Chaotic Parallel Support Vector Machine and its application for fault diagnosis of hydraulic pump

Abstract

Hydraulic pump is the critical part of a hydraulic system. The diagnosis of hydraulic pump is very crucial for reliability. This paper studies on a Chaotic Parallel Support Vector Machine (CPSVM) and employs it for fault diagnosis of hydraulic pump. The CPSVM combines the chaos theory and a number of SVMs connected in parallel. Phase-space reconstruction of chaos theory is utilized to determine the dimension of input vectors for each SVM. Each SVM has an output. A weighted sum of each output is considered as the output of the CPSVM. To diagnose faults of hydraulic pump, a residual error generator is designed based on the CPSVM. This residual error generator is firstly trained using data from normal state. Then, it can be used for fault clustering by analysis of the residual error. Its performance and effectiveness has also been validated via a plunger pump test-bed.