Fault diagnosis method based on sensitive feature selection and manifold learning dimension reduction

Abstract

A fault diagnosis method based on feature selection( FS) and linear local tangent space alignment( LLTSA) was proposed,aiming at solving the problem that there are non-sensitive features and over-high dimensions in the feature set of a fault diagnosis. Firstly,improved kernel distance measurement feature selection method( IKDM-FS) was proposed considering both the distance between classes and the dispersion within a class,and the selected sensitive features were weighted with their sensitive-values. The weighted sensitive feature subset was compressed with LLTSA to reduce its dimensions and get the compressed more sensitive feature subset. Then,the feature subset was fed into a weighted k nearest neighbor classifier( WKNNC) to recognize the fault type,its recognition accuracy was more stable compared with that of a k nearest neighbor classification( KNNC). At last,the validity of the proposed method was verified with fault diagnosis tests of a rolling bearing.