Advanced feature selection for broken rotor bar faults in induction motors

Abstract

This paper presents an effective fault detection approach for broken rotor bar (BRB) faults in induction motors using machine learning. Three methods, Fast Fourier Transform (FFT), Yule Walker Estimate by Auto Regression (YUL-AR), and Matching Pursuit (MP), are considered for feature selection purpose. These methods are implemented on stator current signals of an induction motor under healthy and different number of broken rotor bars (BRBs) conditions simulated by the finite element analysis software ANSYS. It is found that the proposed MP method offers the most effective feature selection among the three methods, and is able to classify BRB faults accurately through support vector machine (SVM) and artificial neural network (ANN).