Support vector machines-based fault diagnosis for turbo-pump rotor
942
S.-F.Yuan,F.-L.Chu/MechanicalSystemsandSignalProcessing20(2006)939–952
recognitionstage,anewsamplexisinputtotheclassi erbetweenthemclassandthenclass:
()lX
mnmn
fmnðxÞ¼sgnlmn.iyiKðx;xiÞþb
i¼1
(5)
Iftheclassi ershowsthatxbelongstothemclass,thenaballotiscastforthemclass.AfterrecognisedbyallthesekðkÀ1Þ=2classi ers,xcanbejudgedtobelongtotheclassthathasthemostballots.
Somedisadvantagesofthisalgorithmare:(1)thenumberkðkÀ1Þ=2oftwo-classSVMclassi ersincreasesgreatlywiththeclassnumberk,calculationincreasesgreatlyaswell,andtherateoftrainingandrecognitionisveryslow;(2)whentwoormoreclasseshavethesameballots,itishardtojudgewhichclassthenewsamplexbelongsto;(3)thereisatleastaclasswhoseballotsarethemost,soanewsamplexthatdoesnotreallybelongtoanyofthesekclasseswouldbemisjudgedtobelongtothemandtheclassifyingerrorappears.
Theothersub-methodisthe‘onetorest’algorithm[1].Thisalgorithmtakesanyoneclassmofthesekclassesasacategory,takestherestkÀ1classesasanothercategory,constructsatwo-classSVMclassi erandnamesitSVMm.Inthisway,ktwo-classSVMclassi erscanbeconstructedinallforacaseofkclasses.TheclassifyingfunctionofSVMmis
()lX
mm
fmðxÞ¼sgnlm.(6)iyiKðx;xiÞþb
i¼1
Intherecognitionstage,thenewsamplexisinputtoallthesekclassi ers.Therearekoutputsinall.Thenewsamplexisjudgedtobelongtotheclasswhoseclassi erhasthelargestoutput.
Somedisadvantagesofthisalgorithmare:(1)allsampleshavetobetakenintothetrainingofallthesekclassi ers;intherecognitionstage,onlyafterallthesekclassi ershaverecognisedthenewsamplexcantheclassi cationresultbeobtained.Thereforethecalculationisverytime-consumingandtherateoftrainingandrecognitionisslowaswell;(2)thereisatleastatwo-classSVMclassi erwhoseoutputisthelargest,soanewsamplexthatdoesnotreallybelongtoanyofthesekclasseswouldbemisjudgedtobelongtothemandtheclassifyingerrorappears;(3)whentwoormoreclassi ershavethesameoutputs,itishardtojudgewhichclassthenewsamplexbelongsto.
3.Onetoothersmulti-classSVMalgorithm
Themulti-outputSVMalgorithmiscomplexandshortofpracticability;andthemethodsthatcombinesometwo-classSVMclassi ersformulti-classpatternrecognitionhaveyetanotherdisadvantage:rejectregion.
Adatasetof3classesisclassi edbythe‘onetoone’andthe‘onetorest’algorithmsrespectivelyasshowninFig.2.Rings,diamondsandtrianglesstandforthese3classesofsamplepointsrespectively,andthereallinesstandforoptimalseparatinghyper-planesofSVM.Itcanbeseenthatthereareshadowzonesinwhichthefaultsamplescannotbeclassi edtoanyclassesbytheoptimalseparatinghyper-planesofSVM;theseshadowzonesarerejectregions.
(a)

(b)
Fig.2.(a)Classi cationwithonetoonealgorithm.(b)Classi cationwithonetorestalgorithm.
Combination of Multi-class Probability Support Vector Machines for Fault Diagnosis_专业资料。维普资讯 http://www.cqvip.com Jra nhaUnvriy(noulDogu iestEg.E)...
Fault diagnosis using a ... 2人阅读 5页 2.00元 Support vector ... Hybrid Support Vector Machines-Based Multi-fault Classification 隐藏>> 维普...
Rotor broken bar fault... 6人阅读 10页 2.00 Fault diagnosis of rec... 12人阅读 6页 2.00 SIMULATION OF CRACK... 2人阅读 4页 2....
Rotor broken bar fault diagnosis for induction motors based on double PQ transformation 隐藏>> 维普资讯 http://www.cqvip.com 12 30 Hunt1/ hin n ...
Fault diagnosis of rotating machinery based on multi-class support vector machines Fault diagnosisFault diagnosis隐藏>> 846 Journal of Mechanical Science and ...
湖南省 长沙市 410082) A Fault Diagnosis Approach for Rotor Systems Based on Empirical Mode Decomposition Method and Support Vector Machines YU De-jie,CHEN...
2009 Fault Diagnosis of Rotor Rub based on Ensemble EMD Yang Yu1, Huihui...INTRODUCTION Rotor rub is a common fault when rotary machine is running, ...
based fault diagnosis method, the improved support vector machine model not ...ZR 图 3.1 感应电机转子电路图 Fig.3.1 Induction Motor Rotor Circuit 3....
Intelligent Fault Diagnosis & Prognosis_机械/仪表...CBM+: Maintenance-Centric Logistics Support for ...pump control valve Rotor mechanical damage Motor ...
Based on the Vibration Signal Xiangyang Jina*,... to achieve the fault diagnosis for rotor system...At the same time, the fault-related machine ...
热门文档