搜索
华中科技大学博士学位论文产品表面缺陷在线检测方法研究及系统实现姓名：彭向前申请学位级别：博士专业：机械电子工程指导教师：陈幼平;周祖德20080510IBPRTTIIAbstractWiththeadvancementofmanufactureandtechniques,theproductqualitybecomesincreasinglyimportant,andtheonlinequalityinspectionsystembasedonmachinevisionhasbeenanimportantqualitycontrolmethod.Thispaperstudiestheonlinedefectsinspectiontheoryandalgorithmsforproductsurfacebasedondistributionmachinevision,andimplementstheonlinedefectsinspectionsytemofmodelproductandnon-modelproductusingthedefectsinspectionsofpressworkandfloatglassastheinstances.Firstly,tomeettherequirementoflargedimensionalityandhighprecisioninproductsurfaceinspection,aninspectionsystembasedondistributionmachinevisionisdesigned,andseveralinspectionsub-systemsareusedtocompletetheinspectiontaskcooperatively.Forsettlingimageregistrationofmodelimage,afastcontoursub-spacebasedimageregistrationalgorithmispresentedtoaimatthelargecomputingcomplexityoftraditionalimageregistration.Todetectimagecontourfastandcorrectly,adirectionbasedmulti-resolutionmorphologycontourdetectionalgorithmispresented,andaforecastbasedbi-thresholdingvaluecontoursegmentationisusedtothresholdingthecontourpixels.Anevaluativemethodofcontoursub-spaceisestablished,andtheimagesareregistratedthroughthecontoursub-spaceforimprovingtheefficiencyofimageregistration.Astofulfillingthedetectiontasks,differentdefectssegmentationmethodsaredesignedformodelimageandnon-modelimage,andafastdefectmergingmethodbasedonsequenceandjourneyspaceispresentedinordertorealizedefectclusteringfastly.Adefectsegmentationmethodbasedonimagesubtractionisdesignedtorealizedefectsegmentationformodelimage.Adefectsegmentationmethodbasedonthresholdingsurfaceispresentedtosegmentdefectsfornon-modelimage,andafastthresholdingsurfaceconstructionmethodbasedonthestatisticofgreylevelispresentedtosolvethegreatcomplexityandinveracityoftraditionalinterpolationmethod.Acontoursubtractionbasedfakedefectdiscriminationmethodisdesignedtoeliminatecontourfakedefectsformodelimageinspection,andatexturebasedfakedefectdiscriminationmethodispresentedtotideoverthefakedefectssuchasinsects,smearinessanddust.Basedontheanalysisofseveralpatternrecognitionmethods,allkindsofdefectscharactersarepickedup.AccordingtothedifferentclassificationrequirementsofdifferentIIIproductiondefects,arulebaseddefectsclassificationmethodisdesignedforpressworkdefects,andanimprovedneuralnetworkbaseddefectclassificationmethodisrealizedforfloatglassdefects.Thesynchronizationandnetworkcongestionofdistributionmachinevisionsystemarealsostudiedhere.Formelioratingthenetworkcongestionindistributionsystem,animprovedcongestioncontrollingmechanismbasedonthepredictableRTTisproposed.Futhermore,aself-diagnosisandself-recoverymechanismisrealizedfordistributionmachinevisionsystem.Atlast,basedontheaforementionedtheoryandalgorithms,twopatentproduction,suchasanonlinepressworkdefectinspectionsystembasedondistributionmachinevisionandanonlinefloatglassdefectinspectionsystembasedondistributionmachinevision,aredesignedandimplemented.Theyhavebeenappliedtomanufacturing,andachievedgoodeconomicandsocialbenefit.Keywords:DistributionMachineVisionSurfaceDefectsImageRegistrationDefectSegmentationFeaturePickedDefectClassification111.11.1.12005ABA2692004AA101B0120045006071-26200606W-071.1.210020052720.5mm0.1mm0.1mm0.15mm/0.10mm/0.25mm100%1001.1.3IC3”1.24(SMESSocietyofManufacturingEngineers)(RIARoboticIndustriesAssociation)[1]1.2.1[2]2060Roberts2070MarrMarrMarr2.53MarrMarrPLCPLCLEDCCS5KISAVMEPCIDataTranslationCorecoPCI[3]NICognexPROIMAGEBOBSTVMTHexsightDACTOKIMECKeyence[4]1.2.220201[5]/[6][7~8][9][10][11][12][13]6[14][14~17][18][19][20~21][22][23][24][25]VisuShrink[26]BayesBayesshrink[27][28][29]MihcakLAWMLHELAWMAP[30]2[31][32][33][34][35]RobertSobelPrewittKirschCanny[36~37]-[38][39~41]37[42][43][44][45]K[46][47~48][49~51][52][53]4[54][55][56]Hopfield[57][58][59]BPIntelNSPIPPMMXSSE[60]81.2.3[61~63]Kang[64]Isra[65~66]1.2.4[67]AOIAreaOfInterestingDALSACORECO9DafgårdBillyICNICognexPROIMAGEBOBSTVMTHexsightDACTOKIMECKeyence100[68]TOKIMECDVT/10[69]0.2mm[70~74]Isra[75]Dr.SchenkImageAutomationPilkington1.3(1)(2)4001000mm30006000mm0.3m(1)11200m/min40960.3mm/pixel100MByte(2)(3)(4)(5)1.48121234OTSU5BP6RTT78131.1127345861.11422.12.1.1/(C/S)2.1151NiNnN1NiNnN2.1162.1.2[76]42.22.12.3(1)[77]2.217LEDLED2.1(2)()2.32.1LEDLED18CamlinkPCI-X(3)2.1.319YN2.4YN2.62.5202.42.52.62.1.4cTiTpToT2.7coipTTTT++cpTTpcTT/oT1oT2oT3oT321ooooTTTT++=32oT3oT1oTpTcTcoipTTTT++iToT2.721copTTT+12.22.2.1(1)(2)AOI(3)(4)2.2.2222.12.2[78]2.823[79][80]2.8132424443555BP252.3CPU6C/S6RTTRTT6262.42733.13.1.13.1),(yxf),(yxg)','(),(yxgkyxf×=(3-1)⎩⎨⎧Δ++Δ+=Δ++Δ+=)sin)(cos)((')sin)(cos)(('qqlqqlxxyyyyyxxxyx(3-2)28kxlylxΔyΔq2||)','(),(||yxgkyxfMinD×-=(3-3)[81][82][83][84][85][86][87][88][89]3.1.229(a)(b)(c)(d)3.1.30.1mm0.2mm0.2mm301-22-324-51,1,0===yxllq(3-2)⎩⎨⎧Δ+=Δ+=yyyxxx''(3-3)1=k),(yxΔΔyyxxqypqxpΔΔ,xpxqxΔypyqyΔ2,2,4,4=-==-=yyxxqpqp3.23.2.1[90][91]3.1[92]31),(yxf),(yx2))/,/((],[),(yfxfGGyxfTyx∂∂∂∂==∇(3-3)3-3xGyGxy(f∇)),(yxf2/122)()(yxGGfmagf+=∇=∇(3-4))/arctan(),(yxGGyx=f(3-5