sensors

Article Thermographic Inspection of Internal Defects in Steel Structures: Analysis of Signal Processing Techniques in Pulsed

Yoonjae Chung 1, Ranjit Shrestha 2 , Seungju Lee 3 and Wontae Kim 4,* 1 Department of Mechanical Engineering, Kongju National University, 1223-24 Cheonan-daero, Seobuk-gu, Cheonan-si 31080, Korea; [email protected] 2 Department of Mechanical Engineering, School of Engineering, Kathmandu University, Dhulikhel, P.O. Box, Kathmandu 6250, Nepal; [email protected] 3 Department of Convergence Mechanical Engineering, Kongju National University, 1223-24 Cheonan-daero, Seobuk-gu, Cheonan-si 31080, Korea; [email protected] 4 Division of Mechanical & Automotive Engineering, Kongju National University, 1223-24 Cheonan-daero, Seobuk-gu, Cheonan-si 31080, Korea * Correspondence: [email protected]; Tel.: +82-41-521-9289; Fax: +82-41-555-9123

 Received: 27 September 2020; Accepted: 21 October 2020; Published: 23 October 2020 

Abstract: This study performed an experimental investigation on pulsed thermography to detect internal defects, the major degradation phenomena in several structures of the secondary systems in nuclear power plants as well as industrial pipelines. The material losses due to wall thinning were simulated by drilling flat-bottomed holes (FBH) on the steel plate. FBH of different sizes in varying depths were considered to evaluate the detection capability of the proposed technique. A short and high energy pulse was deposited on a sample surface, and an infrared camera was used to analyze the effect of the applied heat flux. The three most established signal processing techniques of thermography, namely thermal signal reconstruction (TSR), pulsed phase thermography (PPT), and principal component thermography (PCT), have been applied to raw thermal images. Then, the performance of each technique was evaluated concerning enhanced defect detectability and signal to noise ratio (SNR). The results revealed that TSR enhanced the defect detectability, detecting the maximum number of defects, PPT provided the highest SNR, especially for the deeper defects, and PCT provided the highest SNR for the shallower defects.

Keywords: internal defects; ; pulsed thermography; thermal signal reconstruction; pulsed phase thermography; principle component analysis; signal to noise ratio

1. Introduction Steel is one of the world’s most versatile materials with excellent qualities of strength, corrosion resistance, weldability, formability, workability, and attractiveness. Also, having a long-life cycle and being 100% recyclable, it is the ideal material for a wide range of applications, such as architecture and construction, automotive and transportation, aircraft, chemical, food and beverage, energy, medical, and a variety of other applications. Steel is used in the construction of reactor vessels and vessel liners, reactor building structure, pressure channels, fuel cladding, heat exchanger and condenser tubes, and fuel pool liners in the nuclear industry as well as industrial pipelines [1–3]. A major challenge seen in these systems is the wall thinning due to flow accelerated corrosion and liquid droplet impingement. Such degradation mechanism affects the structural health and can to disastrous failures of the structure; as well as a massive economic loss [4–6]. This calls for a reliable testing and evaluation method to ensure that the functionality of the system is not compromised due to thickness reduction.

Sensors 2020, 20, 6015; doi:10.3390/s20216015 www.mdpi.com/journal/sensors Sensors 20202020,, 2020,, 6015x FOR PEER REVIEW 2 of 17

to thickness reduction. Recently, numerous nondestructive testing (NDT) techniques, such as x‐ray Recently,[7–9], numerous [10–12], nondestructive ultrasonic testing [13–16], (NDT) magnetic techniques, flux leakage such as [17–19], x-ray [7 –acoustic9], microwave emission [10 –[20–12], ultrasonic22], and eddy [13– 16current], magnetic [23–25], flux have leakage been [17 used–19], for acoustic the measurement emission [20 of–22 pipe], and wall eddy thinning. current Each [23–25 of], havethese beentechniques used for have the measurement their own advantages of pipe wall and thinning. limitations. Each ofThe these lists techniques of NDT havetechniques their own are advantagescontinuously and increasing limitations. as Theresearchers lists of NDT around techniques the globe are continuouslyare striving increasingfor the development as researchers of aroundtechniques the and globe methods are striving to guarantee for the development market superiority of techniques in terms andof quality methods and to productivity. guarantee market In this superioritycontext, infrared in terms thermography of quality and (IRT) productivity. plays a dynamic In this context,role to improve infrared the thermography quality and (IRT)productivity plays a dynamicof a scheme role from to improve creation the to qualitycompletion. and productivity of a scheme from creation to completion. IRT is is an an optical optical measurement measurement technique, technique, evolving evolving rapidly rapidly with withthe development the development of high of spatial high spatialresolution resolution and sensitivity and sensitivity detectors detectors and improved and improved computation computation power [26–28]. power [ IRT26–28 has]. IRTbeen has gaining been gainingincreased increased recognition recognition in recent in years recent because years because of its peculiar of its peculiar advantages advantages over its over counterparts. its counterparts. These Theseadvantages advantages include include high‐speed high-speed contactless contactless functioning, functioning, higher higherlevel of level safety, of safety,and portability and portability with a withpotential a potential to encompass to encompass a large a inspection large inspection area [29–31]. area [29 –IRT31]. utilizes IRT utilizes an infrared an infrared camera camera to extract to extract and andanalyse analyse a thermal a thermal pattern pattern based based on the on principle the principle that thatall objects all objects above above absolute absolute zero zero(−273 ( °C)273 emit◦C) − emitinfrared infrared energy. energy. The infrared The infrared radiation emitted emitted by an by object an object is sensed is sensed by the by infrared the infrared camera camera and andtransformed transformed into intoan electronic an electronic signal, signal, which which is then is then processed processed to produce to produce a thermal a thermal image image [32]. [ 32As]. Asshown shown in Figure in Figure 1, 1there, there exist exist several several types types of of thermographic thermographic approaches which cancan bebe classifiedclassified based on the procedure for testingtesting thethe structuralstructural components,components, the origin of the source ofof excitationexcitation energy, andand thethe temperaturetemperature didifferencesfferences ofof thethe constituents.constituents. Based on the experimental procedure, thermography can be classifiedclassified intointo activeactive andand passivepassive thermography.thermography. PassivePassive thermographythermography doesdoes not require any external source to excite surface thermal gradients as it is naturally at a higher or lowerlower temperature comparedcompared toto thethe background. In contrast, active thermography utilizes an external heating source to produce thermal contrast in the region of interest.interest. Based on the exciting method, active thermography can be further classifiedclassified asas pulsedpulsed thermographythermography (PT),(PT), lock-inlock‐in thermographythermography (LIT),(LIT), vibrothermographyvibrothermography (VT), (VT), and and step step heating heating thermography thermography (SHT) (SHT) [33 –[33–35].35]. PT andPT and LIT LIT are theare two the primarytwo primary conventional conventional practices. practices. In both In techniques, both techniques, thermal thermal energy isenergy delivered is delivered into the objectinto the in whichobject thein which heat propagates the heat propagates by diffusion by diffusion through through the material. the material. Then, the Then, thermal the thermal response response recorded recorded by the infraredby the infrared camera iscamera observed is observed to reveal theto reveal presence the of presence the defect. of Thethe experimentdefect. The can experiment be performed can be in twoperformed distinct in ways. two Indistinct transmission ways. In mode, transmission an infrared mode, camera an infrared and the excitation camera and source the excitation are kept opposite source theare samplekept opposite under investigation.the sample under In contrast, investigation. in reflection In contrast, mode, in an reflection infrared mode, camera an and infrared the excitation camera sourceand the are excitation positioned source identically are topositioned the sample identically being investigated to the sample [36,37]. being Furthermore, investigated various [36,37]. signal processingFurthermore, and various filtering signal techniques processing can be and used filtering to access techniques defects information, can be used including to access the defects small internalinformation, discontinuities including the and small material internal characteristics. discontinuities and material characteristics.

Figure 1. ClassificationClassification of common infrared thermography techniques used in NDT.

In the past few decades, research has demonstrated IRT to be useful in detection, identification, identification, and classification classification of of defects defects in composites, in composites, metals, metals, polymers polymers as well as as ceramic well as materials. ceramic G. materials. Busse et G.al. Busse[38,39] et proposed al. [38,39 ]aproposed four‐point a method four-point for methodLIT to extract for LIT amplitude to extract and amplitude phase angle and phaseimages angle and imagesalso demonstrated and also demonstrated that the phase that theangle phase image angle at imageeach pixel at each provides pixel provides the depth the information depth information of flat ofbottom flat bottom holes holes(FBH) (FBH) drilled drilled from from the rear the rear side side in a in polymer a polymer as as well well as as in in inhomogeneous and anisotropic materials such as CFRP. D. Wu and G. Busse [40] later revealed that a phase angle image

Sensors 2020, 20, 6015 3 of 17 anisotropic materials such as CFRP. D. Wu and G. Busse [40] later revealed that a phase angle image can eliminate the influences of uneven heating, surrounding reflections, and local disruptions like infrared emission coefficient and surface optical absorption. B.B. Lahri et al. [41] investigated the quantification of artificially produced FBH defects in composite, rubber, and aluminium structures using LIT and concluded that defects with a radius to depth ratio greater than 1.0 are detectable, which was in agreement with the empirical rule of thumb developed by V. Vavilov and R. Taylor [42]. C. Wallbrink et al. [43] observed the influence of defect size on the assessment of defect depth on a steel structure with FBH using LIT and concluded that the defect size had a significant effect on the phase angle which subsequently has consequences on the defect depth. C. Manyong et al. [44] and L. Junyan et al. [45] used LIT to evaluate the FBH defects in steel structure and showed that the phase angle differences between the defective regions and sound regions provide the defect location and size. D. Sharath et al. [46] studied on the blind frequency and phase contrast methods to evaluate the artificial square, rectangular and circular defects in austenitic stainless steel using LIT. It was concluded that blind frequency is dependent on the defect size and shape and could not be applied for depth quantification unless a normalization method is used. It was also concluded that phase contrast is a function of defect size and shape, apart from depth. S. Ranjit [47] investigated the effect of modulation frequency on the defect detectability on stainless steel with artificial FBH defects using LIT and concluded that the correct excitation frequency range must be selected for the inspected material to detect the defects. Although there has been increasing attention with LIT based structural health monitoring, longer observation time could be the main disadvantage of LIT as the experiment needs to be performed at multiple frequencies. To address this shortcoming, the PT played the important role in NDT. X. Maldague et al. [48] used the PT to inspect different kinds of defects in aluminum structures and showed that the heat pulse of energy ranging from 5 kJ to 20 kJ is enough to excite the object under inspection and emphasize on the frame rate with a high frequency to resolve the thermal process. C. Deemer [49] used temperature contrast method based PT to examine the carbon/carbon and continuous fiber ceramic composite components with drilled FBH at the back surface. C. Ibarra-Castanedo et al. [50] investigated the fact that PT raw data is commonly affected by problems, such as uneven heating, emissivity variations, reflections from the environment, and surface geometry sharp variations. Nevertheless, a wide variety of PT data processing techniques were developed to carry out qualitative and quantitative inspections. O. Wysocka-Fotek et al. [51] used standard thermal contrast method based on PT to estimate the size and depth of FBH defects in austenitic steel. X. Maldague et al. [52,53] combined the facets of PT and LIT techniques and proposed pulsed phase thermography (PPT). It was demonstrated that PPT can detect the deeper defects with other features such as less influence of optical characteristics and surface infrared, and the possibility to inspect test sample with higher thermal conductivity. S.M. Shepard et al. [54] proposed the thermal signal reconstruction (TSR) technique and analyzed the experimental PT data to detect the FBH defects in steel structure. It was also concluded that the time derivative of the reconstructed data detects the defect with high sensitivity at earlier times with better SNR. N. Rajic [55] applied a singular value decomposition method based principal component thermography (PPT) to the NDT of the composite structure and demonstrated the evidence of reduction in noise to a considerable extent with the high level of defect contrast. A. Vageswar et al. [56] investigated the peak contrast slope method using PT in transmission mode to estimate the depth of FBH defects in steel structures. Z. Zeng et al. [57] used PT to investigate the steel sample with FBH defects filled with different materials to replicate different non-air interfaces. The comparison between peak slope time method, absolute peak slope time method, logarithmic peak second-derivative method, and peak contrast time method showed that the later one is extremely influenced by defect size and interface while the other three techniques provided accurate defect depth. P. Simon and A. Darryl [58] used PT and LIT techniques to detect the FBH defects in CFRP and compared their results with respect to SNR. The comparison revealed the better performance of PT with a higher SNR for shallower defects. However, in the case of deeper defects, both the PT and LIT provided approximately the same SNR. D. Duan [59] investigated the Sensors 2020, 20, 6015 4 of 17

Sensors 2020, 20, x FOR PEER REVIEW 4 of 17 detection capability and reliability of PT and LIT techniques to detect FBH defects in aluminium foams introducingand the results probability of each technique of detection were analysis. compared C. Krishnendu in terms of et SNR. al. [ 60It was] researched concluded the that testing PT provided of carbon fibrehigher composite SNR for the sample shallower using PT,defects LIT, whereas and modulated for the deeper thermography, defects all and the the three results techniques of each techniqueprovided werethe approximately compared in same terms SNR. of SNR. This It fact was showed concluded that thatthe selection PT provided of the higher most adequate SNR for theIRT shallowerapproach defectsdepends whereas on the particular for the deeper application defects alland the the three available techniques experimental provided and the expertise approximately resources. same SNR. ThisIn fact this showed work, that an theexperimental selection ofthermographic the most adequate inspection IRT approach process of depends detecting on wall the particular thinning applicationdefects in steel and structure the available is presented. experimental A test and sample expertise with FBH resources. was considered as a best‐case scenario to representIn this work, real wall an experimental thinning defects thermographic as the difference inspection between process real of wall detecting thinning wall defects thinning and defects FBH indefects steel are structure relatively is presented. small. The Aprimary test sample purpose with of FBH the study was considered was to investigate as a best-case the capabilities scenario toof representthe PT to detect real wall the defects thinning in defectsthe referenced as the disample.fference The between second purpose real wall was thinning to employ defects the andstandard FBH defectssignal processing are relatively techniques, small. The namely, primary TSR, purpose PPT, of and the PCT, study and was then to investigate evaluate their the capabilities performance of the in PTterms to detectof defect the detectability defects in the and referenced SNR. sample. The second purpose was to employ the standard signalThe processing remaining techniques, part of the namely,research TSR, article PPT, is arranged and PCT, as and per thenthe followings: evaluate their Section performance 2 introduces in termsa quick of review defect detectabilityof PT. Section and 3 SNR.provides the principle and mathematical representation of signal processingThe remaining techniques, part namely, of the research TSR, PPT article and isPCT. arranged Section as 4 per presents the followings: the details Section of the2 testintroduces sample aand quick experimentation. review of PT. Section Section 53 discusses provides the the results principle and andanalysis mathematical of each signal representation processing technique. of signal processingFinally, Section techniques, 6 outlines namely, the conclusions TSR, PPT and possible PCT. Section future4 presents works. the details of the test sample and experimentation. Section5 discusses the results and analysis of each signal processing technique. Finally,2. Pulsed Section Thermography6 outlines the conclusions and possible future works. Pulsed thermography (PT) is among the most prevalent thermographic techniques in NDT. 2. Pulsed Thermography Figure 2 depicts the principle of PT where a high‐power pulse heating is employed to the sample underPulsed inspection thermography and the (PT)response is among of the the sample most prevalentis recorded thermographic with an infrared techniques camera. in NDT. The Figurecontinuance2 depicts of thethe principlepulse depends of PT where on the a thermal high-power conductivity pulse heating of the is employedsample under to the inspection sample under and inspectionruns between and 2 themilliseconds response ofto the10 milliseconds, sample is recorded making with PT anwell infrared known camera. for its quickness The continuance in testing. of theAs soon pulse as depends the sample on the got thermal excited, conductivity the surface absorbs of the sample the light under energy inspection and temperature and runs between increases 2 millisecondsinstantly. Due to to 10 the milliseconds, propagation making of a thermal PT well wave known inside for the its sample, quickness the surface in testing. temperature As soon asstarts the sampleto decay. got Imperfections excited, the surface could be absorbs seen if the there light is energy variance and in temperature the thermal increases decay rate instantly. across the Due sample to the propagationsurface [61–63]. of a thermal wave inside the sample, the surface temperature starts to decay. Imperfections could be seen if there is variance in the thermal decay rate across the sample surface [61–63].

Figure 2. Experimental configuration of a pulsed thermography inspection system. Figure 2. Experimental configuration of a pulsed thermography inspection system.

The one‐dimensional solution of the Fourier equation for the propagation of a Dirac heat pulse in a semi‐infinite isotropic solid by conduction can be expressed by Equation (1) [64].

Sensors 2020, 20, 6015 5 of 17

The one-dimensional solution of the Fourier equation for the propagation of a Dirac heat pulse in a semi-infinite isotropic solid by conduction can be expressed by Equation (1) [64]. ! Q z2 T(z, t) = T0 + exp (1) e √πt −4αt where T (K) is the temperature rise at the time (t) after the flash heating, T (K) the initial temperature, q 0 2 1/2 2 Q (W/m ) the energy deposited on the surface, e = kρcp (J/s m K) is the material’s thermal effusivity to exchange heat with its surrounding, α = k (m2/s) is the material’s thermal diffusivity ρ.Cp 3 with k (W/m.K) being the thermal conductivity, ρ (kg/m ) the density, and Cp (J/kg.K) the specific heat capacity. At the surface (z = 0 mm), Equation (1) can be rewritten as Equation (2) [65].

Q T(t) = (2) e √πt

3. Signal Processing Techniques PT results are often evaluated by selecting the unprocessed thermal image with the maximum contrast between the sound and defective area. Recently, many signal processing algorithms have been applied to raw thermal images to access the defects.

3.1. Thermography Signal Reconstruction Thermography signal reconstruction (TSR) is a renowned signal processing technique to improve and distinguish pulsed thermographic images, sorting out spatial and temporal non-uniform noise and suppressing the later one. Furthermore, it provides the log-log thermal response of each pixel of sequence with a polynomial of the nth degree. The logarithmic transform of Equation (2) can be expressed by Equation (3) [54,66,67]   Q 1 2 n ln(T) = ln ln(πt) = a + a ln(t) + a [ln(t)] + ... + an[ln(t)] (3) e − 2 0 1 2 where T is the increment in temperature as a function of time and a0, a1, ... ,an are the polynomial coefficients. In NDT, fifth or sixth order polynomial was found effective in temporal noise reduction and the optimum degree of the polynomial in use was found up to ninth order. The first log-time derivative dln(T)/dln(t) and the second derivative d2 ln(T)/dln2(t) of the polynomial enhanced the defect detectability and SNR [68,69].

3.2. Pulsed Phase Thermography In pulse phase thermography, the discrete Fourier transform is applied to transform each pixel of thermal image sequence from a time domain to frequency domain and can be expressed by Equation (4) [52,53,70]: N 1 X− j2πnk Fn = ∆t T(k∆t) exp N = Ren + Imn (4) k=0 where Re and Im are, respectively, the real and imaginary parts of the transform sequence, Fn, N is the number of thermal images, j2 = 1 is the imaginary number, and ∆t is the sampling time interval. − The phase component (∅) of the transformed data can be obtained using Equation (5) [53,70].   1 Imn ∅n = tan− (5) Ren Sensors 2020, 20, 6015 6 of 17

3.3. Principle Component Thermography Principle component thermography (PCT) facilitates in the reduction of undesirable data while preserving the major features of thermal image sequences. The recorded 3-D matrix of the thermal image sequence is transformed into a 2-D matrix, A of dimension M N. Then, the singular value × decomposition method is used to extract the spatial and temporal information and can be expressed by Equation (6) [55,71]. A = USVT (6) where U is the orthogonal matrix of dimension M N and comprises a set of empirical orthogonal × functions (EOFs) representing spatial variation. S is a diagonal matrix of dimension N N and × possesses the singular values of matrix A on the diagonals. VT is the transpose matrix of the N N × orthogonal matrixSensors representing2020, 20, x FOR PEER REVIEW characteristic time. 6 of 17

4. Methods anddecomposition Materials method is used to extract the spatial and temporal information and can be expressed by Equation (6) [55,71]. 4.1. Test Sample A USV (6) where U is the orthogonal matrix of dimension MxN and comprises a set of empirical orthogonal A samplefunctions made of(EOFs) austenitic representing stainless spatial variation. steel (SUS S is a 316)diagonal with matrix the of flat-bottomed dimension NxN and holes (FBH) was considered to simulatepossesses the wall singular thinning values of defect. matrix A Figureon the diagonals.3 depicts VT is the the schematictranspose matrix illustration of the NxN of the sample, orthogonal matrix representing characteristic time. along with the geometry and location of artificial FBH. It is a square-shaped plate with dimensions 180 180 mm and4. Methods thickness and Materials of 10 mm. Artificial FBHs of different sizes at varying depth levels were × milled at the rear4.1. Test surface Sample to evaluate the detection capability of the proposed technique. The FBHs in each column representsA sample the made defects of austenitic with stainless a constant steel (SUS diameter, 316) with the but flat‐ ofbottomed different holes depth(FBH) was from uppermost to lowermost row.considered Referring to simulate to wall Figure thinning3, defect. ‘A’, ‘B’, Figure ‘C’, 3 depicts and the ‘D’ schematic represent illustration defects of the alongsample, the horizontal along with the geometry and location of artificial FBH. It is a square‐shaped plate with dimensions rows and have180 di ff× 180erent mm defectand thickness diameters. of 10 mm. Artificial Similarly, FBHs ‘1’, of different ‘2’, ‘3’, sizes and at varying ‘4’ represent depth levels the were defect size along with vertical columnsmilled at the and rear have surface di toff evaluateerent the depths. detection All capability the sixteen of the proposed defects technique. are indexed The FBHs within a distinctive each column represents the defects with a constant diameter, but of different depth from uppermost ‘Defect ID’ usingto lowermost these representations. row. Referring to Figure For 3, ‘A’, instance, ‘B’, ‘C’, and defect ‘D’ represent ID of defects defect along in the the horizontal first row and the first column is ‘A1’rows which and have has different a diameter defect diameters. of 16 mm Similarly, and ‘1’, depth ‘2’, ‘3’, of and 2 ‘4’ mm; represent defect the defect ID of size defect along in second row and second columnwith vertical is ‘B columns’ which and hashave different a diameter depths. of All 4 the mm sixteen and defects depth are indexed 5 mm; with defect a distinctive ID of defect in third ‘Defect ID’ using2 these representations. For instance, defect ID of defect in the first row and the first row and third columncolumn is ‘A is1’ givenwhich has as a ‘Cdiameter3’ which of 16 mm has and a diameterdepth of 2 mm; of defect 8 mm ID of and defect depth in second of 3row mm; defect ID of and second column is ‘B2’ which has a diameter of 4 mm and depth 5 mm; defect ID of defect in third defect in the last row and the last column is given as ‘D4’ which has a diameter of 12 mm and depth of row and third column is given as ‘C3’ which has a diameter of 8 mm and depth of 3 mm; defect ID of 4 mm, and so on.defect The in the front last row side and of the the last samplecolumn is given was as painted ‘D4’ which black has a diameter using of KRYLON 12 mm and depth flat paint having an emissivity of 0.95of 4 tomm, create and so on. a uniform The front side emissive of the sample surface. was painted Figure black4 usingdepicts KRYLON the flat photographs paint having of the sample an emissivity of 0.95 to create a uniform emissive surface. Figure 4 depicts the photographs of the considered forsample the study. considered for the study.

Figure 3. SchematicFigure 3. illustration Schematic illustration of austenitic of austenitic stainless stainless steel steel (SUS (SUS 316) sample 316) along sample with the along geometry with the geometry and location of artificial flat‐bottomed holes representing locally wall thinned areas of different sizes and location ofat artificial varying depth flat-bottomed levels. holes representing locally wall thinned areas of different sizes at varying depth levels. Sensors 2020, 20, 6015 7 of 17 Sensors 2020, 20, x FOR PEER REVIEW 7 of 17

Figure 4. Photographs of the sample used for the study, (a) front side with black paint and (b) rear side Figure 4. Photographs of the sample used for the study, (a) front side with black paint and (b) rear with flat-bottomed holes. side with flat‐bottomed holes. 4.2. Experimentation 4.2. Experimentation In this research, the experimental setup consisted of a test sample, short and high power heating sourceIn with this theresearch, system the controller, experimental an infrared setup consisted camera and of a computer.test sample, As short a thermal and high excitation power source,heating asource flash lamp with (Universal the system BALCAR, controller, Rungis, an infrared Paris, camera France) and of power a computer. 6400 W-s As wasa thermal used and excitation controlled source, by aa power flash lamp box (BALCAR (Universal Light BALCAR, System, Rungis, Nexus AParis, 6400, France) France). of An power infrared 6400 camera W‐s was (SC655, used and FLIR controlled Systems, Danderyd,by a power Sweden) box (BALCAR with a spectralLight System, range 7.5–13Nexus µAm, 6400, thermal France). sensitivity An infrared (NETD) camera<50 mk, (SC655, accuracy FLIR Systems,2 C, and Danderyd, a 640 480 Sweden) focal plane with array a spectral detector range with 7.5–13 the frame μm, rate thermal of 50 sensitivity Hz was used (NETD) to record <50 themk, ± ◦ × thermalaccuracy response ±2 °C, and of the a 640 test × sample.480 focal A plane computer array (MSIdetector GE620DX) with the was frame used rate to of control 50 Hz the was infrared used to camerarecord andthe thermal to store response the experimental of the test data. sample. FLIR A R&D computer software (MSI was GE620DX) used for was the used pre-processing to control the of rawinfrared thermal camera data. and FLIR to R&D store also the includes experimental the features data. of FLIR camera R&D control, software high-speed was used data for recording, the pre‐ imageprocessing analysis, of raw and thermal data sharing. data. DataFLIR recordingR&D also options includes include the features start time, of camera end time, control, and the high number‐speed ofdata frames recording, to acquire. image analysis, and data sharing. Data recording options include start time, end time, and the number of frames to acquire. 5. Results and Discussions 5. Results and Discussions The experiment was conducted in a dark and closed chamber at Thermal Design and Infrared LaboratoryThe experiment of Kongju Nationalwas conducted University, in a Korea.dark and The closed temperature chamber of at the Thermal chamber Design was set and 22 Infrared◦C and relativeLaboratory humidity of Kongju was maintained National University, 48%. To evaluate Korea. The the thermaltemperature response of the of chamber the front was surface, set 22 the °C test and samplerelative was humidity set upright, was maintained and reflection 48%. mode To evaluate was chosen. the thermal The sample response front of the surface front was surface, excited the test by asample pulse heat was forset aupright, duration and of reflection approximately mode 10was ms. chosen. Then, The the sample thermal front response surface of thewas sample excited was by a recordedpulse heat for for 5 s. a duration of approximately 10 milliseconds. Then, the thermal response of the sample wasFigure recorded5a depicts for 5 s. the thermal image at time 0 s recorded before the application of a pulse heating. FigureFigure5b depicts 5a depicts the thermal the thermal image image at time at time 0.1 s0 recordeds recorded after before the the deposition application of of heat a pulse flux heating. where theFigure detected 5b depicts defects the exhibit thermal the image higher at signal-to-background time 0.1 s recorded after contrast the deposition without performingof heat flux where any data the processingdetected defects techniques. exhibit Figure the5 higherc depicts signal the thermal‐to‐background image which contrast is a resultwithout of subtractingperforming Figure any data5a fromprocessing Figure 5techniques.b. The results Figure showed 5c depicts that the the shallower thermal image defects which with is larger a result size of were subtracting detected Figure clearly, 5a whereasfrom Figure the deeper 5b. The defects results with showed smaller that size the wereshallower detected defects faintly. with To larger investigate size were the detected effects of clearly, defect geometrywhereas onthe the deeper applied defects heat with flux, smaller the thermal size response were detected of the defectsfaintly. withTo investigate different aspect the effects ratio (ratioof defect of diametergeometry to on depth) the applied was analyzed. heat flux, Defects the Athermal1, A4,C 1,responseand C4 withof the an defects aspect ratiowith of different 8.0, 6.0, 5.3,aspect and ratio 4.0, respectively,(ratio of diameter were chosen to depth) for thewas investigation. analyzed. Defects The ROI A1, ofA4 4, C1,4 and pixels C4 at with the an centre aspect of each ratio defect of 8.0, was 6.0, × considered,5.3, and 4.0, and respectively, the average were mean chosen temperature for the wasinvestigation. measured. The Figure ROI6 depictsof 4 × 4 thepixels thermal at the response centre of ofeach the defect selected was defects considered, concerning and the the average time of 5mean s. It temperature was observed was that measured. the temperature Figure 6 attained depicts bythe thethermal sample response was faster of thanthe itsselected decay defects because concerning of short impulse the time time. of As 5 time s. It passed, was observed the temperature that the ontemperature the surface attained tends to by reach the sample equilibrium. was faster It was than worth its decay noting because that the of shallowershort impulse defect time. A1 Aswith time a highpassed, aspect the ratio temperature causes significant on the surface temperature tends ditoff erencereach equilibrium. when compared It was with worth other noting defects that with the a shallower defect A1 with a high aspect ratio causes significant temperature difference when compared with other defects with a smaller aspect ratio. The variation in the surface temperature

Sensors 2020, 20, x FOR PEER REVIEW 8 of 17 Sensors 2020, 20, 6015 8 of 17 Sensorsdistribution 2020, 20, x was FOR enoughPEER REVIEW to distinguish the defects. However, the processing of thermal images8 of was 17 done to reduce noise and enhance the defect detectability. The algorithms stated in Section 3 were distribution was enough to distinguish the defects. However, the processing of thermal images was smallercompiled aspect by ratio.using TheMATLAB variation® R2019a. in the surface temperature distribution was enough to distinguish thedone defects. to reduce However, noise and the enhance processing the of defect thermal detectability. images was The done algorithms to reduce stated noise in and Section enhance 3 were the defectcompiled detectability. by using MATLAB The algorithms® R2019a. stated in Section3 were compiled by using MATLAB ® R2019a.

Figure 5. Experimental pulsed thermal images on the front surface of the flat‐bottomed holes sample Figureafter pre 5. ‐processing,Experimental (a) pulsedthermal thermal image at images time 0 s on before the the front application surface of of thepulse flat-bottomed heating, (b) thermal holes Figure 5. Experimental pulsed thermal images on the front surface of the flat‐bottomed holes sample sampleimage after at time pre-processing, 0.1 s after the (a) thermaldeposition image of heat at time flux 0 swhere before defect the application exhibit the of higher pulse heating,signal‐to‐ after pre‐processing, (a) thermal image at time 0 s before the application of pulse heating, (b) thermal (bbackground) thermal image contrast at time and 0.1(c) thermal s after the image deposition which is of result heat of flux subtracting where defect Figure exhibit 5a from the Figure higher 5b image at time 0.1 s after the deposition of heat flux where defect exhibit the higher signal‐to‐ signal-to-background[Frequency = 50 Hz, contrastnumber andof frames (c) thermal = 250, image truncation which window is result = of 5 subtractings]. Temperature Figure scales5a from are in background contrast and (c) thermal image which is result of subtracting Figure 5a from Figure 5b Figuredegree5b Celsius. [Frequency = 50 Hz, number of frames = 250, truncation window = 5 s]. Temperature scales [Frequency = 50 Hz, number of frames = 250, truncation window = 5 s]. Temperature scales are in are in degree Celsius. degree Celsius.

25.00 Defect A ; size = 16 mm, depth= 2 mm, aspect ratio = 8.0 1 Defect A ; size = 12 mm, depth= 2 mm, aspect ratio = 6.0 24.75 4 25.00 Defect C ; size = 16 mm, depth= 3 mm, aspect ratio = 5.3 1 Defect A ; size = 16 mm, depth= 2 mm, aspect ratio = 8.0 24.50 Defect1 C ; size = 12 mm, depth= 3 mm, aspect ratio = 4.0 4 Defect A ; size = 12 mm, depth= 2 mm, aspect ratio = 6.0 24.75 4 24.25 Defect C ; size = 16 mm, depth= 3 mm, aspect ratio = 5.3 1 C) 0

24.50(

Defect C ; size = 12 mm, depth= 3 mm, aspect ratio = 4.0 24.00 4

24.25 C)

0 23.75 (

24.00

23.50 Temperature 23.75 23.25 23.50 Temperature 23.00 23.25 22.75 23.00 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Time (s) 22.75 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Figure 6. Defect detection from temperature profilesTime of (s) flat-bottomed holes in a steel plate subjected to Figure 6. Defect detection from temperature profiles of flat‐bottomed holes in a steel plate subjected pulse heating [Frequency = 50 Hz, number of frames = 250, truncation window = 5 s]. to pulse heating [Frequency = 50 Hz, number of frames = 250, truncation window = 5 s]. Figure 6. Defect detection from temperature profiles of flat‐bottomed holes in a steel plate subjected 5.1. Thermalto pulse Signalheating Reconstruction [Frequency = 50 Results Hz, number of frames = 250, truncation window = 5 s]. 5.1. Thermal Signal Reconstruction Results The TSR processing was applied to each pixel of the raw thermal images to reconstruct the logarithmic5.1. ThermalThe TSR Signal time processing dependence Reconstruction was and applied Results the corresponding to each pixel first of the and raw second thermal derivatives. images to The reconstruct logarithmic the logarithmic time dependence and the corresponding first and second derivatives. The logarithmic polynomialThe TSR coe processingfficients up was to 8th applied degree to were each evaluated. pixel of the To raw perceive thermal the eimagesffects of to TSR reconstruct process, the polynomial coefficients up to 8th degree were evaluated. To perceive the effects of TSR process, the thermallogarithmic image time at timedependence 0 s (Figure and5a) the and corresponding the thermal image first atand time second 0.1 s (Figurederivatives.5b) were The selected logarithmic for thermal image at time 0 s (Figure 5a) and the thermal image at time 0.1 s (Figure 5b) were selected thepolynomial further investigation. coefficients up Figure to 8th7 depicts degree thewere results evaluated. of TSR To processing perceive ofthe the effects selected of TSR thermal process, images. the for the further investigation. Figure 7 depicts the results of TSR processing of the selected thermal Inthermal Figure image7, the at first time column 0 s (Figure depicts 5a) the and TSR the results thermal for image the thermal at time image0.1 s (Figure at time 5b) 0 s;were the selected second images. In Figure 7, the first column depicts the TSR results for the thermal image at time 0 s; the columnfor the further depicts investigation. the TSR results Figure for the 7 thermal depicts image the results at time of 0.1 TSR s; the processing third column of the depicts selected the thermal second column depicts the TSR results for the thermal image at time 0.1 s; the third column depicts imageimages. acquired In Figure after 7, subtractingthe first column the thermal depicts image the TSR of first results column for fromthe thermal the second image column; at time the 0 fourth s; the the thermal image acquired after subtracting the thermal image of first column from the second columnsecond column depicts thedepicts first derivativethe TSR results of the for first the column thermal thermal image image at time and 0.1 5th s; column the third depicts column the depicts second column; the fourth column depicts the first derivative of the first column thermal image and 5th derivativethe thermal of image the second acquired column after thermal subtracting image; forthe dithermalfferent degreesimage of of first polynomial column coe fromfficients. the second It was column depicts the second derivative of the second column thermal image; for different degrees of observedcolumn; the that fourth the selection column of depicts an appropriate the first polynomial derivative degreeof the first in the column regression thermal process image plays and a vital 5th polynomial coefficients. It was observed that the selection of an appropriate polynomial degree in rolecolumn to fit depicts the thermal the second data. Toderivative measure of the the quality second of column TSR fitting, thermal root image; mean square for different error (RMSE) degrees for of the regression process plays a vital role to fit the thermal data. To measure the quality of TSR fitting, eachpolynomial degree coefficients. of polynomial It was fitting observed was evaluated. that the Forselection this purpose, of an appropriate the defect Apolynomial1 with a high degree aspect in root mean square error (RMSE) for each degree of polynomial fitting was evaluated. For this purpose, ratiothe regression of 8 was process considered, plays and a vital the role temperature to fit the thermal at the centre data. To was measure measured. the quality Figure 8of depicts TSR fitting, the the defect A1 with a high aspect ratio of 8 was considered, and the temperature at the centre was residuals,root mean and square Table error1 presents (RMSE) the for RMSE each degree as a function of polynomial of the numbers fitting was of coe evaluated.fficients For for thethis defect purpose, A1. measured. Figure 8 depicts the residuals, and Table 1 presents the RMSE as a function of the numbers Itthe was defect found A1 that with the a high TSR regressionaspect ratio with of 8 two was coe considered,fficients has and large the residuals temperature at the at beginning the centre of was the of coefficients for the defect A1. It was found that the TSR regression with two coefficients has large measured. Figure 8 depicts the residuals, and Table 1 presents the RMSE as a function of the numbers

of coefficients for the defect A1. It was found that the TSR regression with two coefficients has large

Sensors 2020, 20, 6015 9 of 17 Sensors 2020, 20, x FOR PEER REVIEW 9 of 17 coolingresiduals process. at the With beginning three coe of thefficients, cooling the process. residuals With andthree the coefficients, RMSE began the residuals to reduce and andthe RMSE improved withbegan the increasing to reduce one. and Onimproved the contrary, with the the increasing first and one. second On derivativesthe contrary, provided the first and the bestsecond results derivatives provided the best results with just three degrees of a polynomial function, as noticed in with just three degrees of a polynomial function, as noticed in Figure7b. Figure 7b.

FigureFigure 7. Comparison 7. Comparison of experimentalof experimental pulsed pulsed thermographythermography images images processed processed with with thermal thermal signal signal reconstructionreconstruction for thefor the times times corresponding corresponding toto order of of polynomial polynomial fitting, fitting, (a) (2nda) 2nd order, order, (b) 3rd (b )order, 3rd order, (c) 4th(c) order, 4th order, (d) 5th (d) order,5th order, (e) 6th(e) 6th order, order, (f) ( 7thf) 7th order order and and ( g(g)) 8th 8th orderorder [Frequency == 5050 Hz, Hz, number number of framesof frames= 250, = truncation 250, truncation window window= 5 = s]. 5 s]. Temperature Temperature scales scales are are inin degree Celsius. Celsius.

Sensors 2020, 20, x FOR PEER REVIEW 10 of 17 Sensors 2020, 20, 6015 10 of 17

23.5

23.4

23.3 C) o (

23.2

23.1

23.0 Temperature Unprocessed raw thermal 2nd order polynomial 3rd order polynomial 4th order polynomial 22.9 5th order polynomial 6th order polynomial 7th order polynomial 8th order polynomial 22.8

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Time (s)

Figure 8. Residuals along the time as a function of the numbers of coefficient in the regression of a flat-bottomedFigure 8. Residuals hole of along size 16the mm time and as a depth function 2 mm of [Frequencythe numbers= of50 coefficient Hz, number in the of frames regression= 250, of a truncationflat‐bottomed window hole= 5of s]. size 16 mm and depth 2 mm [Frequency = 50 Hz, number of frames = 250, truncation window = 5 s]. Table 1. Root mean square error as a function of the numbers of coefficient in the regression for a flat-bottomedTable 1. Root hole mean of size square 16 mm error and as a depth function 2 mm. of the numbers of coefficient in the regression for a flat‐ bottomed hole of size 16 mm and depth 2 mm. Degree of Polynomial Coefficient Root Mean Square Error Degree of Polynomial2nd Coefficient Root Mean 0.0503 Square Error 3rd 0.0396 4th2nd 0.03950.0503 5th3rd 0.03260.0396 6th 0.0263 7th4th 0.02550.0395 8th5th 0.02000.0326 6th 0.0263 5.2. Pulsed Phase Thermography Results 7th 0.0255 Figure9 depicts the results of PPT concerning frequency spectra. It was observed that defect 8th 0.0200 detectability with PPT is dependent on the chosen frequency spectra. Figure9a depicts the phase image at frequency 0.2 Hz, where no defects were detected. Figure9b depicts the phase image at 5.2. Pulsed Phase Thermography Results frequency 0.4 Hz, where defects started to be detected but affected by the noise. Figure9c depicts the phaseFigure image 9 depicts at frequency the results 8 Hz, of where PPT concerning a maximum frequency number of spectra. defects It were was detectedobserved with that highdefect phasedetectability contrast. with Figure PPT9d represents is dependent the phaseon the image chosen at afrequency frequency spectra. of 25 Hz, Figure where 9a most depicts of the the defect phase informationimage at frequency was lost. 0.2 Figure Hz,9 wheree depicts no thedefects phase were image detected. at frequency Figure 42.69b depicts Hz, where the phase a maximum image at numberfrequency of defects 0.4 Hz, were where detected defects with started high to phase be detected contrast. but It affected was worth by the noting noise. that Figure phase 9c inversion depicts the takesphase place image at frequency at frequency 42.6 8 Hz, Hz wherewhere thea maximum defects which number were of defects blue at were frequency detected 8 Hz with changed high phase to orange.contrast. Figure Figure9f depicts 9d represents the phase the image phase atimage frequency at a frequency 50 Hz, where of 25 defects Hz, where were most a ffected of the by defect the noiseinformation again. was lost. Figure 9e depicts the phase image at frequency 42.6 Hz, where a maximum number of defects were detected with high phase contrast. It was worth noting that phase inversion takes place at frequency 42.6 Hz where the defects which were blue at frequency 8 Hz changed to orange. Figure 9f depicts the phase image at frequency 50 Hz, where defects were affected by the noise again. To select the most appropriate frequency for high defect detectability and noise reduction, phase contrast analysis for the defect A1 was performed. For this purpose, two ROIs of 4 × 4 pixels, one at the centre of the defect and another in the adjacent sound region were considered. Figure 10 depicts

Sensors 2020, 20, x FOR PEER REVIEW 11 of 17 the phase contrast for the defect A1 concerning frequency spectra. It was found that there exist two favourable frequencies where the defect exhibits maximum phase contrast. Defect exhibits the maximum positive phase contrast at a frequency of 8 Hz and maximum negative phase contrast at a frequencySensors 2020, of20, 42.6 6015 Hz. 11 of 17 Sensors 2020, 20, x FOR PEER REVIEW 11 of 17

the phase contrast for the defect A1 concerning frequency spectra. It was found that there exist two favourable frequencies where the defect exhibits maximum phase contrast. Defect exhibits the maximum positive phase contrast at a frequency of 8 Hz and maximum negative phase contrast at a frequency of 42.6 Hz.

Figure 9. Phase images obtained with pulsed phase thermography for a steel plate with flat bottomed Figureholes after 9. Phase pulse images heating obtained at diff erentwith pulsed frequency phase spectra, thermography (a) 0.2 Hz, for (ab steel) 0.4 plate Hz, ( withc) 8 Hz, flat bottomed (d) 25 Hz holes(e) 42.6 after Hz andpulse (f )heating 50 Hz, numberat different of frames frequency=250 spectra, frames, truncation(a) 0.2 Hz, window(b) 0.4 Hz, 5 s. (c Phase) 8 Hz, angle (d) 25 scales Hz are(e) 42.6in radian. Hz and (f) 50 Hz, number of frames =250 frames, truncation window 5 s. Phase angle scales are in radian. To select the most appropriate frequency for high defect detectability and noise reduction, phase contrast analysis for the defect A was performed. For this purpose, two ROIs of 4 4 pixels, one 1 × at theFigure centre 9. of Phase the images defect obtained and another with pulsed in the phase adjacent thermography sound region for a steel were plate considered. with flat bottomed Figure 10 holes after pulse heating6 at different frequency spectra, (a) 0.2 Hz, (b) 0.4 Hz, (c) 8 Hz, (d) 25 Hz (e) depicts the phase contrast for the defect A1 concerning frequency Defective spectra. Region It was found that there exist 5 two favourable42.6 Hz and frequencies (f) 50 Hz, number where of the frames defect =250 exhibits frames, maximum truncation Sound Region window phase contrast. 5 s. Phase Defect angle scales exhibits are the 4 Phase Difference maximumin radian. positive phase contrast at a frequency of 8 Hz and maximum negative phase contrast at a 3 frequency of 42.6 Hz. 2

1 (radian) 0 6 Defective Region ‐1 5 angle

Sound Region ‐2 4 Phase Difference

Phase ‐3 3 ‐4 2 ‐5 1 (radian) ‐6 0

‐1

angle 0 5 10 15 20 25 30 35 40 45 50

‐2 Frequency (Hz) Phase ‐3 ‐4 Figure 10. Phase profiles‐ 5of defective and sound regions for a flat‐bottomed hole of size 16 mm and depth 2 mm on a steel plate‐6 obtained by pulse phase thermography concerning frequency; frequency rate =50 Hz, number of frames =250 frames, truncation window 5 s. 0 5 10 15 20 25 30 35 40 45 50 Frequency (Hz)

Figure 10. Phase profiles of defective and sound regions for a flat-bottomed hole of size 16 mm and Figure 10. Phase profiles of defective and sound regions for a flat‐bottomed hole of size 16 mm and depth 2 mm on a steel plate obtained by pulse phase thermography concerning frequency; frequency depth 2 mm on a steel plate obtained by pulse phase thermography concerning frequency; frequency rate =50 Hz, number of frames =250 frames, truncation window 5 s. rate =50 Hz, number of frames =250 frames, truncation window 5 s.

Sensors 2020, 20, x 6015 FOR PEER REVIEW 12 of 17

5.3. Principle Component Thermography Results 5.3. Principle Component Thermography Results Figure 11 shows the results of PCT concerning EOFs. Unlike in TSR and PPT, the results of the PCT Figureare less 11 dependentshows the results on the of number PCT concerning of frames EOFs. and Unlike frequency in TSR spectra. and PPT, The the most results meaningful of the PCT informationare less dependent was found on the in numberthe first oftwo frames EOFs. and frequency spectra. The most meaningful information was found in the first two EOFs.

Figure 11. Experimental pulsed thermography images processed by principle component thermography Figureconcerning 11. EOFs,Experimental (a) 1st EOF, pulsed (b) 2ndthermography EOF, (c) 3rd EOF,images (d ) 4thprocessed EOF, ( e)by 5th principle EOF and (componentf) 6th EOF.

thermographyThe scales are in concerning “digital level” EOFs, units. (a) 1st EOF, (b) 2nd EOF, (c) 3rd EOF, (d) 4th EOF, (e) 5th EOF and (f) 6th EOF. The scales are in “digital level” units. 5.4. Comparison of Processing Techniques 5.4. Comparison of Processing Techniques The performance of all the three processing techniques was measured in terms of enhanced defectThe detectability performance and of SNR. all Thethe three optimum processing results providedtechniques by was each measured technique werein terms considered of enhanced in the defectmeasurement. detectability The resultsand SNR. of TSR The with optimum the 8th results degree provided of the polynomial by each technique fitting, the were result considered of PPT at in a thefrequency measurement. of 0.8 Hz The and results the result of TSR of PCT with with the 1st8th EOFdegree were of comparedthe polynomial with thefitting, raw the thermal result image of PPT at attime a frequency 0.1 s. of 0.8 Hz and the result of PCT with 1st EOF were compared with the raw thermal image at time 0.1 s. 5.4.1. Comparison Based on Defect Detectability 5.4.1.The Comparison raw thermal Based image on Defect at time Detectability 0.1 s (Figure 5b) can detect 8 out of 16 defects. The detected defectsThe were raw of thermal size 16 mmimage and at 12 time mm 0.1 with s (Figure an aspect 5b) ratio can of detect more 8 than out 3. of TSR 16 defects. with the The 8th degreedetected of defectspolynomial were fitting of size (Figure 16 mm7e) and can 12 detect mm 16with out an of aspect 16 defects. ratio Even of more the smallerthan 3. defectsTSR with with the a 8th size degree 4 mm ofand polynomial an aspect ratiofitting of (Figure 0.8 were 7e) also can detected detect 16 faintly. out of PPT 16 defects. at a frequency Even the of 0.8smaller Hz (Figure defects9 c)with and a PCT size 4with mm 1st and EOF an (Figureaspect ratio11a) wereof 0.8 able were to also detect detected 14 out offaintly. 16 defects. PPT at Both a frequency PPT and PCTof 0.8 techniques Hz (Figure were 9c) andfound PCT insensitive with 1st to EOF the smaller(Figure defects11a) were of size able 4 mmto detect within 14 an out aspect of 16 ratio defects. of 0.8–1.0. Both PPT and PCT techniques were found insensitive to the smaller defects of size 4 mm within an aspect ratio of 0.8– 5.4.2. Comparison Based on Signal to Noise Ratio 1.0. Two ROIs of 4 4 pixels, one at the centre of the defect and another in the adjacent sound area, × 5.4.2.were consideredComparison for Based each defect.on Signal ROI to in Noise the defective Ratio area was considered as “signal” (DROI), and ROI in theTwo sound ROIs area of was4 × 4 considered pixels, one as at “noise”the centre (SROI). of the Then, defect SNR and was another acquired in the by adjacent Equation sound (7) [72 area,,73]. were considered for each defect. ROI in the defective area was considered as “signal” (DROI), and DROImean SROImean SNR = 20 log | − | (7) ROI in the sound area was considered as “noise”10 (SROI).σ Then, SNR was acquired by Equation (7) [72,73]. where DROImean, SROImean, and σ represent, respectively, the arithmetic mean of ROI in the defective |DROI SROI| area, the arithmetic mean of ROISNR in the20 sound log area, and standard deviation of ROI in the sound area.(7) σ

Sensors 2020, 20, 6015 13 of 17

The results obtained using Equation (7) to the data are presented in Table2. The results demonstrated that all three signal processing techniques greatly improved the SNR. PPT and PCT dominated the TSR in terms of SNR. It was worth noting that PCT dominates PPT for the shallower defects of depth 2 and 3 mm, whereas PPT dominates for the deeper defects of depth 4 mm and 5 mm. For example, in Table2 for shallower defect A 1; the raw thermal image provided the SNR of 33.26 dB; TSR provided the SNR of 40.46 dB, which is an increment of 21.64%; PPT provided the SNR of 45.24 dB, which is an increment of 33.26%; PCT provided the SNR of 47.78 dB which is an increment by 43.65%; when compared with the raw thermal image. Similarly, for the deeper defect B1; the raw thermal image provided the SNR of 19.18 dB; TSR provided the SNR of 31.18 dB, which is an increment of 62.56%; PPT provided the SNR of 46.28 dB, which is an increment of 141.29%; PCT provided the SNR of 41.6 dB which is an increment of 116.89%; when compared with the raw thermal image.

Table 2. SNR values for each processing techniques and flat-bottomed holes.

SNR Defect ID. Raw Image TSR PPT PCA

A1 33.26 40.46 45.24 47.78 A2 - 32.39 42.15 44.56 A3 - 39.18 43.92 47.15 A4 28.17 39.47 44.56 47.72 B1 19.18 31.18 46.28 41.60 B2 - 26.15 -- B3 - 31.75 33.65 31.90 B4 12.19 28.78 40.76 38.85 C1 32.14 38.98 46.55 46.58 C2 - 32.29 36.52 34.91 C3 - 39.47 45.41 45.58 C4 28.39 38.28 45.90 46.76 D1 23.96 28.78 47.12 44.52 D2 - 15.62 -- D3 - 27.82 45.91 40.26 D4 21.44 27.64 47.10 44.16 Note: The symbol ‘-’ is the representation of non-detected defects.

6. Conclusions and Future Works In this work, pulsed thermography was applied to a steel structure with the aim to detect wall thinning defects. The fundamental concepts of TSR, PPT, and PCT have been reviewed for the processing and analysis of thermal images. The performance of each signal processing technique was evaluated in terms of enhanced defect detectability and SNR. Based on the results, it was determined that a considerable improvement in the defect detection capability and SNR values could be obtained after the implementation of techniques. TSR provided the less noisy image when compared with the raw thermal image and bring essential improvements in defect detectability due to an increase of temporal and spatial resolution and the ability to produce time derivative images. However, variation in the results in different instants of time or frames could be disadvantages of TSR. PPT has been found highly sensitive to the defect depth and provided the best SNR for the smaller and deeper defects. However, the results of PPT are also sometimes difficult to handle since it also provides several data in different frequency spectra. The PCT has been found useful for detecting the larger and shallower defects. The independence of the results on the number of frames was found as the major advantage of the PCT. Future work will be directed towards implementing pulsed thermography signal processing techniques to detect real defects in the pipe structure. Furthermore, the maximum depth detection capability of pulsed thermography will be investigated. Additionally, TSR based analysis will be done Sensors 2020, 20, 6015 14 of 17 in high order polynomial fitting. Finally, the error analysis techniques will also be considered for the obtained results.

Author Contributions: Conceptualization, Y.C. and R.S.; Data curation, R.S.; Formal analysis, R.S. and S.L.; Funding acquisition, W.K.; Investigation, Y.C., R.S., and S.L.; Methodology, Y.C., R.S., and S.L.; Project administration, W.K.; Resources, W.K.; Software, Y.C., R.S., and S.L.; Supervision, W.K.; Validation, R.S.; Visualization, R.S.; Writing—original draft, R.S.; Writing—review & editing, W.K. All authors have read and agreed to the published version of the manuscript. Funding: This research were funded by the National Research Foundation of Korea (NRF–2019R1F1A1061328) grant by the Korean Government, Ministry of Education, Science and Technology (MEST) and by the research grant of the Kongju National University in 2020. Conflicts of Interest: The authors declare no conflict of interest.

References

1. McGuire, M.F. Stainless Steels for Design Engineers; ASM International: Materials Park, OH, USA, 2008. 2. Baddoo, N. Stainless Steel in Construction: A Review of Research, Applications, Challenges and Opportunities. J. Constr. Steel Res. 2008, 64, 1199–1206. [CrossRef] 3. Lister, D.H.; Cook, W.G. Nuclear Plant Materials and Corrosion. Available online: https: //www.unene.ca/essentialcandu/pdf/14%20-%20Nuclear%20Plant%20Materials%20and%20Corrosion.pdf (accessed on 2 June 2020). 4. Okada, H.; Uchida, S.; Naitoh, M.; Xiong, J.; Koshizuka, S. Evaluation Methods for Corrosion Damage of Components in Cooling Systems of Nuclear Power Plants by Coupling Analysis of Corrosion and Flow Dynamics (V) Flow-Accelerated Corrosion Under Single-and Two-Phase Flow Conditions. J. Nucl. Sci. Technol. 2011, 48, 65–75. [CrossRef] 5. Ahmed, W.H. Flow accelerated corrosion in nuclear power plants. In Nuclear Power-Practical Aspects; Anonymous, Ed.; IntechOpen: Rijeka, Croatia, 2012. 6. Naitoh, M.; Okada, H.; Uchida, S.; Yugo, H.; Koshizuka, S. Evaluation Method for Pipe Wall Thinning due to Liquid Droplet Impingement. Nucl. Eng. Des. 2013, 264, 195–202. [CrossRef] 7. Onel, Y.; Ewert, U.; Willems, P. Radiographic Wall Thickness Measurement of Pipes by a New Tomographic Algorithm. In Proceedings of the 15th WCNDT, Roma, Italy, 15–21 October 2000; pp. 1–6. 8. Edalati, K.; Rastkhah, N.; Kermani, A.; Seiedi, M.; Movafeghi, A. The use of for Thickness Measurement and Corrosion Monitoring in Pipes. Int. J. Press. Vessel. Pip. 2006, 83, 736–741. [CrossRef] 9. Rakvin, M.; Markuˇciˇc,D.; Hižman, B. Evaluation of Pipe Wall Thickness Based on Contrast Measurement using Computed Radiography (CR). Procedia Eng. 2014, 69, 1216–1224. [CrossRef] 10. Ju, Y. Remote measurement of pipe wall thinning by . In Advanced Nondestructive Evaluation II: Volume 2; Anonymous, Ed.; World Scientific: Busan, Korea, 2008; pp. 1128–1133. 11. Liu, L.; Ju, Y. A High-Efficiency Nondestructive Method for Remote Detection and Quantitative Evaluation of Pipe Wall Thinning using Microwaves. NDT E Int. 2011, 44, 106–110. [CrossRef] 12. Liu, L.; Ju, Y.; Chen, M. Optimizing the Frequency Range of Microwaves for High-Resolution Evaluation of Wall Thinning Locations in a Long-Distance Metal Pipe. NDT E Int. 2013, 57, 52–57. [CrossRef] 13. Nishino, H.; Takemoto, M.; Chubachi, N. Estimating the Diameter Thickness of a Pipe using the Primary Wave Velocity of a Hollow Cylindrical Guided Wave. Appl. Phys. Lett. 2004, 85, 1077–1079. [CrossRef] 14. Leonard, K.R.; Hinders, M.K. Lamb Wave of Pipe-Like Structures. Ultrasonics 2005, 43, 574–583. [CrossRef] 15. Cheong, Y.; Kim, K.; Kim, D. High-Temperature Ultrasonic Thickness Monitoring for Pipe Thinning in a Flow-Accelerated Corrosion Proof Test Facility. Nucl. Eng. Technol. 2017, 49, 1463–1471. [CrossRef] 16. Alobaidi, W.M.; Kintner, C.E.; Alkuam, E.A.; Sasaki, K.; Yusa, N.; Hashizume, H.; Sandgren, E. Experimental Evaluation of Novel Hybrid Microwave/Ultrasonic Technique to Locate and Characterize Pipe Wall Thinning. J. Press. Vessel Technol. 2018, 140, 011501. [CrossRef] 17. Jinfeng, D.; Yihua, K.; Xinjun, W. Tubing Thread Inspection by Magnetic Flux Leakage. NDT E Int. 2006, 39, 53–56. [CrossRef] 18. Zhang, Y.; Yan, G. Detection of Gas Pipe Wall Thickness Based on Electromagnetic Flux Leakage. Russ. J. Nondestruct. Test. 2007, 43, 123–132. [CrossRef] Sensors 2020, 20, 6015 15 of 17

19. Dutta, S.M.; Ghorbel, F.H. Measurement of Pipe Wall Thickness Using Magnetic Flux Leakage Signals. U.S. Patent 8,134,360, 13 March 2012. 20. Nam, K.W.; Ahn, S.H. Fracture Behaviors and Acoustic Emission Characteristics of Pipes with Local Wall Thinning. In Key Engineering Materials; Trans Tech Publications: Baech, Swizerland, 2004; pp. 461–465. 21. Kosaka, D.; Kojima, F.; Yamaguchi, H. Quantitative Evaluation of Wall Thinning in Pipe Wall using Electromagnetic Acoustic Transducer. Int. J. Appl. Electromagn. Mech. 2010, 33, 1195–1200. [CrossRef] 22. Kojima, F.; Nguyen, T.; Yamaguchi, H. Shape Identification of Pipe-wall Thinning Using Electromagnetic Acoustic Transducer. Available online: https://aip.scitation.org/doi/abs/10.1063/1.3591913 (accessed on 2 June 2020). 23. Nestleroth, J.B.; Davis, R.J. Application of Eddy Currents Induced by Permanent Magnets for Pipeline Inspection. NDT E Int. 2007, 40, 77–84. [CrossRef] 24. Cheng, W. Pulsed Eddy Current Testing of Carbon Steel Pipes’ Wall-Thinning through Insulation and Cladding. J. Nondestruct. Eval. 2012, 31, 215–224. [CrossRef] 25. Xie, S.; Chen, Z.; Takagi, T.; Uchimoto, T. Quantitative Non-Destructive Evaluation of Wall Thinning Defect in Double-Layer Pipe of Nuclear Power Plants using Pulsed ECT Method. NDT E Int. 2015, 75, 87–95. [CrossRef] 26. Rogalski, A. Infrared Detectors: An Overview. Infrared Phys. Technol. 2002, 43, 187–210. [CrossRef] 27. Williams, T. Thermal Cameras: Characteristics and Performance; CRC Press: Boca Raton, FL, USA, 2009. 28. Czichos, H. Handbook of Technical Diagnostics: Fundamentals and Application to Structures and Systems; Springer Science & Business Media: Berlin/Heidelberg, Germany, 2013. 29. Maldague, X.P. Introduction to NDT by Active Infrared Thermography. Mater. Eval. 2002, 60, 1060–1073. 30. Ranjit, S.; Choi, M.; Kim, W. Quantification of Defects Depth in Glass Fiber Reinforced Plastic Plate by Infrared Lock-in Thermography. J. Mech. Sci. Technol. 2016, 30, 1111–1118. [CrossRef] 31. Ranjit, S.; Kim, W. Evaluation of Coating Thickness by Thermal Wave Imaging: A Comparative Study of Pulsed and Lock-in Infrared thermography–Part II: Experimental Investigation. Infrared Phys. Technol. 2018, 92, 24–29. 32. Usamentiaga, R.; Venegas, P.; Guerediaga, J.; Vega, L.; Molleda, J.; Bulnes, F. Infrared Thermography for Temperature Measurement and Non-Destructive Testing. Sensors 2014, 14, 12305–12348. [CrossRef][PubMed] 33. Maldague, X.P. Nondestructive Testing Handbook. 3. Infrared and Thermal Testing; American Society for Nondestructive Testing: Arlingate Lane, OA, USA, 2001. 34. Bagavathiappan, S.; Lahiri, B.; Saravanan, T.; Philip, J.; Jayakumar, T. Infrared Thermography for Condition monitoring–A Review. Infrared Phys. Technol. 2013, 60, 35–55. [CrossRef] 35. Ranjit, S.; Kang, K.; Kim, W. Investigation of Lock-in Infrared Thermography for Evaluation of Subsurface Defects Size and Depth. Int. J. Precis. Eng. Manuf. 2015, 16, 2255–2264. [CrossRef] 36. Maldague, X. Theory and Practice of Infrared Technology for Nondestructive Testing; John Wiley & Sons: New York, NY, USA, 2001. 37. Waugh, R.; Dulieu-Barton, J.; Quinn, S. Modelling and Evaluation of Pulsed and Pulse Phase Thermography through Application of Composite and Metallic Case Studies. NDT E Int. 2014, 66, 52–66. [CrossRef] 38. Busse, G.; Wu, D.; Karpen, W. Thermal Wave Imaging with Phase Sensitive Modulated Thermography. J. Appl. Phys. 1992, 71, 3962–3965. [CrossRef] 39. Gleiter, A.; Spiessberger, C.; Busse, G. Phase Angle Thermography for Depth Resolved Defect Characterization. Available online: https://aip.scitation.org/doi/abs/10.1063/1.3114300 (accessed on 2 June 2020). 40. Wu, D.; Busse, G. Lock-in Thermography for Nondestructive Evaluation of Materials. Rev. Générale Therm. 1998, 37, 693–703. [CrossRef] 41. Lahiri, B.; Bagavathiappan, S.; Reshmi, P.; Philip, J.; Jayakumar, T.; Raj, B. Quantification of Defects in Composites and Rubber Materials using Active Thermography. Infrared Phys. Technol. 2012, 55, 191–199. [CrossRef] 42. Vavilov, V.; Taylor, R. Theoretical and Practical Aspects of the Thermal Nondestructive Testing of Bonded Structures. Acad. Press Res. Tech. Nondestruct. Test. 1982, 5, 239–279. 43. Wallbrink, C.; Wade, S.; Jones, R. The Effect of Size on the Quantitative Estimation of Defect Depth in Steel Structures using Lock-in Thermography. J. Appl. Phys. 2007, 101, 104907. [CrossRef] 44. Choi, M.; Kang, K.; Park, J.; Kim, W.; Kim, K. Quantitative Determination of a Subsurface Defect of Reference Specimen by Lock-in Infrared Thermography. NDT E Int. 2008, 41, 119–124. [CrossRef] Sensors 2020, 20, 6015 16 of 17

45. Junyan, L.; Qingju, T.; Xun, L.; Yang, W. Research on the Quantitative Analysis of Subsurface Defects for Non-Destructive Testing by Lock-in Thermography. NDT E Int. 2012, 45, 104–110. [CrossRef] 46. Delanthabettu, S.; Menaka, M.; Venkatraman, B.; Raj, B. Defect Depth Quantification using Lock-in Thermography. Quant. InfraRed Thermogr. J. 2015, 12, 37–52. [CrossRef] 47. Shrestha, R.; Park, J.; Kim, W. Application of Thermal Wave Imaging and Phase Shifting Method for Defect Detection in Stainless Steel. Infrared Phys. Technol. 2016, 76, 676–683. [CrossRef] 48. Maldague, X.P.; Shiryaev, V.V.; Boisvert, E.; Vavilov, V.P. Transient Thermal Nondestructive Testing (NDT) of Defects in Aluminum Panels. In Proceedings of the Thermosense XVII: An International Conference on Thermal Sensing and Imaging Diagnostic Applications, Orlando, FL, USA, 19–21 April 1995; pp. 233–243. 49. Deemer, C. Front-Flash Thermal Imaging Characterization of Continuous Fiber Ceramic Composites. Available online: https://www.osti.gov/biblio/10924 (accessed on 2 June 2020). 50. Ibarra-Castanedo, C.; Genest, M.; Piau, J.; Guibert, S.; Bendada, A.; Maldague, X.P. Active infrared thermography techniques for the nondestructive testing of materials. In Ultrasonic and Advanced Methods for Nondestructive Testing and Material Characterization; Anonymous, Ed.; World Scientific: Toh Tuck Link, Singapore, 2007; pp. 325–348. 51. Wysocka-Fotek, O.; Oliferuk, W.; Maj, M. Reconstruction of Size and Depth of Simulated Defects in Austenitic Steel Plate using Pulsed Infrared Thermography. Infrared Phys. Technol. 2012, 55, 363–367. [CrossRef] 52. Maldague, X.; Marinetti, S. Pulse Phase Infrared Thermography. J. Appl. Phys. 1996, 79, 2694–2698. [CrossRef] 53. Maldague, X.; Galmiche, F.; Ziadi, A. Advances in Pulsed Phase Thermography. Infrared Phys. Technol. 2002, 43, 175–181. [CrossRef] 54. Shepard, S.M.; Lhota, J.R.; Rubadeux, B.A.; Wang, D.; Ahmed, T. Reconstruction and Enhancement of Active Thermographic Image Sequences. Opt. Eng. 2003, 42, 1337–1343. [CrossRef] 55. Rajic, N. Principal Component Thermography for Flaw Contrast Enhancement and Flaw Depth Characterisation in Composite Structures. Compos. Struct. 2002, 58, 521–528. [CrossRef] 56. Vageswar, A.; Balasubramaniam, K.; Krishnamurthy, C. Wall Thinning Defect Estimation using Pulsed IR Thermography in Transmission Mode. Nondestr. Test. Eval. 2010, 25, 333–340. [CrossRef] 57. Zeng, Z.; Li, C.; Tao, N.; Feng, L.; Zhang, C. Depth Prediction of Non-Air Interface Defect using Pulsed Thermography. NDT E Int. 2012, 48, 39–45. [CrossRef] 58. Pickering, S.; Almond, D. Matched Excitation Energy Comparison of the Pulse and Lock-in Thermography NDE Techniques. NDT E Int. 2008, 41, 501–509. [CrossRef] 59. Duan, Y.; Huebner, S.; Hassler, U.; Osman, A.; Ibarra-Castanedo, C.; Maldague, X.P. Quantitative Evaluation of Optical Lock-in and Pulsed Thermography for Aluminum Foam Material. Infrared Phys. Technol. 2013, 60, 275–280. [CrossRef] 60. Chatterjee, K.; Tuli, S.; Pickering, S.G.; Almond, D.P. A Comparison of the Pulsed, Lock-in and Frequency Modulated Thermography Nondestructive Evaluation Techniques. NDT E Int. 2011, 44, 655–667. [CrossRef] 61. Carlomagno, G.M.; Meola, C. Comparison between Thermographic Techniques for Frescoes NDT. NDT E Int. 2002, 35, 559–565. [CrossRef] 62. Shrestha, R.; Kim, W. Modelling of Pulse Thermography for Defect Detection in Aluminium Structures: Assessment on Reflection and Transmission Measurement. World J. Model. Simul. 2017, 13, 45–51. 63. Shrestha, R.; Kim, W. Evaluation of Coating Thickness by Thermal Wave Imaging: A Comparative Study of Pulsed and Lock-in Infrared thermography–Part I: Simulation. Infrared Phys. Technol. 2017, 83, 124–131. [CrossRef] 64. Carslaw, H.S.; Jaeger, J.C. Conduction of Heat in Solid, 2nd ed.; Clarendon Press: Oxford, UK, 1959. 65. Ibarra-Castanedo, C.; González, D.A.; Galmiche, F.; Bendada, A.; Maldague, X.P. On Signal Transforms Applied to Pulsed Thermography. Recent Res. Dev. Appl. Phys. 2006, 9, 101–127. 66. Balageas, D.; Chapuis, B.; Deban, G.; Passilly, F. Improvement of the Detection of Defects by Pulse Thermography Thanks to the TSR Approach in the Case of a Smart Composite Repair Patch. Quant. InfraRed Thermogr. J. 2010, 7, 167–187. [CrossRef] 67. Alvarez-Restrepo, C.; Benitez-Restrepo, H.; Tobón, L. Characterization of Defects of Pulsed Thermography Inspections by Orthogonal Polynomial Decomposition. NDT E Int. 2017, 91, 9–21. [CrossRef] 68. Balageas, D.L.; Roche, J.; Leroy, F.; Liu, W.; Gorbach, A.M. The Thermographic Signal Reconstruction Method: A Powerful Tool for the Enhancement of Transient Thermographic Images. Biocybern. Biomed. Eng. 2015, 35, 1–9. [CrossRef][PubMed] Sensors 2020, 20, 6015 17 of 17

69. Wang, Z.; Tian, G.; Meo, M.; Ciampa, F. Image Processing Based Quantitative Damage Evaluation in Composites with Long Pulse Thermography. NDT E Int. 2018, 99, 93–104. [CrossRef] 70. Ibarra-Castanedo, C.; Maldague, X. Pulsed Phase Thermography Reviewed. Quant. Infrared Thermogr. J. 2004, 1, 47–70. [CrossRef] 71. Daryabor, P.; Safizadeh, M. Comparison of Three Thermographic Post Processing Methods for the Assessment of a Repaired Aluminum Plate with Composite Patch. Infrared Phys. Technol. 2016, 79, 58–67. [CrossRef] 72. Shrestha, R.; Kim, W. Non-Destructive Testing and Evaluation of Materials using Active Thermography and Enhancement of Signal to Noise Ratio through Data Fusion. Infrared Phys. Technol. 2018, 94, 78–84. [CrossRef] 73. Shrestha, R.; Chung, Y.; Kim, W. Wavelet Transform Applied to Lock-in Thermographic Data for Detection of Inclusions in Composite Structures: Simulation and Experimental Studies. Infrared Phys. Technol. 2019, 96, 98–112. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).