Comparative Evaluation of Machine Learning Methods for the Detection of the Eccentricity Faults of Induction Motors by Using Vibration Signals

Abstract

Abstract One of the most critical tasks to ensure continuous operation in most industrial applications is electric machines' fault and condition monitoring. Induction motors are widely used electrical machines. They are more prone to eccentricity faults due to the short air-gap length. Recently, machine learning techniques have been developed to diagnose the faults of induction motors. This study presents an experimental comparison of the performance of four commonly used machine learning techniques in detecting eccentricity faults of induction motors. The detection of the eccentricity faults is conducted by using vibration signals. The three-axis vibration signals were collected for two cases, healthy and faulty, under different loading levels of a three-phase, 3-kW, two-pole induction motor. The performance of each machine learning method in detecting eccentricity was tested with the vibration signals and compared with each other. The purpose of the study is to assess the performance of each machine learning method and find the most effective features. The results show that rms and p2p features of the vibration signals provide the highest accuracy rates in all four ML methods.