%8f机和特征信息融合的水电机组故障诊断

华中科技大学博士学位论文基于改进支持向量机和特征信息融合的水电机组故障诊断姓名：彭兵申请学位级别：博士专业：水利水电工程指导教师：周建中20080608IRBFIID-SD-SSVMIIIABSTRACTWithhigh-speededgrowthofChineseeconomy,demandforelectricityincreasedandrequirementforreliableelectriciybecamestronger.Inthepasttwentyyears,waterresourceandhydropowerwasexploitedwidelyandcapacityofhydro-electricityplayedamoreandmoreimportantroleincapacityofpowersystem.DevelopmentofmanufacturetechnologypromotedhugemazationofHydropowerGeneratingUnit(HGU).SecurityandreliabilityofHGUbecameoneofthemostattentionalobjectsinresearchandapplicationofhydropowerscience.HGUisacomplexnonlinearsystemwithstrongcouplingbetweenpartions.ItisdifficulttocommentconditionofHGUwithsingleandlinearmethod.Byinformationfusion,allkindsofHGU’sinformationwithdifferentdistructionandfromdifferentresourceareintegrated.AndSupportVectorMachine(SVM)withkernelresultsHGU’snonlinearproblemwithmappinginputdataintofeaturespace.ItisvaluableforHGU’sfaultdiagnosisandcondition-basedmentanencetoapplythesemethods.Faultdiagnosismethodsaregenerallyintheconditionofstabilization.Vibrationsignalinstartingprocessissocomplexthatitisnotusedtofaultdiagnosisinstartingprocess.Infact,morefaultinformationcouldbegotinstartingprocess.Theideaoffaultditectionandanalysisinstartingprocessisexpoundedandthemethodoffaultdignosisbymulti-featureinformationfusionisproposedinthisthesis.HGUstartingprocessincludesrising-speedphase,rising-excitationphaseandrising-loadphase,whichmeetsconditionofHGUvibrationexperiments(altering-speedexperiment,altering-excitationexperimentandaltering-loadexperiment).Frenquencyfeature,temporalfeatureandspatialfeatureofvibrationcouldbegainedinHGUstartingprocess.Itcouldraiseveracityoffaultdiagnosiswithmulti-featureinformationfusion.ShaftorbitistheresultofHGUrunninginformationfusionintimedomainofdatalevel.Itisanopenline.LinearmomentinvarintisakindofgraphicsfeatureofshaftorbitandcouldreflectconditionofrunningHGU.However,discretelinearmomentinvariantischangeinzoom.Improvedlinearmomentinvariantispresentedinthisthesis.Experimentresultshowsimprovedlinearmomentsareinvariantinmovement,rotationandzoom.IVInordertomapsamplefrominputspacetofeaturesapceeffectively,geometryofRBFkernelusedinSVMisanalyzedanddatadependentmethodbasedonRiemanniangeometryisusedinthisthesis.ResultsofanalysisandexperimentsshowdelectingredandentsupportvectorsisonereasonforraisingfaultdiagnosisspeedandveracityofSVM.Inaddition,resultsofanalysisandexperimentsdemonstrateSVMhasstronggeneralizationcapabilityandlearningcapabilityfromlittlesampleset,whichmeetsconditionofHGUfaultdiagnosiswithalittlepriorknownledge.AfaultdiagnosisexampleexhibitimprovedSVMcouldraisespeedandveracityofHGUfaultdiagnosis.Fortheproblemofcombinationexplosionininforemationfusion,exchangetheoremandconjunctiontheoremincombinationprocessamongeevidencesareresearchedinthisthesis.ResultsofanalysisoncalculatecomplexofcombineamongevidencesshowinformationfusionbyD-Stheorycouldbereducedwithexchangetheoremandconjunctiontheorem.Resultsofexperimentsdemonstratethesetheorems.Inaddition,informationfusionisappliedinHGUfaultdiagnosis.Byfusingfrequencyfeatrureinformation,temporalfeatureinformationandspatialfeatureinformation,HGUisdiagnosedwithloweruncertaintyandhigherreliability.Atlast,aHGUconditionmonitoringandfaultdiagnosissystemisdesigned.Hardwareandsoftwaredesignofeachmoduleisfinishedbasedongeneraldesignofthesystem.Atthesametime,FaultDiagnosisExpertSubsystemandSVMAssistantDecisionSubsystemarebuilt.ThesystemcouldsatisfymanyrequirementsofHGUconditionmonitoringandfaultdiagnosis.Keywords:HydropowerGeneratingUnit,faultdiagnosis,startingprocess,shaftorbit,informationfusion,SupportVectorMachine,kernel_____111.121101GW338GW[1]20056.94KW6.08KW·h5.42KW2.47KW·h4.02KW1.75KW·h[2]200410256KW10826KW570KW3280KW·h1829250KW357004KW1860KW1260KW420KW420KW420KW360KW360KW300KW184KW175KW150KW70%18.923.35088.7%23.836.518.42.711.77.50.3[3]22007113000200012001000[4]“”“”2005201020152020“”1980KW4440KW8750KW10650KW2004“”1470KW“”510KW2970KW7280KW9180KW[3][5][6][7][8]3[9]SVM1.2[10]250-1100%51.2.11984BillintonR.4CentralElectricityGenerationBoardCEGBLowM.B.J.CEGB1985ClevelandJ.W.Motor-ColumbusHartmannO.WestinghouseKessingerJohnP.TechnologyforEnergyPietyK.R.7Intelli-TrendIBM-XTCEGBHorneB.E.1987LofeJ.J.FaveriGailR.1997LukasMalteSchadeH.8040REMAFEXRemoteMaintenanceforFacilityExploitationIBERDROLAEDPSCADAMIS5CMMS[11]Vibrocontrol4000VibroSystMAGMSZOOM2000VIBRO-METERVM600HydroVUSiemensScardEPRIIREQORE[12]1.2.2209031.2.2.110NW6231FFTTD4047CDMS[13]HSJ/HSJHSJ2004620002004PSTA3060199720051231.2.2.2[14][15]7[16][17][18]8(1)KumamaruKullbackKDI(KullbackInformationCriterion)[19][20](2)[21][22][23](3)[24][25][26]SVD1.2.2.39160MW1981520049232134866h7676h3642h60MW238,218,723MWh8,150,223MWh68,890.34[27][28]1.31.3.100101.3.2DFTFFTDFTFFT23[29][30][31][32]1.3.3FourierSFFTWignerVilleWVDWaveletFourier[33]WignerVille[33][34][35][36]111.3.490°xy[37]7[38][37][39][26][40]SVD[41][42]1.3.5[43][44][45][46][47](PCA)mnm(mn)[48]PCA[49][50][51]121.4(SupportVectorMachineSVM)(StatisticalLearningTheorySLT)1.4.1