Contents lists available at ScienceDirect

Measurement

journal homepage: www.elsevier.com/locate/measurement

Aircraft integrated structural health monitoring using lasers,piezoelectricity, and fiber optics

Yunshil Choi, Syed Haider Abbas, Jung-Ryul Lee⁎

Department of Aerospace Engineering, Korea Advanced Institute of Science and Technology, Daejoen, Republic of Korea

A R T I C L E I N F O

Keywords:Integrated structural health monitoringLaser ultrasonicFiber Bragg GratingPiezoelectric

A B S T R A C T

Various structural health monitoring (SHM) systems that have been developed based on laser ultrasonics andfiber optics are introduced in this paper. The systems are used to realize the new SHM paradigm for ground SHM,called the Smart Hangar. Guided wave ultrasonic propagation imaging (G-UPI) technology is implemented in theSmart Hangar in the form of built-in and mobile G-UPI systems. The laser-induced guided waves generated bypulsed beam scanning in the wings and fuselage of an aircraft are captured by a fiber optic, piezoelectric, or laserultrasonic sensor, and their propagation is visualized. The wave propagation video is further processed to vi-sualize damage in the presence of multiple sources using a multi-time-frame ultrasonic energy mapping (mUEM)method. For in-flight SHM, laser scanning-based smartification of the structure with sensors is presented, and anevent localization method based on fiber optics and piezoelectric sensing is also introduced. Optic sensors arealso utilized to reconstruct the wing deformation from the measured strain. The wing deformation and impactlocalization information is transferred to a ground pilot in the case of unmanned aerial vehicles (UAV), and theground pilot can feel the wing deformation and impact by using a pilot arm-wearable haptic interface, whichmakes it possible to achieve human-UAV interactive decision making.

1. Introduction

Composite materials as aerospace engineering materials have var-ious advantages, including an optimal strength-to-weight, a good stiff-ness-to-weight ratio, and resistance to corrosion and fatigue. The usageof composite materials has been increasing rapidly in aerospace in-dustrial fields due to these benefits. In the past decade, aircraft manu-facturers have extended the ratio of composite uses so that its usage hasexceeded 50 percent in recently developed aircraft structures, such asfuselages and wing skins [1]. However, metallic materials are stillcomprising the main structures of aircrafts to prevent the critical issuesof composite materials; they are more brittle than wrought metals andcause more complex failure modes [2]. Particularly, aluminum alloys(Al-alloys) are utilized as the main materials for constructing airframes,such as skins, spars and ribs, and these alloys allow the structures toreduce the overall weight [3]. However, the structures with Al-alloysmay also experience both low and high cycle fatigue cracking duringtheir lifetime due to excessive and repeated loading environments. Inthe case of aircraft accidents by structural defects or damage, it is veryfatal not only in terms of cost but also for human safety. Thus, structuralhealth monitoring (SHM) technology is essential for reducing the risk ofcatastrophic failures and for improving the safety of structures by

detecting damage before it reaches a critical state [4].SHM for unmanned aerial vehicles (UAVs) also has been advanced

due to the increasing UAV employments in recent years. UAVs havemany advantages regarding safety. However, UAV accidents occur at amuch higher rate than accidents in most manned aircrafts; thus, it isessential to address the issue of UAV safety and performance [5].

Since a UAV is an aircraft without human pilots aboard, abnormaldynamic behaviors of the UAVs may be ignored, which can cause morefatal accidents such as destruction of the entire UAV structure. As arepresentative case, the in-flight break-up of NASA’s Helios wing wasreported in 2003 [6]. Various approaches have been studied regardingSHM for UAVs. Scalea et al. suggested a SHM system for a compositewing skin-to-spar joint in UAVs with ultrasonic guided waves and amacro-fiber composite transducer [7], and Park et al. presented animpact monitoring system with optical fiber sensors [8].

In this paper, smartification techniques for aircraft SHM are pro-posed in which the SHM systems are based on laser-scanning, piezo-electric (PZT) sensors, and Fiber Bragg Grating (FBG) sensors. InSection 2, laser-scanning-based G-UPI systems for fuselage structuresare introduced with an inspection result for simulated multisite cracks(MSCs). Impact localization with a laser-based training technique andits proof-of-concept is presented as a strategy of smartification of the

https://doi.org/10.1016/j.measurement.2018.04.067Received 23 June 2017; Accepted 18 April 2018

⁎ Corresponding author.E-mail address: leejrr@kaist.ac.kr (J.-R. Lee).

Measurement 125 (2018) 294–302

Available online 20 April 20180263-2241/ © 2018 Elsevier Ltd. All rights reserved.

T

fuselage. In Section 3, a wing smartification technique for an UAV isproposed based on pilot-UAV interactive decision making for SHM.

2. Smartification of the fuselage structure

2.1. Q-switched laser stimulation-based multisite damage visualization

Several fatal aircraft crashes have been reported over the last70 years; the de Havilland Comet crashes (1954), the Japan AirlinesFlight 123 Boeing 747SR crash (1985), the Aloha Airlines Flight 243Boeing 737-297 crash (1988), the China Airlines Flight 611 Boeing 747-209B crash (2002), and the Southwest Airlines Flight 2294 and the 812Boeing 737-3H4 crashes (2009 and 2011). These crashes were causedby MSCs on fuselage structures with metal. MSCs are typically origi-nating from riveted joints in a loaded fuselage and tears in the aircraft’sskin [9]. Various diagnostic techniques for SHM have been studied forfuselages structures. Because PZT transducers have several advantages,which include the small size, inexpensive cost and ability to be affixedin a minimally intrusive way to structures, these are one of the widelyselected transducers for SHM studies and applications. However, pre-vious studies showed the critical disadvantage that the suggested sys-tems implemented SHM only at the hotspots, such as at fastener holes[10]. This approach requires many transducers and sensing channels forthe general and wide structural areas.

In this section, a serially connected PZT sensor network (SCSN) andQ-switched laser stimulation near the PZTs is introduced for smartifi-cation of the airplane structure. A real metallic fuselage is used fordemonstration of this approach. The fuselage section with the SCSNresponds to the Q-switched laser beam stimulation and eventually vi-sualizes the damage. UPI is a laser ultrasonic testing technique that hasshown high feasibility for overcoming the drawbacks of previous stu-dies for global damage monitoring [11]. The UPI systems have beendeveloped into different versions, such as G-UPI using a surface waveand Lamb wave and a bulk wave (B)-UPI using pulse-echo and through-transmission modes. The G-UPI system monitors multisite damage si-multaneously by covering a wide area during a one-time inspection.Furthermore, the G-UPI system is able to cover a wider area by adaptingthe SCSN [12]. As shown in Fig. 1(a) and (b), a demonstration model(Cessna C-150) was prepared, and a SCSN was installed on the innersurface of the fuselage. The SCSN has many advantages, such as a widercoverage and low cost DAQ channel operation cost (a single channel is

used). Five artificial back-surface cracks were simulated for proof-of-concept at the five locations depicted in Fig. 1(c). Cracks A and B werefabricated with a length of 15mm and 10mm, respectively, as copies ofhidden cracks on the fuselage back-surface skin. Cracks C to E werefabricated with a length of 10mm each as copies of fatigue cracksemanating from the rivet holes in the lab joints. All of the cracks had adepth of 0.3 mm from the back surface.

Since it is hard to visualize the damage in the presence of multiplehigh energy sources inside the inspection area, a multi-source removalmethod was applied. The multi-source removal technique has two steps.First, multiple sources are automatically localized. This is done byfinding the first high energy point inside the inspection area as shownin Eqs. (1) and (2). We store the location of the highest energy point,which is the location of the first sensor because only the sensor locationpossess the energy of this high magnitude. To find the next source lo-cation, we apply a zero mask to our previous location and find the nextsensor location up to the point where all sensor locations are detectedas shown in the flow chart in Fig. 2.

∑==

−

E x y S i( , ) | ( ) |SLi

N

x y0

1

,2

(1)

=x y E x y( , ) argmax{ ( , )}SL SLx y

SL( , ) (2)

where Sx,y is the time domain signal, xSL and ySL are coordinates of thesensor location, and ESL(x,y) is the energy matrix of the complete scanarea. When the source locations are identified, multi-source removalmasks are applied at each source location by multiplying the originalraw data with a matrix mask having the same size as the scan area witha 0 value inside the radius (r) from the center at the sensor location and1 elsewhere. Mathematically, this is represented by

= ⎧⎨⎩

− + − ⩽− + − >

Mif i x r j y rif i x r j y r

0 ( / ) ( / ) 11 ( / ) ( / ) 1i j

SL SL

SL SL,

2 2

2 2(3)

The source removal algorithm was implemented in a mobile G-UPIfor inspecting a target, and the inspection results, which include a videoclip of the multisource ultrasonic wave propagation imaging (mUWPI)video clip and a multi-time-frame ultrasonic energy map (mUEM) [13],are presented in Fig. 3. Fig. 3(a) shows a freeze-frame of the multi-source UWPI video clip for the entire scan area. All MSCs shown inFig. 1(c) were successfully visualized by generating the scattering wave

Fig. 1. (a) Demonstration model (Cessna C-150) and health management area, (b) Serially connected PZT sensor network on the back surface of the fuselage, (c)Details of the artificial MSCs on the fuselage [11].

Y. Choi et al. Measurement 125 (2018) 294–302

295

out of the anomalous wave as enlarged freeze-frames in Fig. 3(b)–(g),which were captured at different time windows, respectively. ThemUEM result with the source removal technique was adapted based onthe freeze-frame results in Fig. 3(b)–(g). This method has the advantage

of a fast and automatic multiple source location and removal, which notonly helps to realize the damage in the presence of a strong source butalso enables the enlargement of the inspection area. Moreover, damagethat is visible in different time frames can be visualized in a single map

Fig. 2. Flowchart for automatic multi-source location identification.

Fig. 3. UWPI freeze frames at (a) 16.0 μs of the overall area, (b) 21.6 μs for crack A, (c) 21.6 μs for crack B, (d) 27.2 μs for crack E, (e) 37.7 μs for crack C, (f) 37.6 μsfor crack D, (g) 66.4 μs for crack C, and mUEM results (h) of the overall area, (i) detected damage dimensions for cracks A–E.

Y. Choi et al. Measurement 125 (2018) 294–302

296

without spending any additional processing time on the UEM result.Fig. 3(h) and (i) show the UEM result within the overall scan area andthe enlarged images for each MSC, respectively. Consistent sizing for allof the detected MSCs were implemented based on a threshold method[12]. This image result provides a far clearer visibility for detectedMSCs because the locations of the MSCs are able to be recognized im-mediately from the surrounding area in a single image map.

For this study, the mUEM was implemented in field programmablegate array (FPGA)-based processing hardware. FPGA has the advantageof performing multiple operations simultaneously. Fig. 4 shows theperformance graphs for the mUEM. Fig. 4(a) shows that the parallelismoffered by FPGA accelerated the mUEM results by processing 100,000scan points in less than 9 s. Similarly, it requires almost 0.5 s to find 8sensor locations in the scan area having 100,000 points as shown inFig. 4(b). This acceleration is made possible by performing parallelprocessing for multi-dimensional Fourier transform, source removal,and multi-time-frame wave energy computations.

2.2. Q-switched laser training for impact event localization

Conventional impact localization techniques have been studied withmultiple data acquisition channels for multiple transducers. Acousticemissions induced by an impact are mostly utilized for extracting ar-rival time differences or similarity comparison with pre-training signals[14,15]. A laser scanning application, which is introduced in this sec-tion, is also equipped with the identical G-UPI system for pre-training ofthe target structure shown in Fig. 1(a) and (b). Advantages of laser-based training are the high reliability and repeatability of the trainingdata and the significant decrease in training time compared withmanual training via manpower. Furthermore, the training data areeasily modulated up to the characteristics of real impact signals becauselaser ultrasonic signals are composed of a broad range of frequencycomponents.

Training data are acquired and stored by performing laser pulsescanning using the G-UPI system onto the target structure, which isembedded with the SCSN. This SCSN captures not only training data butalso impact-induced signals. Fig. 5 shows the impact localization al-gorithm with the laser-based training technique, which is triggered assoon as the impact signal is detected. After the training is completed,the SCSN passively monitors any impact-induced signal. During eventmonitoring, a high-pass filter is used to isolate high-frequency signalsfrom parasite vibrations, and a low-pass filter is also used together toreject high-frequency noise. Once an impact wave is detected, itsspectrum is obtained by fast Fourier transform and enveloped to findthe central frequency of fs. A certain band centered on fs is furthernumerically band-pass filtered one more time because a narrower bandprovides a more accurate comparison with laser impact signals. The

stored laser scanned signals are processed by exactly the same filteringprocess, and each processed laser impact signal and the processedmechanical impact signal are compared via cross correlation. Eachcross-correlation coefficient is divided by the geometric mean of thefirst norm value of each signal to eliminate the error caused by theamplitude difference depending on the propagation distance. Maximumcoefficient values are extracted and mapped on the scanned area. Theraw cross correlation map is still noisy, and a spatial averaging is ap-plied to smooth the map. Fig. 6 is the final cross correlation map, and(96, 73) is detected as the impact location where another second peak isshown in the right side of the map. The coefficient is smaller than thatof the detected impact location. Since the SCSN is actually a singlechannel with multiple sensing elements, high coefficients are symme-trically observed with respect to the sensor locations. However, sincethe reflected waves on the left and right sides are different, one of thetwo sides has a higher coefficient. Therefore, the signal channel systemis sufficient for this algorithm to localize the impact event.

More impact tests were performed for the proof-of-concept, andFig. 7 and Table 1 present the experimental results of the impact lo-calization with laser-based training data and the SCSN. Table 1 includesthe coordinates of attempted and detected locations within the scanarea and errors in distance for each attempt, and Fig. 7 is a grid plotshowing the attempted impact locations and detected impact locations.Twenty arbitrary impacts were applied onto the front surface of theCessna-150 fuselage structure, within the identical area depicted inFigs. 1 and 3. Laser-based training was also implemented with a spatialinterval of 1mm, and all of the laser-induced signals were filteredthrough an analogue programmable filter within the band-pass filteringrange of 5–180 kHz. As shown in Table 1, all trials were localized withsmall errors within a distance of 3.61–7.1mm. According to the results,laser-based training is clearly able to provide training data with a highspatial resolution of complex structures.

3. Smartification of the wing structure

The UPI system mentioned in Section 2 has many advantages, suchas a rapid inspection speed and non-contacting scanning. Long distancescanning is also one advantage of the UPI system. This ability is ac-complished by collimating the laser beam using beam expanders [16].The Smart Hangar, which is for the proof-of-concept of the SHM systemfor in-service aircraft structures, was built with a multi-area scanning G-UPI system and a long distance scanning configuration as shown inFig. 8 [17]. The system, which was built in the smart hangar, is capableof both long distance and dual area scanning via optical collimators,and the two tilting mirror systems were installed into the ceiling of thebuilding. Laser scanning was implemented on two different areas si-multaneously, and all generated ultrasonic signals were stored

Fig. 4. Performance graphs for the mUEM method. (a) Processing time vs. number of scan points to process mUEM with four time frames, (b) Time taken to find eightsensor locations for various scan points.

Y. Choi et al. Measurement 125 (2018) 294–302

297

separately. This system is also able to be utilized for SHM and laser-based training as mentioned in Section 2. Long distance and dual areascanning allows for inspections or training in inaccessible areas such asthe upper skin of wing structure.

In the previous smartification methods, the PZT transducers aremostly used to convert general structures into smart structures. TheFBG sensor is also an important selection for aircraft SHM depending onthe application. Since FBG sensors are very thin and light, they can beeasily embedded into structures. Furthermore, FBG sensors have im-munity from electro-magnetic, high-intensity radiated fields, and tran-sient signal interference or potential jamming; thus, they are widelyused in SHM of specific structures that are equipped with variouselectronic components such as in aircrafts [18].

This section describes the smartification of UAV structures usingFBG sensors and a pilot arm-wearable haptic interface. The systemaccomplishes its objective via human-robot interactions by acquiringspectral variations caused by deflection or impaction on the wingstructures. In general, ground pilots monitor massive amounts of in-formation from UAVs through the auditory or visual senses. If theground system feeds too much information to the pilots in the tradi-tional manner, it may reduce the multitasking efficiency of ground pi-lots. As an alternative sense for transferring extra information, thetactile sense is the fastest way to transfer any information to the humanbrain. Fig. 9 shows a concept for a SHM system for a UAV wingstructure. The objective of the system is to allow the ground pilot torecognize information about the structural health via stimulation fromthe arm-wearable interface and allow ground pilots make decisionsthemselves. Fig. 10 shows the arm-wearable haptic interface for sti-mulating human subjects. A total of nineteen vibro-motors were at-tached on an arm sleeve, and all vibro-motors were connected to a

microcontroller, which is for modulating the pulse width of the inputsignal [19]. Twelve motors represent the wing displacement in upwardand downward directions, respectively, and the other seven motorscorresponding to seven arbitrary divisions on the leading edge of thewing.

In this system, four FBG sensors were installed onto a half wingspecimen with an aluminum alloy, to measure the dynamic strains. FBGspectra were measured using an interrogator [20]. As shown in Fig. 11,the installed FBG sensors and the interrogator captured the wavelengthshift from each FBG sensor that was caused by deflection or impact.

3.1. Estimation of wing deflection

The first step of estimating wing deflection from wavelength shifts isa strain calculation from the wavelength shifts. Wing deflection dis-placement is estimated based on the strain data using a dis-placement–strain transformation matrix [21]. Six pieces of wing dis-placement data for certain locations were extracted from the resultantmatrix of the above estimation process and were converted into eightbits of binary data to be transferred to each corresponding vibro-motoronto the arm-wearable haptic interface. Positive wing deflection dis-placements were allocated within the range of 254–128, and negativewing deflection displacements were allocated within the range of128–0, where zero displacement is 127 in the binary form. Finally, thesix pieces of converted binary data for each location were transmittedto the haptic interface using a trigger signal, which is 255 for syn-chronization with the UAV wing deflection information.

Human subject testing was performed to verify the feasibility of theproposed system. The first test was performed to verify that wing de-flection information is recognizable to the participants. Fifteen

Start(Event detection)

Band-pass filtering in AE regime.FFT and find the central freq. Laser scanned training set

Narrow band-pass filtering for cross-correlation ( )

between event and laser spectra

Cross-correlation betweenevent and laser signals, in time

domain

Display location

Cross-correlation coefficient map with spatial averaging

Take maximum coefficient location as the event location

Fig. 5. Flow chart of the impact localization algorithm with laser-based training.

Fig. 6. Cross-correlation coefficient map after localization algorithm processing with laser-based training data (Detected location: (96, 73)).

Y. Choi et al. Measurement 125 (2018) 294–302

298

participants were given a brief instruction regarding the arm-wearablehaptic interface, and each participant conducted arbitrary wing de-flection detection tests 10 times. All tests were conducted as blind tests.All participants were able to sense downward or upward deflectioninformation 100% for 150 deflection testing times.

3.2. Impact localization

Wavelength shifts caused by an impact event are transferredthrough the serial communication and go through an impact localiza-tion algorithm. Seven imaginary areas are defined as shown in Fig. 12;thus, an index of area where the event location is included is trans-mitted to the arm-wearable haptic interface. It is assumed that all im-pact locations are at the leading edge of the wing structure becauseimpact events occur while the UAV is operating; thus, the algorithmwill solve simple linear equations. Fig. 13 shows the impact localizationalgorithm on the leading edge. The algorithm detects the arrival time ofthe signal from each FBG sensor based on Akaike information criterion(AIC) [22],

= + − − +AIC i k k var s i k N k var s i k N( , ) log ( [ ,1: ]) 1log ( [ , 1: ]) (4)

where i and k are indices of the FBG sensors and the length of the

discrete wavelength data, respectively, and s is ×i N of the wavelengthshifts from the interrogator. AICs are calculated for each wavelengthshift, and they are used for determining the global arrival time, ATG,i.e.,

∗ =AIC i k AT iargmin ( , ) Δ t ( )k

G (5)

where tΔ is the sampling interval of a discrete wavelength shift. Thealgorithm searches the minimum ATG, that corresponds to the ATG ofthe first arrival wavelength shift used for self-triggering. Thus, the re-levant arrival time of ATΔ G is calculated as

= −AT i AT i ATΔ ( ) ( ) min{ }G G G (6)

Relevant distance differences (Drel) are estimated by multiplying thewave propagation speed in shear (Vs) or longitudinal (Vl) to the relevantarrival times, and they are compared with relevant distance differencescalculated from the structural geometry (Dgeo). As a result of the com-parison, M(i), the coordinates along the x-axis, with the smallest errorsin distance are stored, and Mavg, a mean value of all detected M valuesis printed as the final result of the localization process. The arm-wearable interface receives an index of area, which corresponds to thelocation, Mavg, and stimulates human subjects according to the

0 100 200 300 400 500 600 700 800 900

0

50

100

150

200

: Attempted impact locations: Detected impact locations

: PZT patches

Fig. 7. Impact localization results for twenty attempts (Grid plot).

Table 1Impact localization results for the twenty attempts (Coordinates and distance errors).

Location 1 [100, 70] Location 2 [300, 100] Location 3 [500, 130] Location 4 [850, 190]

Detected location Error in Distance Detected location Error in Distance Detected location Error in Distance Detected location Error in Distance

1 [98, 73] 3.61 [296, 106] 7.21 [499, 126] 6.08 [847, 187] 7.622 [98, 75] 5.39 [301, 104] 4.12 [500, 126] 6.00 [848, 185] 5.393 [98, 73] 3.61 [295, 106] 7.81 [498, 126] 6.32 [849, 187] 7.074 [96, 76] 7.21 [299, 105] 5.10 [499, 127] 7.07 [851, 183] 3.165 [96, 73] 5.00 [297, 107] 7.62 [497, 126] 6.71 [851, 186] 6.08

Fig. 8. Demonstration setup used for the dual scanning UPI system with two tilting mirror systems installed on the hangar ceiling [17].

Y. Choi et al. Measurement 125 (2018) 294–302

299

localization result.A second test was also performed to verify that participants are able

to distinguish the impact localization results. Seven impacts were ap-plied on seven different locations corresponding to specific locations ineach area. Note that all impacts were applied along the edge of the half-wing specimen to simulate an impact event on the leading edge. Table 2

shows the localization results and answers from a participant for all ofseven attempts. The algorithm succeeded to localized event locationscorrectly in all attempts with distance errors from 9.3mm to 24.2mm.A participant received information through the arm-wearable interfaceand made identical decisions to the algorithm output results.

4. Conclusion

Aircraft SHM has become essential to reduce the budget and timerequired for guaranteeing structural safety. Four SHM technologies andtheir real-world implementation strategies were proposed in this paper.The mobile G-UPI system and SCSN was applied for Cessna-150 fu-selage smartification. The laser pulses from the mobile G-UPI systemperformed an inspection of a large area that was smartified by theSCSN. Thus, the system was able to perform MSC monitoring and laser-based structure training by presenting crack visualization and impactlocalization results. This technique showed its potential to smartify in-service aircraft structures under the SHM scheme. A FBG sensor-basedSHM system also has been introduced as a SHM solution for UAV.Wavelength shifts were measured by FBG sensors and were used to

Fig. 9. Concept of UAV wing smartification with FBG sensors and arm-wearable haptic interface (deflection and impact sensing on the wings) [19].

Fig. 10. An arm-wearable interface with vibro-tactile motors and micro-controller for impact monitoring.

Fig. 11. Estimation procedures for wing deflection and impact location.

Y. Choi et al. Measurement 125 (2018) 294–302

300

calculate deflection or impact locations. The arm-wearable haptic in-terface stimulated participants based on the estimated deflection orlocalization results. In the human subject testing, all participants wereable to recognize the vibrations clearly from the wearable interface andmake their own decisions based on the transmitted information. In thefuture, these suggested systems will contribute to reducing the oper-ating expenses for in-service commercial aircraft or UAVs.

Acknowledgments

This work was supported by the Technology Innovation Program(10074278) funded by the Ministry of Trade, Industry & Energy (MI,Korea).

References

[1] G.L. Dillingham, Aviation Safety-Status of FAA’s Actions to Oversee the Safety of

Composite Airplanes, United States Government Accountability Office, WashingtonDC, USA, 2011.

[2] K. Armstrong, L. Bevan, W. Cole, Care and Repair of Advanced Composites, Seconded., SAE International, Warrendale, USA, 2005.

[3] P. Jakab, Wood to metal: the structural origins of the modern airplane, J. Aircraft36 (1999) 914–918.

[4] C. Sbarufatti, A. Manes, M. Giglio, Application of sensor technologies for local anddistributed structural health monitoring, Struct. Control Health Monit. 21 (2014)1057–1083.

[5] J. Menda, J.T. Hing, H. Ayaz, P.A. Shewokis, K. Izzetoglu, B. Onaral, P. Oh, Opticalbrain imaging to enhance UAV operator training, evaluation, and interface devel-opment, J. Intell. Robot. Syst. 61 (1–4) (2011) 423–443.

[6] T.E. Noll, J.M. Brown, M.E. Perez-Davis, S.D. Ishmael, G.C. Tiffany, M. Gaier,Investigation of the Helios Prototype Aircraft Mishap volume I mishap report, NASALangley Research Center, USA, 2004.

[7] F.L.D. Scalea, H. Matt, I. Bartori, S. Coccia, G. Park, C. Farrar, Health monitoring ofUAV wing skin-to-spar joints using guided waves and macro fiber compositetransducers, J. Intell. Mater. Syst. Struct. 18 (4) (2007) 373–388.

[8] C.Y. Park, B.W. Jang, J.H. Kim, C.G. Kim, S.M. Jun, Bird strike event monitoring ina composite UAV wing using high speed optical fiber sensing system, Compos. Sci.Technol. 72 (4) (2012) 498–505.

[9] J. He, X. Guan, T. Peng, Y. Liu, A. Saxena, J. Celay, K. Goebel, A multi-feature

Fig. 12. Defined imaginary areas on the half-wing specimen.

Fig. 13. Flow chart of the impact localization algorithm with relevant arrival time differences.

Table 2Impact localization results and answers from participants.

Attempted locations [28.5, −52.9] [85.7, −48.7] [142.9, −44.4] [200.0, −40.2] [257.2, −36.0] [314.3, −31.7] [371.5, −27.5]Area index # 1 2 3 4 5 6 7Detected impact locations [50.4, −51.3] [98.8, −47.7] [167, −42.6] [181.7, −41.6] [278.7, −34.4] [323.6, −31.1] [350.4, −29.1]Detected area index # 1 2 3 4 5 6 7Errors in distance 21.8 13.1 24.2 18.3 21.5 9.3 21.1Answers from the human subject 1 2 3 4 5 6 7

Y. Choi et al. Measurement 125 (2018) 294–302

301

integration method for fatigue crack detection and crack length estimation inriveted lap joints using Lamb waves, Smart Mater. Struct. 22 (2013) 105007.

[10] E. Kostson, P. Fromme, Fatigue crack growth monitoring in multi-layered structuresusing guided ultrasonic waves, J. Phys.: Conf. Ser. 195 (2009) 012003.

[11] C. C. Chia, J. R. Lee, J. S. Park, et al., New design and algorithm for an ultrasonicpropagation imaging system, in: Proceedings of Defektoskopie, Brno, CzechRepublic, 2008, pp. 63–70.

[12] D.Y. Bae, J.R. Lee, Development of single channeled serial-connected piezoelectricsensor array and damage visualization based on multi-source wave propagationimaging, J. Intel. Mat. Syst. Str. 1–10 (2015).

[13] T.C. Truong, J.R. Lee, A versatile inspection system for pipe structure using ultra-sonic waves propagation imager, J. Phys.: Conf. Ser. 628 (2015) 012015.

[14] C.Y. Park, B.W. Jang, J.H. Kim, C.G. Kim, S.M. Jun, Bird strike event monitoring ina composite UAV wing using high speed optical fiber sensing system, Compos. Sci.Technol. 72 (2012) 498–505.

[15] L. Qiu, S. Yuan, H. Mei, W. Qian, Digital sequences and a time reversal-based impactregion imaging and localization method, Sensors 13 (2013) 13356–13381.

[16] J.R. Lee, H.J. Shin, C.C. Chia, D. Dhital, D.J. Yoon, Y.H. Huh, Long distance laserultrasonic propagation imaging system for damage visualization, Opt. Lasers Eng.

49 (2011) 1361–1371.[17] H.J. Shin, J.R. Lee, Development of a long range multi-area scanning ultrasonic

propagation imaging system built into a hangar and its application on an actualaircraft, Struct. Health Monit. Int. J. (2016) 1–15 online published.

[18] J.R. Lee, C.Y. Ryu, B.Y. Koo, S.G. Kang, C.S. Hong, C.G. Kim, In-flight healthmonitoring of a subscale wing using a fiber Bragg grating sensor system, SmartMater. Struct. 12 (1) (2003) 147–155.

[19] S.C. Hong, H.C. Kim, J.R. Lee, L.H. Kang, C.Y. Park, D.L Mascarenas, C.R. Farrar,Development of wireless and pilot arm-wearable haptic interface for UAV wingdeformation sensing, in: Proc. of Advanced in Structural Health Management andComposite Structures 2014, vol. 1, 2014, pp. 38-1–5.

[20] B.W. Lee, M.S. Seo, H.G. Oh, C.Y. Park, Real-time instrument for structural healthmonitoring, in: Proc. of the 18th International Conference on Composite Materials,2011.

[21] L.H. Kang, D.K. Kim, J.H. Han, Estimation of dynamic structural displacementsusing fiber Bragg grating strain sensors, J. Sound Vib. 305 (3) (2007) 534–542.

[22] P. Sedlak, Y. Hirose, M. Enoki, J. Sikula, Arrival time detection in thin multilayerplates on the basis of akaike information criterion, J. Acoust. Emission 26 (2008)182–188.

Y. Choi et al. Measurement 125 (2018) 294–302

302