X射线探测焊缝及机械损伤方法概述----中英文翻译.docx
Originaltext:X-RAYDETECTSWELDSANDMECHANICALSTRUCTUREDAMAGEMATHODS,SUMMARIZEThemovingsmallobjectdetectioninimageisalwaysadifficultprobleminfieldofimageprocessing,whichappliesinmanyfields,suchasindustrialdetectionandmedicaldetection.Thedefects,suchasblowholesandincompletepenetration,occasionallyappearintheweldingprocess.Thesedefectscanaffectthequalityandthesecurityofproducts.Therefore,defectsdetectioninweldingseamisextremelyimportant.Now,theon-linedetectionofdefectsintheweldisstilldonebyhumaninterpreter.However,thisprocessissubjective,inconsistent,laborintensiveandfatigueofinterpreter.Itisdesirabletofindaneffectiveautomaticdefectsdetectionmethodtoassisthumaninterpreterinevaluatingthequalityofweldandtomaketheon-linedetectionobjective,standardandintelligent.Ourresearchisbasedonthis.Wehavestudiedtheautomaticdefectsdetectionintheweldseamandmainlydonethefollowingresearch:(1)Thereismuchredundantbackgroundinformationforthedefectsdetectionintheimage.Thereforeweuseanautomaticallyabstractingmethodofweldareabasedontheauto-adaptedthresholdsegmentation.Thismethodcanreducethecomputationandincreasetheprecision.(2)TheSUSANalgorithmhasgoodanti-noiseability,whichcanrecognizetheimageedgeverywell.SowehavestudiedadefectsdetectionmethodbasedonSUSANalgorithm,whichassociatedwiththemorphologyoperation.Theresultsindicatethatthismethodiseffective.(3)Waveletanalysismethodhasaverygoodlocalizationcharacteristic,whichcanfocusonthearbitrarydetailoftheanalyzedobject.Therefore,Westudiedamethodusingwaveletdecompositiontogettheshapeandpositioninformationofthedefects.Thenweusethewienerfilterandmorphologymethodtocompletethedetection.Theautomaticflawdetectionofweldedtubesisoneofthemostimportantstepstoensurethequalityofthetubes.Nondestructiveinspectiononweldingseamoftubeisrequiredinthetubeproduction,andrealtimeX-Rayradiographyinspectionisaneffectivemeans.Alongwithcontinuousimprovementoftheproductiveratio,thedemandfortheautomaticinspectiontotheweldingseambecomesmoreandmorepressing,soimplementationoftheautomaticinspectionpossessesimportantsignificanceonboththeoryandreality.Wavelettransformisapowerfultoolinthesignalandimageprocessing,anditsfundamentaltheoryhasbeenformed.Fromtheviewofengineeringapplications,however,thewavelettransformisstillintheelementarystage,thefurtherResearchesarerequiredforthepracticaluses.Inthisthesis,Weconcentratemainlyonusingwaveletanalysisforweldingseamimageprocessingandrecognition,andsomerelatedtechniquesaredeveloped.Forconstructingweldingseampositioninganddetectioncontrolsystem,themultiplecomputersconfigurationforweldseamimagerecognitionisproposed.ThesystemadoptsthearchitectureinwhichmultipleCPUsprocessparallelsunderthecontrolofthemasterIPCcomputer.Thesystemcanperformstoringtheweldseamimages,positioning,flawsrecognizingandqualityprejudging.TheWatch-Doginterfacecardissuccessfullydeveloped;itcanimprovethesystemreliabilitybyredundanciestechniqueofsavingbreakpointdataandrestoringthem.ThehardwaresupportingthesystemmakesuseofthehighspeeddigitalsignalprocessorTMS320C30fromTaxaxInstrumentsCompany.Theframegrabbercancapture25framesofweldingseamimagepersecondcontinuouslyandmakeitpossibletofulfilltherealtimeweldingseamimageprocessingAndrecognition.TheonekindofimprovedFWT(FastWaveletTransform)algorithmforafinitesequenceisproposedafterstudyingtheoryofmustiersolutionanalysisandanalyzingtechnicalcharacteristicsofDSP.TheimplementationoftheperiodicextensionoftheFWTonDSPisdescribedindetailandthecorrespondingFWTassemblycodeisdescribedfortheDSPTMS320C3Xseries.Thisdissertationsuggestsschemeofimagedemonizingbasedontwo-dimensionaldiscretewavelettransform.Thedemonizingalgorithmisdescribedwithsomeoperators.Bythresholdthewavelettransformcoefficients,ofnoisyimages,theoriginalimagecanbereconstructedcorrectly.Differentthresholdselectionsandthresholdmethodsarediscussed.Thenewrobustlocalthresholdschemeisproposed.Quantifyingtheperformanceofimagedemonizingschemesbyusingthemeansquareerror,theperformanceoftherobustlocalthresholdschemeisdemonstratedandiscomparedwiththeuniversalthresholdscheme.Theexperimentshowsthatimagedemonizingusingtherobustlocalthresholdperformsbetterthanthatusingtheuniversalthreshold.Inordertoimprovetheaccuracyandtherealtimeperformanceofedgedetection,amethodneedtobefoundtomatchthedetectionoflowcontrastblurredweldingseamimage.Thisdissertationanalyzedthemainsourcesofnoiseaswellasthedifferentcharacteristicsofnoiseandsignalunderwavelettransform,andproposedaMoultriesolutionedgedetectionmethodbasedonwavelettransform.Theexperimentalresultsshowtheeffectofthisalgorithmisadvantageousoverthatoftraditionaledgedetectionalgorithm.Thegeometricalrelationofellipticimagingisstudiedforweldingseamimageofthebuttweldsinstraighttubes.Theregionmodelofweldingseamimageisproposed,Itfurnishesaevidencetheorytofurtherprocesstoweldingseamimage.Combiningwiththeregionmodel,amodel-basedadaptivetargetsegmentationalgorithmisproposed.OnebasisofthealgorithmisOtsu,sdiscriminatescriterion.Theadaptivetargetsegmentationofweldingseamimageisrealized.Theeffecto