Bearing Fault Classification Based on Residual Component of Motor Current Signal and Machine Learning

This paper introduces an algorithm for bearing detection using the residual component of stator current signals of induction motors and machine learning. Firstly, the fundamental component and its harmonics in the measured motor current signal corresponding to different bearing states is estimated using extended Kalman filtering. Next, the residual component is generated from the measurement signal and the estimated signal. Eventually, we use root mean square feature of the residual component to train a support vector machine classifier and to classify bearing faults. Comparing the suggested technique to existing methods, experimental results show that it achieves high accuracy and short execution time. Specifically, the proposed algorithm obtains an average accuracy of 87.63% and its execution time is 0.003 s to process a data frame.