Vous êtes sur la page 1sur 6

NDTnetApril1996,Vol.1No.

04

Ultrasonictestingandimageprocessingforinprogressweld
inspection*
byShaunLawson

Introductionaresumeofultrasonicweldinspection
TableofcontentsofthisPage:
Semiautomatedpulseechoscanningand'inprogress'inspection
Timeofflightdiffraction(TOFD)technique
Summary
References
Semiautomatedpulseechoscanningand'inprogress'inspection
Amajordevelopmentinultrasonictestinginrecentyearshasbeenuseofmechanicalmanipulatorsfor
thescanningofcomponentswithsuitablesurfacegeometries.Manualinspectionisvariableinthatthe
positionoftheprobemaymoveinarandom,uncoordinatedfashionwiththeresultthatdefectechoes
maybemissed.Theuseofamechanicalsystemtogeneratethemotionoftheprobeformalisesthe
inspectionvolumeandensuresauniformcouplingwiththecomponentsurface.Inaddition,itisa
relativelystraightforwarddevelopmenttotransformthemanipulatorcontrollerintoasophisticateddata
acquisitionsystem,whichisabletorecordsignalamplitudeandrangeforeachpositionofascanned
probe.FromthisinformationcanbederivedCartesianpositionandsignalamplitudedatawhichcontains
fullthreedimensionalinformationaboutthecomponent.Furthermore,multipleprobesmaybefixedto
themanipulatordeviceenablingsimultaneousscanningwitharangeofincidentangles.

Typically,recordedechoesfromthescansarepresentedoverlaidonaCADdrawingofthecomponentin
end,toporsideelevationsrespectivelytermedBscan,CscanandDscanplots.Inadditionto
visualisationmethods,sophisticatedcalibrationroutinesforrangefindingandattenuationmayalsobe
programmedinthescancomputer.Thistypeofsystem,howeverisrelativelyexpensive,andistypically
onlyusedtogeneratecomprehensiveNDTdatafromweldsinsafetycriticalcomponents,suchasnuclear
pressurevessels,whereevenverysmallflawsmayprovesignificant.Thedrawbacktoautomated
scanningtechniques,somewhatparadoxically,isthesheervolumeofdatawhichisproducedascan
withmultipleprobeswillgenerateaseriesofscanplotswhichrequireinvolvedmanualinterpretation.
Furthermore,ifasuspiciousdefectareaislocatedwhichrequiresfurtherinvestigation,theentirescan
proceduremayhavetoberepeated.Howeverautomaticscanninghasanotheradvantageovermanual
probepositioninginthatitmayenable'online'weldscanningsincehightemperaturesshouldnotoverly
affectitsuse.
Howeverthereareproblemsassociatedwithpulseechotestingofincompleteweldgeometries.With
componentsofpeculiargeometries,evensophisticatedthreedimensionalplotsmaynotbesufficientto
detectalldefectsthecharacteristic'sparseness'ofaplotwillbemaskedwithlargedeflectionsfromthe
componentgeometries,manyofwhichmayregisterover100%FSH.Theseareaswillservetomerge
withdefectsandrenderthetaskofflawdetectionverydifficult.Flawclassificationislikewisemade
moredifficult.Fig(2)forexampleillustratesasimilarscenariotothatshowninFig.1onlythistime
theweldisonly50%complete.Confusingsignalswillbeprevalentduetotheopenfacesoftheunfused
weldsurfaces.ABscantakenwitha60degreeprobeofa50%completeweldwithalargedefect.The
largemassofsignalsfromtheopenfaceoftheplateandthecornerofweldregionservetoalmost
completelyobscurethedefectinanycaseitisalmostimpossibletodiscerntheactualextentofthe
defectarea,evenifitisactuallydetected.Othermethodsmustthereforebesoughttodealwiththese
problems.

Timeofflightdiffraction(TOFD)technique
ArelativelyrecentultrasonicNDTtechnique,istheTimeofFlightDiffraction(TOFD)method,which
wasfirstdescribedbySilk(1977).Thismethodreliesonthediffractionofultrasonicenergiesfrom
'corners'and'ends'ofinternalstructures(primarilydefects)inacomponentundertest.Thisisincontrast
toconventionalpulseechomethods,describedabove,whichrelyondirectlyreflectedsignals.The
typicalapparatusforaTOFDweldexaminationisshowninFig.3.Usuallyatwoprobe(onetransmitter
onereceiver)arrangementisusedthechosentransmitterproducingarelativelywidebeamspreadto
maximisetheextentofthescan.Thetwoprobesarealignedgeometricallyeithersideoftheweldandan
Ascantakenatsequentialpositionsalongthelengthofthebead.Thetimetakentoscanalengthofweld
isthereforeveryshortsincenorasterscanningateachpositionisnecessary.

AschematicofanAscanforthearrangementshowninFig.3isgiveninFig.4.Thisshowsthe
detectionoffourmainsignals(1)thesurfaceorlateralsignalwhichtravelsalongthesurfaceofthe
componentandhastheshortestarrivaltime(2)thetoptipofthedefect(3)thebottomtipofthedefect
and(4)thebackwallecho,whichhasthelongesttransittime.AlsopresentintheAscanarereflections
frommodeconvertedsignalswhichhaveaslowerspeedofpropagationthroughthespecimenandhence
alongertransittime.AlthoughmodeconvertedsignalsarenotusuallyexaminedinTOFDinspection,
theycanoftenduplicatethemainbodyoftheAscan.IfbothtipsofadefectcanberesolvedintheA
scanthentheactualdepthand'throughwall'thicknessoftheflawcanbeaccuratelycalculated(Silk,
1977Carter1984).Thiswasthemainthrustbehindthedevelopmentofthetechnique,sinceother
methods,suchasstandardpulseechotechniquescouldoftenlocatebutnotsizedefectsaccurately.

TOFDhassomedrawbacksinthatitisraretoachieveclearsignalsintheAscansduetothenatureof
theweakdiffractedsignals,coupledwithnoise,andinterferencefromsignalsderivedfromverysmall
poresornonuniformityintheweldregion.However,oneofthereasonswhyTOFDexaminationhas
attainedwidespreaduseistheassociateddataacquisitionandpresentationmethodologiesproposedvery
earlyinitsdevelopmentbySilk(1977).EachAscanisdigitisedasitiscollectedbythereceiver,toa
resolutionof8bitsthesignalisnotrectifiedandinsteadismappeddirectlytothe256possiblegrey

levels.AseriesofAscans,forasectionofweld,isthenpresentedinBorDscanfashionasshownin
theexamplescaninFig.5.Inthisexamplescanwecanclearlyseethelateralwave(atthetop),andthe
massofbackwallechoes(atthebottom).Alsopresentbetweenthesesignalsisadefect,D,aroundwhich
arepeculiardowncurvedarcswhicharetypicalofTOFDsignatures.Thearcsarecausedbythe
interactionofthedefectwiththewidebeamspreadwhentheprobesarenotexactlyinlinewiththeflaw
hencethesignalbecomesprogressivelystrongernearthecentreofthearcastheactualpositionofthe
defectreachesthecentreoftheultrasonicbeam.Sincethesignalsassociatedwiththepresenceofdefects
areoftenquitecharacteristicwhenshowninSilk'sDscanplot,theTOFDtechniqueisalsonow
extensivelyemployedfordefectdetectionaswellassizing(Silk,1984),andhasalsorecentlybeenused
tocompletelyreplaceotherNDTmethods(Verkooijen,1995)
TOFDscansarecommonlyundertakenwithautomatedscanningmanipulators,andtheobtainedscans
canbecomparedwithpulseechoscansusingthesameapparatus.Currentlythereareseveralcommercial
digitalultrasonicsuchsystemsthatcanacquiredatainthisway(includingtheconfusinglynamedMicro
pulseandMicroplussystems)Allofthesesystemsnowemployvariousdataenhancementalgorithms
suchaspseudocolouring,geometricstretching,andmostnotablytheSyntheticApertureFocusing
Technique(SAFT)whichcanbeusedtoeliminatebeamspreadeffectsandthusimprovedefect
characterisation.

AutomaticinterpretationofTOFDdataforonlineweldinspection
TOFDtestingisoneofthemostpromisingNDTtechniquesforonlineweldinspectionsincethemethod
isvirtuallyimmunetomanyoftheproblemsassociatedwithUTtestingofweldswithincomplete
geometries.Thisisbecausetheprobeseparationandanglecanbealteredtoallowforalmostfullweld
regioncoveragewhilstavoidingsignalsfromeithertheopenparentmetalfacesortheweld/metalapex.
Furthermore,iftheinterpretationtaskcanbeautomatedtheninlineprocesscontrolofthewelding
operationisaforeseeableentity.

InadditionTOFDscansare,bytheirnature,continuousandcontainallinformationrecordedinthe
individualAscansthisservestoeasethedetectionofdefectsbyahumanobserversinceheisableto
employintuitivetwodimensionalreasoningtolocatetheextentofadefectarea.Ifinterpretationistobe
automatedthenthemanualinterpretationproceduremustbemimickedbycomputeralgorithms.Theuse
ofimageprocessingandpatternrecognitiontechnologiesinautomaticultrasonicinspectionarenotnew
(Windsor,1995)thoughthetaskhas.inmostcases,provedextremelydifficultasisindicatedbythe
plethoraofdifferentmethodswhichhavebeendescribedintheliterature(Chen,1994McNaband
Dunlop,1995).Themajorityofsuchworkshavebeenappliedtopulseechotestingandmostlyfordefect
classificationwithrelativelylittleworkdirectedatdefectdetectionandTOFDanalysis.AnECfunded

BRITEEURAMprojectNDTMethodsforFlawDetectionDuringWeldingiscurrentlyaddressinga
numberofautomaticdefectdetectionmethodsincludingmethodsforTOFDinspection.

TheapproachadoptedfortheanalysisofTOFDimagesistofirstdetect'interesting'areasofthescan
i.e.flawsandcomponentsignals.Theinitialprobleminlocatingthesesignalshoweverliesinthe
removalofbackgroundsignalsduetonoiseandsmallinsignificantartefacts,suchasairpores.General
segmentationalgorithmssuchasedgedetection,orthresholdingcannotcopewiththevaryinggrey
intensitiestypicalofTOFDscans.Amoreintelligentmethod,usingtextureanalysisandaneuralnetwork
hasthereforebeendevelopedseeFig.6.TheeffectofprocessingwiththismethodisillustratedinFig.
7btherawscanshowninFig.7aissegmentedpixelbypixeltheneuralnetworkintothreeclassesof
interestwithhighinterestareaslabelledwhite.Componentsignals(i.e.backwallandlateralwaves)are
thenremovedbyrulebasedmethodstorevealdefectareas.Thesecanbehighlightedtoaidmanual
inspection,asisshowninFig.7c,orbefurtheranalysedtogiveanindicationofthedefecttype,sizeand
significance.Thesemethodshaveprovensuccessfulinautomaticallyidentifyingmostrepresentative
defecttypesinweldsofvariousstagesofcompletionandathightemperatures.

Summary
PulseechotestingisthemostcommonlyusedultrasonicmethodinweldNDTandcanbemechanisedto
producehighqualitythreedimensionalinformationonascannedcomponent.However,theinterpretation
ofpulseechoscansisfrequentlyfraughtwithdifficulty,particularlysowhenpeculiargeometries(such
asincompletewelds)arepresent,orifthecomponentisverylarge.Therefore,analternativetechniques,
suchasTOFD,areofinterestforpeculiarordemandingapplicationssuchasonlineweldinspection.
TheBRITEEURAMprojectNDTMethodsforFlawDetectionDuringWelding*iscurrently
addressingmethodsfortheimplementationofonlineautomaticTOFDinspection.Novelpattern
recognitionmethodshavebeendevelopedtodistinguishbetweendefectandnondefectsignalsinthe
highlyvaryingbackgroundofthescan.Oncethisisachievedthentheextentandpositionofthedefect
maybeautomaticallymeasuredtogiveanindicationofitsseverity.Thesetechniquesarefast,robustand
capableofdetectingdefectsinweldsofvariousstagesofcompletionandathightemperatures.

References
1. Carter,P.,"Experiencewiththetimeofflightdiffractiontechniqueandanaccompanyingportable
andversatileultrasonicdigitalrecordingsystem",Brit.J.ofNDT,Sept.1984,pp354361.

2. Chen,C.H.,"Patternrecognitioninnondestructiveevaluationofmaterials",intheHandbookof
PatternRecognitionandComputerVision,ed.byChen,PauandWang,WorldScientific,1992.
3. McNab,A.,andDunlop,I.,"Areviewofartificialintelligenceappliedtoultrasonicdefect
evaluation",INSIGHT,vol.37(1),pp.1116,1995.
4. Silk,M.G.,"Sizingcracklikedefectsbyultrasonicmeans",inResearchTechniquesinNon
destructiveTesting,vol.III,ed.byR.S.Sharpe,AcademicPress,1977.
5. Silk,M.G.,"Theuseofdiffractionbasedtimeofflightmeasurementstolocateandsizedefects",
Brit.J.ofNDT,vol.26,1984,pp208213.
6. Verkooijen,J.,"TOFDusedtoreplaceradiography",INSIGHT,vol.37(6),pp.433435,June
1995.
7. Webb,S.,"Inthebeginning",inThePhysicsofMedicalImaging,editedbyWebb,S.,IOP
Publishing,London,1988,pp.719.
8. Wild,J.J.,"Theuseofultrasonicpulsesforthemeasurementofbiologicaltissueandthedetection
oftissuedensitychanges",Surgery,vol.27,pp.183188,1950.
9. Windsor,C.G.,"Canwetrainacomputertobeaskilledinspector?",INSIGHT,vol.37(1),pp.
3649,January1995.

Author
|Frontpage||Toptothispage|
RolfDiederichs1.April.1996,info@ndt.net
/DB:Article/AU:Lawson_S/IN:SurreyUni/CN:UK/CT:UT/CT:TOFD/CT:weld/ED:199604