基于拒绝式转导推理M-SVDD的机械故障诊断

Abstract

针对现有多类支持向量域数据描述(MulTI-ClASS SuPPOrT dATA dESCrIPTIOn,M-SVdd)存在的不足和局限性,提出一种拒绝式转导推理多类支持向量域数据描述(rEJECTEd TrAnSduCTIVE InfErEnCE M-SVdd,rTIM-SVdd)方法,并将该方法应用于机械加工故障诊断当中。首先,rTIM-SVdd通过训练寻求一个尽可能包含所有数据样本的最小超球体作为拒绝检测面,来整体学习样本知识或数据描述,并以一定的拒绝度判别新的测试样本,达到故障检测的目的;其次,应用分别包含各个类别样本的多个超球体,来判别满足一定拒绝度的样本,实现多分类问题。对于模糊样本点归属判别关键问题,本文采用一种新的转导推理规则来进行决策。最后,通过一个仿真实验进行验证,结果证明了rTIM-SVdd的可行性和有效性。

To improve the performance of M-SVDD(multi-class support vector data description),a RTIM-SVDD (rejected transductive inference M-SVDD)method is presented and used in the application of fault diagnosis of manufacturing process.Firstly,a hyper-sphere with minimum volume,named rejected detection decision boundary,is used to enclose all labeled data set so as to globally learn the knowledge of existed data set.And a rejected measure index is used to detect the new unlabeled test data.Secondly,apply a similar hyper-sphere to enclose each class data set,and then combine these one-class hyper-spheres as multi-class SVDD classifier so as to identify the labels of the test data justified by rejected detection.To identify fuzzy samples,this paper presents a transductive decision rule to decide the labels of fuzzy samples.Finally,the feasibility and effectiveness of the presented method is illustrated in the application of two simulation experiments