资源描述：
《汽轮机轴系振动故障诊断中的信息融合方法研究》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、华中科技大学博士学位论文汽轮机轴系振动故障诊断中的信息融合方法研究姓名：张燕平申请学位级别：博士专业：热能工程指导教师：黄树红20060508地融合了波形的时域信息，可以定量地评估该时间序列在基线附近波动的不规则性，因而可实现对故障的识别。在对关联分形维计算研究的基础上，提出利用信号的二进小波高频重构信号计算关联维数。对转子试验台故障模拟试验信号的原始采样信号和高频重构信号分别计算了关联维数并进行了对照分析，结果表明，经过高频重构后的关联维数能够更好地进行故障识别，是对汽轮机典型故障进行定量诊断的有效时域特征，值得进
2、行深入的理论和应用研究。最后，研究了小波灰度图及小波灰度矩在实际诊断系统中的应用。研究了小波算法的程序实现，将小波灰度图及小波灰度矩作为独立的小波诊断模块，应用在某省的“汽轮机组振动远程监测及故障诊断专家系统”中。作为故障诊断的辅助模块，小波诊断模块在该远程诊断系统中发挥了应有的作用。关键词：信息融合汽轮发电机组小波分析小波灰度矩最小距离分类器概率神经网络关联维数IIAbstractInthispaper,theproblemofinformationfusionandquantitativediagnosisofl
3、arge-scaleturbinegeneratorswereexplored.Thefusionsoffaultysignalswerecarriedfromtimedomain,frequencydomain,andtime-frequencydomainrespectively,andsomeresearchwereappliedonactualdiagnosissystem.Atfirst,thefaultsimulationrotortestrigofturbinerotorshaftsystemwasde
4、signed,andseveraltypicalfaultysignalsduringspeedrisingwerecollectedfromthisrotortestrig.Thisestablishedtheanalysisfoundationinthispaper.Tocheckthecorrectnessofcollectedfaultysignals,theFourierways,3-DamplitudespectrumandBodediagramwasusedtoanalyzethesesignals.A
5、tthesecond,thecontinuouswavelettransformscalogramwasexplored,andtwofeatures,waveletgraymomentandfirst-orderwaveletgraymomentvector,areproposedforfaultclassificationofsteamturbinerotor.Theanalysisindicatesthatfirst-orderwaveletgraymomentcanrevealthetime-frequenc
6、yfeaturesofviberationsignalswell,andcouldbeusedtodiagnosefaultsquantitatively.Theeffectivenessofthefirst-orderwaveletgraymomentvectorisalsodemontratedbyexperimentaldata.Resultshowthatthefirst-orderwaveletgraymomentvectorissuitabletoreflectthelocalinformationofs
7、calogram,andwouldbeaeffectivemethodofvibrationsignalanalysisforfaultdiagnosticsofrotatingmachinery.Atthethird,basedonthemethodofinformationentropy,fusionresearchonfourinformationentropy:singularspectrumentropy,powerspectrumentropy,waveletenergyspectrumentropyan
8、dwaveletspacestatespectrumentropywerecarried.Twomethodswereexploredtofusethefourinformationentropyabove,oneistheMinimumDistanceClassifier(MDC),anotherisProbabilityNeuralNetw
