Abstract
Herein, an automated diagnostic and predictive system has been proposed for low-voltage induction motors. The stator Inter-Turn Shorts-Circuits (ITSC) faults has been considered due to their impact on unexpected failures for this critical component. The system consists of non-invasive Motor Current Signature Analysis (MCSA) and machine learning techniques. The evaluation was based on a dataset representing three phase stator current signals from a 0.75 hp motor. Two conditions have been chosen, healthy and ITSC at four different severity levels. The Random Forest algorithm includes a number of parameters, such as variance, peak-to-peak amplitude, root-mean-square value of spectral amplitudes, and envelope energy. Simulation results showed that the accuracy of binary classification of healthy and faulty devices reached 94.12%, while the accuracy of multi-class fault severity prediction was only 76.47%.