基于軟計(jì)算和信息融合的故障診斷方法研究.pdf

1、申請(qǐng)上海交通大學(xué)工學(xué)博士學(xué)位論文i基于軟計(jì)算和信息融合的故障診斷方法研究摘要隨著現(xiàn)代化大生產(chǎn)的發(fā)展和科學(xué)技術(shù)的進(jìn)步機(jī)械設(shè)備的結(jié)構(gòu)越來(lái)越復(fù)雜功能越來(lái)越完備自動(dòng)化程度也越來(lái)越高目前在石化能源冶金電力航空機(jī)械等各領(lǐng)域中廣泛使用的旋轉(zhuǎn)機(jī)械由于工作條件的變化及使用壽命的限制一旦在某處發(fā)生故障就可能會(huì)引起鏈鎖反應(yīng)導(dǎo)致整臺(tái)設(shè)備甚至整個(gè)生成過(guò)程不能正常工作造成巨大的經(jīng)濟(jì)損失甚至造成災(zāi)難性的人員傷亡產(chǎn)生不良的社會(huì)影響所以有計(jì)劃有組織有針對(duì)地對(duì)關(guān)鍵設(shè)備進(jìn)行

2、實(shí)時(shí)監(jiān)測(cè)與診斷做到盡早地發(fā)現(xiàn)設(shè)備在運(yùn)行過(guò)程中的各種隱患從而防止災(zāi)難性事故的發(fā)生成為機(jī)械設(shè)備故障診斷系統(tǒng)面臨和解決的首要問(wèn)題隨著現(xiàn)代計(jì)算機(jī)技術(shù)測(cè)試技術(shù)信號(hào)分析處理技術(shù)人工智能技術(shù)以及軟計(jì)算和信息融合理論等學(xué)科的發(fā)展機(jī)械設(shè)備狀態(tài)監(jiān)測(cè)與故障診斷技術(shù)進(jìn)入了一個(gè)嶄新的階段本文在詳細(xì)總結(jié)基于人工智能軟計(jì)算和信息融合技術(shù)的故障診斷的現(xiàn)狀和發(fā)展趨勢(shì)的基礎(chǔ)上針對(duì)故障診斷中知識(shí)獲取的瓶頸問(wèn)題診斷信息的不完整不精確模糊和不確定性的特點(diǎn)對(duì)基于軟計(jì)算和信息融合的

3、設(shè)備故障診斷方法系統(tǒng)級(jí)診斷知識(shí)的組織和推理知識(shí)的自動(dòng)獲取與診斷基于軟計(jì)算和信息融合技術(shù)的故障診斷進(jìn)行了深入的研究知識(shí)模型是描述某一領(lǐng)域產(chǎn)品相關(guān)專(zhuān)家知識(shí)的信息模型隨著軟計(jì)算和信息融合技術(shù)用于故障診斷便產(chǎn)生了新的診斷知識(shí)而傳統(tǒng)的知識(shí)表達(dá)和推理方法難以適應(yīng)這些知識(shí)在研究各種知識(shí)模型的基礎(chǔ)上提出了故障診斷集成知識(shí)模型并統(tǒng)一以面向?qū)ο蟮目蚣芤?guī)則表示這種知識(shí)模型不但能處理顯式知識(shí)隱式知識(shí)而且能夠處理信息融合知識(shí)在診斷系統(tǒng)的系統(tǒng)級(jí)推理中基于分而治之的

4、任務(wù)分解思想給出了基于黑板模型的診斷知識(shí)組織和推理方法該方法利用多個(gè)知識(shí)源通過(guò)任務(wù)調(diào)度和黑板控制獲得最終的推理診斷結(jié)果知識(shí)獲取被公認(rèn)為是專(zhuān)家系統(tǒng)建造中的瓶頸問(wèn)題本文基于矩陣的奇異值分解模糊C均值聚類(lèi)和粗糙集理論研究機(jī)械振動(dòng)的連續(xù)數(shù)值特征的自動(dòng)知識(shí)獲取并使用獲得的知識(shí)進(jìn)行故障診斷在知識(shí)的自動(dòng)獲取中對(duì)含有重復(fù)和沖突對(duì)象的離散決策表提出了一種基于啟發(fā)式約簡(jiǎn)和屬性頻率函數(shù)的規(guī)則獲取方法該方法能夠一定程度上降低計(jì)算復(fù)雜度把若干振動(dòng)征兆數(shù)據(jù)和其對(duì)應(yīng)

5、的故障類(lèi)型數(shù)據(jù)看作矩陣和構(gòu)成決策表利用矩陣的奇異值分解模糊C均值聚類(lèi)和粗糙集學(xué)習(xí)技術(shù)建立了連續(xù)屬性值決策表的規(guī)則獲取方法該方法獲取的診斷規(guī)則不僅有很好的知識(shí)歸納能力而且還有很好的知識(shí)泛化(推廣)能力在利用獲取的知識(shí)進(jìn)行故障診斷中分析了故障診斷中的各種匹配模式建立了一種基于征兆重要度考慮的彈性故障診斷規(guī)則匹配模式該模式根據(jù)診斷對(duì)象和診斷規(guī)則的匹配程度規(guī)則強(qiáng)度和預(yù)定的診斷結(jié)論閾值確定診斷結(jié)論從而使得故障診斷能夠根據(jù)不同情況確定診斷結(jié)論這更符

6、合實(shí)際的需要單一神經(jīng)網(wǎng)絡(luò)故障診斷存在輸出神經(jīng)元數(shù)目較多診斷率較低和發(fā)生新故障時(shí)神經(jīng)網(wǎng)絡(luò)的重復(fù)訓(xùn)練等問(wèn)題借鑒分而治之的思想采用組合式神經(jīng)網(wǎng)絡(luò)和基于模糊積分的信息融合技術(shù)進(jìn)行故障診斷不同的組合式神經(jīng)網(wǎng)絡(luò)負(fù)責(zé)對(duì)不同故障類(lèi)型進(jìn)行識(shí)別和診斷對(duì)它們的診斷結(jié)果進(jìn)行模糊積分從而確定診斷對(duì)象的故障狀態(tài)其中在模糊積分的模糊密度計(jì)算中對(duì)模糊C均值聚類(lèi)技術(shù)進(jìn)行改進(jìn)建立了一種基于模糊C均值聚類(lèi)的模糊劃分的方法基于此方法提出申請(qǐng)上海交通大學(xué)工學(xué)博士學(xué)位論文iiiR

7、esearchonApproachesofSoftComputingInfmationFusionbasedFaultDiagnosisABSTRACTWiththedevelopmentoflargemodernproductiontechnologymechanicalstructuresbecomememecomplicatedwithmemecomprehensivefunctionhigherautomatization.Th

8、epresentrotatingmachinessubstantivelyusedinsuchfieldsaspetrifactionenergymetallurgyelectricpowernavigationmachinerywouldbefallenfaultfdiversificationofwkingconditionslimitedlongevityprobablytochainreacttocausethewholeass

9、embleevenalltheproductionprocesstowkabnmallyresultinginenmouseconomiclossevencatastrophiccasualtyillsocialinfluence.Consequentlyitistheprimarytaskfacedtobesolvedffaultdiagnosticsystemofmachinerytoonlinedetectdiagnosethos

10、ekeyequipmentsdesignedlyganizationallypertinentlytofindoutallkindsofhiddentroubleofrunningmachinesassoonaspossibletopreventdisasterfromhappening.Withtheadvancementofsuchdisciplinesasmoderncomputertechnologytestingtechnol

11、ogysignalprocessingtechnologysoftcomputinginfmationfusiontheiesthetechnologyofconditiondetectiondiagnosisofmachinescomesintoabrnewera.Onthebasisofsummarizingthepresentstatusdevelopingtrendoffaultdiagnosisbasedonartificia

12、lintelligencesoftcomputinginfmationfusiontechnologiessuchdiagnosticapproachesofmachinesbasedonsoftcomputinginfmationfusionasganizationreasoningofdiagnosticknowledgeindiagnosticsystemautomotiveknowledgeacquirementsoftcomp

13、utinginfmationfusionbasedfaultdiagnosishavebeenresearchedindetailsconsideringthebottleneckproblemofdiagnosticknowledgeacteristicsofincompleteinaccuratefuzzyuncertaindiagnosticinfmation.Knowledgemodelistheinfmationmodelto

14、describeexpertknowledgerelativetoonecertainfield.Withtheapplicationtothefieldoffaultdiagnosisofsoftcomputinginfmationfusionbasedtechnologiesnewtypeofdiagnosticknowledgearisesbutthetraditionalknowledgeganizationreasoningm

15、ethodscannotmeettheincomingknowledge’sneed.Therefeafterinvestigationofadiversityofknowledgemodelsthehybridknowledgemodelisproposedinthisdissertationwhichcandealwithnotonlyexplicitimplicitknowledgebutalsotheknowledgeoninf

16、mationfusionisrepresentedbyobjectientedframerulefmulation.Asfthesystematicreasoningbasedondivideconquerstrategytheblackboardbasedmodelisusedtoganizereasonknowledgetoobtainfinalreasoningconclusionbytaskmaneuverofallkindso

17、fknowledge.Itiswellknownthatknowledgeacquirementisthebottleneckprobleminconstructingexpertsystem.Basedonthesingularvaluesdecomposition(SVD)ofmatrixfuzzyCmeansclustering(FCM)roughsetthey(RST)anapproachofautomotiveknowledg

18、eacquirementofcontinuousnumericalfeaturesofmechanicalvibrationisinvestigatedaccdinglyusedtofaultdiagnosis.Intheautomotiveknowledgeacquirementfdecisiontablewithreduplicateconflictingobjectsaninnovativeapproachofdecisionru

19、leacquirementisproposedbasedonheuristicreductionfrequenciesofattributeswhichincertaindegreedecreasethecomputationalcomplexity.Thedecisionruleacquirementmethodofdecisiontablewithcontinuousdecisionattributevaluesisputfward