Development of Detection and Prediction System for Induction Motor Faults Using Linear Regression

Abstract

Abstract: Induction motors are one of the commonly used electrical machines in the industry because of its greatness in the performance. They face various stresses during operating conditions that lead of occurring unexpected faults. In order to avoid the risk of unexpected failures and to improve the performance of the induction motor, condition monitoring becomes necessary. Hence, condition monitoring of the machines has become a more important strategy in structural health monitoring (SHM) research. Therefore, the aim of this study is to perform fault analysis of the induction motor and to study the failure identification techniques. The analysis for fault identification is categorized into two significant main components, which are feature extraction and prediction of the error. The first is used to extract the data from the signal and used to make prediction for the error in the IM. This paper uses a combination between Empirical Wavelet Transform (EWT) and Linear Regression (LR), in order to develop effective strategic method for detection and prediction using programming code in MATLAB software. The wavelet filter bank is created by EWT in order to extract the amplitude modulated-frequency modulated component of signal. The time domain features (TDFs) are then added to the reconstructed signal to extract the fault features. Then, it be the data for LR to developed and categorize the error and coefficient of induction motor faults. Table 1 shows the coefficient of determination, R² for bearing fault. As observed from Table 1, R² for the spectral kurtosis is 0.9955, for spectral crest factor is 0.9967 and for spectral entropy is 0.9986. The coefficient of determination is quite similar for the various features but in details, the coefficient of entropy is higher which is almost closed to ideal value of 1. It indicated an excellent coefficient if the value of R² is close or same value with 1. The experimental results indicate that, under various condition, the techniques can reliably extract and diagnose the IM fault. In addition, the efficiency of the EWT and LR indicates the superiority of the techniques suggested.