外文翻譯-- 基于小波的神經(jīng)探測器分析電機的低頻振動信號 （英文）

1、AppliedSoftComputing13(2013)2683–2691ContentslistsavailableatSciVerseScienceDirectAppliedSoftComputingjournalhomepage:www.elsevier.com/locate/asocWaveletbasedNeuro-Detectorforlowfrequenciesofvibrationsignalsinelectric mo

2、torsDuyguBayram ?, SerhatS ¸ ekerIstanbulTechnicalUniversity,ElectricalEngineeringDepartment,34469Istanbul,Turkeya r t i c l e i nf oArticlehistory:Received23September2010Receivedinrevisedform1September2012

3、Accepted24November2012Availableonline11December2012Keywords:AutoAssociativeNeuralNetworkMultiResolutionWaveletTransformElectricmotorAgingVibrationBearingdamageab s t r a c tThis study presentsa Waveletbased Neur

4、o-Detectorapproachemployedto detect the aging indicationsofan electricmotor. Analysisofthe aging indications,which can be seen in the low frequencyregion,isperformedusing vibrationsignals.Morespecifically

5、,two vibrationsignalsare observedfor healthyandfaulty (aged)cases whichare measuredfrom the sameelectricmotor.Multi ResolutionWaveletAnalysis(MRWA)is appliedin order to obtain low and highfrequencybands of

6、 the vibrationsignals.Thusfor detectingthe aging propertiesin the spectra, the PowerSpectralDensity (PSD) of the subbandforthe healthycase is used to train an Auto AssociativeNeural Network(AANN).The

7、PSD amplitudes,whichare computedfor the faulty case, are appliedto input nodesof the trained networkfor the re-callingprocessof AANN. Consequently,the simulationresults show that some spectralproperti

8、esdefinedin lowfrequencyregion are determinedthrough the error responseof AANN. Hence,somespecific frequenciesofthe bearingdamagerelated to the aging processare detectedand identified.©2012ElsevierB.V.

9、 All rightsreserved.1.IntroductionInductionmotoristhemostpopularelectricmotortypebecauseofitssimpleconstructionandlonglifewithoutmaintenanceneed.Intheliterature,therearelotsofstudiesrelatedtothedevelopmentandresearchofi

10、nductionmotors.Theperformancestudiesaregen-erallybasedonelectrical,mechanicalandthermalparameterslikevoltage,current,torque,temperature,noise,etc.Theseparametersreflectveryimportanthintsabouttheperformanceandworkingcondi

11、tionoftheelectricmotors.Theprognosticanddiagnosticstudiesoninductionmotorsareanotherimportantaspectinprovidingtheoperationalcontinuityoftheindustrialprocess[1–3].Andalso,conditionmonitoringstud-iesgainimportanceinearlyde

12、tectionoffailureinsomecriticalsystemslikenuclearpowerplantsandpetrochemicalprocesses[4,5].Inthissense,therearesomanystudiesaimingfindingoutthesourcesofdegradationsoninductionmotors.Mechanicalfailuresrelatedtothemanufactu

13、ringhavethegreatestmajorityintermsofAbbreviations:MRWA,MultiResolutionWaveletAnalysis;PSD,PowerSpectralDensity;AANN,AutoAssociativeNeuralNetwork;DWT,DiscreteWaveletTrans-form;BPF,BallPassFrequency;EDM,ElectricalDischarge

14、Machining. ? Correspondingauthorat:IstanbulTechnicalUniversity,ElectricalandElectron-icsFaculty,ElectricalEngineeringDepartment,34469Maslak,Istanbul,Turkey.Tel.:+902122856736;fax:+902122856700.E-mailaddresses:bayramd@itu

15、.edu.tr (D.Bayram),sekers@itu.edu.tr(S.S ¸ eker).theencounteredfaults[1]. Thesearerespectivelybearing,balanceandalignmentdefects[6–9].Inthismanner,somesignalprocessingmethodsareusedtoextractthehiddenhintsandinforma

16、tionofthefaultsignatures[10–14]. Vibrationsignalsusuallyshowcrucialindicationsaboutagingoftheelectricmotorwhereaselectricalsignalscarryhintsonly.Becauseofthis,vibrationsignalsareusedinvariousstudiesintheliterature[13–22]

17、. Forexample,theycanbeusedtodefineatransferfunctionofagingprocessinelectricmotor[15].Intheliterature,somestatisticaltechniquesareappliedtothevibrationsignalsinordertodetectagingeffectsandhenceitisfiguredoutthatsomestati

18、sticalparameterschangebyaging[16].Andalso,itispointedoutsomecorrelationsbetweenvibration’sspectraldensityandagingdefects[17–20]. Motorvibrationsignalsarealsousedtoextractsomefeaturesofbearingagingbywavelettransforms.Fro

19、mthispointofview,majoreffectsofthebearingdamagearedetectedinthehighfrequencyband(between2and4kHz)[21,22].Theaimofthisstudyistoanalyzetheelectricmotorvibrationsignal’slowfrequencybandusingawaveletbasedneuro-detectorapproa

20、ch,inordertoextractitsagingeffectsanddefects.However,thereisnoencounteredstudywhichobservestheroleofthelowfrequencybandintheinvestigationoftheagingeffects.Thisisthemostimportantcontributionofthisstudyintermsofextractiono

21、ftheagingeffects,aswellasusedmethodology.Anexperimentalsetupisrealizedtoobtainthevibrationsignaltogetherwithelectricalsignals.Intheexperiment,thevibrationsignalsofa5HPelectricmotorarerecordedbytwoaccelerometers.Inorderto

22、arrangethedata,asignalconditionerisusedbeforestoringthedataonaregular1568-4946/$–seefrontmatter©2012ElsevierB.V.Allrightsreserved.http://dx.doi.org/10.1016/j.asoc.2012.11.019D.Bayram,S.S ¸ eker/AppliedSoftCompu

23、ting13(2013)2683–26912685Fig.2.MRWAatnthlevel.processtheoutputsofthehighfrequencyfiltersarenamedasdetailsDj,theoutputsofthelowfrequencyfiltersarenamedasapproximationsAj.Theschematicinterpretationofthen-levelmultiresoluti

24、onwaveletanalysisisshowninFig.2,herejisthedecompo-sitionlevel.ThesignalrepresentationcanbegivenasinEq.(4).s(t)=D1 +D2 +D3 +···+Dj +Aj (4)2.3.AutoAssociativeNeuralNetworksAutoAssociativeNeuralNetwork(AANN)i

25、safeedforward,fullyconnected,multilayerperceptronnetwork.ThedimensionsoftheinputandoutputlayersareequaltoeachotherintheAANNtopol-ogy.Also,thenumberofthehiddennodesforonehiddenlayerislessthanthenumberofinputandoutputnodes

26、.HencehiddenlayeroftheAANNiscalledas“bottleneck”,whichcompressestheinformationtoobtainacorrelationmodel[23].Sigmoidfunctionsareusedtoprovidethenonlinearityonthehiddenlayer.Theneedofnonlinearfunctionisindispensablebecause

27、AutoAssociativeNeuralNetworkisexpectedtoproduceitsinputattheoutputlayer[30–33].Atthetrainingprocess,thehiddenlayerappliesanencodingbycompressingtheinformationappliedtothenetworkastheinputsignal.Thenthenetworkdecodesbydec

28、ompressingthecarriedinformationtoproducethetargetsignal,whichisthesamewiththeinputsignal.ThestructureofAutoAssociativeNeuralNetworkleadstheusageareaofthenetworktodetectthefailurebycomparingtherealtimeoutputandtheoutputof

29、thenetwork.Forthisreason,AutoAssociativeNeuralNetworkisusedinsensorvalidation,detectionandmonitoringapplications[23,34,35].ThebasictopologyofanAutoAssociativeNeuralNetworkcanbegivenasshowninFig.3.Fig.3.Representativetopo

30、logyoftheAANN.AsatrainingalgorithmoftheAANN,BackPropagationalgo-rithmcanbeusedwhichisawell-knownalgorithmintherelatedliterature[36].3.ExperimentalstudyandmeasurementsystemAnexperimentalsetupisdesignedtoacquirethevibratio

31、nsig-nals.Twotypesofdataarecollectedusingthesetup,thesearevibrationdatainhealthycaseandthevibrationdatainfaultycase.Duringthisagingprocesstwotechniqueshavebeenexecutedonthemotor,theseareElectricalDischargeMachining(EDM)a

32、ndtheThermalAging,respectively.Asanaturalresultofthehighspeedoperationofelectricmotor,unexpectedshaftvoltageisinduced.Theshaftvoltagelevelincre-mentcancausebreakingdownofthegreasefilmbetweentherollingelementsandinner/out

33、erracesofthebearing.Hence,ran-domarcingoccursandthendischargecurrentsstarttoflowthroughtherollingelements.Asaresultofthisdischargemode,bearingflutingcomesinexistenceandthenbearingfaultsareencountered.TheElectricalDischar

34、geMachining(EDM)isasimulationofthisnaturalprocess.Forthispurpose,anexternalshaftvoltageandcur-rentisappliedtothemotorfor30minutesat30VACand27A.Theappliedexternalvoltageandcurrentcausethedischargesandbearingflutings.Alsot

35、hechemicalandthermalagingstepsareappliedtothemotorinordertoacceleratetheagingaftertheEDM.Chemicalandthermaleffectscausetothecorrosionandthermaldamageofthematerial.Intermsofthechemicalandthermalapplication,themotorisimmer

36、sedintothewatertankandputintotheovenat140 ?C.Inthissensetherearethreeaspectsoftheagingpro-cess.Theseareelectrical,chemicalandthermal.Thisprocedureisappliedforseventimes(7cycles)duringtheexperiment.Aftereachagingcycle,the

37、motorwasseparatedfromtheconnectionplatformbeforethechemicalandthermalapplication.However,foreachcycle,itwasfixedontheplatformagain,accordingtotheacceptablevibrationlevel.ThedetailsoftheexperimentcanbefoundinthePhDthesiso

38、fDr.A.S.ErbayandrelevantstudieswerementionedinthetextbygivenRefs.[37–39].Afterthesesteps,thevibrationdataaretakenbytheaccelero-metersplacedonthemotoraswellasthemotorcurrent/voltageinformationandthermaldata.However,electr

39、icalsignalsarewell-knowninformationsourcesfortrackingthemotorconditionintheliterature,thisinformationisfocusedonlinefrequency(fundamen-talfrequency)anditssidebands.Thermalvariationisalsoveryslowtoextractthemechanicaleffe

40、cts.Thereforethebestoneisthevibrationforthemechanicalandstructuralfaults.Forthisreason,inthestudyvibrationdataareused[37–39].Motortypeusedintheexperimentalstudyisaninductionmotorof5HP,three-phase,four-poles,designedfor60

41、Hz sup-plyfrequencyandwiththenominalspeed1742rpm(rotationperminute).InFig.4measurementanddataacquisitionsystemcanbeseen.Twoidenticalaccelerometersareusedtorecordthevibrationmeasurementsattheprocessend.Therefore,sensor#2

42、isonlyusedinthisstudy.Thesamplingfrequencyoftherecordeddatais12kHz.Besides,inordertoavoidfromthehighfrequencynoiseinterferedintherecordeddata,anantialiasingfilter,withcutofffrequencyat4kHz,isused.4.Applicationtovibration