Multiple features extraction and selection for detection and classification of stator winding faults

In this study, a new effective approach for detection and classification of stator winding faults in induction motors is presented. The approach is based on current analysis. It uses multiple features extraction techniques, where Park transform, zero crossing time signal, and the envelope are extracted from the three-phase stator currents. Then, statistical features are calculated from time and frequency domains of each extracted signal. The Features selection techniques (ReliefF, minimum redundancy and max relevancy, and support vector machine approach based on recursive feature elimination) are used to select from the extracted features the most relevant ones. As a classifier, the self-organising map neural network is used. The proposed procedure is experimentally studied using stator current signals obtained from various faulty cases and a healthy induction motor at different load variations. The experimental results verify that the proposed strategy is able to distinguish the faulty cases from the healthy ones. Also, it effectively identifies the faulty phase in addition to the extent of the fault.