Comparison of Machine Learning Algorithms for Induction MotorRotor SingleFault Diagnosis using Stator Current Signal

This quantitive researchwas conducted to compare the efficiency between three groups of machine learning' classification techniques for detecting broken rotor bar (BRB)fault in induction motor using stator currents signals with two different signal processing method. Thus, the main purpose of the article is to find out the most suitable method of distributing and extracting data for the fault diagnosis problems. Two of the most common used signal processing method –Discrete Wavelet Transform (DWT) and Fast Fourier Transform (FFT)has been implemented to extract the statiscal features of the faults. Then in the next logical steps, also the most important step, fault diagnosis, three classification algorithms: Support Vector Machines (SVM), K-nearest Neighbors (KNN), and Ensembles are chosen to evaluate the performance and the impact of those different classifiers for induction motor fault diagnosis. Hence, the study found there are five classifiers (Fine Gaussian SVM, Fine KNN, Weighted KNN, Bagged Trees and Subspace KNN) are best suited for the proposed problem when providing nearly 100% classification accuracy for all fault that the other 12 classifiers can not perform well.