FAULT DIAGNOSIS BASED ON IMPROVED LOCALITY PRESERVING PROJECTIONS ALOGRITHM

Abstract

Aiming at the problem that accuracy of orthogonal locality preserving projections（ LPP） for fault diagnosis is not high enough,a fault diagnosis method based on none parameter supervised kernel locality preserving projections（ NPSKLPP） for dimension reduction is proposed. In NPSKLPP,firstly,by changing the Euclidean distance to the Cosine distance which is more robust to outline,and constructing a none parameter nearest-neighbor graph which combined sample label information. And then use the nonlinear mapping to map the high dimension fault feature into an implicit feature space to dimension reduction. Thus a linear transformation is performed to preserve locality geometric structures of the fault feature,which solves the difficulty of parameter selection in computing affinity matrix,as a result,better fault diagnosis accuracy can achieved. The experiment results of gear fault diagnosis verified the effectiveness of the method.

Keywords

• Fault diagnosis;Locality preserving projections;None parameter;Supervised;Gear