基于神经网络的异步电机故障诊断研究

太原理工大学硕士学位论文基于神经网络的异步电机故障诊断研究姓名：兰宇飞申请学位级别：硕士专业：@指导教师：田慕琴20100401IMatlabBPBPBPPSOBPPSOIIPSOBPPSOBPPSOIIIRESEARCHONFAULTDIAGNOSISOFASYNCHRONOUSMOTORBASEDONNEURALNETWORKABSTRACTAsynchronousmotoriswidelyappliedandplaysanimportantroleinindustrialandagriculturalproductionasoneofdrivefacilities.Withtherapiddevelopmentofmodernindustrialsystemunitcapacityofmotorcontinuouslyincreasesandtheloaditdrivesgrowsmoreandmorecomplex.Motorfaultsnotonlydamagemotoritselfinsomeseriouscasesbutalsostopmotorinasuddenandmaketheproductionlinecollapsewhichcausesgreateconomiclosses.Moreoveritmayevenposeseriouslythreatentopersonalsafety.Thereforeitisofgreatsignificancetopreciselyfindanddiagnosefaultsofrunningmotorsintheindustrialproduction.Thethesisanalysesthestatusquoandtheexistentialproblemsoffaultdiagnosisofmotorshomeandabroadatpresentintensivelystudiesfaultcharacteristicsandthemechanisticofseveralcommonfaultsincludinginductionmotorsstatorfaultbrokenrotorbarsfaultbearingfaultinsulationfaultetc.BasedondetailedanalysisofFouriertransformtheorythepaperconcludestheweakpointswhenitisusedtoexaminenon-stationarysignalsproposesanextractionmethodoffaultsignalscharacteristicsbaseduponwavelettransformusingenergyvalueextractedintermsofwaveletasfaultIVfeaturevector.Withthedevelopmentofneuralnetworktechnologyneuralnetworkhasalreadybeenextensivelyusedinthefieldoffaultdiagnosisofmotors.Firstlythepaperutilizesthree-layer-neuralnetworktrainingfromthestandardBPalgorithminthesimulationplatformofMATLABtodiagnosethemotorfaults.Thenitproposestwoimprovementmethodswhenthetestshowslessimpressiveresults.OneisincreasingmomentumtermtostandardBPalgorithminordertobeoutoflocalminimawheniterationoccursstagnationwhicheffectivelysolvestheproblemofBPalgorithmbeingeasytofallintolocalminimaandmakesitconvergenttoabsoluteminimum.AnotheristoexaminefaultsofasynchronousMotorsintermsofBPneuralnetworkbasedonParticleSwarmOptimizationPSO.Tobemorespecificitistoimprovethevalueofweightandthresholdofinputlayer-hiddenlayerandhiddenlayer-outputlayerofneuralnetworkbyPSO.ItisconfirmedthatPSOcouldovercomeintrinsicshortcomingsofBPneuralnetworkincludinglowlearningefficiencyslowconvergenceratebeingeasytofallintolocalminimaetc.FinallythepaperextractsthefaultcharacteristicsofmotorvibrationandelectriccurrentsignalsconcludescorrespondingfaultsamplestrainswhichbyBPneuralnetworkbasedonPSOandappliesthetrainednetworktothefaultdiagnosisofmotors.Aftertestingtheactualfaultdatatheprecisionofthisdiagnosismethodcouldbesubstantiallyimprovedsoastosolvethefaultdiagnosisproblemsofmotorsandmorepreciselyandintelligently.ThismethodVillustratesagoodfeasibilityandabroadprospectofapplication.KEYWORDS:asynchronousmotorswaveletanalysisfaultdiagnosisbpneuralnetworkparticleswarmoptimization11.1[1][2]200012%19952001988200Mw814()21.2(TechnicalDiagnostics)(EngineeringDiagnostics)20601965207020801987TavnerP.J.PenmanJ.[3]70-803(FastFourierTransformFFT)PHILIPSRMS700ENTKEIRDMPULSEGEMULTILNI80Mo-YuenChowS.O.Yee604[4]PertiPetri[5]BPRBFMCSA(MotorCurrentSignatureAnalysis)[6][7]51.3(1)(2)(3)(4)(5)BPBPBPPSOBPBPBP(6)6(7)MATLAB72.11n1nn2-12-1Fig.2-1rotationschematicdiagramofasynchronousmotor1nn8s1ns[8]1100%nnsn−=×2-12.22.2.1123940f002fsf+Hargis[9]P0fm101111sin()mKNItPωθ=−2-21K1N1Iωθrtφθω=−2-3rω(P=1)1111sin[()]rmKNItωωφ=−−2-42222sin[()]rmKNItωωφ=−−2-52K2N2Isin2φ2222sin[()]sin2rmKNItωωφφ=−−2-62-3rsωωω−=2-72221{cos[(32)3]cos[(12)]}2KNImststωθωθ=−−−−−2-82-83tω3θ2sω2sω0(12)sf−11(,)DQii2.2.21222Q=1P≠2-2122-2Fig.2-2rotorcurrentdistributioninthecaseofnofault2-32-345Fig.2-3rotorcurrentdistributioninthecaseofbrokenNo.4&No.5rotorbars[10]2-4132-41234Fig.2-4rotorcurrentpharosdiagramsinthecaseofbroken1345conductingbar(s)11brokenbar23brokenbars34brokenbars45brokenbars22-52-62QP22QP()142-5(5)12Fig.2-5rotorcurrentinthecaseofanendringfracture(No.5bar)(1)Conductingbarcurrent¤tofendringfractureside(2)conductingbarcurrent¤tofunfaultedendringside2-6()(5)Fig.2-6rotorcurrentofendringfracturebefore&after(supposeNo.5barbroken)22QP22QP315()2-9021[()()]eccdsffkQNnpω−=±±2-9dNnωk1,2,3k=⋅⋅⋅2-1001[1()]eccsffkp−=±[11]2-102.2.3162-12-1Tab.2-1mainfaultsandsymptomsofmotors⁄173.13.1.11183-11234XYZ7.5Kw1432DEWETRON2010XYZ3-1Fig.3-1Signalacquisitionofmotorfaults3-13-1Tab.3-1accelerometerdirectionandchannelsconfiguration1x-ch10y-ch2z-ch92x-15y-ch11z-ch13x-ch4y-ch5z-ch34x-ch6y-ch7z-ch82Hz193.1.23553-23-2Tab.3-2vibrationandcurrentcharacteristicfrequencycorrespondingtovariousfaultsf0f()()0200211rddffskQnvfpfQnspnω±−±±±−±0kf20()()0000112rkspsffkfffsf−±±±±b()012sf±0000246ffff()()021kmsf±−0.51cosrdZfDβ+0.51cosrdZfDβ−220.51cosrDdfdDβ−(),,......23,2,3......rrrrrrrrffffkfffkf0bfkf±0.51cosrdfDβ−rf0dn=11,3,5v=±±±⋅⋅⋅bfb0,1,2,3m=⋅⋅⋅ZdDβ213.2FFT224.14.1.11870FourierFourierFourierFourierFourier()xt()Xω4-11()()2()()ititxtXedXxtedtωωωωπω∞−∞∞−−∞==∫∫4-123()xtNyqrtist[12]1()ft234.1.20-512FFT4-14-10-512Fig.4-1time-domainwaveformoforiginalsignals(0-512samplepoint)FFT4-2244-2Fig.4-2frequency-domainwaveformof