Induction motor rotor bars faults diagnosis based on multiple features extraction and selection with self-organising map neural network

Abstract

This paper presents high performances fault detection and diagnosis approach for broken rotor bar (BRB) and severity evaluation in squirrel cage induction motors. The proposed approach is based on combination of multiple features extraction techniques from the three-phase stator currents, features selection, and self-organising maps (SOM) as classifier in the BRB fault diagnosis process. For feature extraction, the envelope and the zero crossing times (ZCT) signals are extracted from stator currents, then, statistical parameters from time and frequency domains, in addition to fault-related frequencies are calculated from the current waveform, the envelope, and the ZCT signals. The most relevant features are then selected using the relief feature selection algorithm. Finally, the SOM is used for the decision-making step. Conducted experimental investigations on a healthy and faulty machines, have exposed the robustness and accuracy of the proposed BRB fault detection technique.