research article nondestructive inspection of thin...
TRANSCRIPT
Research ArticleNondestructive Inspection of Thin Basalt FiberReinforced Composites Using Combined Terahertz Imaging andInfrared Thermography
Przemyslaw Lopato Grzegorz Psuj and Barbara Szymanik
Department of Electrical and Computer Engineering Faculty of Electrical Engineering West Pomeranian University of TechnologyAl Piastow 17 70-310 Szczecin Poland
Correspondence should be addressed to Grzegorz Psuj gpsujzutedupl
Received 8 March 2016 Accepted 7 June 2016
Academic Editor Fridon Shubitidze
Copyright copy 2016 Przemyslaw Lopato et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited
The inspection of thin basalt fiber reinforced compositematerials was carried out using two nondestructivemethods terahertz timedomain imaging and infrared thermography In order to combine the information about the defects arising in examined materialsthe inspection results were parametrized In order to acquire more information content new approximation based features areproposedThen a knowledge extraction based multivariate analysis of preselected featuresrsquo vector was carried out Finally in orderto integrate features distributions of representing different dynamic level of information amultiresolutionwavelet based data fusionalgorithm was applied The results are presented and discussed
1 Introduction
The increasing application of composite materials in modernstructures of the aerospace construction and automotiveindustries puts new demands on the quality control unitsMostly nondestructive testing (NDT) methods relate tometals inspectionTherefore there is a need for developmentof new or adaptation of existing methods to compositematerials properties [1]The selection of themost appropriatemethod for testing of composite materials depends on thenature of the fibers and the polymer matrix [2] However nosingle NDT technique allows a full assessment of thematerialunder test structurersquos integrity In essence each methodpresents some limits of detection and determination of thenonhomogeneity of the material [3] In order to increase theprobability of defectsrsquo detection and a proper evaluation of thematerialrsquos structure a need of fusion of data obtained fromvarious testing methods is arising [4 5]
In this paper the inspection of thin basalt fiber reinforcedcomposite (BFRC) materials (shortly described in Section 2)is carried out using two electromagnetic NDT methodsThe selected techniques allow assessing different aspects of
the defects arising in specimen First the acquired data wereprocessed and parameterized The analysis of each methodrsquosresults allowed evaluating the usability of each applied testingtechnique for the inspection of basalt composite Then inorder to enhance the proper evaluation of the material stageand combine information supplied by eachmethod a knowl-edge extraction based multivariate analysis of features vectorwas performed followed by multiresolution decompositionbased data fusion Finally the performance of the data fusionwas assessed and compared with the results obtained for eachmethod separately
2 Basalt Fiber Reinforced Composite
Basalt fiber is a perspective polymer-reinforcingmaterial andcan be applied in polymer matrix composites instead of glassfiber Fabrics of varying surface densities aremade dependingupon the application type and are in the range from 160 gm2
to 1100 gm2 [6] The comparison of selected properties ofbasalt and glass fibers is shown in Table 1 One can observe
Hindawi Publishing CorporationAdvances in Materials Science and EngineeringVolume 2016 Article ID 1249625 13 pageshttpdxdoiorg10115520161249625
2 Advances in Materials Science and Engineering
Table 1 Selected properties of basalt and glass fibers [7 8]
Properties Unit Basalt E-Glass S-GlassDensity gcm3 27 257 248Thermal linear expansion coefficient ppm∘C 80 54 29Tensile strength GPa 484 345 471Elastic modulus GPa 89 77 89Elongation at break 315 47 56Compression strength MPa 3792 3033 3033Maximum application temperature ∘C 982∘ 650∘ 650∘
Sustained operating temperature ∘C 820∘ 480∘ 480∘
Minimum operating temperature ∘C minus260∘ minus60∘ minus60∘
Melting temperature ∘C 1450∘ 1120∘ 1120∘
Absorption of humidity (65 RAH) lt01 lt01 lt01Emissivity at 20∘C mdash 092 065 065Thermal conductivity WmK 0031ndash0038 0034ndash004 0034ndash004
40
40
Thickness tbc asymp 10ndash12mm
(a) (b)
Figure 1 Photo of the examined specimen 119904IMC 5 J (basalt reinforced composite material exposed to impact energy) (a) general view and (b)enlargement of measured region
an advantage of basalt over glass fibers in case of mechanicaland thermal properties
The photo of the utilized basalt fibers reinforced com-posite specimen is presented in Figure 1 The materialconsists of six layers of basalt fiber fabric within a polyesterresin (Polimal 1094 AWTP-1) The thickness of the sampleswas 1ndash12mm thus thickness of single layer was less than200120583mThe evaluated basalt test specimens were exposed tomechanical impacts of 2 and 5J energy (sample 119904IMP 2 J and119904IMP 5 J) During samples production process artificial inclu-sions (mica film of 145 120583m thickness) were also introducedunder selected layers (samples 119904INC A and 119904INC B)
3 Multisource Inspection Procedure
In the process of inspection of thin basalt fiber reinforcedcomposite materials two nondestructive methods were uti-lized The measurements were carried out using terahertz(THz) inspection technique and active infrared thermogra-phy (IRT) This approach enabled more efficient detectionof defects based on changes of both electrical and thermalparameters
Terahertz electromagnetic radiation enables noninva-sive nonionizing and noncontact examination of dielectricmaterials such as plastics dry wood explosives ceramicsfoams and composites [9 10] Any defect which notice-ably disturbs refractive index can be detected for exam-ple void delamination inclusion material inhomogeneity(fibermatrix distribution) surface roughness fiberwavinessand internal interfaces between layers (in layered structures)Principle of operation is similar to radar electromagneticwave in terahertz frequency range is emitted focused onsurface of examined material and acquired after interaction(transmission or reflection) with material In most casesdefects are detected by reflection and transmission imagingbased on pulsed terahertz TDS (TimeDomain Spectroscopy)[11] In this case electromagnetic excitation is in form of veryshort (order of picoseconds) pulsesThemethod is well suitedfor evaluation of layered materials [12] Due to the differentintrinsic impedance values presence of defect or interfacebetween separate layers causes reflection of the incident THzpulse and attenuation of the transmitted one Differences indelays of the propagated pulses and their echo (delayed layerreflections) enable characterization of the inner structure
Advances in Materials Science and Engineering 3
THz beam splitter
THz sourceTHz receiver
Evaluated material(BFRC)
THz receiver
Focusing lens
Focused THz beam
(a)
Halogen lamps
Evaluated material (BFRC)
IR camera
Halogen lamps
IR camera
(b)
Figure 2 The measurement setups (a) terahertz inspection system and (b) IRT system
state Very short pulses contain wide frequency bandwidth(005ndash3 THz) and therefore it is possible to carry one singlepoint broadband measurements [13] Photo of utilized tera-hertz measuring setup is presented in Figure 2(a)
Active infrared technique (IRT) finds its applicationin inspection of variety of materials not only compositeIn this method thermal nonequilibrium of the system isobtained using external energy source [2 3]The temperaturedifference in examined samples can be induced using severalexcitationmethods [3]The choice of propermethod dependson the tested materialsrsquo properties For conductive materialsthe induction heating can be used while in case of noncon-ductive composites materials other energy sources have to beutilized for example halogen lamps convective heating andmicrowaves The temperature distribution at the specimenrsquossurface can be observed during both heating and coolingstage using thermovision camera (in this study the FLIRA325 camera was used) Defects are detected as (dependingon the damage type) under- or overheated spots In this studythe halogen lamp heating system was used to induce the tem-perature differences within the examined composite samplesIn this method the specimen is placed in front of two halogenlamps of maximum power 2000W each positioned in the
manner which ensures uniform illumination of the samplePhoto of IRT measuring setup is presented in Figure 2(b)
4 Results of Measurements andParameters Calculation
41 Terahertz Inspection Results Terahertz inspection wasperformed using imaging system based on TRay4000 pulsedspectroscope of Picometrix Measuring head was scannedover the surface of a sample and for each (119909 119910) positiontime domain signal 119904(119905) was acquired An exemplary resultof pulsed terahertz inspection of thin basalt fiber reinforcedcomposite without defects in given (119909 119910) position is shownin Figure 3 One can distinguish seven peaks caused by stepchange of wave impedance 119885
119908on layers boundaries First
peak caused by front surface reflection (FSR) is negativeand the othermdashincluding the last one caused by back surfacereflection (BSR)mdashis positive An existence of defects changesrefractive index distribution andor introduces additionalboundaries Those effects influence obtained signal 119904(119905)Results of line scan (B-scan) over defected area are presentedin Figure 4
4 Advances in Materials Science and Engineering
BSR
FSR
FS BSTHz source
detectorFocusing
lens
s(t)
(au
)
td (ps)0 5 10 15 20 25 30 35
minus012
minus01
minus008
minus006
minus004
minus002
0
002
004
Figure 3 Exemplary A-scan signal 119904(119905) acquired over thin basaltfiber sample without defects FSR pulse caused by front surfacereflection BSR pulse caused by back surface reflection
In order to detect hidden defects inside of examinedstructure parametrization procedure was utilized The fol-lowing parameters of 119904(119905) signal were calculated for each(119909 119910) position
Maxval (119909 119910) = max (119904 (119905)(119909119910)
)
Minval (119909 119910) = min (119904 (119905)(119909119910)
)
Sumabs (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
119904 (119905)
(119909119910)
1003816
1003816
1003816
1003816
1003816
119889119905
Sumderiv (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889119905
Maxderiv (119909 119910) = max(1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
)
(1)
where 119904(119905)(119909119910)
is the A-scan signal acquired in given position(119909 119910) 119905 is a time and 119905
1 1199052are start and stop times for
integration (time gating of A-scan signal)Additionally approximation procedure was utilized The
material response signal 119904(119905) is approximated using thefollowing function
119904app (119905) =2
sum
119896=0
119901
119896119905
119896+
7
sum
119897=1
119886
119897119890
minus((119905minus119887119897)119888119897)2
(2)
where 119905 is a time 119901119896is a polynomial component coefficient
and 119886119897 119887119897 119888119897are Gaussian component coefficients
The approximation function is dedicated to thin materialconsisting of 6 layers It consists of combination of poly-nomial and Gaussian components The resulting numberof coefficients is high in case of proposed approximationfunction but it enables very precise evaluation of materialsstate at different depths (layers) Spatial distributions ofselected parameters in case of various defects are presentedin further sections Such distributions will be utilized by datafusion algorithm
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 35
BSR
FSR
(a)
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 30
BSR
FSRInclusion
(b)
Figure 4 Selected B-scan results of pulsed terahertz inspection (a)5J impact caused defect and (b) thin inclusion under 1th layer
42 Infrared Thermography Inspection During active ther-mographic inspection the halogen lamps heatingwas utilizedTaking into consideration the possibility of heating phaseobservation the Pulsed Phase Thermography (PPT) wasapplied in order to extract the information about defectsThis method combines the experimental procedure used inPulsed Thermography (PT) with signal analysis used inModulated Thermography (MT) First the thermogramsrsquosequence is recorded while the heat pulse is applied toexamined specimen and for the certain time after the heatingphase to observe the cooling phase as well In this study thesample was heated for 30 seconds and then natural process ofcooling was observed for additional 30 secondsThe recordedsequence consisted of 60 thermograms (recording frequencywas set to 1 image per second)Then the analysis of obtainedsequence is based on Discrete Fourier Transform (DFT)which allows evaluating the output as the combinationof phase (PhaIRT
119899(119909 119910)) and amplitude (AmpIRT
119899(119909 119910))
images of 119899th harmonic of IRT image sequence [14ndash16] Theselected results of the thermographic inspection of the basaltfiber reinforced composite material achieved for halogenlamps heating method are presented in Figure 5
Advances in Materials Science and Engineering 5
Posit
iony
(mm
)
35
0
Position x (mm)0 35
Posit
iony
(mm
)
35
0
(a) (b)Position x (mm)
0 35
DefectAmplitude for frequency number = 4 Phase for frequency number = 4
Figure 5 Selected results of thermographic inspection (a) amplitude image AmpIRT4(119909 119910) and (b) phase image PhaIRT
4(119909 119910) obtained
using halogen lamps heating system
Nor
mal
ized
tem
pera
ture
Tno
rm
084
086
088
09
092
094
096
098
1
5020 30 4010 600Heating time t (s)
Figure 6Measured temperature changes of selected (119909 119910) position
Obtained amplitude AmpIRT(119909 119910) and phase PhaIRT(119909119910) distributions can be utilized as parameters for fur-ther analysismdashdata fusion and defects detection proce-dure Another interesting parameter that can be calculatedfrom frequency transformed time sequence of images is totalharmonic distortion defined as
THD (119909 119910) =
radic
sum
119873
119899=2[AmpIRT
119899(119909 119910)]
2
AmpIRT1(119909 119910)
(3)
where 119899 is the harmonics number of image sequencesTemperature changes in case of selected (119909 119910) position
are shown in Figure 6 During first 31 seconds temperature isgrowing semilinearly (heating phase) After this heat source isturned off and the sample temperature is decreasing (coolingphase) Based on heating phase observation the followinglinear approximation model is proposed
119879 (119909 119910 119905) = 120572IRT (119909 119910) sdot 119905 + 120573IRT (119909 119910) (4)
where 119879(119909 119910) is the temperature at given position (119909 119910)120572IRT(119909 119910) is themultiplicative coefficient (slope of the heatingphase line) and 120573IRT(119909 119910) is the additive coefficient
Selected parameters will be utilized by data fusion proce-dure presented in next sections
5 Evaluation Using MultipleFeatures Data Fusion
None of the methods allow us to fully assess the integrationstage of the structure Both inspection techniques bringcrucial information in the process of the nondestructivestructural integrity imaging Even within a single methoddifferent features of acquired data can provide one withunique information pertaining to current stage of the mate-rialrsquos structure that is time response allows monitoringmaterialrsquos condition at different depth Therefore in orderto fully conduct the process of nondestructive imaging ofthe structure state data fusion of multiple features extractedfrom both methods results was carried out The multiplesourcesrsquo inspection using the methods coming from differentphysical origins and utilizing various phenomena to visualizethe structure state make the evaluation system more robustfor the unwanted disturbances and increase the overallperformance of the accession process [4 5 17]
The block diagram of the algorithm utilized in thispaper is presented in Figure 7 First features extracted fromboth inspection methods are transformed into commonrepresentation format undergoing the spatial and resolu-tion registration process Then a representative set of alltransformed features allowingmonitoring of different aspectsof material structure were chosen for final definition ofdatabase feature vector Finally a knowledge extraction for thepurpose of vector dimension reduction was carried out and amultiresolution data fusion was proceeded for final materialrsquosstructure imaging
51 Spatial Registration of Extracted Features In order totransform the features of both methods the data registrationprocess must be carried outTherefore several aspects have to
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
2 Advances in Materials Science and Engineering
Table 1 Selected properties of basalt and glass fibers [7 8]
Properties Unit Basalt E-Glass S-GlassDensity gcm3 27 257 248Thermal linear expansion coefficient ppm∘C 80 54 29Tensile strength GPa 484 345 471Elastic modulus GPa 89 77 89Elongation at break 315 47 56Compression strength MPa 3792 3033 3033Maximum application temperature ∘C 982∘ 650∘ 650∘
Sustained operating temperature ∘C 820∘ 480∘ 480∘
Minimum operating temperature ∘C minus260∘ minus60∘ minus60∘
Melting temperature ∘C 1450∘ 1120∘ 1120∘
Absorption of humidity (65 RAH) lt01 lt01 lt01Emissivity at 20∘C mdash 092 065 065Thermal conductivity WmK 0031ndash0038 0034ndash004 0034ndash004
40
40
Thickness tbc asymp 10ndash12mm
(a) (b)
Figure 1 Photo of the examined specimen 119904IMC 5 J (basalt reinforced composite material exposed to impact energy) (a) general view and (b)enlargement of measured region
an advantage of basalt over glass fibers in case of mechanicaland thermal properties
The photo of the utilized basalt fibers reinforced com-posite specimen is presented in Figure 1 The materialconsists of six layers of basalt fiber fabric within a polyesterresin (Polimal 1094 AWTP-1) The thickness of the sampleswas 1ndash12mm thus thickness of single layer was less than200120583mThe evaluated basalt test specimens were exposed tomechanical impacts of 2 and 5J energy (sample 119904IMP 2 J and119904IMP 5 J) During samples production process artificial inclu-sions (mica film of 145 120583m thickness) were also introducedunder selected layers (samples 119904INC A and 119904INC B)
3 Multisource Inspection Procedure
In the process of inspection of thin basalt fiber reinforcedcomposite materials two nondestructive methods were uti-lized The measurements were carried out using terahertz(THz) inspection technique and active infrared thermogra-phy (IRT) This approach enabled more efficient detectionof defects based on changes of both electrical and thermalparameters
Terahertz electromagnetic radiation enables noninva-sive nonionizing and noncontact examination of dielectricmaterials such as plastics dry wood explosives ceramicsfoams and composites [9 10] Any defect which notice-ably disturbs refractive index can be detected for exam-ple void delamination inclusion material inhomogeneity(fibermatrix distribution) surface roughness fiberwavinessand internal interfaces between layers (in layered structures)Principle of operation is similar to radar electromagneticwave in terahertz frequency range is emitted focused onsurface of examined material and acquired after interaction(transmission or reflection) with material In most casesdefects are detected by reflection and transmission imagingbased on pulsed terahertz TDS (TimeDomain Spectroscopy)[11] In this case electromagnetic excitation is in form of veryshort (order of picoseconds) pulsesThemethod is well suitedfor evaluation of layered materials [12] Due to the differentintrinsic impedance values presence of defect or interfacebetween separate layers causes reflection of the incident THzpulse and attenuation of the transmitted one Differences indelays of the propagated pulses and their echo (delayed layerreflections) enable characterization of the inner structure
Advances in Materials Science and Engineering 3
THz beam splitter
THz sourceTHz receiver
Evaluated material(BFRC)
THz receiver
Focusing lens
Focused THz beam
(a)
Halogen lamps
Evaluated material (BFRC)
IR camera
Halogen lamps
IR camera
(b)
Figure 2 The measurement setups (a) terahertz inspection system and (b) IRT system
state Very short pulses contain wide frequency bandwidth(005ndash3 THz) and therefore it is possible to carry one singlepoint broadband measurements [13] Photo of utilized tera-hertz measuring setup is presented in Figure 2(a)
Active infrared technique (IRT) finds its applicationin inspection of variety of materials not only compositeIn this method thermal nonequilibrium of the system isobtained using external energy source [2 3]The temperaturedifference in examined samples can be induced using severalexcitationmethods [3]The choice of propermethod dependson the tested materialsrsquo properties For conductive materialsthe induction heating can be used while in case of noncon-ductive composites materials other energy sources have to beutilized for example halogen lamps convective heating andmicrowaves The temperature distribution at the specimenrsquossurface can be observed during both heating and coolingstage using thermovision camera (in this study the FLIRA325 camera was used) Defects are detected as (dependingon the damage type) under- or overheated spots In this studythe halogen lamp heating system was used to induce the tem-perature differences within the examined composite samplesIn this method the specimen is placed in front of two halogenlamps of maximum power 2000W each positioned in the
manner which ensures uniform illumination of the samplePhoto of IRT measuring setup is presented in Figure 2(b)
4 Results of Measurements andParameters Calculation
41 Terahertz Inspection Results Terahertz inspection wasperformed using imaging system based on TRay4000 pulsedspectroscope of Picometrix Measuring head was scannedover the surface of a sample and for each (119909 119910) positiontime domain signal 119904(119905) was acquired An exemplary resultof pulsed terahertz inspection of thin basalt fiber reinforcedcomposite without defects in given (119909 119910) position is shownin Figure 3 One can distinguish seven peaks caused by stepchange of wave impedance 119885
119908on layers boundaries First
peak caused by front surface reflection (FSR) is negativeand the othermdashincluding the last one caused by back surfacereflection (BSR)mdashis positive An existence of defects changesrefractive index distribution andor introduces additionalboundaries Those effects influence obtained signal 119904(119905)Results of line scan (B-scan) over defected area are presentedin Figure 4
4 Advances in Materials Science and Engineering
BSR
FSR
FS BSTHz source
detectorFocusing
lens
s(t)
(au
)
td (ps)0 5 10 15 20 25 30 35
minus012
minus01
minus008
minus006
minus004
minus002
0
002
004
Figure 3 Exemplary A-scan signal 119904(119905) acquired over thin basaltfiber sample without defects FSR pulse caused by front surfacereflection BSR pulse caused by back surface reflection
In order to detect hidden defects inside of examinedstructure parametrization procedure was utilized The fol-lowing parameters of 119904(119905) signal were calculated for each(119909 119910) position
Maxval (119909 119910) = max (119904 (119905)(119909119910)
)
Minval (119909 119910) = min (119904 (119905)(119909119910)
)
Sumabs (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
119904 (119905)
(119909119910)
1003816
1003816
1003816
1003816
1003816
119889119905
Sumderiv (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889119905
Maxderiv (119909 119910) = max(1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
)
(1)
where 119904(119905)(119909119910)
is the A-scan signal acquired in given position(119909 119910) 119905 is a time and 119905
1 1199052are start and stop times for
integration (time gating of A-scan signal)Additionally approximation procedure was utilized The
material response signal 119904(119905) is approximated using thefollowing function
119904app (119905) =2
sum
119896=0
119901
119896119905
119896+
7
sum
119897=1
119886
119897119890
minus((119905minus119887119897)119888119897)2
(2)
where 119905 is a time 119901119896is a polynomial component coefficient
and 119886119897 119887119897 119888119897are Gaussian component coefficients
The approximation function is dedicated to thin materialconsisting of 6 layers It consists of combination of poly-nomial and Gaussian components The resulting numberof coefficients is high in case of proposed approximationfunction but it enables very precise evaluation of materialsstate at different depths (layers) Spatial distributions ofselected parameters in case of various defects are presentedin further sections Such distributions will be utilized by datafusion algorithm
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 35
BSR
FSR
(a)
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 30
BSR
FSRInclusion
(b)
Figure 4 Selected B-scan results of pulsed terahertz inspection (a)5J impact caused defect and (b) thin inclusion under 1th layer
42 Infrared Thermography Inspection During active ther-mographic inspection the halogen lamps heatingwas utilizedTaking into consideration the possibility of heating phaseobservation the Pulsed Phase Thermography (PPT) wasapplied in order to extract the information about defectsThis method combines the experimental procedure used inPulsed Thermography (PT) with signal analysis used inModulated Thermography (MT) First the thermogramsrsquosequence is recorded while the heat pulse is applied toexamined specimen and for the certain time after the heatingphase to observe the cooling phase as well In this study thesample was heated for 30 seconds and then natural process ofcooling was observed for additional 30 secondsThe recordedsequence consisted of 60 thermograms (recording frequencywas set to 1 image per second)Then the analysis of obtainedsequence is based on Discrete Fourier Transform (DFT)which allows evaluating the output as the combinationof phase (PhaIRT
119899(119909 119910)) and amplitude (AmpIRT
119899(119909 119910))
images of 119899th harmonic of IRT image sequence [14ndash16] Theselected results of the thermographic inspection of the basaltfiber reinforced composite material achieved for halogenlamps heating method are presented in Figure 5
Advances in Materials Science and Engineering 5
Posit
iony
(mm
)
35
0
Position x (mm)0 35
Posit
iony
(mm
)
35
0
(a) (b)Position x (mm)
0 35
DefectAmplitude for frequency number = 4 Phase for frequency number = 4
Figure 5 Selected results of thermographic inspection (a) amplitude image AmpIRT4(119909 119910) and (b) phase image PhaIRT
4(119909 119910) obtained
using halogen lamps heating system
Nor
mal
ized
tem
pera
ture
Tno
rm
084
086
088
09
092
094
096
098
1
5020 30 4010 600Heating time t (s)
Figure 6Measured temperature changes of selected (119909 119910) position
Obtained amplitude AmpIRT(119909 119910) and phase PhaIRT(119909119910) distributions can be utilized as parameters for fur-ther analysismdashdata fusion and defects detection proce-dure Another interesting parameter that can be calculatedfrom frequency transformed time sequence of images is totalharmonic distortion defined as
THD (119909 119910) =
radic
sum
119873
119899=2[AmpIRT
119899(119909 119910)]
2
AmpIRT1(119909 119910)
(3)
where 119899 is the harmonics number of image sequencesTemperature changes in case of selected (119909 119910) position
are shown in Figure 6 During first 31 seconds temperature isgrowing semilinearly (heating phase) After this heat source isturned off and the sample temperature is decreasing (coolingphase) Based on heating phase observation the followinglinear approximation model is proposed
119879 (119909 119910 119905) = 120572IRT (119909 119910) sdot 119905 + 120573IRT (119909 119910) (4)
where 119879(119909 119910) is the temperature at given position (119909 119910)120572IRT(119909 119910) is themultiplicative coefficient (slope of the heatingphase line) and 120573IRT(119909 119910) is the additive coefficient
Selected parameters will be utilized by data fusion proce-dure presented in next sections
5 Evaluation Using MultipleFeatures Data Fusion
None of the methods allow us to fully assess the integrationstage of the structure Both inspection techniques bringcrucial information in the process of the nondestructivestructural integrity imaging Even within a single methoddifferent features of acquired data can provide one withunique information pertaining to current stage of the mate-rialrsquos structure that is time response allows monitoringmaterialrsquos condition at different depth Therefore in orderto fully conduct the process of nondestructive imaging ofthe structure state data fusion of multiple features extractedfrom both methods results was carried out The multiplesourcesrsquo inspection using the methods coming from differentphysical origins and utilizing various phenomena to visualizethe structure state make the evaluation system more robustfor the unwanted disturbances and increase the overallperformance of the accession process [4 5 17]
The block diagram of the algorithm utilized in thispaper is presented in Figure 7 First features extracted fromboth inspection methods are transformed into commonrepresentation format undergoing the spatial and resolu-tion registration process Then a representative set of alltransformed features allowingmonitoring of different aspectsof material structure were chosen for final definition ofdatabase feature vector Finally a knowledge extraction for thepurpose of vector dimension reduction was carried out and amultiresolution data fusion was proceeded for final materialrsquosstructure imaging
51 Spatial Registration of Extracted Features In order totransform the features of both methods the data registrationprocess must be carried outTherefore several aspects have to
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 3
THz beam splitter
THz sourceTHz receiver
Evaluated material(BFRC)
THz receiver
Focusing lens
Focused THz beam
(a)
Halogen lamps
Evaluated material (BFRC)
IR camera
Halogen lamps
IR camera
(b)
Figure 2 The measurement setups (a) terahertz inspection system and (b) IRT system
state Very short pulses contain wide frequency bandwidth(005ndash3 THz) and therefore it is possible to carry one singlepoint broadband measurements [13] Photo of utilized tera-hertz measuring setup is presented in Figure 2(a)
Active infrared technique (IRT) finds its applicationin inspection of variety of materials not only compositeIn this method thermal nonequilibrium of the system isobtained using external energy source [2 3]The temperaturedifference in examined samples can be induced using severalexcitationmethods [3]The choice of propermethod dependson the tested materialsrsquo properties For conductive materialsthe induction heating can be used while in case of noncon-ductive composites materials other energy sources have to beutilized for example halogen lamps convective heating andmicrowaves The temperature distribution at the specimenrsquossurface can be observed during both heating and coolingstage using thermovision camera (in this study the FLIRA325 camera was used) Defects are detected as (dependingon the damage type) under- or overheated spots In this studythe halogen lamp heating system was used to induce the tem-perature differences within the examined composite samplesIn this method the specimen is placed in front of two halogenlamps of maximum power 2000W each positioned in the
manner which ensures uniform illumination of the samplePhoto of IRT measuring setup is presented in Figure 2(b)
4 Results of Measurements andParameters Calculation
41 Terahertz Inspection Results Terahertz inspection wasperformed using imaging system based on TRay4000 pulsedspectroscope of Picometrix Measuring head was scannedover the surface of a sample and for each (119909 119910) positiontime domain signal 119904(119905) was acquired An exemplary resultof pulsed terahertz inspection of thin basalt fiber reinforcedcomposite without defects in given (119909 119910) position is shownin Figure 3 One can distinguish seven peaks caused by stepchange of wave impedance 119885
119908on layers boundaries First
peak caused by front surface reflection (FSR) is negativeand the othermdashincluding the last one caused by back surfacereflection (BSR)mdashis positive An existence of defects changesrefractive index distribution andor introduces additionalboundaries Those effects influence obtained signal 119904(119905)Results of line scan (B-scan) over defected area are presentedin Figure 4
4 Advances in Materials Science and Engineering
BSR
FSR
FS BSTHz source
detectorFocusing
lens
s(t)
(au
)
td (ps)0 5 10 15 20 25 30 35
minus012
minus01
minus008
minus006
minus004
minus002
0
002
004
Figure 3 Exemplary A-scan signal 119904(119905) acquired over thin basaltfiber sample without defects FSR pulse caused by front surfacereflection BSR pulse caused by back surface reflection
In order to detect hidden defects inside of examinedstructure parametrization procedure was utilized The fol-lowing parameters of 119904(119905) signal were calculated for each(119909 119910) position
Maxval (119909 119910) = max (119904 (119905)(119909119910)
)
Minval (119909 119910) = min (119904 (119905)(119909119910)
)
Sumabs (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
119904 (119905)
(119909119910)
1003816
1003816
1003816
1003816
1003816
119889119905
Sumderiv (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889119905
Maxderiv (119909 119910) = max(1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
)
(1)
where 119904(119905)(119909119910)
is the A-scan signal acquired in given position(119909 119910) 119905 is a time and 119905
1 1199052are start and stop times for
integration (time gating of A-scan signal)Additionally approximation procedure was utilized The
material response signal 119904(119905) is approximated using thefollowing function
119904app (119905) =2
sum
119896=0
119901
119896119905
119896+
7
sum
119897=1
119886
119897119890
minus((119905minus119887119897)119888119897)2
(2)
where 119905 is a time 119901119896is a polynomial component coefficient
and 119886119897 119887119897 119888119897are Gaussian component coefficients
The approximation function is dedicated to thin materialconsisting of 6 layers It consists of combination of poly-nomial and Gaussian components The resulting numberof coefficients is high in case of proposed approximationfunction but it enables very precise evaluation of materialsstate at different depths (layers) Spatial distributions ofselected parameters in case of various defects are presentedin further sections Such distributions will be utilized by datafusion algorithm
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 35
BSR
FSR
(a)
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 30
BSR
FSRInclusion
(b)
Figure 4 Selected B-scan results of pulsed terahertz inspection (a)5J impact caused defect and (b) thin inclusion under 1th layer
42 Infrared Thermography Inspection During active ther-mographic inspection the halogen lamps heatingwas utilizedTaking into consideration the possibility of heating phaseobservation the Pulsed Phase Thermography (PPT) wasapplied in order to extract the information about defectsThis method combines the experimental procedure used inPulsed Thermography (PT) with signal analysis used inModulated Thermography (MT) First the thermogramsrsquosequence is recorded while the heat pulse is applied toexamined specimen and for the certain time after the heatingphase to observe the cooling phase as well In this study thesample was heated for 30 seconds and then natural process ofcooling was observed for additional 30 secondsThe recordedsequence consisted of 60 thermograms (recording frequencywas set to 1 image per second)Then the analysis of obtainedsequence is based on Discrete Fourier Transform (DFT)which allows evaluating the output as the combinationof phase (PhaIRT
119899(119909 119910)) and amplitude (AmpIRT
119899(119909 119910))
images of 119899th harmonic of IRT image sequence [14ndash16] Theselected results of the thermographic inspection of the basaltfiber reinforced composite material achieved for halogenlamps heating method are presented in Figure 5
Advances in Materials Science and Engineering 5
Posit
iony
(mm
)
35
0
Position x (mm)0 35
Posit
iony
(mm
)
35
0
(a) (b)Position x (mm)
0 35
DefectAmplitude for frequency number = 4 Phase for frequency number = 4
Figure 5 Selected results of thermographic inspection (a) amplitude image AmpIRT4(119909 119910) and (b) phase image PhaIRT
4(119909 119910) obtained
using halogen lamps heating system
Nor
mal
ized
tem
pera
ture
Tno
rm
084
086
088
09
092
094
096
098
1
5020 30 4010 600Heating time t (s)
Figure 6Measured temperature changes of selected (119909 119910) position
Obtained amplitude AmpIRT(119909 119910) and phase PhaIRT(119909119910) distributions can be utilized as parameters for fur-ther analysismdashdata fusion and defects detection proce-dure Another interesting parameter that can be calculatedfrom frequency transformed time sequence of images is totalharmonic distortion defined as
THD (119909 119910) =
radic
sum
119873
119899=2[AmpIRT
119899(119909 119910)]
2
AmpIRT1(119909 119910)
(3)
where 119899 is the harmonics number of image sequencesTemperature changes in case of selected (119909 119910) position
are shown in Figure 6 During first 31 seconds temperature isgrowing semilinearly (heating phase) After this heat source isturned off and the sample temperature is decreasing (coolingphase) Based on heating phase observation the followinglinear approximation model is proposed
119879 (119909 119910 119905) = 120572IRT (119909 119910) sdot 119905 + 120573IRT (119909 119910) (4)
where 119879(119909 119910) is the temperature at given position (119909 119910)120572IRT(119909 119910) is themultiplicative coefficient (slope of the heatingphase line) and 120573IRT(119909 119910) is the additive coefficient
Selected parameters will be utilized by data fusion proce-dure presented in next sections
5 Evaluation Using MultipleFeatures Data Fusion
None of the methods allow us to fully assess the integrationstage of the structure Both inspection techniques bringcrucial information in the process of the nondestructivestructural integrity imaging Even within a single methoddifferent features of acquired data can provide one withunique information pertaining to current stage of the mate-rialrsquos structure that is time response allows monitoringmaterialrsquos condition at different depth Therefore in orderto fully conduct the process of nondestructive imaging ofthe structure state data fusion of multiple features extractedfrom both methods results was carried out The multiplesourcesrsquo inspection using the methods coming from differentphysical origins and utilizing various phenomena to visualizethe structure state make the evaluation system more robustfor the unwanted disturbances and increase the overallperformance of the accession process [4 5 17]
The block diagram of the algorithm utilized in thispaper is presented in Figure 7 First features extracted fromboth inspection methods are transformed into commonrepresentation format undergoing the spatial and resolu-tion registration process Then a representative set of alltransformed features allowingmonitoring of different aspectsof material structure were chosen for final definition ofdatabase feature vector Finally a knowledge extraction for thepurpose of vector dimension reduction was carried out and amultiresolution data fusion was proceeded for final materialrsquosstructure imaging
51 Spatial Registration of Extracted Features In order totransform the features of both methods the data registrationprocess must be carried outTherefore several aspects have to
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
4 Advances in Materials Science and Engineering
BSR
FSR
FS BSTHz source
detectorFocusing
lens
s(t)
(au
)
td (ps)0 5 10 15 20 25 30 35
minus012
minus01
minus008
minus006
minus004
minus002
0
002
004
Figure 3 Exemplary A-scan signal 119904(119905) acquired over thin basaltfiber sample without defects FSR pulse caused by front surfacereflection BSR pulse caused by back surface reflection
In order to detect hidden defects inside of examinedstructure parametrization procedure was utilized The fol-lowing parameters of 119904(119905) signal were calculated for each(119909 119910) position
Maxval (119909 119910) = max (119904 (119905)(119909119910)
)
Minval (119909 119910) = min (119904 (119905)(119909119910)
)
Sumabs (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
119904 (119905)
(119909119910)
1003816
1003816
1003816
1003816
1003816
119889119905
Sumderiv (119909 119910) = int1199052
1199051
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889119905
Maxderiv (119909 119910) = max(1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
119889
2119904 (119905)
(119909119910)
(119889119905)
2
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
1003816
)
(1)
where 119904(119905)(119909119910)
is the A-scan signal acquired in given position(119909 119910) 119905 is a time and 119905
1 1199052are start and stop times for
integration (time gating of A-scan signal)Additionally approximation procedure was utilized The
material response signal 119904(119905) is approximated using thefollowing function
119904app (119905) =2
sum
119896=0
119901
119896119905
119896+
7
sum
119897=1
119886
119897119890
minus((119905minus119887119897)119888119897)2
(2)
where 119905 is a time 119901119896is a polynomial component coefficient
and 119886119897 119887119897 119888119897are Gaussian component coefficients
The approximation function is dedicated to thin materialconsisting of 6 layers It consists of combination of poly-nomial and Gaussian components The resulting numberof coefficients is high in case of proposed approximationfunction but it enables very precise evaluation of materialsstate at different depths (layers) Spatial distributions ofselected parameters in case of various defects are presentedin further sections Such distributions will be utilized by datafusion algorithm
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 35
BSR
FSR
(a)
Tim
e del
ayt d
(ps)
35
0
Position x (mm)0 30
BSR
FSRInclusion
(b)
Figure 4 Selected B-scan results of pulsed terahertz inspection (a)5J impact caused defect and (b) thin inclusion under 1th layer
42 Infrared Thermography Inspection During active ther-mographic inspection the halogen lamps heatingwas utilizedTaking into consideration the possibility of heating phaseobservation the Pulsed Phase Thermography (PPT) wasapplied in order to extract the information about defectsThis method combines the experimental procedure used inPulsed Thermography (PT) with signal analysis used inModulated Thermography (MT) First the thermogramsrsquosequence is recorded while the heat pulse is applied toexamined specimen and for the certain time after the heatingphase to observe the cooling phase as well In this study thesample was heated for 30 seconds and then natural process ofcooling was observed for additional 30 secondsThe recordedsequence consisted of 60 thermograms (recording frequencywas set to 1 image per second)Then the analysis of obtainedsequence is based on Discrete Fourier Transform (DFT)which allows evaluating the output as the combinationof phase (PhaIRT
119899(119909 119910)) and amplitude (AmpIRT
119899(119909 119910))
images of 119899th harmonic of IRT image sequence [14ndash16] Theselected results of the thermographic inspection of the basaltfiber reinforced composite material achieved for halogenlamps heating method are presented in Figure 5
Advances in Materials Science and Engineering 5
Posit
iony
(mm
)
35
0
Position x (mm)0 35
Posit
iony
(mm
)
35
0
(a) (b)Position x (mm)
0 35
DefectAmplitude for frequency number = 4 Phase for frequency number = 4
Figure 5 Selected results of thermographic inspection (a) amplitude image AmpIRT4(119909 119910) and (b) phase image PhaIRT
4(119909 119910) obtained
using halogen lamps heating system
Nor
mal
ized
tem
pera
ture
Tno
rm
084
086
088
09
092
094
096
098
1
5020 30 4010 600Heating time t (s)
Figure 6Measured temperature changes of selected (119909 119910) position
Obtained amplitude AmpIRT(119909 119910) and phase PhaIRT(119909119910) distributions can be utilized as parameters for fur-ther analysismdashdata fusion and defects detection proce-dure Another interesting parameter that can be calculatedfrom frequency transformed time sequence of images is totalharmonic distortion defined as
THD (119909 119910) =
radic
sum
119873
119899=2[AmpIRT
119899(119909 119910)]
2
AmpIRT1(119909 119910)
(3)
where 119899 is the harmonics number of image sequencesTemperature changes in case of selected (119909 119910) position
are shown in Figure 6 During first 31 seconds temperature isgrowing semilinearly (heating phase) After this heat source isturned off and the sample temperature is decreasing (coolingphase) Based on heating phase observation the followinglinear approximation model is proposed
119879 (119909 119910 119905) = 120572IRT (119909 119910) sdot 119905 + 120573IRT (119909 119910) (4)
where 119879(119909 119910) is the temperature at given position (119909 119910)120572IRT(119909 119910) is themultiplicative coefficient (slope of the heatingphase line) and 120573IRT(119909 119910) is the additive coefficient
Selected parameters will be utilized by data fusion proce-dure presented in next sections
5 Evaluation Using MultipleFeatures Data Fusion
None of the methods allow us to fully assess the integrationstage of the structure Both inspection techniques bringcrucial information in the process of the nondestructivestructural integrity imaging Even within a single methoddifferent features of acquired data can provide one withunique information pertaining to current stage of the mate-rialrsquos structure that is time response allows monitoringmaterialrsquos condition at different depth Therefore in orderto fully conduct the process of nondestructive imaging ofthe structure state data fusion of multiple features extractedfrom both methods results was carried out The multiplesourcesrsquo inspection using the methods coming from differentphysical origins and utilizing various phenomena to visualizethe structure state make the evaluation system more robustfor the unwanted disturbances and increase the overallperformance of the accession process [4 5 17]
The block diagram of the algorithm utilized in thispaper is presented in Figure 7 First features extracted fromboth inspection methods are transformed into commonrepresentation format undergoing the spatial and resolu-tion registration process Then a representative set of alltransformed features allowingmonitoring of different aspectsof material structure were chosen for final definition ofdatabase feature vector Finally a knowledge extraction for thepurpose of vector dimension reduction was carried out and amultiresolution data fusion was proceeded for final materialrsquosstructure imaging
51 Spatial Registration of Extracted Features In order totransform the features of both methods the data registrationprocess must be carried outTherefore several aspects have to
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 5
Posit
iony
(mm
)
35
0
Position x (mm)0 35
Posit
iony
(mm
)
35
0
(a) (b)Position x (mm)
0 35
DefectAmplitude for frequency number = 4 Phase for frequency number = 4
Figure 5 Selected results of thermographic inspection (a) amplitude image AmpIRT4(119909 119910) and (b) phase image PhaIRT
4(119909 119910) obtained
using halogen lamps heating system
Nor
mal
ized
tem
pera
ture
Tno
rm
084
086
088
09
092
094
096
098
1
5020 30 4010 600Heating time t (s)
Figure 6Measured temperature changes of selected (119909 119910) position
Obtained amplitude AmpIRT(119909 119910) and phase PhaIRT(119909119910) distributions can be utilized as parameters for fur-ther analysismdashdata fusion and defects detection proce-dure Another interesting parameter that can be calculatedfrom frequency transformed time sequence of images is totalharmonic distortion defined as
THD (119909 119910) =
radic
sum
119873
119899=2[AmpIRT
119899(119909 119910)]
2
AmpIRT1(119909 119910)
(3)
where 119899 is the harmonics number of image sequencesTemperature changes in case of selected (119909 119910) position
are shown in Figure 6 During first 31 seconds temperature isgrowing semilinearly (heating phase) After this heat source isturned off and the sample temperature is decreasing (coolingphase) Based on heating phase observation the followinglinear approximation model is proposed
119879 (119909 119910 119905) = 120572IRT (119909 119910) sdot 119905 + 120573IRT (119909 119910) (4)
where 119879(119909 119910) is the temperature at given position (119909 119910)120572IRT(119909 119910) is themultiplicative coefficient (slope of the heatingphase line) and 120573IRT(119909 119910) is the additive coefficient
Selected parameters will be utilized by data fusion proce-dure presented in next sections
5 Evaluation Using MultipleFeatures Data Fusion
None of the methods allow us to fully assess the integrationstage of the structure Both inspection techniques bringcrucial information in the process of the nondestructivestructural integrity imaging Even within a single methoddifferent features of acquired data can provide one withunique information pertaining to current stage of the mate-rialrsquos structure that is time response allows monitoringmaterialrsquos condition at different depth Therefore in orderto fully conduct the process of nondestructive imaging ofthe structure state data fusion of multiple features extractedfrom both methods results was carried out The multiplesourcesrsquo inspection using the methods coming from differentphysical origins and utilizing various phenomena to visualizethe structure state make the evaluation system more robustfor the unwanted disturbances and increase the overallperformance of the accession process [4 5 17]
The block diagram of the algorithm utilized in thispaper is presented in Figure 7 First features extracted fromboth inspection methods are transformed into commonrepresentation format undergoing the spatial and resolu-tion registration process Then a representative set of alltransformed features allowingmonitoring of different aspectsof material structure were chosen for final definition ofdatabase feature vector Finally a knowledge extraction for thepurpose of vector dimension reduction was carried out and amultiresolution data fusion was proceeded for final materialrsquosstructure imaging
51 Spatial Registration of Extracted Features In order totransform the features of both methods the data registrationprocess must be carried outTherefore several aspects have to
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
6 Advances in Materials Science and Engineering
Extractedfeaturesof THzimaging
Extractedfeaturesof IRT
imaging
Dataregistration
Featureselection
Featureselection
Finaldefinitionof feature
vector
Multiresolutionbased data
fusion
Fusedresults
Figure 7 The functional block diagram of the data fusion process
IRT image
THz image
Controlpoints
detection
Control points
matching
Imagesresampling and
spatialtransformation
Transformation modelmatrix
Registration process
Inputfeatures
Registeredimages
IRT image
THz image
Trans-formation
matrixcalculation
CCD image
Figure 8 The functional block diagram of the data registration process
be considered to carry out the registration process The mostimportant factors that should be taken under considerationare each methodrsquos sensing element geometrical distortionsand position (alignment and rotation) with respect to theevaluated material or measuring resolution The block dia-gram of the applied algorithm is presented in Figure 8 Highresolution photo of the sample was used as the referenceimage First dedicated metallic markers (control points CP)detectable by bothmethodswere usedThen theCPmatchingprocess was proceeded Taking into consideration that duringthe experiments different position of the sensing elementin respect to the examined composite was applied differenttransformation model had to be considered Visualization ofthe measuring procedure and setup of the sensing deviceswith respect to sample surface were presented in Figure 9 Incase of terahertz inspection both wave source and detectorwere placed parallel to the composite material (parallelprojection) while in case of the thermographic inspectionthe infrared camera was observing a sample from someperspective (perspective diametric projection) Thereforeconsidering the projection type the similarity and the pro-jective transformation of data were applied respectively infirst and second case
Both conversions are a standard geometrical transforma-tion utilized in image processing algorithms and are definedas follows [18 19]
(i) Similarity transformation
119909
1015840= 119904 (119909 cos120572 minus 119910 sin120572 + 119905
119909)
119910
1015840= 119904 (119909 sin120572 + 119910 cos120572 + 119905
119910)
(5)
where (119909 119910) are original coordinates (1199091015840 1199101015840) are newcoordinates (119905
119909 119905
119910) are translation coefficients which
specify the movement of the systemrsquos center 119904 is ascaling factor and 120572 corresponds to a rotation angle
(ii) Perspective transformation
119909
1015840=
119886
11119909 + 119886
12119910 + 119905
119909
119887
1119909 + 119887
2119910 + 1
119910
1015840=
119886
21119909 + 119886
22119910 + 119905
119910
119887
1119909 + 119887
2119910 + 1
(6)
where 11988611 11988612 11988621 11988622are coefficients responsible for
rotation and scaling and 119887
1and 119887
2are coefficients
defining the projection
The similarity is shape and angles preserving transfor-mation allowing scaling rotation and reflection operationThe projective one is used to transform an image perspectiveIt preserves collinearity and incidence however it affectsparallelism length and angle
After the procedure of matching of the control points(Figure 10 presents the projection of each methodrsquos resultson CCD image of the sample) the images geometrical dis-tortions were eliminated Selected results of the registrationprocess are shown in Figure 11
52 Final Feature Vector Definition After the data regis-tration process the features distributions obtained for eachsingle method were analyzed and preselected During thepreselection process two aspects were taken into account firstto minimize the amount of similar information constancyand second to preserve the possibility of observation ofdifferent aspects of structure state assessment
Finally the set of 20 features representing both methodswere used to define the featuresrsquo vector119865 for processing of the
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 7
IRTcamera
THzhead
THzinspection
area
IRTrecording
area
Basalt composite
Markers
Halogenlamps
y-axis
x-axis
(a)
IRTrecording
area
THzhead
scanningrange
THzhead
IRTcamera
(b)
Figure 9 Schematic view of the experiments setup and spatial relationship between both methods sensing units (a) top view and (b) sideview
(a) (b) (c)
Figure 10 Results of data to inspection area matching (a) photo (b) photo with depicted exemplary THz result and (c) photo with depictedexemplary IRT result
(a) (b) (c) (d)
Figure 11 Results of the data registration process IRT exemplary results (a) before and (b) after the registration process THz exemplaryresults (c) before and (d) after the registration process dpi before registration process 40 (IRT) and 25 (THz) and after 500
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
8 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 12 Results of the data registration process carried out for features vector obtained for sample impacted by 2J
data fusion algorithm The chosen feature list is presented inTable 2 while the features distributions obtained for impacted(119904IMP 2 J 119904IMP 5 J) and with inclusions (119904INC A 119904INC B) samplesare shown in Figures 12 13 14 and 15 respectively
The THz signals features allow observing rapid changesin the structure of the examined composite materials Inparticular inclusions are very well visible It is possible to pre-cisely localize rectangular shaped inclusions in case of bothsamples in distributions of features 4 6 7 8 and 10 In caseof sample 119904INC A during the production process additionaldefects were unintentionally introduced beside the inclusionThese defected areas can be visible as straight lines rangingfrom top left corner to bottom right in the distributions offeatures 1 3 8 and 11 For impacted samples the THz featuresrather allow indicating the localization where the highestdamage occurred rather than the range of the defected area
Contrary to THz the IRT imaging features are character-ized by the low dynamics of the heating system In case ofcomposites which generally have low value of thermal diffu-sivity factor the heat conduction process within material isrelatively slow Therefore the indications of defects observedat the samplersquos surface are in most cases blurred This effectcan be noticed for samples with inclusions (it is not possibleto identify the shape of the inclusions) On the other hand
the obtained features distributions make it possible to fullyassess the range of the impacted area even in the sampleimpacted with lower energy Despite the lower ratio betweenthe response to defects and the background values than incase of the THz features it is possible also to observe theadditional defects in included sample 119904INC A (see feature 1618 and 19 distributions)
53 Multiresolution Decomposition Based Data Fusion Algo-rithm The results of bothmethods represent different spatialdynamics of the feature changes In case of the THz theinspecting wave is focused on local point of the material sothe response is also from that specific point while in caseof IRT testing obviously the heat generated by the halogenlamps cannot be focused at one point but is radiated to thesamplersquos surface and then is conducted evenly throughoutthe material volume Therefore the thermal response in thespecific point is disturbed by the surround This results inlower spatial dynamics of the IRT imaging in comparisonto THz one Therefore in order to preserve the informationcontent of both rapid and slowly changing responses amultiresolution decomposition (MRD) based data fusionalgorithm was applied to the features vector
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 9
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 13 Results of the data registration process carried out for features vector obtained for sample impacted by 5J
The block diagram of the procedure is presented inFigure 16 First in order to reduce the dimensionality of theproblem the knowledge extraction was carried out for thefeaturesrsquo vector The purpose of this stage is to transformthe original featuresrsquo vector into a new coordinate systemof reduced dimensions level by extracting the most usefulinformation from the original data In order to carry out themultivariate analysis two algorithms were used IndependentComponent Analysis (ICA) and Principal Component Anal-ysis (PCA)
The general idea of the ICA is to extract from multidi-mensional set of data the independent components underthe assumption that the sources of data are statisticallyindependent [19]Thismeans that each feature value gatheredin the vector 119865 is a combination of the original sources dataThen the ICA analysis allows estimating the original sourcesfrom the collected features vector
The objective of PCA is to convert the database of originalpossibly correlated features into a database having newstatistically independent variables using orthogonal lineartransformation [19 20] The new parameters called principalcomponents pc are obtained from eigenvectors of covariancematrix of the original database and ordered in accordancewith quantity of their variance (described by eigenvalues)in the database The greater the variance of the variable isthe more significant it is Typically the first few pc variables
express the majority of database variance Therefore they canbe used to represent the whole database without losing asignificant amount of information
In both cases the analysis of the two new components(combined features cf
1and cf
2) was taken for further pro-
cessing two independent and first two principal componentsrepresenting the greatest variance of the whole set
After the new components extraction MRD was carriedoutThe general idea ofMRD is to represent the source of datawith a collection of hierarchical representations of basis func-tion at different resolutions (frequency bands) The originaldata is decomposed into a set of spatial frequency bandpassrepresentatives obtained by convolving and subsamplingoperations The data can be decomposed using variousmethods [5] In this paper the wavelet decomposition WTwas utilized WT is one of the most effective decompositionmethods allowing us to obtain a good resolution in both timeand frequency domain [21 22] After the decomposition thetwo distributions inputs at given sublevel are fused using thespecified fusion rule Then the inverse transform DWTminus1 iscomputed and the fused distribution is reconstructed Thewhole process can be described by
119865DF (119909 119910) = DWTminus1 (120601 (DWT (119865C1 (119909 119910))
DWT (119865C2 (119909 119910)))) (7)
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
10 Advances in Materials Science and Engineering
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 14 Results of the data registration process carried out for features vector obtained for sample with inclusion A
where 120601 is the fusion rule 119865C1 and 119865C2 are components 1and 2 obtained after extraction process and 119865DF is the fuseddistribution
The result achieved for the proposed algorithm withthe utilization of Daubechies 120595D10 wavelet function for thedecomposition of cf
1and cf
2and the maximum selection
as the fusion rule [22] are presented in Figures 17 and18 respectively for ICA and PCA knowledge extractionprocedure The general advantage of the used data fusionalgorithm is that no matter which feature would provide theinformation about the defect it would be indicated in thefused distribution Analyzing the obtained results one cannotice that the fused distributions carry crucial informationabout the defects and the structure delivered by all featuresAll defects were indicated in single fused distributions whichcan be especially seen in case of samples with inclusions119904INC A and 119904INC B Besides clearly visible rectangular shapedinclusions it is possible to observe additional defect of thestructure in the neighborhood of the inclusions (Figures17 18(c) and 18(d)) Also in case of impacted samples theachieved results allow distinguishing the areas of differentdamage level The applied knowledge extraction proceduresallowed compressing information inmuch less data represen-tations and the multiresolution decomposition fused bothrapid and slowly changing signalsThis was useful once again
especially in case of samples with inclusions Inclusions arecharacterized mainly by the high dynamic signalsrsquo compo-nents while additional anomalies in the composite structurebeing results of faulty production process are indicated in thelow dynamic components
6 Conclusions
Methods of inspection of the composite materials still needto be developed In comparison to the solid materials suchas steel the structure of composite materials results incomplicated response even in no defect cases Therefore aneed of implementation of the various techniques arises Twodifferent nondestructive techniques pulsed THz inspectionand active IR thermography were utilized Both of thesemethods are contactless thus their common implementa-tion is simpler Proposed techniques have various contrastmechanisms THz method as a pure electromagnetic one issensitive to electromagnetic parameters like permittivity (orrefractive index) changes Due to the very small power ofpicosecond pulses utilized as excitation temperature changesin the evaluated material can be omitted In case of IRTfrom the other hand the heat transfer within the examinedstructure is crucial and observed temperature values are con-nected directly to the changes of different physical properties
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 11
Feature 1 Feature 2 Feature 3 Feature 4 Feature 5
Feature 6 Feature 7 Feature 8 Feature 9 Feature 10
Feature 11 Feature 12 Feature 13 Feature 14 Feature 15
Feature 16 Feature 17 Feature 18 Feature 19 Feature 20
Figure 15 Results of the data registration process carried out for features vector obtained for sample with inclusion B
Feature 1
Feature 2
Feature 20
Knowledgeextraction
Combinedfeature 1
Combinedfeature 2
DWT
W coefficients
W coefficientsFusion
rule
Fused Wcoefficients
DWT
Fused data
ICA PCA
DWTminus1
Figure 16 Multiresolution decomposition based data fusion algorithm diagram DWT and DWTminus1 wavelet multiresolution decompositionand its inverse
like thermal conductivity or material density It was shownthat different features of THz and IRT inspection methodssignals can provide strong indication of different defects thatarose in thin basalt fiber reinforced composites Howeverno single feature allows observing all aspects of materialrsquosstructure state Therefore the next step in inspection proce-dures which was presented in this paper is to combine the
information from various techniques to present the resultsin common format It was shown that utilization of methodsbased on both electromagnetic and thermal parametersenables better detection of various types of defects in basaltfiber reinforced compositesThe appliedmultiresolution datafusion algorithm allowed preserving information represent-ing diverse dynamic ranges of signals Different defects types
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
12 Advances in Materials Science and Engineering
(a) (b) (c) (d)
Figure 17 Results of the ICA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
(a) (b) (c) (d)
Figure 18 Results of the PCA-DWT data fusion algorithm obtained for sample (a) 119904IMP 2 J (b) 119904IMP 5 J (c) 119904INC A and (d) 119904INC B
Table 2 Featuresrsquo vector 119865 of the database
Feature number Name Measuring method1 Maxval THz2 Minval THz3 Sumabs THz4 Sumderiv THz5 Maxderiv THz6 119886
1THz
7 119887
2THz
8 119887
4THz
9 119904app(1198871) THz10 119904app(1198873) THz11 119904app(1198877) THz12 AmpIRT
1IRT
13 AmpIRT2
IRT14 AmpIRT
3IRT
15 PhaIRT2
IRT16 PhaIRT
3IRT
17 PhaIRT4
IRT18 THD IRT19 120572IRT IRT20 120573IRT IRT
reflect different spatial frequency of the obtained parametersdistribution Moreover both inspection techniques are alsorepresenting different defects response projections on theacquired signals Therefore the utilized MRD with supportof knowledge extraction procedures can result in an effectiveevaluation of the examined materialsrsquo structure state
The performance of data fusion algorithms can beassessed both qualitatively and quantitatively [23] Thesubjective estimation of quality (qualitative assessment) of
achieved results confirmed that the fusion of few inspectionmethods can be effective in detection of different defectssuch as resin and fibers degradation or inclusions The com-plementary information about the defects details gatheredby single inspection methods is fused into the commonrepresentation Nevertheless quantitative evaluation of thedata fusion results is a complex process because thereexists no ground truth reference distribution contacting allinformation about the structure state
Competing Interests
The authors declare that they have no competing interests
Acknowledgments
The authors would like to express special thanks to ProfessorTomasz Chady and Dr Krzysztof Goracy fromWest Pomera-nian University of Technology in Szczecin for providingsamples
References
[1] D K Hsu ldquoNondestructive inspection of composite structuresmethods and practicerdquo in Proceedings of the 17th World Confer-ence on Nondestructive Testing Shanghai China 2008
[2] MRaoPawar andDRao ldquoReviewof nondestructive evaluationtechniques for FRP composite structural componentsrdquo ProblemReport College of Engineering and Mineral Resources at WestVirginia University Morgantown WVa USA 2007
[3] A Kapadia Non Destructive Testing of Composites MaterialsTWI National Composites Network 2006
[4] X E Gros Application of NDT Data Fusion Kluwer AcademicNew York NY USA 2001
[5] Z Liu D S Forsyth J P Komorowski K Hanasaki andT Kirubarajan ldquoSurvey state of the art in NDE data fusion
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Advances in Materials Science and Engineering 13
techniquesrdquo IEEE Transactions on Instrumentation and Mea-surement vol 56 no 6 pp 2435ndash2451 2007
[6] K Singha ldquoA short review on basalt fiberrdquo International Journalof Textile Science vol 1 no 4 pp 19ndash28 2012
[7] R ParnasM Shaw andQ Liu ldquoBasalt fiber reinforced polymercompositesrdquo Tech Rep NE TCR63 Institute of Materials Sci-ence University of Connecticut 2007 httpwwwnetcumassdedunetcr63 03-7pdf
[8] T Bhat V Chevali X Liu S Feih and A P Mouritz ldquoFirestructural resistance of basalt fibre compositerdquo Composites PartA Applied Science and Manufacturing vol 71 pp 107ndash115 2015
[9] N Palka and D Miedzinska ldquoDetailed non-destructive evalu-ation of UHMWPE composites in the terahertz rangerdquo Opticaland Quantum Electronics vol 46 no 4 pp 515ndash525 2014
[10] D Mittelman Ed Sensing with Terahertz Radiation SpringerBerlin Germany 2010
[11] D M Mittleman M Gupta R Neelamani R G Baraniuk JV Rudd and M Koch ldquoRecent advances in terahertz imagingrdquoApplied Physics B Lasers and Optics vol 68 no 6 pp 1085ndash1094 1999
[12] P Lopato and T Chady ldquoTerahertz detection and identificationof defects in layered polymer composites and composite coat-ingsrdquo Nondestructive Testing and Evaluation vol 28 no 1 pp28ndash43 2013
[13] P Lopato ldquoTerahertz inspection of static loaded compos-ite materialsrdquo in Electromagnetic Nondestructive Evaluation(XVIII) Z Chen S Xie and Y Li Eds pp 184ndash191 IOS Press2015
[14] X Maladegue Theory and Practice of Infrared Technology forNondestructive Testing John Wiley and Sons New York NYUSA 2001
[15] B Szymanik S Unnikrishnakurup and K BalasubramaniamldquoBackground removal methods in thermographic non destruc-tive testing of compositematerialsrdquoThee-Journal ofNondestruc-tive Testing vol 20 no 6 2015
[16] S Marintetti Y A Plotnikov W Winfree and A BraggiottildquoPulse phase thermography for defect detection and visualiza-tionrdquo in Nondestructive Evaluation of Aging Aircraft Airportsand Aerospace Hardware III vol 3586 of Proceedings of SPIEpp 230ndash238 January 1999
[17] C Kohl M Krause CMaierhofer and JWostmann ldquo2D- And3D-visualisation of NDT-data using data fusion techniquerdquoMaterials and StructuresMateriaux et Constructions vol 38 no283 pp 817ndash826 2005
[18] B Zitova and J Flusser ldquoImage registration methods a surveyrdquoImage and Vision Computing vol 21 no 11 pp 977ndash1000 2003
[19] T Stathaki Image Fusion Algorithms and Applications ElsevierAcademic Press 2008
[20] C Wang W Guan J Gou F Hou J Bai and G Yan ldquoPrin-cipal component analysis based three-dimensional operationalmodal analysisrdquo International Journal of Applied Electromagnet-ics and Mechanics vol 45 no 1ndash4 pp 137ndash144 2014
[21] V P S Naidu and J R Raol ldquoPixel-level image fusion usingwavelets and principal component analysisrdquo Defence ScienceJournal vol 58 no 3 pp 338ndash352 2008
[22] P Hill N Canagarajah and D Bull ldquoImage fusion usingcomplex waveletsrdquo in Electronic Proceedings of the 13th BritishMachine Vision Conference (BMVC rsquo02) pp 2ndash5 University ofCardiff September 2002
[23] S Li Z Li and J Gong ldquoMultivariate statistical analysis ofmea-sures for assessing the quality of image fusionrdquo InternationalJournal of Image and Data Fusion vol 1 no 1 pp 47ndash66 2010
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials
Submit your manuscripts athttpwwwhindawicom
ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CorrosionInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Polymer ScienceInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CeramicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CompositesJournal of
NanoparticlesJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Biomaterials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
NanoscienceJournal of
TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Journal of
NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
CrystallographyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CoatingsJournal of
Advances in
Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Smart Materials Research
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MetallurgyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
MaterialsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Nano
materials
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal ofNanomaterials