Original Article

Structural Health Monitoring

2017, Vol. 16(4) 382–399

Improved acoustic emission source Ó The Author(s) 2016

location during fatigue and impact Reprints and permissions:

sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1475921716672206
events in metallic and composite journals.sagepub.com/home/shm

structures

Matthew R Pearson, Mark Eaton, Carol Featherston, Rhys Pullin and

Karen Holford

Abstract
In order to overcome the difficulties in applying traditional time-of-arrival techniques for locating acoustic emission
events in complex structures and materials, a technique termed ‘Delta-t mapping’ was developed. This article presents a
significant improvement on this, in which the difficulties in identifying the precise arrival time of an acoustic emission sig-
nal are addressed by incorporating the Akaike information criteria. The performance of the time of arrival, the Delta-t
mapping and the Akaike information criteria Delta-t mapping techniques is assessed by locating artificial acoustic emis-
sion sources, fatigue damage and impact events in aluminium and composite materials, respectively. For all investigations
conducted, the improved Akaike information criteria Delta-t technique shows a reduction in average Euclidean source
location error irrespective of material or source type. For locating Hsu–Nielsen sources on a complex aluminium speci-
men, the average source location error (Euclidean) is 32.6 (time of arrival), 5.8 (Delta-t) and 3 mm (Akaike information
criteria Delta-t). For locating fatigue damage on the same specimen, the average error is 20.2 (time of arrival), 4.2
(Delta-t) and 3.4 mm (Akaike information criteria Delta-t). For locating Hsu–Nielsen sources on a composite panel, the
average error is 19.3 (time of arrival), 18.9 (Delta-t) and 4.2 mm (Akaike information criteria Delta-t). Finally, the Akaike
information criteria Delta-t mapping technique had the lowest average error (3.3 mm) when locating impact events
when compared with the Delta-t (18.9 mm) and time of arrival (124.7 mm) techniques. Overall, the Akaike information
criteria Delta-t mapping technique is the only technique which demonstrates consistently the lowest average source
location error (greatest average error of 4.2 mm) when compared with the Delta-t (greatest average error of 18.9 mm)
and time of arrival (greatest average error of 124.7 mm) techniques. These results demonstrate that the Akaike informa-
tion criteria Delta-t mapping technique is a viable option for acoustic emission source location, increasing the accuracy
and likelihood of damage detection, irrespective of material, geometry and source type.

Keywords
Acoustic emission, source location, structural health monitoring, damage detection, signal processing

Introduction non-operational ground time is due to scheduled and

periodic inspections.2 SHM systems allow maintenance
Structural health monitoring (SHM) systems allow in- to be conducted when required rather than based on
service structures to be continuously monitored using operational hours. This increases asset use and reduces
the permanently mounted sensors and are akin to the operating costs, resulting in the most efficient use of
human nervous system.1 These systems allow for the the aircraft while still maintaining reliability and safety.
early detection of damage and in the future, if applied
appropriately, would enable damage characterisation
and the calculation of the remaining life. With the Cardiff School of Engineering, Cardiff University, Cardiff, UK
increasing use of composite materials and ageing air-
Corresponding author:
craft, the aerospace industry is one particular sector
Matthew R Pearson, Cardiff School of Engineering, Cardiff University,
that would benefit from wide-scale application of Queen’s Buildings, The Parade, Newport Road, Cardiff CF24 3AA, UK.
SHM systems. A substantial proportion of aircraft Email: [email protected]
Pearson et al. 383

Composites materials are susceptible to large inter- arrays, statistical, beam-forming and mapping
nal defects which may not be visually detectable, signif- techniques.
icantly affecting the designed strength and component Dispersion curves or experimentally derived wave
life.3 This leads to conservative designs resulting in hea- velocities for different propagation angles are often uti-
vier components and reducing the strength-to-weight lised for location techniques that require prior knowl-
ratio advantage of composite materials. The introduc- edge of material wave speeds. Mostafapour et al.5 used
tion of SHM systems at the design stage would remove wavelet decomposition, cross-time–frequency spec-
the need for conservative design, leading to more opti- trums and the most energetic frequency velocity from
mised structures, reducing aircraft weight and improv- dispersion curves to locate sources in a steel plate.
ing performance. SHM systems can also be retrofitted Often the wave velocities in two principal directions are
to ageing aircraft and it is thought that the long-term used to locate sources in the composite materials, such
benefit to maintenance could easily offset the initial as the elliptical wave front triangulation method6 which
outlay.3 These systems can be beneficially used to mon- was combined with Rayleigh maximum likelihood esti-
itor inaccessible hotspot areas which are labour- mate7 to locate impacts in composite aircraft panels.8
intensive to inspect using conventional techniques due Koabaz et al.9 used an experimentally derived wave
to the need for disassembly of components. velocity and developed the work of others10 to realise
One particular non-destructive technique (NDT) an improved objective function in an optimisation
that is particularly suited to SHM is acoustic emission scheme to locate impacts on a composite panel. For
(AE). AE is described as the rapid release of energy that structures with increasing complexity, Geiger’s method
propagates in a material in the form of an elastic stress and a variable velocity model were used to locate
wave due to permanent internal changes. A variety of micro-cracks in cemented total hip arthroplasty.11
mechanisms can give rise to AE, including crack initia- Often the use of dispersion curves or experimentally
tion and propagation, plastic deformation, fretting, derived wave velocities is not practical in reality.
matrix cracking and fibre breakage. These stress waves Techniques have been developed that require no fur-
cause minute displacements at the surface of the mate- ther knowledge of the material properties. Ciampa and
rial that can be detected using piezoelectric sensors and Meo12 used a specific triangular layout of six sensors
are recorded on an acquisition system based on a user- arranged in three closely spaced pairs, reducing the
defined threshold. AE is seen as a passive approach, number of non-linear equations, which were solved
essentially ‘listening’ to active sources within the using Newton’s method. This technique was further
structure. applied to composite sandwich panels with an arbitrary
layup.13
All the techniques discussed thus far have focussed
AE source location on using sparsely distributed sensors and locating
One advantage of AE for SHM is its ability to locate within the sensor array. Researchers have developed
the source; the most common commercial algorithm techniques that utilise specific closely spaced sensor
for this is the time of arrival (TOA) method.4 However, arrays to enable global location outside the sensor
due to the assumptions in the algorithm, significant array. Aljets et al.14 used three closely spaced sensors in
errors can be introduced. One of these assumptions is a triangular array to locate sources globally in a com-
that the structure is homogenous, and that the wave posite plate. The propagating angle from the array was
velocity is the same in all directions. This is certainly determined using the A0 wave arrival time using a
not the case in composite materials and complex struc- wavelet transform (WT) at a specific frequency.
tures where velocity varies with propagation angle, due Knowing the maximum and minimum S0 wave veloci-
to the fibre layup in composite materials or geometric ties for different propagating angles enabled the formu-
features which give rise to an interrupted propagation lation of a numerical model. This, combined with the
path. The user-defined threshold technique to deter- temporal separation of the two wave modes, deter-
mine arrival times, common to commercial AE acquisi- mined the distance along the propagation angle using
tion systems, can introduce significant errors in single sensor location techniques.15 Matt and Lanza Di
location accuracy due to attenuation, source amplitude Scalea16 used a rosette arrangement of three macro
and dispersion which make it difficult to determine the fibre composites sensors, exploiting the directional
wave arrival time with precision. response of these sensors to determine the propagating
There has been a vast amount of research focussed angle of the flexural mode. Utilising at least two
on addressing these assumptions and therefore improv- rosettes allowed source location in two dimensions.
ing source location. The proposed techniques can be Kundu et al.17 used two clusters of a specific closely
categorised into the following groups: wave speed is spaced triangular array of three sensors to determine
known or unknown, specific closely spaced sensor the source location based on the intersection of the two
384 Structural Health Monitoring 16(4)

calculated propagating angles. This technique required matched point search method was developed by
no prior knowledge of the wave speed. The introduc- Scholey et al.;30 an array of points was used to repre-
tion of an optimisation scheme using the initial esti- sent the geometry. Matching the theoretical and experi-
mate from the intersection further improved the mental difference in arrival times, by minimising the
technique.18 error, it was possible to estimate source location.
Delay and sum beam-forming is another technique Hensman et al.31 used a laser to generate artificial AE
that incorporates a closely spaced sensor array to deter- sources on a structure and the Akaike information cri-
mine global source locations which has been used on teria (AIC) picker to determine the arrival times;
large-scale bridge structures19 and in metallic plates.20 Gaussian processes were used to directly relate the dif-
Xiao et al.21 used two linear sensor arrays to perform ference in arrival times to Cartesian coordinates.
delay and sum beam-forming to give accurate source
locations irrespective of wave velocity.
The majority of recent research has focussed on Onset picking
probabilistic source location estimates rather than
In order to reduce source location errors, accurate arri-
deterministic solutions. This has allowed calculation of
val time estimation is crucial. Traditionally, the first
the source location confidence, and many researchers
crossing threshold technique is used and consideration
have incorporated Bayesian statistics. A Bayesian
of the chosen threshold is of paramount importance.
approach was used to estimate source locations in a
Reducing the threshold level can improve location
concrete beam.22 A simplified predictive model was cre-
accuracy; however, early triggering could occur.
ated which described the source location estimations
using four parameters represented by a probability den- Filtering, WTs and statistical approaches have been
sity function (PDF). Bayesian statistics and Markov used to improve source location accuracy. Cross-corre-
chain Monte Carlo (MCMC) methods have been used lation has been used to calculate difference in arrival
with a ray tracing model to locate AE sources in liquid- times by modulating the sensor output to emphasise
filled storage tanks23 and with data fusion techniques the phase difference of a single frequency.32 Knowing
to locate sources on a stiffened aluminium panel.24 that the peak of the wavelet coefficient at a particular
Kalman filters have also been utilised for probabilistic frequency could be used to determine the group velo-
AE source location. Extended Kalman filters (EKFs) city paved the way for WT arrival time estimation.33
alongside parameter extraction, binary hypothesis test- Jeong and Jang34 used this principle to locate Hsu–
ing and data weighting were used as a data fusion Nielsen (H–N) sources35 on a large aluminium panel.
approach to locate in noisy environments.25 Both EKF However, the accuracy of WT arrival time picks
and unscented Kalman filters (UKFs) were used to becomes far less reliable in the presence of reflections
locate impacts on a composite plate for conditions from structural features and boundaries and its perfor-
where the wave velocity was known and unknown.26 mance is poor at low signal-to-noise ratios.36 Ding
The UKF is seen as a performance improvement over et al.37 used wavelet decomposition to identify energetic
the EKF.27 The EKF can suffer from poor perfor- frequencies within a signal at which they band-pass fil-
mance and diverge for highly non-linear problems due tered the signal prior to arrival time detection by a
to the distribution being represented by a first-order threshold. The filter bands used were quite wide
linearisation of the system. The UKF addresses these (150 kHz), and arrival times could still be picked at a
problems by achieving the approximation through eval- range of different frequencies (and hence velocities);
uating the non-linear function with carefully chosen therefore, threshold crossing errors due to attenuation
sample points.28 Niri et al.29 used unscented transfor- can still occur. Shehadeh et al.38 presented a sliding
mation for source location and compared the results window energy technique that uses a change in the
with MCMC methods using Kullback–Leibler diver- ratio of signal energy above and below a user-defined
gence. The technique gave accurate results and required frequency level. The sliding window energy technique
a lower number of samples. Although there are advan- was compared with wavelet, cross-correlation and
tages of being able to determine the confidence bounds threshold arrival time determination techniques to mea-
of the estimated source locations, these statistical sure the wave speed in a long steel pipe. The sliding
approaches often take a large amount of processing window energy technique was shown to compare most
time and at the moment would not be suited to real- favourably to manual arrival time picks. However, the
time monitoring of actual damage mechanisms. authors showed that the technique is sensitive to the
The final approach is the use of mapping techniques. splitting frequency selected and this is likely to vary for
Most solutions presented so far assume an uninter- differing materials, sensors and sources. A method of
rupted source to sensor propagation path which may automatically detecting the observed changes is also
not be applicable in complex structures. The best required.
Pearson et al. 385

Other researchers have focussed on the use of statisti- monitor large structures but more focus on ‘hotspot’
cal methods to determine the first signal motion above areas with complex geometry or known stress concen-
background noise. Lokajicek and Kilma39 utilised trations. A summary of the main steps in the technique
higher order statistics within a sliding window. When is given in Figure 1.
the window contained noise prior to signal onset, the Baxter et al. showed that for locating H–N sources
distribution was normal. When signal samples were on an aircraft component, a reduction in error from
present in the sliding window, the distribution is dis- 4.81% to 1.77% was observed when compared with the
torted and the statistical moments are seen to change. TOA technique. Further robustness testing and loca-
The change in the derivative of the sixth-order statistical tion of fatigue damage in composite structures have
moment was detected using the ratio of the short-time been demonstrated.50 The major disadvantage of this
average to long-time average (STA/LTA) approach40 technique is that the first threshold crossing technique
and an arrival time estimation accuracy of 62 sample defines the arrival times calculated at the sensors.
points was demonstrated for 95% of events when com-
pared with manual picks. Kurz et al. utilised a modified
form of AIC41,42 proposed by Maeda for direct applica- Improved AIC Delta-t mapping
tion to raw transient signals. The technique compares
It is proposed that significant improvement can be
signal entropy before and after each point t in a signal
made to the Delta-t mapping technique by introducing
and returns a minimum at the point of signal onset.
a more robust approach to arrival time estimation. The
This occurs when high-entropy uncorrelated noise prior
AIC-based arrival time estimation discussed above is
to signal onset is compared with low-entropy waveform
selected for use in this work for a number of reasons.
showing marked correlation after signal onset. A simple
Researchers have shown AIC to compare very favour-
minimum finding function can then be used to deter-
ably to manually picked arrival times and to consis-
mine the signal arrival time. Further details on the
tently outperform other arrival time estimators (as
implementation of the AIC technique are given later in
discussed above). The AIC function returns a single
this article. The AIC technique was compared to the
minimum at signal onset that can be easily identified,
Hinkley43 criteria, proposed by Grosse,44 which com-
with no ambiguity, by a simple minimum finding func-
putes the partial energy of the signal for each point i,
tion. For many other arrival time estimators that detect
and by applying a negative trend, a minimum is
a change in a particular metric at the signal onset time,
observed at the point of signal arrival. Kurz et al.45 con-
there remains a requirement to automatically identify
cluded that the AIC function was superior to the
the point of change which commonly relies on thresh-
Hinkley criteria across a range of signal-to-noise ratios
olds that are inherently ambiguous. The AIC function
and showed that 90% of events were located within
operates with relatively low computational cost com-
5 mm of those computed from manual onset picking,
pared with more complex arrival time estimators, for
compared with only 40% for the Hinkley criteria.
example, those based on WTs. It has been shown to be
Carpinteri et al.46 showed similar performance of the
robust across a range of signal-to-noise ratios and sig-
AIC onset picker in concrete beam tests. Sedlak and
nal types. The main challenge in implementing the AIC
colleagues47,48 proposed a two-step AIC approach that
approach is the selection of an appropriate window
resulted in computed locations that were on average
length over which to apply the function; too long and
within 1 mm of those computed from manual picks. It
multiple minima may occur, and too short and sensitiv-
is questionable if the increased processing required is
ity may be reduced. A detailed discussion of the imple-
necessary, given the accuracy reported by others for the
mentation of the AIC function follows.
standard implementation of the AIC function.
This work uses an adaptation of the AIC implemen-
tation described by Kurz et al.45 Whereas Kurz utilised
Hilbert and WTs to approximate the signal onset time
Delta-t mapping
prior to applying the AIC function to a reduced win-
The traditional Delta-t mapping technique was devel- dow, here the first threshold crossing from the raw
oped by Baxter et al.49 and was developed for source transient was used as an initial approximate, thus fur-
location in complex structures. It uses H–N sources to ther reducing computational demand. Kurz et al. then
generate artificial AE signals in the structure in order took a window of 400 samples prior to the approximate
to generate arrival times at the sensors. This creates onset and 150 samples after approximate onset within
maps of the difference in arrival time (Delta-t) between which to apply the AIC function (equation (1)). Here,
sensor pairs. Test data are compared to the training the same function is applied to a window from the
maps in order to estimate the source location. It was waveform start (500 ms prior to threshold crossing) to
not envisaged that the technique would be used to 150 sample points after the threshold crossing.
386 Structural Health Monitoring 16(4)

Figure 1. Delta-t mapping technique.

Figure 2. Arrival time comparison for (a) high- and (b) low-amplitude AE signals.

AIC(t) = t log10 ðvarðfx(1 : t)gÞÞ source and then trying to locate a lower amplitude
ð1Þ sources. Kurz et al. demonstrated their implementation
+ (T  t  1)log10 ðvarðfx(t : T )gÞÞ
to be robust across a range of signal-to-noise ratios and
Equation 1 shows the AIC function where var is the similar is found here with favourable comparison to
classic variance of a given vector. As can be seen, the manual picks observed for a variety of signals from dif-
AIC function splits the signal into two vectors, {x(1:t)} fering sources, materials and sensor types. Table 1 com-
and {x(t:T}, and describes the similarity between them. pares 45 signals from a range of tests in an aluminium
When point t is aligned with the signal onset, the vector and composite specimen detailed in the experimental
prior to t contains only high-entropy uncorrelated noise procedure. In practice, it is only the window size after
and the vector after t contains only low-entropy signal the signal onset that is important for the AIC process,
with marked correlation and the AIC function therefore with 150 samples after onset coupled with varying sam-
returns a minimum. Figure 2 shows an implementation ple numbers prior to the onset, that is, 500 ms at varying
of this approach on a high-amplitude signal from an H– samples rates, still yielding accurate results (Table 1).
N source and a low-amplitude signal from a fatigue test The steps in the improved AIC Delta-t mapping are
with threshold crossing errors of a few samples, and outlined in Figure 3 with the improvements highlighted
approximately 100 samples are seen, respectively, fur- in grey. This article showcases the improvement of the
ther highlighting the potential errors that may occur AIC Delta-t technique over conventional TOA and the
when training Delta-t maps with a high-amplitude H–N traditional Delta-t mapping technique for a variety of
Pearson et al. 387

Table 1. Arrival time estimation in terms of samples for TOA and AIC when compared with manual picking

Experimental test TOA AIC

In terms of samples
Average error Standard deviation Average error Standard deviation

Aluminium specimen 65.5 129.9 2.6 3.0

Composite specimen 23.8 17.5 1.6 2.5

TOA: time of arrival; AIC: Akaike information criteria.

Figure 3. Improved AIC Delta-t mapping.

situations: locating H–N source in the composite and overall specimen dimensions were 370 3 200 mm with
metallic structures, fatigue damage in an aluminium a thickness of 3.18 mm. A series of different diameter
specimen and impacts in a composite specimen; it is the holes were machined into the specimen to ensure a
first time comparisons for all three technique have been complex and interrupted wave propagation path. For
made for locating a variety of source mechanisms in a both the Delta-t mapping techniques, an area of inter-
variety of materials. est was defined by placing a 200 3 160 mm grid on the
structure with a nominal node spacing of 10 mm. In
Experimental procedure order to monitor any subsequent AE activity, four
Mistras Group Limited (MGL) Nano 30 sensors (oper-
Three experimental investigations were undertaken to
ating frequency of 125–750 kHz) were coupled to the
demonstrate the performance enhancement of the AIC
specimen using silicon RTV (Loctite 595) and left to
Delta-t mapping over the Delta-t mapping and TOA
cure for 24 hours. MGL in-line preamplifiers with a
techniques. Typical aerospace damage mechanisms
gain of 40 dB were used alongside an MGL PCI-2
were investigated: fatigue cracking in aluminium and
acquisition system. AE waveforms were recorded using
impact events in carbon fibre composite. An additional
a 40-dB threshold, 100–1200 kHz analogue filters and
large-scale test was undertaken using H–N sources on a
a 2-MHz sample rate. Prior to the testing, training data
complex geometry aluminium aerospace component.
were collected for the Delta-t techniques. The training
These tests were devised to show the robustness of the
process for the Delta-t maps was completed once on
AIC Delta-t for locating damage irrespective of mate-
the pristine sample and the subsequent location calcula-
rial type, damage mechanism and component scale.
tion of all damage events was undertaken using this sin-
gle set of training data, without further updating. A
Aluminium fatigue specimen series of five H–N sources were made at each grid posi-
A complex geometry specimen was manufactured from tion and this generated artificial AE sources enabling
aerospace grade aluminium (2024-T3; Figure 4). The sensor arrival times to be determined. Load
388 Structural Health Monitoring 16(4)

Figure 4. 2024-T3 complex geometry specimen.

distribution into the specimen was achieved using 5-

mm steel clamping plates, and a 20-mm pin attached
the specimen to the hydraulic load machine. The speci-
men was loaded under tension–tension fatigue with a
minimum load of 0.25 kN and a maximum load of
24 kN using a 2-Hz sinusoidal loading profile. AE data Figure 5. Composite impact specimen.
were continuously recorded throughout the fatigue
testing. impacts (3, 4, 5 and 5 J energy) at the same position of
(40,65) mm from the bottom left-hand corner of the
grid using an Instron Dynatup 9250HV impact test
Composite impact specimen machine. During the impact testing, the sample was
Figure 5 shows the composite specimen used for the supported between two square clamping frames placed
impact investigation, which was manufactured from on either side of the specimen and centred about the
Cytec MTM28-1/HS-135-34%RW with a ((0,90)4)s impact location. The frames provide a 60 3 60 mm
layup. The overall specimen dimensions were unsupported area and have an overall outer dimension
370 3 185 mm and 2.15 mm thick. Figure 5 shows the of 100 3 100 mm. The rest of the sample remained
area of interest (180 3 160 mm, 20 mm grid resolu- unclamped throughout the impact testing. AE data
tion), sensor positions (red circles) and impact site (red were recorded during the impact events when the speci-
cross). Four Pancom Pico-Z sensors, with an operating men was clamped in the arrangement described.
frequency of 125–750 kHz, were mounted to the speci-
men using cyanoacrylate. In accordance with the alu-
minium fatigue investigation, the same in-line Aluminium aircraft panel
amplifiers and AE acquisition system were used; how- Figure 6(a) and (b) shows the complex geometry alumi-
ever, a 50-dB threshold, 20–1200 kHz analogue filters nium wing rib under investigation, with a 400 mm
and a 5-MHz sample rate were used to record AE data. 3 400 mm Delta-t grid applied at a grid resolution of
Prior to the impacts, five H–N sources were used at 10 mm. The overall specimen dimension was
each node position within the grid to create the neces- 1.5 3 0.48 m. Figure 6(b) shows the reverse of the
sary data for the Delta-t mapping techniques. As with panel and shows the structural features present and
the aluminium fatigue specimen, the training data for their position with respect to the grid (dashed square).
the Delta-t maps were collected with the specimen in Four Mistras Nano30 sensors were mounted at the cor-
pristine condition and the single data set was used for ners of the grid with coordinates of (215,28) mm,
all subsequent location calculations. The effectiveness (413,24) mm, (27,405) mm and (415,402) mm with
of the training data for both techniques was evaluated respect to the (0,0) grid position. The mechanical and
by H–N sources at random grid positions. An ultraso- acoustic coupling of the sensors was achieved using sili-
nic C-scan was used to determine there were no manu- con RTV (Loctite 595) which was allowed to cure for
facturing defects. The specimen was subjected to four 24 hours prior to testing. The sensors was connected to
Pearson et al. 389

Experimental results and discussion

H–N source investigations
In advance of using the Delta-t mapping techniques for
damage location, the accuracy of the training data was
assessed using H–N sources at random node positions.
This allowed assessment of the training data quality
and provided the opportunity to compare location
techniques.

Complex fatigue specimen. Figure 7(a) shows the esti-

mated source locations for three H–N sources at six
random positions for the TOA, Delta-t and AIC Delta-
t mapping techniques. Figure 7(b) shows the results of
the average Euclidean distance error for each location
Figure 6. Aluminium aircraft panel showing the a) Delta-t grid position. The error bars in the figure represent the max-
and b) the structural features on the reverse of the panel. imum and minimum errors for each position. This is
the case for all the remaining figures. Both the Delta-t
mapping techniques are able to resolve all six locations
an AE acquisition system via amplifiers with a 40-dB within the specimen geometry. The TOA technique is
gain and a 20–1200 kHz analogue filter. AE data was only able to locate four of the six locations within the
collected using a 40-dB threshold level and waveforms specimen with locations for positions 1 and 6 located
were recorded at a sample rate of 2 MHz. Training 52 and 90 mm outside of the structure boundary,
data were collected from five H–N sources performed respectively. This demonstrates the robustness of
at each node within the grid. To assess the effect of Delta-t mapping over the TOA technique and allows
training data resolution, the data was then separated the user to be confident that all possible sources will be
into training sets with grid resolutions of 10, 20, 50 and sensibly located. The AIC Delta-t mapping technique
100 mm, and Delta-t maps were prepared for each has located all but position 5 with an accuracy of less
training data set. To assess the location performance, than 3 mm and in four of the six cases has outper-
eight positions were selected within the grid that did formed the standard Delta-t approach, the exceptions
not correspond to grid nodes (i.e. training data points) being positions 1 and 3. In the case of positions 1 and
and five H–N sources were collected from each. 3, the AIC Delta-t mapping technique has an accuracy
Locations were computed using TOA, traditional of less than 2.5 mm, with the standard Delta-t tech-
Delta-t and AIC Delta-t for comparison. nique errors of 1.0 and 0.8 mm, respectively. However,

Figure 7. Fatigue specimen: (a) estimated H–N source location and (b) average error for each location.
390 Structural Health Monitoring 16(4)

Figure 8. Composite impact specimen: (a) estimated H–N source location and (b) average error for each location.

if these differences are considered relatively to the shows an improvement of 0.22 and 7 mm, respectively,
uncertainty in the measurement chain, then they appear for locations 3 and 4. For location 4, the improvement
to be well within the expected noise. For example, pla- of 7 mm in average error in the Delta-t mapping over
cement of the H–N source is expected to be within 1– the AIC Delta-t mapping requires further discussion.
2 mm of the desired location (due to human error) and This difference in error is larger than the uncertainty in
the Nano30 sensors used in the example have a radius the measurements due to the sample rate and sensor
of 4 mm. Additionally, assuming a wave speed of face diameter which for this specimen were 1–1.5 and
5400 m/s (fastest propagating mode in this material) 4 mm, respectively, meaning another factor has con-
and a sample rate of 2 MHz, one sample error in arri- tributed to this error. The reasons for this are most
val time is equivalent to 2.7 mm. So even if the first likely caused by the grid data for the AIC Delta-t tech-
sample of the arriving wave is accurately picked, an nique. The training maps for the Delta-t between chan-
error of up to 2.7 mm could still be expected. Coupling nels 1 and 2 and channels 3 and 4 in the proximity of
these together means that any fluctuations of less than (60,140) mm show a local peak and a change in gradi-
;8 mm can be considered to be within the limits of ent either side. This means that there are positions
uncertainty for this measurement. The largest error along these contours which are equal, meaning any var-
observed for the AIC Delta-t is seen at position 5 which iation in arrival time estimation could lead to two pos-
is still within the suggested 8 mm uncertainty. The sible source locations and hence a larger error for this
overall average Euclidean location errors for the three particular position. This is not the case for the Delta-t
techniques are 32.6, 5.8 and 3.0 mm for the AIC Delta- mapping contours where, due to the threshold crossing
t, Delta-t and threshold approaches, respectively, technique, the arrival time is less accurate. This there-
demonstrating the improved overall performance of the fore means the local variations are not present and
AIC Delta-t technique. results in a more accurate source estimation for this
particular position. Another interesting result is the rel-
Composite impact specimen. Figure 8 shows the located atively high location inaccuracy for the Delta-t map-
H–N source events and the average Euclidean distance ping for positions 1 and 5. For position 1, for each H–
error for each position for all the location methods. N source, the errors are 4, 4 and 168 mm. For location
Again the Delta-t techniques are able to locate sensibly 5, similar results are observed with errors of 8, 8 and
all H–N sources within the specimen’s geometric 113 mm from the source. The same training and H–N
boundary when compared with just half of the posi- source AE data were used for both techniques showing
tions with the TOA technique. The comparison of both the sensitivity in using arrival times calculated by
Delta-t techniques shows that the improved AIC loca- threshold crossing for AE source location. This rein-
tion algorithm is able to locate with a higher level of forces the use of the AIC Delta-t technique and shows
accuracy for 67% of the locations (positions 1, 2, 5 and the ability to confidently determine accurate source
6), with improvements of 51, 1, 40 and 3 mm, respec- locations. In addition, no sensible source location was
tively. The traditional Delta-t mapping technique determined by the TOA algorithm for H–N location 1.
Pearson et al. 391

Figure 9. Estimated H–N source location for aluminium

aircraft panel. Figure 10. Average Euclidean distance error for each H–N
source location.
The average Euclidean distance error for this position
was 375 m and hence was not shown in the figure. The
reasons for this large error are due to the H–N source
position being located on the edge of the sensor array
which is known to give large errors in TOA source
location. Also, this particular H–N position is located
close to a sensor, where the contours of Delta-t can
change sharply resulting in small arrival time errors
leading to large source location errors. These errors are
further compounded by the fact that two of the source
to sensors’ paths are on the fibre direction and hence
the wave travels fastest in these directions, while the
other two are off the fibre direction and hence travel-
ling the slowest. The TOA algorithm is unable to
account for the difference in wave speed for propaga-
tion angle and hence why there is significant error for
this particular location.

Figure 11. Average Euclidean distance error source location

Aluminium aircraft panel. Figure 9 shows the estimated for all H–N locations for different grid resolutions.
H–N source locations for all the source location tech-
niques. All estimated H–N source positions were sensi- actual positions. For the AIC Delta-t mapping, all H–
bly located within the boundary of the grid for this N source estimations are located within 10 mm from
particular specimen. It is evident from the figure that their actual position. While for the Delta-t mapping
the AIC Delta-t mapping technique shows a consistent two of the eight locations have errors greater than
average improvement of 9.3 and 5.1 mm over the TOA 10 mm (positions 4 and 8). Also, at two positions (3
and Delta-t mapping techniques, respectively. The fig- and 6), the Delta-t mapping technique has a minor
ure also shows that for H–N source positions in close improvement of the AIC mapping technique of 0.4 and
proximity to the hole which have multiple source to 0.5 mm, respectively. The average Euclidean distance
sensor paths affected by this structural feature, the errors for all H–N locations are 3.6, 8.7 and 13 mm for
accuracy and precision of the estimated source loca- the AIC Delta-t mapping, Delta-t mapping and TOA
tions are worse when compared with the AIC Delta-t techniques, respectively. The potential source location
technique. It also demonstrated the robustness of the errors due to the sample rate and sensor face diameter
AIC Delta-t mapping technique. Figure 10 shows the are 2.6 and 8 mm, respectively, for this specimen.
average Euclidean distance error for each H–N source Figure 11 presents a comparison of the traditional
location for all three location techniques. All source Delta-t mapping, which utilises a one-dimensional (1D)
location estimations are located within 25 mm from the interpolation strategy, and the AIC Delta-t mapping
392 Structural Health Monitoring 16(4)

Figure 12. Resulting damage from the fatigue testing showing a) the damage locations and b) a close up image of the damage region.

using both the same 1D interpolation and a two-

dimensional (2D) interpolation strategy. The three
approaches are also compared across a range of grid
resolutions. It is clear that the interpolation strategy
has little effect on the performance of the Delta-t map-
ping approach with very similar accuracy observed for
AIC Delta-t with both 1D and 2D interpolations.
Considering that the Delta-t contour maps commonly
change gradually and are generally smooth, this is rela-
tively unsurprising. Hence, the observed improvements
over the traditional Delta-t algorithm are predomi-
nantly related to the improved robustness in arrival Figure 13. Events located throughout the fatigue testing.
time estimation. When considering the effect of grid
resolution, the accuracy is only affected when the reso-
lution reaches 100 mm, in this example, and even then indication of how the algorithms will perform when
not significantly. This also relates to the smoothness of subjected to a test environment. Therefore, a compari-
the Delta-t contour maps, which here are relatively son of the techniques is presented for detecting fatigue
smooth and change gradually, meaning that interpola- crack and impact events.
tion works very effectively. If more complex structural
features are present that cause local discontinuities
within the contour maps, then a grid resolution is Aluminium fatigue study. Figure 12 shows the damage
required that is suitable to capture the nature of these regions after the specimen had been subjected to 96,000
discontinuities. It should also be noted that should a fatigue cycles. No damage was observed at the last
local discontinuity in the contour maps not be suitably visual inspection at 76,000 cycles. Closer visual
captured, it will only affect the location accuracy within observation after testing showed that a small fatigue
that region and not globally. crack initiated at the right-hand side thin section
(Figure 12 – solid arrow). This propagated a small dis-
tance before leading to a rapid failure of that section,
Aerospace damage mechanism case studies which, in turn, caused a rapid failure of the left-hand
Two further case studies are presented to demonstrate side thin section (Figure 12 – dashed arrow).
the ability of the TOA and Delta-t mapping and the Figure 13 shows the number of located events
novel AIC Delta-t mapping to detect actual damage recorded throughout the fatigue test. The figure shows
mechanisms which occur in the aerospace environment. the last visual inspection (black dashed line) and the
Although validation of the technique using H–N point at which the rate of located events increased at
sources is perfectly adequate, it does not give an 88,000 cycles (red dashed line). This identifies the point
Pearson et al. 393

Figure 14. (a) Visual observation of the damage regions and the associated events’ spatial binning for all fatigue cycles for the (b)
TOA, (c) Delta-t and (d) AIC Delta-t mapping.

at which significant damage occurred in the specimen. mapping improve on those of the TOA technique, with
This suggests that the damage initiated and propagated significantly higher numbers of located events occur-
within the final 8,000 cycles, and that approximately ring in the top half of the specimen. Again from these
15,000 events were located within this timeframe. results, it would be difficult to confidently determine
Although not conclusive, this gives a clear indication that damage had occurred in the structure. The spatial
when damage initiated and propagated in the structure, bin with the highest concentration of located events is
a clearer understanding of the deterioration of the at (100,135) mm and contains 381 events. Averaging
structure, and shows the advantages of AE for SHM the locations in this bin results in a Euclidean distance
systems. error of 24 mm. There are 50–100 events located on the
Figure 14 illustrates the damage regions in the speci- right damage region; however, this event bin concentra-
men as well as the results of spatially binned located tion also occurs frequently in the top half of the speci-
events. The number of events that were located within men, which adds further difficulty in confidently
each 5 3 5 mm spatial bin are shown. The results pre- determining damage in the specimen. The AIC Delta-t
sented enable a comparison between the TOA, conven- mapping spatial binning results are very promising and
tional and AIC Delta-t mapping techniques. The TOA due to the close grouping of located events show a clear
results show three location groups, with the closest indication that damage has occurred in the specimen.
being approximately 15 mm away from the right-hand The results would allow for confident determination of
side thin section damage region. From these results, it an area of interest to investigate further. There are two
would be difficult to conclude that damage had areas of higher activity (greater than 250 events). The
occurred in the specimen. The results for the Delta-t first and most active area contains 2117 events and is
394 Structural Health Monitoring 16(4)

Figure 15. AE energy spatial binning for all fatigue cycles for the (a) TOA, (b) Delta-t and (c) AIC Delta-t mapping and (d) the
Euclidean distance area for spatial bins above the energy threshold.

located 19 mm directly north of the right-hand damage on the left-hand thin section, the accuracy of the AIC
region. A lower intensity of located events (217 events) Delta-t mapping technique would enable the user to
occurs at the damage site. The relative size of this have greater confidence that damage had occurred. It
region of higher located events (2117 events) is not sur- should also be stated that in total, both damage sur-
prising as it is possible that the AE is being detected faces were 5 mm in length and the ability to accurately
from plastic deformation and the damage region. The locate damage of this size is an achievement. Overall,
other area of higher activity is located at (55,130) mm the AIC Delta-t mapping shows a clear area of higher
and contains 381 events. This particular region could activity which could direct an operator to investigate
be from the left-hand side damage region. There are a the damage area further.
variety of reasons why this area maybe mis-located; the Figure 15 shows the results of energy spatial binning
first is that the higher AE activity of the right-hand side where individual bins represent the energy summation
damage region could have masked signals from the of the first hit sensor in the event. The energy approach
left-hand side region, if the right-hand side damage can be another useful tool for determining significant
region propagated all the way across this section before AE activity. The figure also shows the average
the left-hand section failed, causing alterations of the Euclidean distance error for located events in spatial
training grid resulting in erroneous locations. The final bins with energy levels greater than 7.2 3 107 aJ. The
possible reason could be the rapid failure of this sec- location events within these bins were averaged to gain
tion, which was evident from the crack surfaces, arising a representative evaluation of the accuracy of each
in a short sudden release of AE activity. Although all technique. In addition, the figure shows the number of
three techniques have been unable to locate the damage events contained within these higher energy spatial
Pearson et al. 395

and contains 410 events. Both the traditional and AIC

Delta-t mapping technique show an improvement over
the TOA technique where a reduction in the average
error of 15.97 and 16.74 mm was observed, respec-
tively. The comparison of the traditional and AIC
Delta-t mapping technique shows similar performance
in the average error when detecting aluminium fatigue
damage. However, a significant improvement is that
the AIC Delta-t technique has located a further 197
events when compared with traditional technique.
From a statistical approach, this can further improve
the confidence of damage location, showing that the
AIC Delta-t mapping is able to locate lower energy
events more consistently than the Delta-t mapping
technique. Therefore, this has the potential to provide
an earlier indication of the damage presence.

Figure 16. Ultrasonic C-scan of composite impact specimen. Composite impact specimen. Ultrasonic C-scans were per-
formed pre and post impact testing; the results for the
pre-impact C-scan showed that there was no manufac-
bins. The results for the TOA technique show a region turing damage in the panel. Figure 16 shows the C-scan
of higher energy located 20.2 mm away from the right- results for the specimen after it was subjected to four
hand side damage region; the total AE energy recorded impacts with respective impact energies of 3, 4, 5 and
is 3.99 3 108 aJ from 102 events. The results for Delta- 5 J. The figure clearly shows an area of delamination at
t mapping again show one specific region of higher AE the impact location, which had an approximate size of
activity; this area on average is located 4.19 mm away 188 mm2 using ‘26 dB drop method’.51,52 The other
from the damage region and contains a total energy of higher attenuation areas correspond to the four sensors.
3.51 3 108 aJ from 213 events. Finally, the results for Figure 17 shows the results of the TOA, Delta-t and
the AIC Delta-t mapping show that it is the only tech- AIC Delta-t mapping locations for the impact events
nique where the significant energy bins are located at overlaid on the C-scan image. The TOA technique
the damage region. The higher AE activity region is located all four impact events; however, none was
located within an average error of 3.42 mm from the located within the delamination area. The Delta-t tech-
damage region and totals a region of 3.91 3 108 aJ nique located three of the four impacts and was unable

Figure 17. Composite impact specimen estimated H–N source locations showing the a) identified area of interest and b) a close
up of the damage region.
396 Structural Health Monitoring 16(4)

Table 2. Summary of the average Euclidean error in source location.

Average Euclidean source location error (mm)

TOA Delta-t AIC Delta-t

Aluminium specimen H–N sources 32.6 5.8 3.0

Fatigue damage 20.2 4.2 3.4
Composite specimen H–N sources 19.3 18.9 4.2
Impact events 124.7 18.9a 3.3
Aircraft panel H–N sources 13.0 8.7 3.6

TOA: time of arrival; AIC: Akaike information criteria.

a
Only three of four events located.

distance errors for all the located impact events for the
TOA, traditional Delta-t and AIC Delta-t mapping
were 124.7, 18.9 and 3.3 mm, respectively. Again these
results show the performance enhancement of the AIC
Delta-t mapping for locating impact events with signifi-
cant improvements in the accuracy and confidence that
all impacts will be located. This is increasingly impor-
tant as impact events are one off events. It is noted that
despite the clamping frames sitting between the impact
site and the sensor positions, they do not appear to
have a detrimental effect on the location accuracy
(Figure 18). The small width of the frames (20 mm) and
the dry interface (i.e. they are not acoustically coupled
to the sample) mean that their effect on propagation is
minimal. Consideration of frequency content and
attenuation of the recorded waves may reveal some
changes, but the concern here is the time of flight (ToF)
from source to sensor which appears to have remained
unaffected.
Figure 18. Euclidean distance error for each impact event.
Results summary
to locate impact 1. The TOA technique located impact Table 2 shows the summary of the average Euclidean
1 at (2362,22) mm; this large error has arisen due to source location error for all the investigations underta-
inaccurate arrival time estimation using the threshold ken for both specimens. For locating H–N sources on
crossing technique, and therefore, erroneous Delta-t is the aluminium specimen, the average Euclidean source
used in the TOA algorithm. This also effects the calcu- location error was 32.6, 5.8 and 3 mm for the TOA,
lation of the source position using the Delta-t mapping traditional and AIC Delta-t mapping techniques,
technique, meaning the impact event Delta-t’s fall out- respectively. For locating fatigue damage, the average
side those contained in the training grids, and therefore, errors were 20.2 mm (TOA), 4.2 mm (Delta-t) and
no source position can be determined. The impact loca- 3.4 mm (AIC Delta-t). In addition, the AIC Delta-t
tions for the AIC Delta-t mapping technique are technique was able to locate a further 197 and 308
extremely promising, locating all four impacts within events in the spatial bins with an accumulated energy
the damage area. Figure 18 shows the Euclidean dis- greater than 7.2 3 107 aJ. For locating H–N sources
tance errors calculated for each impact. Comparing all on the composite specimen, the average error was
three techniques, the TOA technique’s largest and smal- 19.3 mm (TOA), 18.9 mm (Delta-t) and 4.2 mm (AIC
lest location errors were 404 and 19 mm, respectively, Delta-t). Furthermore, for locating impact events, the
while for the Delta-t mapping these values were 39 and average errors were 124.7 mm (TOA), 18.9 mm (Delta-
5 mm. The most accurate technique was the AIC Delta- t) and 3.3 mm (AIC Delta-t). Finally, the results of a
t mapping with the maximum and minimum errors of 5 larger grid (400 3 400 mm) for locating H–N sources
and 2.5 mm, respectively. The average Euclidean on an aircraft panel showed the average errors of
Pearson et al. 397

13.0 mm (TOA), 8.7 mm (Delta-t) and 3.6 mm (AIC significant improvements over the conventional TOA
Delta-t) mapping. Overall, AIC Delta-t mapping shows and Delta-t mapping techniques. The use of the AIC
a reduction in error irrespective of the material or Delta-t mapping for locating damage in an SHM system
source type when compared with the other two tech- would allow confident and more probable detection of
niques and excellent accuracy with average errors less damage irrespective of the threshold used. Often dam-
than 4.2 mm, which was not achieved with the other age location is the first step in any SHM system; this
techniques. In all the presented examples, a single set of needs to be achieved with accuracy and confidence so
training data were captured and used to locate damage that any other technique used, such as classification, can
from subsequent testing. The question still remains as determine the nature of deterioration. The AIC Delta-t
to how much damage or degradation a structure can mapping technique offers a very credible method of
sustain before the training data become invalid. If dam- achieving this and in reality could lead to a greater con-
age occurs at a localised site, such as an impact event fidence in SHM systems.
on a composite panel, then this damage will only affect
subsequent location calculations if it interrupts the
Declaration of Conflicting Interests
shortest path from a new source to any sensors and
hence changes the ToF and therefore Delta-t values. The author(s) declared no potential conflicts of interest with
respect to the research, authorship and/or publication of this
For example, in the case of the composite impact tests
article.
above, multiple impacts were performed at the same
location and all accurately located, without update to Funding
the training data. However, in this example, the dam-
age all occurred at the same position and wave paths The author(s) disclosed receipt of the following financial sup-
port for the research, authorship, and/or publication of this
propagated from the damage site out to sensors, unin-
article: This work was supported by Innovate UK (grant
terrupted, hence accuracy was maintained. Had dam- number 100444).
age occurred at a different position and the resulting
waves had to travel through a previous damage site, References
then a reduction in accuracy would be expected. The
1. Beral B and Speckmann H. Structural health monitoring
extent of this change will depend on both the position (SHM) for aircraft structures: a challenge for system
of the existing and the new damage and also the type developers and aircraft manufacturers. In: Chang FK
and extent of damage present. (ed.) Structural health monitoring 2003: from diagnostics
& prognostics to structural health management: proceed-
ings of the 4th international workshop on structural health
Conclusion monitoring. Lancaster, PA: DEStech Publications, 2003,
TOA, Delta-t and the novel AIC Delta-t mapping tech- pp. 12–28.
niques were used to locate H–N sources, fatigue damage 2. Kapoor H, Boller C, Worden K, et al. Prospective appli-
cations for SHM on commercial aircraft. In: Chang FK
in aluminium and impact events in composite panels.
(ed.) Structural health monitoring 2009: from system inte-
The AIC Delta-t technique was able to consistently
gration to autonomous systems: proceedings of the 7th
show accuracy improvements when compared with the international workshop on structural health monitoring.
TOA and the Delta-t mapping when locating H–N Lancaster, PA: DEStech Publications, 2009, pp. 223–230.
sources in aluminium and composite specimens. The 3. Chang FK and Markmiller JFC. A new look in design of
ability of all three location techniques to detect real intelligent structures with SHM. In: Structural health
damage mechanisms, which could be prevalent when monitoring 2006: proceedings of the third European work-
using an SHM system in reality, was assessed. For locat- shop on structural health monitoring, Granada, 5–7 July
ing fatigue damage, the AIC Delta-t mapping technique 2006, pp. 5–20. Lancaster, PA: DEStech Publications.
produced results where a user could confidently deter- 4. Miller RK, Hill EvK and Moore PO. Nondestructive test-
mine an area of interest. Increasing the confidence dam- ing handbook, third edition: volume 6: acoustic emission
age could be detected and shows that AIC Delta-t testing. Columbus, OH: ASNT, 2005.
5. Mostafapour A, Davoodi S and Ghareaghaji M. Acous-
mapping can accurately detect lower energy events.
tic emission source location in plates using wavelet analy-
Finally, the results for locating impacts showed the AIC sis and cross time frequency spectrum. Ultrasonics 2014;
Delta-t mapping technique located all four impacts with 54: 2055–2062.
the highest level of accuracy, further demonstrating the 6. Paget CA, Atherton K and O’Brien EW. Triangulation
robustness of the AIC Delta-t mapping technique for algorithm for damage location in aeronautical composites
located different damage mechanisms. Overall, the com- structures. In: Chang FK (ed.) Structural health monitor-
bination of the Delta-t mapping technique with an AIC ing 2003: from diagnostics & prognostics to structural
picker is able to provide accurate location and showed health management: proceedings of the 4th international
398 Structural Health Monitoring 16(4)

workshop on structural health monitoring. Lancaster, PA: 24. Yan G and Tang J. A Bayesian approach for localization
DEStech Publications, 2003, pp. 363–370. of acoustic emission source in plate-like structures. Math
7. Flynn EB, Todd MD, Wilcox PD, et al. Maximum-likeli- Probl Eng 2015; 2015: 1–14.
hood estimation of damage location in guided-wave 25. Niri ED, Farhidzadeh A and Salamone S. Adaptive mul-
structural health monitoring. P Roy Soc A: Math Phy tisensor data fusion for acoustic emission source localiza-
2011; 467: 2575–2596. tion in noisy environment. Struct Health Monit 2012; 12:
8. Ward J, Croxford A and Paget C. Passive impact localisa- 59–77.
tion for the structural health monitoring of new airframe 26. Niri ED, Farhidzadeh A and Salamone S. Nonlinear Kal-
materials. J Phys Conf 2013; 457: 1–11. man filtering for acoustic emission source localization in
9. Koabaz M, Hajzargarbashi T, Kundu T, et al. Locating anisotropic panels. Ultrasonics 2014; 54: 486–501.
the acoustic source in an anisotropic plate. Struct Health 27. Julier SJ and Uhlmann JK. New extension of the Kal-
Monit 2011; 11: 315–323. man filter to nonlinear systems. Proc SPIE 1997; 3068:
10. Kundu T, Das S, Martin SA, et al. Locating point of 182–193.
impact in anisotropic fiber reinforced composite plates. 28. St-Pierre M and Gringas D. Comparison between the
Ultrasonics 2008; 48: 193–201. unscented Kalman filter and the extended Kalman filter
11. Li J and Qi G. Improving source location accuracy of for the position estimation module of an integrated navi-
acoustic emission in complicated structures. J Nondest gation information system. In: Proceedings of IEEE intel-
Eval 2009; 28: 1–8. ligent vehicle symposium, Parma, 14–17 June 2004, pp.
12. Ciampa F and Meo M. A new algorithm for acoustic 831–835. New York: IEEE.
emission localization and flexural group velocity determi- 29. Niri ED, Farhidzadeh A and Salamone S. Determination
nation in anisotropic structures. Compos Part A: Appl S of the probability zone for acoustic emission source loca-
2010; 41: 1777–1786. tion in cylindrical shell structures. Mech Syst Signal Pr
13. Ciampa F, Meo M and Baram J. Impact localization in 2015; 60–61: 971–985.
composite structures of arbitrary cross section. Struct 30. Scholey JJ, Wilcox PD, Wisnom MR, et al. A generic
Health Monit 2012; 11: 643–655. technique for acoustic emission source location. J Acoust
14. Aljets D, Chong A, Wilcox S, et al. Acoustic emission Emiss 2009; 27: 291–298.
source location in plate-like structures using a closely 31. Hensman J, Mills R, Pierce SG, et al. Locating acoustic
arranged triangular sensor array. J Acoust Emiss 2010; emission sources in complex structures using Gaussian
28: 85–98. processes. Mech Syst Signal Pr 2010; 24: 211–223.
15. Surgeon M and Wevers M. One sensor linear location of 32. Ziola SM and Gorman MR. Source location in thin
acoustic emission events using plate wave theories. Mat plates using cross-correlation. J Acoust Soc Am 1991;
Sci Eng A: Struct 1999; 265: 254–261. 901: 2551–2556.
16. Matt HM and Lanza di Scalea F. Macro-fiber composite 33. Jeong H. Analysis of plate wave propagation in anisotro-
piezoelectric rosettes for acoustic source location in com- pic laminates using wavelet transform. NDT&E Int 2001;
plex structures. Smart Mater Struct 2007; 16: 1489–1499. 34: 185–190.
17. Kundu T, Nakatani H and Takeda N. Acoustic source 34. Jeong H and Jang YS. Fracture source location in thin
localization in anisotropic plates. Ultrasonics 2012; 52: plates using the wavelet transform of dispersive waves.
740–746. IEEE T Ultrason Ferr 2000; 47: 612–619.
18. Kundu T, Yang X, Nakatani H, et al. A two-step hybrid 35. Hsu NN and Breckenbridge FR. Characterisation and
technique for accurately localizing acoustic source in ani- calibration of acoustic emission sensors. Mater Eval 1981;
sotropic structures without knowing their material prop- 39: 60–68.
erties. Ultrasonics 2015; 56: 271–278. 36. Hamstad MA, O’Gallagher A and Gary J. A wavelet
19. Grosse CU. Acoustic emission localization methods for transform applied to acoustic emission signals: part 2:
large structures based on beamforming and array tech- source location. J Acoust Emiss 2002; 20: 62–82.
niques. In: 7th International symposium on nondestructive 37. Ding Y, Reuben RL and Steel JA. A new method for
testing in civil engineering, Nantes, 30 June–3 July 2009. waveform analysis for estimating AE wave arrival times
20. He T, Pan Q, Liu Y, et al. Near-field beamforming anal- using wavelet decomposition. NDT&E Int 2004; 37:
ysis for acoustic emission source localization. Ultrasonics 279–290.
2012; 52: 587–592. 38. Shehadeh M, Steel JA and Reuben RL. Acoustic emis-
21. Xiao D, He T, Pan Q, et al. A novel acoustic emission sion source location for steel pipe and pipeline applica-
beamforming method with two uniform linear arrays on tions: the role of arrival time estimation. Proc IMechE
plate-like structures. Ultrasonics 2014; 54: 737–745. Part E: J Process 2006; 220: 121–133.
22. Schumacher T, Straub D and Higgins C. Towards a prob- 39. Lokajicek T and Kilma K. A first arrival identification
abilistic acoustic emission source location algorithm: a system of acoustic emission signals by means of a high-
Bayesian approach. J Sound Vib 2012; 331: 4233–4245. order statistics approach. Meas Sci Technol 2006; 17:
23. Zarate BA, Pollock A, Momeni S, et al. Structural health 2461–2466.
monitoring of liquid-filled tanks: a Bayesian approach for 40. Baer M and Kradolfer U. An automatic phase picker for
location of acoustic emission sources. Smart Mater Struct local and teleseismic events. B Seismol Soc Am 1987; 77:
2015; 24: 1–10. 1437–1445.
Pearson et al. 399

41. Akaike H. Markovian representation of stochastic pro- 47. Sedlak P, Hirose Y and Enoki M. Acoustic emission loca-
cesses and its application to the analysis of autoregressive lization in thin multi-layer plates using first-arrival deter-
moving average processes. Ann I Stat Math 1974; 26: mination. Mech Syst Signal Pr 2013; 36: 636–649.
363–387. 48. Sedlak P, Hirose Y, Khan SA, et al. New automatic loca-
42. Kitagawa G and Akaike H. A procedure for the model- lization technique of acoustic emission signals in thin
ing of non-stationary time series. Ann I Stat Math 1978; metal plates. Ultrasonics 2009; 49: 254–462.
30: 351–363. 49. Baxter MG, Pullin R, Holford KM, et al. Delta T source
43. Hinkley DV. Inference about the change-point for cumu- location for acoustic emission. Mech Syst Signal Pr 2007;
lative sum tests. Biometrika 1971; 58: 509–523. 21: 1512–1520.
44. Grosse C. Winpecker version 1.2: instruction manual. 50. Eaton MJ, Pullin R and Holford KM. Acoustic emission
Stuttgart: University of Stuttgart, 2000. source location in composite materials using delta T map-
45. Kurz JH, Grosse CU and Reinhardt HW. Strategies for ping. Compos Part A: Appl S 2012; 43: 856–863.
reliable automatic onset time picking of acoustic emis- 51. Broughton WR, Sims GD and Lodeiro MJ. Overview of
sions and of ultrasound signals in concrete. Ultrasonics DTI-funded programme on ‘Standardised procedures for
2005; 2005: 538–546. ultrasonic inspection of polymer matrix composites’.
46. Carpinteri A, Xu J, Lacidogna G, et al. Reliable onset Insight: Non-Dest Test Cond Monit 1998; 40: 8–11.
time determination and source location of acoustic emis- 52. Smith RA, Jones LD, Willsher SJ, et al. Diffraction and
sions in concrete structures. Cement Concrete Comp 2012; shadowing errors in 26 dB defect sizing of delaminations
34: 529–537. in composites. Insight: Non-Dest Test Cond Monit 1998;
40: 44–49.