
the different techniques. The metallography was performed using a Analytical solutions, the Kirchhoff approximation and a finite
Leica DM2500 microscope (Leica Microsystems GmbH, Wetzlar, element (FE) technique were all considered and were implemented
Germany) and an example image is shown in Figure 3(a). The in MATLAB (MathWorks Inc, Massachusetts, USA). Analytical
sample was polished down in 0.25 mm increments. The images solutions for ultrasonic scattering only exist for simple shapes
obtained with metallography (Figure 3(a)) and dye penetrant such as side-drilled holes[6] and slots[7], which provide important
(Figure 3(b)) had a better contrast than those obtained using X-ray validation cases for other models. The Kirchhoff approximation can
CT (Figure 3(c)), but there was obviously a significantly lower be applied for high frequencies or for scatterers whose details are
resolution in the slicing direction (normal to the images shown in large in size compared to the wavelength[5]. It uses the assumption
Figure 3). In addition, the problem of low contrast with X-ray CT that when a plane wave is incident on a scatterer, the reflected wave
has been overcome by using image processing. Therefore, X-ray from each portion of the scatterer that it strikes directly is the same
CT in conjunction with image processing was selected as the as that from an infinite planar surface with the same normal.
method for extracting crack geometries to input into the computer An FE technique was considered because it was thought that a
model. more flexible and accurate technique than the analytical solutions

Insight Vol 56 No 9 September 2014 493

and Kirchhoff approximation would be needed due to the complex
nature of stress corrosion cracking. A common complaint about FE
techniques is that they are time-consuming, so to overcome this
an efficient, novel FE method that was developed by Velichko and
Wilcox was considered[8]. This method is a further improvement to
the finite element local scattering (FELS) method[9] and both are
performed in the frequency domain. The FELS method was more
efficient than conventional FE methods because only the scatterer
and the immediate space around it were modelled, instead of the
scatterer inside an entire part. However, an absorbing region was
required to stop the ultrasound waves from reflecting back onto the
scatterer and this region can contain up to 90% of all the elements in
the 3D model. A schematic of the FELS method is shown in Figure
5(a). The improved method[8] goes one step further by eliminating
the need for this absorbing region by implementing non-reflecting
boundary conditions. In addition, the elements containing the
loading nodes and monitoring nodes, which apply the incident
ultrasound and measure the scattered ultrasound, respectively, are
now located on the surface of the scatterer. Therefore, just one layer
of elements is required, which lies between the scatterer surface
and the boundary of the model, as shown in Figure 5(b).
Figure 6. The scattering matrix for a slot of length 1 wavelength
obtained using (a) an analytical solution, (b) the Kirchhoff
approximation and (c) the efficient FE method. Note the
discrepancy in (b) near θsc = +90° and θsc = –90°, which
corresponds to the fact that the Kirchhoff approximation does
not correctly simulate tip diffraction
The efficient FE method was selected because it does correctly
simulate the ultrasonic scattering from branched crack shapes.

4. Hybrid model for simulating ultrasonic

Figure 5. (a) Finite element local scattering method and (b) array response of SCC
improved finite element method with the single layer of elements
that is used in this paper 4.1 Implementation of hybrid model
The hybrid model combines the efficient FE method with simple ray
3.2 Comparison of different scattering simulation tracing to obtain a simulated ultrasonic array response. In particular,
methods it is the full matrix capture (FMC) data that is simulated. This is the
The first step was to understand how the FE and Kirchhoff set of time domain signals from every possible transmitter-receiver
methods performed for different-sized scatterers and for different element combination and is stored in an n × n matrix, where n is
inspection angles. These two methods and an analytical solution[7] the number of elements in the array. The FMC data is the complete
were used to simulate the S-Matrix of slots of different lengths. As dataset from an array, so any post-processing techniques can be
expected, the FE method was found to be more accurate, compared performed on it[11]. FMC data was simulated using the hybrid model
to the analytical solution, than the Kirchhoff approximation and so that no additional simulation runs would be necessary during
the accuracy of the latter decreased with decreasing slot length, post-processing.
whilst it was relatively unchanged for FE. With regards to different The hybrid model is implemented entirely in MATLAB, with
scattered angles, the FE method correctly simulated scattering the main calculations performed in the frequency, ω, domain. A
at all angles, while there was a discrepancy with the Kirchhoff five-cycle windowed toneburst is used as the array time domain
approximation for values of θsc close to ±90°, as shown in Figure input signal. At the start of the hybrid model, the frequency
6. This was expected because the approximation does not correctly spectrum of this input signal, uT(ω), is calculated using the fast
simulate the diffraction of ultrasound from the slot tips[10]. Fourier transform. The FE method is used to compute the S-Matrix
With the above studies completed, the work progressed to more for a particular scatterer (for example a crack) for all the component
complex cracks. An analytical solution does not exist for branched frequencies to obtain a stack of 2D S-Matrices or a 3D S-Matrix.
cracks, so a comparison between the FE method and the Kirchhoff Further calculations (shown below) are done in the frequency
approximation was performed to gain an understanding of how domain and, finally, the inverse fast Fourier transform is used to
the methods dealt with these cracks. Results showed that with the obtain the time domain array response from the crack.
Kirchhoff approximation the S-Matrix of a shape made of several Consider an array with a single element operating in transmission
branches was identical to the sum of the separate S-Matrices of each and a single element operating in reception in a 2D space. A
branch, whilst with the FE method this was not the case, as shown in scatterer is located so that the array is in its far field, which means
Figure 7. The FE results are in line with what is observed in practice. it is at a sufficient distance from the scatterer that the scattered field
For example, note the circled points of high amplitude close to exhibits no further radial dependence other than decay due to beam
θin = θsc = 0° in Figure 7(f), which correspond to multiple reflections spread. This scenario is represented in Figure 8, where all values of
between the branches. These points are not observed in Figure 7(e). r and θ are calculated using trigonometry.
The Kirchhoff approximation was dismissed because it does The incident signal, uin(ω), at the scatterer is given by:
not simulate shadowing or multiple reflections between features, ⎛ 1⎞
0.5

neither does it correctly simulate tip diffraction or scattering from uin (ω ) = e ikrT ⎜ ⎟ D (θ T ,ω ) uT (ω ) ...................(1)
small lengths. These are all interactions likely to occur with SCC, ⎝ rT ⎠
so it is important that a method is used that simulates them correctly. where k = ω/c, c is the ultrasonic velocity, rT is the distance

494 Insight Vol 56 No 9 September 2014

 between the centre of the scatterer and the transmitting element,
θT is the angle a line between the scatterer and the transmitting
element makes with the transmitting element normal and D(θ,ω)
is array element directivity for the longitudinal mode. The first
term in Equation (1) caters for the phase shift due to the ultrasound
propagation path and the second term caters for beam spread.
D(θ,ω) accounts for the fact that array elements have the highest
sensitivity when the transmitted or received ultrasound is normal to
their surface (θ = 0°) and this sensitivity decreases as |θ| increases.
It is approximated by:
⎛1 ⎞
sin ⎜ kasin θ ⎟
⎝2 ⎠
D (θ ,ω ) ≈ cosθ .......................(2)
1
kasin θ
2
where a is the array element width.
The scattered signal, usc(ω), at a distance from the centre of the
scatterer equal to 1 wavelength, λ, is given by:
usc (ω ) = S (θ T ,θ R ,ω ) uin (ω ) ........................(3)
where θR is the angle a line between the scatterer and the receiving
element makes with the receiving element normal and S is the
S-Matrix. In other words, the incident signal is multiplied by the
single value from the S-Matrix, which corresponds to the correct
incident angle, scattered angle and frequency.
Finally, the received signal, uR(ω), is given by:
0.5
⎛ 2π ⎞
uR (ω ) = e ikrR ⎜ ⎟ D (θ R ,ω ) usc (ω ) ..................(4)
⎝ krR ⎠
where rR is the distance between the centre of the scatterer and the
receiving element. Note that if the array is in the near field of the
scatterer, additional calculations must be used[9] but the required
information is stored in the same scattering matrix, S.
The above calculations (Equations (1) to (4)) are repeated for all
possible transmitter-receiver element combinations and the inverse
fast Fourier transform is used to obtain the time domain signals. In
this way, the FMC data is simulated.
One post-processing technique that utilises all the FMC data
is the total focusing method (TFM). This method synthetically
focuses ultrasonic energy at every pixel or voxel to produce a high-
quality image[11] and is the method used in this paper.
Figure 7. Scattering matrices for branched shapes; 0° is
vertical as in Figure 4: (a) schematic of case A: scattering
4.2 Comparison of hybrid model and experimental
from each of the two branches is modelled separately and results
the results summed; (b) S-Matrix for case A obtained using The efficient FE method has previously been validated using
Kirchhoff approximation; (c) S-Matrix for case A obtained using
FE method; (d) schematic of case B: scattering from the two simple shapes embedded in an infinite host medium (not close to
branches together is modelled; (e) S-Matrix for case B obtained any surface) by comparison to analytical solutions[8]. The steps of
using Kirchhoff approximation; and (f) S-Matrix for case B the hybrid model that involved ray tracing (but not the FE method)
obtained using FE method were validated as follows. The S-Matrix of a side-drilled hole was
obtained using an analytical solution[6] and was therefore known to
be correct. This was then used in the hybrid model, FMC data was
simulated and a TFM image was computed. The image agreed well
with that obtained experimentally and hence demonstrated that the
ray tracing part of the hybrid model was implemented correctly.
Analytical solutions do not exist for branched shapes, nor for any
shapes that are near-surface or surface-breaking. Therefore, in these
cases the complete hybrid model must be validated by comparison
with experimental results. This is what is done in this paper for
branched embedded shapes (Section 4.2.1) and for branched surface-
breaking shapes (Section 4.2.2). Such experimental validation has
previously been performed for simple surface-breaking shapes[12].
Note that for near-surface and surface-breaking scatterers, the same
FE method that is used for embedded scatterers is used. However,
Figure 8. Schematic diagram for the hybrid model showing the information about the scattering between the scatterer (for example
transmitting element marked with ‘T’, the receiving element a crack) and the surface is stored in the S-Matrix, along with
marked with ‘R’ and the scatterer represented by a circle information about the direct scattering from the crack.

Insight Vol 56 No 9 September 2014 495

4.2.1 Embedded scatterers Table 2. Array parameters
The hybrid model, with the efficient FE method, was run for Parameter Value
a branched shape embedded in aluminium at a depth of 20 mm
from the array described in Table 2. The set-up is shown in Figure Centre frequency (MHz) 5
9(a). Full matrix capture data was simulated and a TFM image was Bandwidth (–6dB) (MHz) 3-7
computed (Figure 9(d)). An experiment was then performed using Number of cycles in input pulse 5
the same array and an aluminium block containing the same shape
Number of elements 64
manufactured with wire cutting (Figure 9(b)). Time signals were
truncated so that reflections from the backwall were not included in Element layout Linear
the calculation of the TFM image (Figure 9(c)), since the backwall Element pitch (mm) 0.63
was not included in the model. The horizontal portion of the
shape appears slightly curved in the simulated TFM image. This
4.2.2 Surface-breaking scatterers
discrepancy is due to the fact that the shape is large so the array is
in its near field at a distance of 20 mm, and an assumption in the A similar procedure to that described in Section 4.2.1 was
hybrid model (Equation (3)) is that the array is in the far field of the performed for a surface-breaking shape using the same array (Table
scatterer. An improvement is obtained when the shape is simulated 2). The set-up is shown in Figure 10(a) and the wire-cut shape is
to be further from the array (Figure 9(e)). The images can be shown in Figure 10(b).
compared quantitatively, for example by measuring the horizontal The TFM image obtained experimentally is shown in Figure
branch indications using the 6 dB drop rule. The horizontal length 10(c) (detail in Figure 10(d)) and the TFM image obtained from the
of the indication is 7.3 mm in the experimental image, 7.1 mm in simulation is shown in Figure 10(e) (detail). The two TFM images
the simulated image at 20 mm depth and 7.2 mm in the simulated show the same salient features, including a reduction in amplitude
image at 40 mm depth. The vertical distance between the centre of in the backwall. The simulation was repeated using the geometry
the branch indication and its highest point is 0.6 mm in both the of the wire-cut shape as it was manufactured (Figure 10(b)), not as
experimental image and in the simulated image at 40 mm depth, it was designed (Figure 10(a)), and an improved image is obtained
whilst it is 0.9 mm in the simulated image at 20 mm depth because (Figure 10(f)). Comparing the images quantitatively, in both the
of the curvature. These values show that the hybrid model results simulated images the centre of the shape is 2.5 mm above the
agree well with experimental results and that an improvement is backwall and the two tips are 5 mm above the backwall, as in the
obtained when the array is further from the scatterer. experimental image. There is a 5 dB difference between the tip
In addition to a comparison with experimental results, indications in the experimental image, a 0 dB difference between
convergence studies were performed to determine the optimum the tip indications in the simulated image of the ‘as designed’ shape
mesh density and optimum number of frequencies, for a given
bandwidth, at which to compute the S-Matrix.

Figure 10. Comparison of hybrid model and experimental

results for surface-breaking scatterers: (a) schematic of the
Figure 9. Comparison of hybrid model and experimental results ultrasonic array and surface-breaking shape (dimensions in
for embedded scatterers: (a) schematic of the ultrasonic array mm); (b) photograph of the wire-cut surface-breaking shape;
and embedded shape (dimensions in mm); (b) photograph of the (c) experimental TFM image of the shape and backwall; (d)
test-block with the wire-cut embedded shape; (c) experimental detail of (c); (e) simulated TFM image of the ‘as designed’ shape
TFM image of the shape; (d) simulated TFM image of the shape; and backwall (detail); and (f) simulated TFM image of the ‘as
and (e) simulated TFM image of the shape at depth 40 mm. manufactured’ shape and backwall (detail). Note that in the TFM
Note that in the TFM images the amplitude scale is in dB and is images the amplitude scale is in dB and is normalised to the
normalised to the maximum maximum

496 Insight Vol 56 No 9 September 2014

and a 4 dB difference between the tip indications in the simulated simulates the response from real cracks. These two advantages are
image of the ‘as manufactured’ shape. This shows that the hybrid now discussed in turn.
model correctly accounts for small changes in shape geometry. With the hybrid model approach, the finite element method
needs to be run just once per crack shape. This takes in the order of
4.3 Use of real crack shapes in hybrid model tens of minutes, with the exact duration depending on the number
Following comparison with experimental results for wire-cut of frequencies at which the scattering matrix is computed. The
shapes, the hybrid model was used to successfully simulate the scattering matrices are stored and inputted into the ray tracing part
ultrasonic array response from a 2D slice of a real stress corrosion of the model, which needs to be run just once for each particular
crack geometry, obtained using X-ray CT (Figure 11(a)) and array because the full matrix capture (FMC) data is simulated.
imported into MATLAB. The outline of the crack shape was This data can be inputted into different array imaging algorithms,
extracted and a second, outer boundary was plotted around it. A such as the total focusing method (TFM). The time taken for these
mesh was created between the crack outline and the outer boundary stages is of the order of seconds or a few minutes, depending on
as required for the efficient FE method within the hybrid model the size of the array and the image resolution required. Preliminary
(Figure 11(b)). A simulated TFM image obtained for the crack comparison work has shown that when using a commercial finite
using the hybrid model with the array described in Table 2 is shown element software package, the time taken to run a similar simulation
in Figure 11(c). is in the order of hours, and when an array parameter is changed,
the entire model must be run again.
The proposed methodology involves moving away from the
common practice of representing cracks as simplified shapes.
Using real crack geometries from X-ray CT images contributes to
more realistic results, which will enable an optimised array to be
designed and will improve the ability to quantify the inspection
performance prior to implementation. In addition, since the
geometry of the cracks is converted into a series of coordinates, it
is possible to apply realistic variations to the geometry. In this way,
a quantitative understanding of how the developed technique will
cope with different cracks will also be understood.
Two-dimensional (2D) arrays can be used to image a 3D space
and would therefore offer advantages in imaging stress corrosion
cracks, which are 3D in nature. The simulation work in this paper
has all been in 2D, since this provided a computationally and
experimentally simpler demonstration of the novel methodology.
However, the methodology could be extended into 3D and used to
design 2D arrays. The efficient FE method has been developed for
problems in 3D[8,12] and its efficiency is even more evident in 3D. It
makes 3D finite element analysis a viable technique for ultrasonic
simulations in a commercial environment.

6. Conclusion
In this paper, the tools required for the efficient simulation of
ultrasonic array data from real stress corrosion crack shapes have
been developed. The Kirchhoff approximation and an efficient
finite element method were compared with regards to their ability
to correctly simulate the ultrasonic scattering from branched crack
Figure 11. (a) Stress corrosion crack marked on an X-ray CT shapes, and the FE method was selected. The FE method was
image; (b) outline of the crack and single layer of elements incorporated into an efficient hybrid model, which simulates the
around it (as inputted into hybrid model); and (c) simulated TFM ultrasonic array response from cracks. The hybrid model results
image of the crack and backwall were compared to experimental results for branched shapes and
showed good agreement. A repeatable way of extracting real crack
geometry from X-ray CT images in a form that can be inputted into
5. Use of methodology for array design the hybrid model was established. The ultrasonic array response
The work described in this paper can be exploited in industry as from a real crack shape was successfully simulated. Finally, an
an array design methodology. The ultrasonic array response from array design methodology, which makes use of the efficient hybrid
real crack geometries for different array parameters and imaging model and real crack shapes, was outlined. The methodology is
techniques can be obtained. These can be compared quantitatively, not only applicable to stress corrosion cracking problems but to
for example by comparing the amplitudes of the crack indications any straight or branched cracking issues. The methodology is
relative to a reference indication, such as that from the backwall. In applicable to industrial scenarios where a reliable and bespoke
this way, the best array parameters and mode of operation can be ultrasonic array is required for a cracking issue.
determined. In the case of in situ inspections, certain parameters,
Acknowledgements
such as array size, are limited by the geometry and assembly of
components. These constraints can easily be included in the M V Felice is supported financially by the Industrial Doctorate
computer model to ensure that no time is wasted developing an Centre in NDE and the Royal Commission for the Exhibition of
impractical array. 1851. The dye penetrant and X-ray CT work mentioned in this
Two advantages of the methodology, which make it viable paper was carried out by members of the NDE team at Rolls-Royce
for use in an industrial setting, are its speed and the fact that it plc, Bristol.

Insight Vol 56 No 9 September 2014 497

References analytically-based computer model for surface measurements
in ultrasonic crack detection’, Wave Motion, Vol 43, No 6, pp
1. W D Callister, Materials Science and Engineering: An 458-473, June 2006.
Introduction, John Wiley and Sons, New York, 2007. 8. A Velichko and P D Wilcox, ‘Efficient finite element modelling
2. W Cheng, S Kanemoto, I Komura and M Shiwa, ‘Depth sizing of elastodynamic scattering with non-reflecting boundary
of partial-contact stress corrosion cracks from ECT signals’, conditions’, AIP Conference Proceedings, 1430, pp 142-149,
NDT&E International, Vol 39, No 5, pp 374-383, July 2006. 2012.
3. B W Drinkwater and P D Wilcox, ‘Ultrasonic arrays for non- 9. P D Wilcox and A Velichko, ‘Efficient frequency domain
destructive evaluation: a review’, NDT&E International, Vol finite element modelling of two-dimensional elastodynamic
39, No 7, pp 525-541, October 2006. scattering’, Journal of the Acoustical Society of America, Vol
4. S J Song, H J Shin and Y H Jang, ‘Development of an ultrasonic 127, No 1, pp 155-165, January 2010.
phased array system for non-destructive tests of nuclear power 10. V Dorval, S Chatillon, B Lu, M Darmon and S Mahaut, ‘A
plant components’, Nuclear Engineering and Design, Vol 214, general Kirchhoff approximation for echo simulation in
No 1-2, pp 151-161, May 2002. ultrasonic NDT’, AIP Conference Proceedings, 1430, pp 193-
5. L W Schmerr, ‘Fundamentals of ultrasonic non-destructive 200, 2012.
evaluation: a modelling approach’, Plenum, New York, 1998. 11. C Holmes, B W Drinkwater and P D Wilcox, ‘Post-processing
6. A L Lopez-Sanchez, H J Kim, L W Schmerr and A Sedov, of the full matrix of ultrasonic transmit-receive array data for
‘Measurement models and scattering models for predicting non-destructive evaluation’, NDT&E International, Vol 38, No
the ultrasonic pulse-echo response from side-drilled holes’, 8, pp 701-711, December 2005.
Journal of Nondestructive Evaluation, Vol 24, No 3, pp 83-96, 12. A Velichko, P D Wilcox and B W Drinkwater, ‘Detection of
September 2005. near-surface and surface-breaking defects using ultrasonic
7. E Glushkov, N Glushkova, A Ekhlakov and E Shapar, ‘An arrays’, AIP Conference Proceedings, 1430, pp 929-936, 2012.

498 Insight Vol 56 No 9 September 2014