A fault classification method using dynamic centered one-dimensional local angular binary pattern for a PMSM and drive system

Abstract

Nowadays, fault classification of the electric motors has become a hot-topic research area. Therefore, many machine learning method has been presented to create an intelligent fault detection system for electric motor. In this work, a novel classification method is presented for five different fault conditions of a motor and its drive system by using dynamic centered one-dimensional local angular binary pattern (DC-1D-LABP). It is proposed a novel multi-leveled feature extraction network, and one-dimensional discrete wavelet transform (1D-DWT) is used in order to create levels. Features are extracted from each level by using the proposed DC-1D-LABP. Neighborhood component analysis (NCA) and ReliefF-based 2-layered feature selector (NCARF) are used to select most discriminative features, and four conventional classifiers are selected for testing. A novel fault dataset is acquired, and this dataset is used for tests. Four cases were defined according to input current signal. The achieved best classification accuracy rates are 0.9692, 0.9571, 0.9650 and 1 for Case 1, Case 2, Case 3 and Case 4, respectively. These results indicate that the proposed DC-1D-LABP-based method is very effective for fault classification. Consequently, it is proposed a highly accurate and cognitive method for a fault classification in this study.