文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 机械/仪表 > Support vector machines-based fault diagnosis for turbo-pump rotor
侵权投诉

Support vector machines-based fault diagnosis for turbo-pump rotor

Support vector machines-based fault diagnosis for turbo-pump rotor

S.-F.Yuan,F.-L.Chu/MechanicalSystemsandSignalProcessing20(2006)939–952

941

wheresgn()isthesignfunction,ifu40thensgnðuÞ¼1,ifup0thensgnðuÞ¼À1,theLagrangecoef cientliisthesolutionofthefollowingquadraticprogramming(QP)problem:

Maximize

L

T

WðKÞ¼ÀKT1þ1KDK

subjecttoKTy¼0,

KÀCp0,ÀKp0,

ð3Þ

whereðKÞi¼li,1i¼1andDij¼yiyjxTixj,Cisapenaltyconstantforthosesamplepointsmisclassi ed

bytheoptimalseparatingplane.Itsroleistostrikeaproperbalancebetweenthecalculationcomplexityandtheclassifyingerror.WhenC-+N,itistheidealcaseinwhichallsamplesaretheoreticallycorrectlyseparatedbytheoptimalseparatingplaneandthereisnoclassifyingerroratall,butthecalculationcomplexityisthebiggest.Itisfoundthatonlyafewcoef cientsliarenotzero,andsinceeverycoef cientlicorrespondstoaparticularsamplepoint,thismeansthatonlythesamplepointswhosecorrespondingliarenotzerodeterminetheoptimalseparatingplane.Onlythesefewsamplepoints,thatarecalledsupportvectors,affecttheclassi cationresult;whileothersamplepointscouldberemovedfromthesamplesetandtheoptimalseparatingplanewouldbealmostunaltered.Thesesupportvectorsareusuallyfewinthesampleset,andVapnik[1]showedthattheirnumberisproportionaltothegeneralisationerroroftheclassi er.

Sincethereexistcertainpracticalproblemsthatarenotlinearlyseparable,SVMhastobedevelopedfortheclassi cationofnonlinearproblems.Byprojectingtheoriginalsamplespaceintoahigh-dimensionaleigenspacewithakernelfunctionKðx;xiÞ,thenonlinearseparableproblembecomeslinearlyseparableintheeigenspace.TheSVMclassi erintheeigenspacehastheclassifyingfunctionasfollows:

()lX

fðxÞ¼sgnliyiKðx;xiÞþb.(4)

i¼1

InTable1somekernelfunctionsKðx;xiÞproposedbyVapnikarelisted,andnoticehowtheyleadtowell-knownclassi ers,whichareknowntohavegoodproperties.

SVMisproposedinitiallyfortwo-classpatternrecognition.ThisSVMcanbecalledtwo-classSVMclassi ers.Therearesomemethodstogeneraliseittomulti-classpatternrecognition[2–24].Theremethodscanbesummarisedintotwokinds.

Oneisthemulti-outputSVMalgorithm.Thisalgorithmhasmulti-outputwithonlyoneSVMclassi er.Itsmainideaistotakeallclassesintoaccountwhenconstructingtheclassifyingfunctionandadifferentclassifyingfunctionisderivedwithrevisedoptimisationfunctionandconstraintcondition.ThisalgorithmhasonlyoneSVMclassi er,butitsclassifyingfunctionisverycomplex,thecalculationisverycomplexaswell,thetrainingandrecognitionisverytime-consuming,andtheclassifyingerrorisbigwhensamplesarelarge[2].Anotherkindofmethodsistocombinesometwo-classSVMclassi ersformulti-classpatternrecognition.Thiskindofmethodsconsistsoftwosub-methodsasfollows.

Oneisthe‘onetoone’algorithm[3].Thisalgorithmconstructsalltwo-classSVMclassi ersbetweenanytwoclasses,sokðkÀ1Þ=2two-classSVMclassi erscanbeconstructedinallforacaseofkclasses.Inthe

Table1

Somepossiblekernelfunctionsandthetypeofclassi erstheyde neKernelfunction

Kðx;xiÞ¼ðxTxiþ1Þd

Kðx;xiÞ¼expðÀkxÀxk2=2s2ÞKðx;xiÞ¼tanhðxTxiþyÞ

Typeofclassi erPolynomialofdegreedGaussianRBF

Multi-layerperceptron

上一页第3页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email