LAUSR

The feature dimension reduction is a useful pre-processing step for fault diagnosis, where the irrelevant and redundant information in the data can be reduced. In this case, not only the computation complexity can be reduced, but also the better classification performance can be obtained. Hence, this paper presents a multiple fault diagnosis method for the three-phase inverter in PMSM drive system, where the hybrid dimensionality reduction method is applied to reduce the dimension of the fault features. First, time-domain fault features are extracted from the line-to-line voltage signals. Second, a feature reduction is performed by combining principal component analysis and linear discriminant analysis, where the linear discriminant analysis is improved by introducing the singular value decomposition and redefining the between-class scatter which solve the issue of small sample and the problem that the similar classes are not easily separated. Finally, the faults are diagnosed by support vector machine. Both the simulation and the experimental results show that the proposed method can improve the discrimination performance of the fault types after the dimension reduction, making it in favor of fault classification.