搜索
-1-()()20-2-BootstrapBootstrapBootstrap-3--1-Subject:InformationextractioninmachinerydiagnosisandconditionpredictionSpecialty:InstrumentscienceandtechnologyName:ZhangHaijun(Signature)Instructor:QuLiangsheng(Signature)AbstractOverthepasttwodecades,thetechnologyofmachinerydiagnosishasachievedsignificantprogress:firstly,thistechnologycanbringindustrygreateconomicbenefits.Manymeasures,suchasraisingtheutilizationratioofequipment,reducingtheshut-downtime,prolongingtheserviceperiodandchangingtheregularforcedmaintenanceintopredictivemaintenance,etc.,havebeenincreasinglytakenastheeffectivemeasures.Secondly,theprogressofmachinerydiagnosiscomesfromtheincorporationofbasicdisciplineswiththefrontiertechnology.Machinerydiagnosisisabranchofmechanology.Itsessenceisthepatternrecognitionofmachineoperatingcondition.However,inthepastdecades,itreliedmainlyontheexperienceandintuitionofmaintenancepersonnel.Therefore,onceitincorporateswiththeinformationscience,systemscience,artificialintelligenceandcomputerscience,thesefrontiertechnologieswillcertainlypenetrateintothetraditionaldiagnosistechnologyandmakeitdevelopinparallelwiththeformer.However,thereexistsmuchdeficiencyinthedevelopmentofmachinerydiagnosis:firstly,comparedwithrotatingmachinery,toolittleattentionhasbeenpaidtoreciprocatingmachinery;secondly,peopleprefertofocustheirresearchonfaultdiagnosticnetworksratherthanonthemethodsofdiagnosisandprediction.Thisdissertationconcentratesonthemethodstoimprovethequalityoffaultdiagnosisandconditionprediction.Themainworkincludes:theinvestigationaboutdiagnostibility,thedecompositionofdiagnosticinformation,extractingdiagnosticfeaturesandsettingupfaultdiscriminant,-2-thepredictabilityofmachinerybehaviorandpredictivetechnology.Diagnosticcertaintyreliesnotonlyonthediagnostibilityofthemachineitself,butalsoonthefourstepsofdiagnosticprocess:signalsampling,signalanalysis,featureextractionanddiagnosticdecision.Themachinerydiagnostibilityrelatestothecomplexityofmachine,thequalityofdesignandproduction,andtheoperatingenvironmentetc.Thefaultentropymayrealisticallydepictmachinerydiagnostibility.However,somemethodsofinformationfusionanddecompositionsuchas:Holo-spectrum,decompositionorbitarethepowerfultoolstoenhancethemachinerydiagnostibility.Theoperatinginformationaboutequipmentlikevibration,acousticsandtemperatureetc.aretheoriginalevidenceinfaultdiagnosis.So,thequalityofthoseinformationisveryimportant.However,signalmixingandnoisesdisturbancealwaysexistinproductionenvironment.Therefore,sometechniquessuchascontinuouswavelet,principalcomponentanalysisanddigitalfilteringareusuallyusedtoimprovethesignalquality.Inchapter3,BlindSourceSeparationisadoptedtodecomposethemixeddiagnosticsignals.Thismethodcangreatlyimprovethequalityofthediagnosticinformation,bothsimulationandpracticeproveditshighefficiency.Featureextractionanddiscriminationfunctionarethekeystepsinmachinerydiagnosis.Withoutdoubt,thosefeaturesandfunctionsareverysensitivetothechangingofequipmentcondition,andlessaffectedbytherunningenvironment.Whereas,undermanycircumstances,itisdifficulttogetenoughdiagnosticsamplesbecausethemonitoringprocessesarealwaysnon-stationary,asaresult,gooddiagnosticfeaturesanddiscriminationfunctionsaredifficulttoobtain.Chapter4and5focusonBootstrapmethod,basedonwhichwecanextractdiagnosticfeaturesandconstructdiscriminationfunctionsundertheconditionthatonlyafewsamplesareavailable.Whatismore,Bootstrapmethodmayalsoquantitativelyevaluatetherecognitionabilityofthefeaturesanddiscriminationfunctions.ItistestifiedthatBootstrapmethodispracticalinmachinerydiagnosis.Conditionpredictionisoneofmostimportantissuesinmachinerydiagnosis.Theessentialobjectiveofmachinerydiagnosisistoachieve-3-predictivemaintenance.Undoubtedly,inordertorealizethisgoal,itissignificanttobeawareofthefutureconditionofequipment.Chapter6concentratesonthestudyofmachinerybehaviorforecasting.Itincludestwoparts:ontheonehand,thepredictabilityoftimeseriesisdiscussed,someeffectivemethodsincludingrecurrenceplots,Lyapunovexponent,correlationstatisticsandredundancyareusedtodescribethepredictability;ontheotherhand,somepredictivemethodsfrequentlyusedsuchastrendanalysis,modelbasedpredictionandartificialneuralnetworkarediscussed.Atlast,conclusionsofthedissertationaresummarizedinChapter7.Keywords:FaultdiagnosisBootstrapBlindsourceseparationPredictionFeatureextractionDissertation:ApplicationFundamentalsI11.11.21.2.11.2.21.2.31.2.41.31.3.11.3.21.3.31.422.12.22.32.3.12.3.22.42.4.12.4.22.4.32.533.13.2II3.2.13.2.23.2.33.33.3.13.3.23.43.544.14.24.2.14.2.24.2.34.2.44.34.3.14.3.24.44.4.14.4.24.54.5.14.5.24.5.34.655.15.2III5.35.45.4.15.4.25.4.35.55.5.15.5.25.65.766.16.26.36.3.16.3.26.3.36.46.4.16.4.26.4.36.5712019852198851985199019761985200030CEGB556075%2550%CEGB293Gravelines40054021988304301967MFPG20803MEMS901021(VirtualInstrument)NI4(TheSoftwareisTheInstrument)NI(NationalInstrument)HP(HewlettPackard)TektronixRacal10%RotVIEWLabWav5─8061.11.178910111948ShannonAn12,,,nAAALA()()()21logniiiHAPAPA==-∑()11niiPA==∑12()HAA()iPAi(2.1)sA()1sPA=()0HA=nn12,,,