FAULT DIAGNOSIS BASED ON MIXED LABEL INFORMATION LOCALITY PRESERVING PROJECTIONS DIMENSION REDUCTION

Abstract

In order to preserve the local structure of high dimension fault feature better, locality preserving projections（LPP） are improved and a fault diagnosis method based mixed label information locality preserving projections（MLILPP） for dimension reduction is proposed. MLILPP constructed similarity matrix and divergence matrix based on label information. The similarity matrix is used to preserve the local structure of the same-label samples before and after dimension reduction, whereas the divergence matrix is used to enlarge the distance between the different-label samples after dimension reduction. As a result, data structure is retained effectively, better low dimension structure can be obtained, and higher fault diagnosis accuracy can achieved. The experiment results of hydraulic pump fault diagnosis verified the effectiveness of the method.

Keywords

• Fault diagnosis;Locality preserving projections;Label information;Hydraulic pump