Broken Rotor Bar Fault Detection Using Machine Learning: Optimal Frequency Resolution

Abstract

This paper explores the optimal frequency resolution of the current spectra for detecting the broken rotor bar fault in induction motors with machine learning and motor current signature analysis. Conventional methods of broken rotor bar detection usually advocate for a higher frequency resolution in the motor current spectrum, which requires longer current signal measurements that are difficult and expensive to conduct. Thus, this work aims to identify the limitations to frequency resolution for successful broken rotor bar diagnosis when applying machine learning algorithms. The study also provides recommendations on the signal processing for feature extraction to enhance machine learning model performance. The machine learning algorithms used in the study are support vector machines, gradient boosting machines, and multilayer perceptron.