This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 173.250.148.152

This content was downloaded on 15/02/2017 at 21:42

Please note that terms and conditions apply.

Visualization of stress wave propagation via air-coupled acoustic emission sensors

View the table of contents for this issue, or go to the journal homepage for more

2017 Smart Mater. Struct. 26 025020

(http://iopscience.iop.org/0964-1726/26/2/025020)

Home Search Collections Journals About Contact us My IOPscience

You may also be interested in:

Solitary wave-based delamination detection in composite plates using a combined granular crystal

sensor and actuator

Eunho Kim, Francesco Restuccia, Jinkyu Yang et al.

A novel impact identification algorithm based on a linear approximation with maximum entropy

N Sanchez, V Meruane and A Ortiz-Bernardin

Real-time location of coherent sound sources by the observer-based array algorithm

Xun Huang

Detection, localization and characterization of damage in plates with an in situ array ofspatially

distributed ultrasonic sensors

Jennifer E Michaels

Impact Damage Detection and Assessment in Composite Panels using Macro Fibre Composites Transducers

M R Pearson, M J Eaton, C A Featherston et al.

A delay-and-Boolean-ADD imaging algorithm for damage detection with a small number of piezoceramic

transducers

Guangtao Lu, Yourong Li and Gangbing Song

Dispersion analysis of Lamb waves and damage detection

Fucai Li, Guang Meng, Lin Ye et al.

Guided wave based structural health monitoring: A review

Mira Mitra and S Gopalakrishnan

Visualization of stress wave propagation viaair-coupled acoustic emission sensors

Joshua C Rivey1,4, Gil-Yong Lee1,4, Jinkyu Yang1, Youngkey Kim2 andSungchan Kim3

1Aeronautics and Astronautics, University of Washington, Seattle, WA 98195, USA2 SM Instruments, Daejon, Korea3Korea Aerospace Research Institute, Daejon, Korea

E-mail: jkyang@aa.washington.edu

Received 21 March 2016, revised 24 August 2016Accepted for publication 14 September 2016Published 16 January 2017

AbstractWe experimentally demonstrate the feasibility of visualizing stress waves propagating in platesusing air-coupled acoustic emission sensors. Specifically, we employ a device that embedsarrays of microphones around an optical lens in a helical pattern. By implementing abeamforming technique, this remote sensing system allows us to record wave propagationevents in situ via a single-shot and full-field measurement. This is a significant improvementover the conventional wave propagation tracking approaches based on laser dopplervibrometry or digital image correlation techniques. In this paper, we focus on demonstratingthe feasibility and efficacy of this air-coupled acoustic emission technique by using largemetallic plates exposed to external impacts. The visualization results of stress wavepropagation will be shown under various impact scenarios. The proposed technique can beused to characterize and localize damage by detecting the attenuation, reflection, and scatteringof stress waves that occurs at damage locations. This can ultimately lead to the development ofnew structural health monitoring and nondestructive evaluation methods for identifying hiddencracks or delaminations in metallic or composite plate structures, simultaneously negating theneed for mounted contact sensors.

Keywords: sound camera, beamforming, acoustic emission, impact identification

(Some figures may appear in colour only in the online journal)

1. Introduction

In aerospace, civil, and mechanical engineering applications,it is imperative to ensure structural integrity through thedetection and characterization of damage and defects. Thepresence of defects, such as cracks, dents, corrosion, dela-minations, or numerous other forms of damage, can sig-nificantly reduce the inherit properties and performance of astructure, thereby increasing the chance of premature failure.Therefore, structural health monitoring (SHM) and non-destructive evaluation (NDE) have been subjects of intensestudies in recent decades. In particular, increases in the use ofadvanced materials, sensors/actuators, and manufacturingprocesses has spurred the development of numerous SHM/

NDE techniques. These methods include—but are not limitedto—thermography, shearography, X-radiography, eddy cur-rent, ultrasonic c-scan, scanning laser vibrometry, and guidedwave-based ultrasound techniques [1–5]. They are based onthermal, electromagnetic, acoustic/mechanical, and othermulti-physical feedback, and each technique offers uniqueadvantages and shortcomings.

Using acoustic emissions for purposes of detecting andlocalizing damage has been one of the widely adoptedmethods in the SHM/NDE community [6–9]. Acousticemissions are manifest when a material is subject to extremestress conditions due to external loads, such that a local pointsource within the material suddenly releases irreversibleenergy in the form of stress waves. The released stress wavesare transmitted to the surface of the material and then prop-agate outwards from the epicenter of the release source.

Smart Materials and Structures

Smart Mater. Struct. 26 (2017) 025020 (10pp) doi:10.1088/1361-665X/26/2/025020

4 These authors contributed equally.

0964-1726/17/025020+10$33.00 © 2017 IOP Publishing Ltd Printed in the UK1

Previous studies involving acoustic emission techniques havefocused largely on the onset of such acoustic emissions tolocate their release sources [10–13]. However, more usefulapplications can be derived from the susceptibility of acousticemissions to attenuation, scattering, or reflection by dis-continuities present in the material [8]. We identify that it isthis particular property of acoustic emissions that can beexploited in order to detect and localize pre-existing damagein an inspection medium.

In this study, we experimentally demonstrate the feasi-bility of visualizing stress waves in an aluminum plate usingacoustic emissions. The focus is whether acoustic emissiontechniques can capture the scattering of stress waves due to thepresence of damage resulting from an applied impact on theplate. For recording and subsequent visualization of suchtransient events, we employ a device referred to as an acousticor sound camera. This device embeds an optical lens at thecenter and arrays of microphones around the optical lens in ahelical pattern [14]. Note that arrays of microphones have beenused in previous studies to identify the locations of vibrationsources [15], but it has not been thoroughly explored yet tovisualize stress waves in structures via air-coupled microphonesensors to the best of the authors’ knowledge. This is becauseof a short characteristic time—on the order of micro-seconds—of stress waves propagating in solids and also due to the dif-ficulty in post-processing measured raw data to visualize thewavefronts of these stress waves.

To address the challenges associated with stress wavevisualization, we implement time-domain delay-sum beam-forming techniques [16] based on acoustic emission infor-mation collected from the arrays of microphones. To enhancethe accuracy of the diagnostic scheme, we conduct parametricstudies on various post-processing conditions, including thetemporal resolution of the sensor data and the spatial reso-lution of the inspection plate. Finally, the developed techni-que is evaluated for SHM/NDE purposes with capabilitiesassessed for detecting the defect location simulated with amass placed in the path of the wave propagation on theinspection plate.

The acoustic emission beamforming technique coupledwith the sound camera device has unique advantages com-pared to the conventional techniques such as ultrasonic orlaser based testing methods [6]. Many of the conventionaltesting methods are founded on the principle of impartingexternal stimuli or excitations on the inspection material andmeasuring differences in the received signal in order to detectwhether damage is present and where it is located. While suchmethods generally provide highly accurate results, they ofteninvolve slow and expensive processes to operate equipment.Conversely, acoustic emission beamforming methods can beconducted in situ and in real time, without necessitatingpermanently mounted contact sensors or baseline data.

Recently, laser Doppler vibrometry has gained significantattention as a means to visualize stress waves in solids andstructures with an unprecedented resolution [17–19]. How-ever, this method requires synchronization and reconstruction

of data measured from every single discretized spatial point ofan inspection medium in order to compose the propagationdevelopment. This is not practical given the difficulties inexciting structures in a repeated, identical fashion. Digitalimage correlation techniques can also visualize extremelydynamic motions, but they require speckle patterns on spe-cimens and their field of view can be often narrow forrecording high speed events [20]. In contrast, the proposedacoustic emission beamforming technique is capable of con-ducting non-contact—yet full-field—stress wave visualizationacross an inspection medium in a single shot measurement.Consequently, we envision that this method can open newavenues to diagnosing the existence of damage in structures ina time- and cost-efficient manner by conducting simple tests.

The contents of this manuscript will address the follow-ing topics. Section 2 contains an introduction to the time-domain delay-sum beamforming theory and a discussion ofhow this concept is incorporated into the study containedherein. Section 3 describes the experimental setup used formonitoring inspection plates and tracking stress waves usingthe sound camera. Section 4 provides an analysis of para-metric studies performed on pristine plates that are used todetermine temporospatial resolutions necessary for cap-abilities-centric analyses. Finally, section 5 concludes thestudy with an analysis of the acoustic emission beamformingmethod for identifying varying impact locations and detectingsimulated damage on the inspection plate.

2. Theoretical background

Traditional methods of source localization via air-coupledacoustic emission rely on the assumption that the recordingdevice is pointed at or aimed in the general vicinity of theemission origin. Acoustic beamforming is an attractivealternative for use in acoustic feedback NDE as it provides theopportunity to detect the direction of unknown acousticemissions associated with failure events across a large spatialdomain [21]. While acoustic beamforming is a relatively newconcept for applications related to damage localization,beamforming using microphone arrays is a standard practicefor spatial isolation of sound sources. The beamformingmethod—also known as microphone antenna, phased array ofmicrophones, acoustic telescope, or acoustic camera—is usedextensively for localizing sounds on moving objects and tofilter out background noise in acoustically active environ-ments with stationary sound sources [22].

Figure 1 gives a visual representation of the basic delay-sum beamforming method. This process can be expressedmathematically by equation (1) in terms of the time-domaindelay-sum beamforming output [16]:

å t

= -

=

B t X a s t X, , 1Pn

N

n n n P1

( ) [ ( )] ( )

where N is the total number of microphones, t is time, andXP

represents an inspection position on the plate with respect to

2

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

the reference position (e.g., center of the microphone arrays asshown in the figure). an is a spatial shading or weight coef-ficient that can be applied to the individual microphones tocontrol mainlobe width and sidelobe levels [16]. In manyinstances, the weighting coefficients are set to unity. sn(t)represents the acoustic emission received by the nth micro-phone emitted by an arbitrary sound source.

After receiving the signal, a specified delay, tn, isimposed to the signal for each individual microphone basedon the spatial domain. Figure 1 shows the operation used forcalculating the microphone delays. First, the spatial domain ofinterest is discretized into ‘pixels’ rendering a superimposedgrid onto the spatial domain. Second, for each point on the

spatial domain the position vectorXp from the predetermined

reference point is calculated. Then, the position vector of thatsame spatial pixel from the nth microphone whose position

relative to the reference point is given by¾Mn is determined as

=

X Xn p −

¾Mn . Finally, the difference of flight time of the

acoustic signal between the two vectors is found by calcu-lating the difference in vector magnitudes and dividing thespeed of sound c. That is, the time delay for each point on the

inspection plate and each nth microphone is defined by

t

=

-

Xc

X X1

. 2n P p n( ) (∣ ∣ ∣ ∣) ( )

The scanning algorithm of the microphone array willperform the delay calculation operation for the entirety of thespatial domain which it is monitoring. After the time delaysfor all microphones are imposed on the signal for a givenspatial pixel, the transformed signal from each microphone issummed as shown in figure 1(b). The location of the soundsource is determined by the delays for a specific spatiallocation which produced the maximum beam output, B(t). It isimportant to note that the microphone delays are time inde-pendent and therefore will remain constant for a givenmicrophone to a given spatial pixel throughout the monitoringperiod assuming the geometry of the test setup remainsconstant. While figure 1 and equation (1) illustrate the basicdelay-sum beamformer in the time-domain, there exist manyvariations and modifications to this algorithm, see refer-ence [16].

Figure 1. Illustrations showing (a) location vectors of reference point, microphones, and inspection points; and (b) concept of calculatingdelays for a specific microphone for a given spatial position.

Figure 2. (a) Arrangement of 30 MEMS microphones and an optical camera. Inset shows a digital image of the sound camera. (b) Proposedexperimental setup for inducing and tracking transient waves in an aluminum plate using the sound camera.

3

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

3. Experimental setup

For the study presented herein, we used a commercialacoustic emission sensing device equipped with microphonearrays and a motion camera (SeeSV-S205 Sound Camera, SMInstruments) [14]. Specifically, the sensing device consistedof a high resolution optical camera with a sampling rate of 25frames per second located in the center of the device. Theoptical camera is surrounded by 30 high sensitivity digitalmicro-electric mechanical system (MEMS) microphones witha sampling frequency of 25.6kHz arranged in five helicalarrays of six microphones each as shown in figure 2(a). Thefigure inset shows a digital image of embedded microphonesalong with the optical lens. The locations of microphonesshown in the inset are flipped horizontally (in x-direction) tomake the microphones see the inspection plate correspondingto the coordinate configuration shown in figure 1(a).

Figure 2(b) shows the experimental setup used in thisstudy to induce and track the transient waves in the aluminumplate. The images show the 1.2 m×1.2 m×1.02 mm 6061-T6 aluminum plate mounted to an optical table using a railsystem to create a fixed boundary condition around the plate.The plate was secured between the angle bars and the squaretube with fasteners placed every 15.24 cm that went throughall components to effectively fix the boundary of the plate andsuspend it above the optical table.

Once the plate was installed, the acoustic camera wasmounted above the center of the plate at a height of 0.52 m asshown in figure 2(b). Impacts were introduced into the platevia manually tapping the plate using the tip of a hexagonalwrench while simultaneously capturing the acoustic signalrecorded by the 30 microphones. This manual tapping issimilar to the conventional pencil lead break (PLB) test [25]in a sense that they all initiate stress wave propagation in aninspection medium. However, the PLB approach mimics thegeneration of stress waves due to the initiation of cracks orsimilar types of internal/external damage, while our manualtapping simulates external impact events, e.g., bird and hailimpact or inspecting crew’s manual tapping. Since the PLB

test generates very low-level acoustic signals it requires ahighly sensitive contact-type acoustic emission sensor torecord the low amplitude signals. In our study, we focus onvisualizing stress waves via non-contact microphone arrays,and thus, the manual tapping is highly efficient for generatingtunable, high amplitude acoustic emission signals. Based onthe pressure waves initiated by the manual tapping andmeasured by the microphone arrays, post-processing wasperformed to calculate the time delays and beamformingoutput by using equations (1) and (2) in order to produce theacoustic images.

4. Parametric studies on pristine plates

In this study different parameters were systematically variedto investigate and assess the wave propagation tracking cap-abilities of the sound camera in addition to determining post-processing parameters used in subsequent analyses. Theseparameters included the temporal and spatial resolutions inpost-processing. This section discusses the results from theseparametric studies given pristine plates. For all images shownof the wave propagation tracking, the area presented in theimage represents that of the entire inspection plate.

4.1. Effects of temporal resolution

The temporal resolution was investigated with the goal ofachieving a smooth propagation of the transient wave front. Inthis study, the sound camera MEMS microphones imposed ahardware limitation on the sampling frequency restricting it to25.6 kHz. This meant the time between individual samples wasslightly greater than 39 μs. Figure 3 shows an exemplary signalcaptured by microphone 1 (see the location in figure 2(a))immediately following a top left impact on the plate.Figure 3(a) shows the time history of sound pressure up to 8 mspost-impact, while figure 3(b) plots the frequency spectrum ofthe same signal. In the temporal signal, the maximum pressuremeasured is around 2.5 Pa (the saturation threshold of eachmicrophone sensor). Note that in experiments, the manual

Figure 3. Sample acoustic emission captured by a single microphone on the sound camera. (a) Acoustic emission amplitude versus timeimmediately following impact on plate. (b) Single-side amplitude versus frequency of the acoustic emission.

4

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

tapping was practiced carefully in a way that the peak ampl-itude of the sound pressure is kept around 2.5 Pa. In figure 3(b),we observe that a peak frequency is located at approximately850 Hz and that very little acoustic activity occurs at highfrequencies, i.e., above 6 kHz.

During investigations of wave propagation velocities, thespeed of major flexural waves propagating in the 1.02 mmthick plate was approximately 1700 m s–1 (to be further dis-cussed below). At this wave speed and the given samplingfrequency, the wave front propagated approximately 66.4 mmbetween each sample or 5.4% of the total plate width,resulting in a very coarse propagation tracking ability. Toreduce the effective propagation distance between sequentialsamples and to improve the resolution of the wave propaga-tion tracking, an interpolation of the raw acoustic signal wasperformed. Figure 4(a) shows a window of the raw signalwith no interpolation applied (left panel), in which weobserve the drastic amplitude changes of pressure measuredfrom a microphone between subsequent samples. The beam-forming results of the stress wave propagation for the rawsignal without interpolation are shown in the three images tothe right of the signal. The inset image at t=0 ms infigure 4(a) represents the top left impact induced in theinspection plate. Note that we used the beamforming power topresent the acoustic images, by summing the microphonesignals of 30 microphones with the calculated delays as inequation (1). So the maximum possible value of beamformingshown in the acoustic images should be in the range of 60∼75 Pa, assuming the peak amplitude of each microphone isaround 2 ∼2.5 Pa.

Figure 4(b) shows the same raw signal as the one aboveit, but with a frequency-domain fast fourier transformationinterpolation applied to reconstruct the signal. An upsamplefactor of 10 was chosen for the interpolation in order todecrease the time between samples to 3.9 μs. This meant thatthe propagation distance between samples was reduced to 6.6mm and only 0.5% of the total plate width. Comparing thetwo sets of images presented in figure 4, we find that bothapproaches successfully identify the impact location with areasonably high accuracy (to be further discussed in the latersection). However, the results from the raw signal do notshow clear boundaries of wave front while the second set ofimages associated with the interpolated data show theincreased definition of the wave front. Thus, the effects of theinterpolation are evident in the smoothing of the wave pro-pagation and acoustic features.

While an improved image quality was achieved byincreasing the upsample factor, a computational penalty wasincurred. Based on parametric studies utilizing differentupsample factors, the upsample-computational time relation-ship was approximately linear with an increase in computa-tional time. Specifically, for the upsample factor of n, thebeamforming computational time is increased by a factor of

´ n1.044 0.014( ) times compared to the original rawsignal. The reduced computational efficiency was deemed anacceptable cost for the increased image quality gained usingthe reconstructed signal and necessary for identifying andlocalizing masses. For all subsequent simulations, we used anupsample factor of 10. We also note that it is ideal that weobtain a higher sampling frequency via hardware, i.e., animproved sampling rate of the microphone. However, despite

Figure 4. Effects of changing temporal resolution using interpolation of results in frequency domain. (a) Original temporal data and the post-processed surface maps showing stress wave propagation at t=0, 0.20, and 0.31 ms. (b) Interpolated temporal data and the post-processedsurface maps showing stress wave propagation at t=0, 0.20, and 0.31 ms. For all surface maps, the spatial resolution was fixed at 30 mmbetween adjacent pixels. The colorbar at the bottom shows the intensity of the beamforming output in Pa, which is based on the summation ofpressure signals measured from all channels.

5

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

the limitation of the sampling frequency of the microphoneused in this study (25.6 kHz), our investigations of theinterpolation (reconstruction of data) enables us to get clearacoustic images in a simple and cost efficient manner withoutfurther increasing the hardware specifications.

4.2. Effects of spatial resolution

Now we investigate the effect of spatial resolution in anattempt to further refine the wave front throughout the pro-pagation time. Figure 5 shows the results of the spatialresolution study. Here the spatial resolution was system-atically halved for each simulation throughout the parametricstudy. Figure 5(a) and subsequent beamforming images showthe results for a spatial resolution (ΔX) of 80 mm between

pixels. While the impact point in the surface map is welllocalized (compare with the actual impact location as shownin the inset image), the wave front definition dissipates as itbegins to propagate into the far-field as seen at times t=0.21ms and t=0.32 ms.

Figures 5(b)–(c) and associated images show spatialresolutions of 40 mm and 20 mm between pixels, respec-tively. Both sets of images show an accurate impact locali-zation and increased definition of the wave front as the spatialresolution increases. Figure 5(d) and corresponding wavepropagation images represent the case of ΔX=10 mmbetween pixels. Compared to all previous results, these ima-ges show a very definitive wave front at t=0.21 ms. Theimage at t=0.32 ms for a spatial resolution of 10 mmvalidates that increased spatial resolution can maintain the

Figure 5. Effects of changing spatial resolution by decreasing step size between ‘pixels’ on virtual inspection plane. Left column showsspatial resolution, while the next three right columns show surface maps of wave propagation for t=0, 0.21, and 0.31 ms under the spatialresolution of (a) ΔX=80 mm; (b) 40 mm; (c) 20 mm; and (d) 10 mm.

6

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

desired and necessary wave front definition into the far-fieldof the plate while post-impact saturation behind the wavefront clearly emanates from the impact location.

While a 10 mm displacement between adjacent pixelsmay be excessive by creating ripples, it increases the chanceof detecting artificially created damage (to be discussed insection 5) and accurately determining its location. Howeverfor each halving of the spatial resolution, the computationaltime is increased by a factor of approximately 4.25, therebyresulting in an approximately power-law relationship equal to

- n 2.08 0.042( ) where n is the reduction factor with respect toΔX (e.g. 80 mm to 20 mm, n=1/4). The power-law rela-tionship is due to the spatial resolution affecting bothdimensions of the inspection plate meaning that the increasein spatial resolution is squared with each iteration. While thisleads to a significant increase in computational time, theincrease was again considered as an acceptable trade-off forthe gained wave propagation resolution. This would provenecessary for detecting and localizing artificially createddamage on the plate in subsequent analyses. Following thetemporal and spatial resolution studies, a temporal upsamplefactor of 10 and a spatial resolution of 10 mm were selectedfor use in all following analyses.

5. Feasibility studies on the applications toNDE/SHM

In this section, we assess the feasibility of using the beam-forming-based sound camera technique for (i) identifyingimpact location for real-time SHM applications and (ii)detecting artificial damage location for real-time NDEapplications.

5.1. Identification of impact location

Once simulation parameters were determined, tests wereperformed to characterize detection and localization cap-abilities of the sound camera. Prior to masses being placed onthe plate, it was necessary to determine if different impactlocations could be identified since a single impact location inthe top left corner of the plate had been used for all previousparametric studies. Additionally, these tests would serve asthe first quantitative indication of the detection and localiza-tion capabilities.

Figures 6(a)–(d) show the beamforming images at themoment of impact for impacts located in the four corners ofthe plate. With the center of the inspection plate above which

Figure 6. Impact locations (top row) and post-processing results (bottom row) showing impact detection capability of the sound camera andbeamforming algorithm. Location of impact was varied in four cases: (a) top left, (b) top right, (c) bottom right, and (d) bottom left.

Table 1. Quantification of impact localization.

Impact Actual Identified Absolute NormalizedLocation Position [m, m] Position [m, m] Difference [m] Error [Num. Pixels]

Top Left (−0.3032, 0.3048) (−0.2743, 0.3251) 0.0353 3.53Top Right (0.3048, 0.3064) (0.3556,0.3150) 0.0515 5.15Bottom Right (0.3056, −0.3064) (0.3150,−0.3251) 0.0209 2.09Bottom Left (−0.3040, −0.3056) (−0.2946, −0.2946) 0.0145 1.45

7

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

the camera is suspended defined as the origin of the plate, theactual measured impact locations are given in table 1. Foreach of the impact locations shown, the images suggest thatthe impact location is properly identified and localized usingthe sound camera and beamforming algorithm. The imagesshown were used to identify the point of maximum amplitudeat the instant of impact which was defined as the impactlocation. These identified locations are given as the identifiedlocations in table 1.

We quantify the normalized error between the actual andidentified impact locations as below

=- + -

DX X Y Y

X

Normalized Error

,3actual identified

2actual identified

2( ) ( ) ( )

where DX represents the pixel size. Looking at actual andidentified positions in table 1 (i.e., (X Y,actual actual) and(X Y,identified identified)), it is observed that all impacts werelocated relatively accurately with regards to identified posi-tions being in the general vicinity of the actual known posi-tion of the impact. Using equation (3), the normalized error interms of number of pixels between the actual and identifiedimpact locations with respect to the spatial discretization wascalculated. The calculated errors are given in table 1. Giventhe ΔX of 10 mm used for this study, the normalized errorbetween the actual and identified impact locations representdifferences ranging from 1.45 to 5.15 pixels as shown intable 1. The largest error was calculated for the top rightimpact location and found to be 5.15 pixels or approximately4.22% of the plate width. This means that, with respect to thetotal spatial domain in this case, the localization was stillrelatively accurate. The larger errors could be attributed to not

a fine enough spatial resolution which leads to a larger errorbetween the actual and identified locations. Additionally, thelarger error values could be due to human error when initi-ating the impacts on the plate. Despite some seemingly largeerror values between actual and identified impact locationsthis study indicates the sound camera and beamformingalgorithm were able to fairly accurately localize differentimpact locations and track the resultant transient wave acrossthe plate.

5.2. Identification of pre-existing artificial damage

Once it was established that the sound camera could detectand sufficiently track the transient wave, the impact locationwas fixed and masses were added to the plate. This is todetermine the capability of the sound camera and beam-forming algorithm to detect the discontinuities which repre-sented psuedo-damage cases. Note that if we successfullyvisualize stress wave scattering at these damage sites, thismeans that our technique can be used for NDE purposes. Thatis, assuming damage is present in a plate-like structure, aninspection crew can conduct tapping tests on the plate, and byvisualizing the attenuation or reflection of stress waves at thisdamage site, he/she will be able to detect the damage in anefficient manner. For the feasibility study, we applied impactsin the top left corner of the plate as shown in figures 7(a)–(b).The mass used for this study was a 0.615 kg, 0.0635 m×0.0635 m× 0.0196 m stainless steel block. The mass wasaffixed to the inspection plate using silicon sealant tape thatwas applied to the entire contact surface of the mass. The useof masses to simulate damage has been used on many occa-sions to assess the capabilities of different SHM/NDE tech-niques [23, 24]. Figure 7(b) shows the mass was placed

Figure 7. Beamforming results for plate without and with mass. Results suggest that sound camera using acoustic beamforming can be usedfor detection of discontinuities.

8

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

halfway between the impact location and the center of theplate meaning the mass was approximately 0.215m from theimpact location. Figure 7 shows the wave propagation resultsfor the inspection plate without and with mass present.

Figure 7 uses a red cross and a red square to mark theapproximate location of the impact and mass, respectively, onthe plate. At t=0 ms, the instant of impact is shown for boththe without- and with-mass cases. Both images feature similaracoustic characteristics and neither image shows an indicationof a mass present on the plate. The images at t=0.25 msshow the wave front just after it has propagated past thelocation of the mass. The first difference is seen in the wavefront definition. While the no-mass case maintains a smoothand symmetric wave front, the mass case displays a moreturbulent wave front with a significant reduction of the wavefront amplitude on the far side of the mass relative to theimpact point. Additionally, the widespread saturation behindthe wave front seen in the no-mass case is not seen in thewith-mass case. In the with-mass case, the area behind thewave front maintains a higher acoustic level throughoutcompared to the no-mass case contributing to the less definedwave front. Finally, there is much more acoustic noise in thefarfield of the plate for the with-mass case compared to theno-mass case where the farfield acoustic level is approxi-mately equal to the amplitude behind the wave front att=0.25 ms.

At t=0.34 ms, the most noticeable difference betweenthe no-mass and with-mass cases is the high amplitudeacoustic event present at the location of the mass in the lattercase. For the no-mass case, the wave front is losing amplitudeand definition as it propagates and dissipates into the farfieldand the area behind the wave front is becoming saturated withlow amplitude acoustic levels. However, the backside of thewave front remains very symmetric with respect to the impactlocation and there is very little excessive acoustic featuresseen on the plate. In the with-mass case, the overall wavefront condition appears very similar to that for the no-masscase. In terms of dissipation and primary acoustic features, thewave fronts for the two cases display very similar amplitudesand structures at t=0.34 ms. However, despite the simila-rities the with-mass case wave front retains a much lessrefined wave front at this instance with much more acousticnoise seen in the farfield of the plate. As aforementioned, themost observable and desirable difference is the presence ofthe acoustic event emanating from the mass location denotedby the red arrow in figure 7 which alludes to the presence of adiscontinuity at this location on the inspection plate.

Similar to the impact localization, the maximum ampl-itude of the acoustic feature near the known location of themass was used as the identifying location of the mass. Thiswas then compared to the actual location of the center of the

mass given in table 2 and using equation (3) the normalizederror was calculated between the actual and identified positionwith respect to the spatial discretization. The calculated errorof only 5 spatial pixels represents a fairly high degree ofaccuracy when identifying the known position of the mass.Given that the mass is 63.5 mm× 63.5 mm, the mass itselfis 6.35 pixels× 6.35 pixels. While the exact coordinates ofthe actual and identified positions do not necessarily correlate,it is likely affected by the acoustic feedback from the mass notemanating from the center of the block. Therefore, if themaximum amplitude were located at one of the points of themass as it seems to be in figure 7 (corner nearest top edge ofplate), this alone would considerably affect the error value.Given the normalized dimensions of the mass, the identifiedposition of the acoustic emission denoting the location of themass is likely within the bounds of the mass despite therelatively large normalized error.

6. Conclusions

This study demonstrated the visualization of stress wavepropagation across an aluminum plate using an air-couplemicrophone array. The scope of the beamforming method forvisualizing stress waves was investigated through variousparametric studies that resulted in the establishment of post-processing parameters of upsample factor and spatial resolu-tion for further testing. Subsequent testing revealed the abilityof the sound camera to accurately identify various impactlocations on the inspection plate. Finally, the beamformingmethod was used to detect and localize a mass on theinspection plate used to simulate the presence of damage. Thesound camera and derived beamforming method were able tolocate the position of the mass through the detection ofacoustic emission structures indicating the existence of adiscontinuity on the inspection plate. This study showed thepotential for an acoustic emission beamforming based methodto be used for damage detection in SHM/NDE applications.While this study focused on the feasibility of the proposedtechnique, further studies need to be conducted to assess itssensitivity, to find more sophisticated beamforming techni-ques, and to optimize their post-processing parameters.Corresponding experiments should be conducted in order tocompare results to other experimental techniques, such aslaser Doppler vibrometry and digital image correlation tech-niques. Lastly, the authors plan to perform computationalstudies of air-coupled acoustic emission events using finiteelement method and probability of detection analysis.

Table 2. Quantification of mass detection and localization capability.

Mass Actual Identified Absolute NormalizedLocation Position (m, m) Position (m, m) Difference (m) Error (Num. Pixels)

Point 1 (−0.1524, 0.1524) (−0.1219, 0.193) 0.0508 5.08

9

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al

Acknowledgments

This research was supported by INNOPOLIS Foundationgrant funded by the Korean government (Ministry of Science,ICT & Future Planning, Grant number: 14DDI084) throughSM Instruments Inc. We also acknowledge the research grantfrom the Joint Center for Aerospace Technology Innovation(JCATI) from Washington State in the USA. We thankresearchers at SM Instruments, including InKwon Kim andJunGoo Kang, for their technical assistance. Lastly, J Racknowledges the support of the U.S. Air Force during thecourse of his Master’s degree at the University ofWashington.

References

[1] Giurgiutiu V 2015 Structural Health Monitoring of AerospaceComposites (New York: Elsevier)

[2] Johnson S B, Gormley T, Kessler S, Mott C, Patterson-Hine A,Reichard K and Scandura P Jr 2011 System HealthManagement: With Aerospace Applications (New York:Wiley)

[3] Wang M L, Lynch J P and Sohn H 2014 Sensor Technologiesfor Civil Infrastructures: Applications in Structural HealthMonitoring (Cambridge: Woodhead Publishing)

[4] Mix P E 2005 Introduction to Nondestructive Testing: ATraining Guide 2nd edn (New Jersey: John Wiley &Sons, Inc.)

[5] Djordjevic B B 2009 Nondestructive test technology for thecomposites Proc. 10th Int. Conf. Slovenian Soc. NDTpp 259–65

[6] Scruby C B 1987 An introduction to acoustic emission J. Phys.E 20 946

[7] Rizzo P and Lanza di Scalea F 2001 Acoustic emissionmonitoring of carbon-fiber-reinforced-polymer bridge staycables in large-scale testing Exp. Mech. 41 282

[8] Hellier C 2012 Handbook of Nondestructive Evaluation 2ndedn (New York: McGraw Hill)

[9] Farhidzadeh A, Mpalaskas A C, Matikas T E,Farhidzadeh H and Aggelis D G 2014 Fracture modeidentification in cementitious materials using supervisedpattern recognition of acoustic emission features Constr.Build. Mater. 67 129–38

[10] McLaskey G C, Glaser S D and Grosse C U 2008 Acousticemission beamforming for enhanced damage detection,

sensors and smart structures technologies for civil,mechanical, and aerospace systems Proc. SPIE 6932 1–9

[11] Grosse C U 2009 Acoustic emission localization methods forlarge structures based on beam forming and array techniquesProc. NDTCE 9 (Nantes, France)

[12] Olson S E, DeSimio M P and Derriso M M 2007 Beamforming of lamb waves for structural health monitoringJ. Vib. Acoust. 129 730–8

[13] Sengupt S, Datta A K and Topdar P 2015 Structural damagelocalisation by acoustic emission technique: a state of the artreview Latin Am. J. Solids Struct. 12 1565–82

[14] SM Instruments 2014 SeeSV-S205 Portable Sound CameraProduct Overview and Specifications (Daejeon, Korea: SMInstruments Inc)

[15] Dougherty R P 2014 Functional beamforming for aeroacousticsource distributions 20th AIAA/CEAS Aeroacoustics Conf.(Atlanta, GA)

[16] Mucci R 1984 A comparison of efficient beamformingalgorithms IEEE Trans. Acoust. Speech Signal Process. 32548–58

[17] Ruzzene M 2007 Frequency-wavenumber domain filtering forimproved damage visualization Smart Mater. Struct.16 2116

[18] Li F, Zhao L, Tian Z, Yu L and Yang J 2013 Visualization ofsolitary waves via laser Doppler vibrometry for heavyimpurity identification in a granular chain Smart Mater.Struct. 22 035016

[19] Yang J, Gonzalez M, Kim E, Agbasi C and Sutton M 2014Attenuation of solitary waves and localization of breathers in1D granular crystals visualized via high speed photographyExp. Mech. 54 1043

[20] Schreier H, Orteu J and Sutton M A 2009 Image Correlationfor Shape, Motion and Deformation Measurements: BasicConcepts, Theory and Applications (New York: Springer)

[21] Chen J C, Yao K and Hudson R E 2003 Acoustic sourcelocalization and beamforming: theory and practice EURASIPJ. Appl. Signal Process. 4 359–70

[22] Michel U 2006 History of acoustic beamforming BerlinBeamforming Conf. 21-22 (Berlin, Germany)

[23] Kody A, Li X and Moaveni B 2013 Identification of physicallysimulated damage on a footbridge based on ambientvibration data American Society of Civil EngineersStructures Congress pp 352–62

[24] Adams D 2007 Health Monitoring of Structural Materials andComponents: Methods with Applications (Chichester:Wiley)

[25] Hamstad M 2007 Acoustic emission signals generated bymonopole (pencillead break) versus dipole sources: finiteelement modeling and experiments J. Acoust. Emiss. 2592–106

10

Smart Mater. Struct. 26 (2017) 025020 J C Rivey et al