An AutoML-based comparative evaluation of machine learning methods for detection of eccentricity faults in induction motors by using vibration signals

Induction motors (IMs) are the most widely used electrical machines in industrial applications. However, they are subject to various mechanical and electrical faults. Eccentricity faults are among the common mechanical faults of IMs. This study compares the performance of four commonly used machine learning (ML) methods, including k-nearest neighbours (k-NN), decision tree (DT), support vector machine (SVM), and random forest (RF) along with the statistical features in detecting eccentricity faults of IMs with an automated machine learning (AutoML) model. The aim of using AutoML in this study is to fully automate the process of detection of eccentricity faults of IMs by selecting the classifier with the highest accuracy rate and shortest computation time along with the most effective feature(s). The eccentricity fault analysed in this study was experimentally implemented in the laboratory. Three-axis vibration signals were collected for healthy and eccentricity-faulty IMs. In the proposed study the three-axis vibration signals are pre-processed to determine the statistical features that are used as input to the ML methods. The proposed study offers the best ML method among the four studied algorithms and the need for expert knowledge of ML and eccentricity fault detection. The proposed AutoML model offers the DT method along with the z-axis rms feature for the highest accuracy rate and the shortest computation time in detecting the eccentricity fault.