基于复杂网络优化的DAG-SVM在滚动轴承故障诊断中的应用.pdf

《基于复杂网络优化的DAG-SVM在滚动轴承故障诊断中的应用.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、振动与冲击第34卷第l2期JOURNALOFVIBRATIONANDSHOCKVo1．34No．122015基于复杂网络优化的DAG—SVM在滚动轴承故障诊断中的应用石瑞敏，杨兆建(太原理工大学机械工程学院，太原030024)摘要：针对滚动轴承故障与其演化程度组合类型数量大，一般模式识别方法难以适应的问题，提出基于复杂网络优化的有向无环图支持向量机(CNDAG．SVM)。该方法引入复杂网络理论中相似性测度概念用以评定各样本类型间的分离性质，并以平均相似性测度作为有效度量样本类型可区分程度的测度对有向无环图叶节点类型进行排序，依次提取对应二元分类

2、器构造较优有向无环图拓扑结构，缓解误差累积效应的同时提高了结构上层节点的容错能力，获得较高的正确识别率。利用局部均值分解方法提取乘积函数(ProductionFunction，PF)分量波峰系数、峭度系数及能量构造特征向量，将其输入CNDAG．SVM分类器中用于区分滚动轴承的故障类型与演化程度。对滚动轴承内圈故障、外圈故障及滚动体故障振动信号的分析结果表明，该方法能准确有效识别故障类型与其演化程度，较之传统多元分类支持向量机具有更高的识别精度和效率。关键词：复杂网络；有向无环图支持向量机；滚动轴承；故障诊断中图分类号：TH133．33文献标志码

3、：ADOI：10．13465／j．cnki．jVS．2015．12．001ApplicationofoptimizeddirectedacyclicgraphsupportvectormachinebasedoncomplexnetworkinfaultdiagnosisofrollingbearingSH／Rui—min，YANGZhao-jian(SchoolofMechanicalEngineering，TaiyuanUniversityofTechnology，Taiyuan030024，China)Abstract：Duetothel

4、argeamountofcrossedcombinationsoffaultpatternsandevolutionstagesofrollingbearings，thegeneralpatternsrecognitionmethodisdifficulttoadapttomultivariateprocess．Inviewoftheproblem，anoptimizeddirectedacyclicgraphsupportvectormachine(DAG—SVM)basedoncomplexnetwork(CN)wasproposed．Ac

5、cordingtothesimilaritymeasureincomplexnetworktheory，theseparatingcharactersofsampleswereevaluated，andthenodesofdirectedacyclicgraphweresequencedbytheaveragesimilaritymeasurewhichwascalculatedasthecriterionfordistinguishingdegreeofsamples．Thenthecorrespondingbinarysupportvect

6、ormachineswereselectedtoconstructanoptimaldirectedacyclicgraph，toachievehighcorrectionidentificationratiobyalleviatingerroraccumulationandimprovingfaulttoleranceoftheuppernodes．Featurevectorswereconstructedofthecrestfactor，kurtosiscoefficientandenergyofproductfunctions，obtai

7、nedbylocalmeandecomposition．AndthenthefeaturevectorswereservedasinputparametersofCNDAG—SVMclassifiertosortfaultpatternsandevolutionstagesofrollingbearings．Byanalyzingthevibrationsignalacquiredfromthebearingswithinner—race，outer—raceorelementsfaults，theexperimentalresultsindi

8、catethattheproposedmethodcanrecognizethefaulttypesandevolutiongradeseffecti