資源描述:
《A COMPARISON OF ARTIFICIAL NEURAL NETWORKS AND OTHER STATISTICAL METHODS FOR ROTATING MACHI》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫。
1、ACOMPARISONOFARTIFICIALNEURALNETWORKSANDOTHERSTATISTICALMETHODSFORROTATINGMACHINECONDITIONCLASSIFICATIONA.C.McCormickandA.K.NandiAbstractStatisticalestimatesofvibrationsignalssuchasthemeanandvariancecanprovideindicationoffaultsinrotatingmachinery.Usingtheseestimatesjoi
2、ntlycangiveamorerobustclassicationthanusingeachindividually.Articialneuralnetworkarchitecturesandsomestatisticalalgorithmsarecomparedwithemphasisontrainingrequirementsandreal-timeimplementationaswellasoverallperformance.IntroductionAnalysisofvibrationscanindicatefaul
3、tconditionsinrotatingmachinery[1]suchasshaftunbalanceorrubbing.Onecommonapproachistoestimatetime-invariantfeaturesfromthevibrationtimeserieswhichchangewhenafaultoccursinthemachine.Thesefeaturescanthenbeinputintosomeformofclassicationsystemtodecidethemachine'scondition
4、.Articialneuralnetworkssuchasmulti-layerperceptrons(MLPs)provideasystemwhichcantheo-reticallyprovideBayesoptimalclassicationoftheconditionbaseduponmanyfeatures[2].Trainingnetworkscanhowevertakeasubstantiallengthoftimeandisnotguaranteedtondtheoptimalso-lution.Radialb
5、asisfunction(RBF)neuralnetworksprovideanalternativearchitecturewhichcanbetrainedinamuchshorterperiodoftime.Traditionalstatisticaldiscriminantanalysis[3]algorithmscanbeverysimpletoimplementanddonotrequireatimeconsumingtrainingalgorithm.Butinmanycases,theyrequirecertaina
6、ssumptionstobemadeabouttheinputdata.Iftheseassumptionsarenotvalid,theymaynotprovideasgoodasolutionasaneuralnetwork.ExperimentalSet-UpThevibrationsweremeasuredfromasmallexperimentalmachineset.Thisconsistedofanelectricmotorwhichdroveashaftwitha
ywheel.Smallweightscouldbe
7、attachedtothe
ywheelunbalancingtheshaftandrubbingcouldbeappliedusingascrewattachedtoaframe.Thevibrationsweremeasuredhorizontallyandverticallyusingaccelerometersattachedtoabearingblock.Thisset-upallowedthecreationoffourmachineconditions:NN-nofaultsapplied;NR-onlytherubf
8、aultisapplied;WN-onlytheunbalancefaultisapplied;WR-bothrubandunbalancefaultsareapplied.TheshaftDepartmentofElectronic