Enhancing Supervised-Learning Fault Classification of Permanent Magnet Synchronous Motors with Feature Selections

Inter turn short circuit (ITSC) faults in the permanent magnet synchronous motors (PMSMs) using applied signal processing methods might not be detected at low severities or in early stage. Four supervised machine learning (ML) algorithms, namely decision tree (DT), random forest (RF), extreme gradient boosting (XGB), and support vector machine (SVM) are used to classify different incipient faults with the healthy condition. Phase current datasets of an in-house PMSM are collected in the healthy, 2% ITSC, and 5% ITSC fault conditions. Statistical features, namely root mean square (RMS) crest, impulse, kurtosis, standard deviation, are calculated based on the measured data for the classification. Wrapper-based feature selection (WFS) technique is employed to optimize the best features to maximize the classification accuracy while reducing computational complexity. It is found that the combination of RMS factor, standard deviation, and crest factor is the best feature for fault classifications of 2% ITSC and 5% ITSC.