基于ARHSMM的设备故障诊断与预测研究

《基于ARHSMM的设备故障诊断与预测研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、上海交通大学硕士学位论文基于AR-HSMM的设备故障诊断与预测研究姓名：杨志波申请学位级别：硕士专业：管理科学与工程指导教师：董明20070201摘要AR-HSMM与HMM的识别率比较，针对本文改进的Markov模型，与普通Markov模型的预测结果进行了比较，实验结果表明，文中所提出的AR-HSMM和基于AR-HSMM的改进的Markov模型预测方法是有效的。关键字：故障诊断，剩余寿命预测，自回归隐半马尔可夫模型，马尔可夫模型ABSTRACTTheResearchonEquipmentDiagnosisandPrognosisBasedonAR-HSMMABSTRACTInrecent

2、years,withtherapiddevelopmentofcomputer,sensor,signalprocessingandAItechnology,CBM(ConditionBasedMaintenance)asanadvancedmaintenancemethodisbecomingmoreandmoreimportant.AprerequisitetowidespreaddeploymentofCBMtechnologyandpracticeinindustryiseffectivediagnosticsandprognostics.Recently,HMM(HiddenM

3、arkovModel)anditsextendedmodeliswidelyusedinmanyfields,but,becauseHMManditsextendedmodelarenotsorealistic,thediagnosisresultsarenotsogood.ThisthesisdevelopsanewextendedmodelofHMMwhichiscalledAR-HSMMAR-HSMM(Auto-RegressiveHiddensemi-MarkovModel).AR-HSMMisbasedonAR-HMM(Auto-RegressiveHiddenMarkovMo

4、del)andHSMM(Hiddensemi-MarkovModel)byrelaxingthelimitationsofHMM.Themajorresearchcontentsinthisthesisareasfollows.First,aVQ(VectorQuatization)algorithmbasedonK-MEANSisdeveloped.Then,byidentifyingthedisadvantageandadvantageofAR-HMMandHSMMandcombiningtheiradvantages,anewextendedmodelnamedAR-HSMMisp

5、roposed.Newforwardandbackwardalgorithmsandreestimationformulaareprovided.TheinitiationprocessoftheAR-HSMMisdiscussed,whichincludsthechoiceoftheAR-HSMMtypeandthenumberofhiddenstate.Finally,thediagnosisprocedureofAR-HSMMisgiven.BasedontheresultsofAR-HSMMbaseddiagnosis,aprognosismodelforpredictingth

6、eremainingusefullifeisgiven.Thismodelusesthediagnosticsresultsasinputs.Theparameterestimationformulaandtheprognosisprocedureareprovidedinthethesis.Theproposedalgorithmsareverifiedthroughtestingontherealdata.InordertochooseABSTRACTthebestnumberofhiddenstateandcomparethediagnosticsresultswithHMM,fo

7、urtestsareexperimented.Theresultsshowthattheproposedmethodsareeffective.InordertocomparetheprognosisresultswiththoseofregularMarkovmodel,anothertestisimplemented.TheresultsshowthatthemodifiedMarkovmodelisbetterthanregu