A Diagnosis Method based on Maximum Information Coefficient and MKL for Open Circuit Fault in PMSM inverter

Abstract

This paper presents a fault diagnosis method for the three-phase inverter in Permanent magnet synchronous motors (PMSM) drive system, which based on the maximum information coefficient and multi kernel SVM model. Firstly, we extract fault characteristics from line-to-line voltage signals. Secondly, all the fault features are sorted based on the maximum information coefficient (MIC) and select the effective features. Thirdly, select a set of basic kernels, use EasyMKL algorithm to assign weight to each kernel function, and cut the inefficient kernel function to get the multi-kernel function. Finally, the open circuit faults are diagnosed by the multi-kernel function and the support vector machine. To verify the effectiveness of diagnosis method. We conduct plenty of experiments by comparing our method with other machine learning methods. The test results show that the fault diagnosis accuracy rate after feature selection is as high as 98%, which is better than the comparison method.