Inter-Turn Fault Diagnosis in Inverter-Fed Induction Motors via MVMD-Based Signal Processing and Machine Learning

Abstract

This paper presents a machine learning-based approach for the detection of interturn faults in inverter-fed induction motors (IM). An interturn fault in the stator winding is a severe fault that is hard to identify at an early stage because of the switching harmonics and non-sinusoidal excitation. This work proposes a method to investigate the inter-turn faults in a 2.2 kW three-phase induction motor operating under open-loop $V / f$ control. An experimental setup is made with the incorporation of different inter-turn fault severities in a coil of IM stator winding. Multivariate variational mode decomposition-based signal processing is utilized to extract features for machine learning analysis. Three machine learning models, Extreme Gradient Boosting (XGBoost), Extra Trees, and CatBoost, are implemented and evaluated for fault classification. The findings show that inter-turn faults cause apparent asymmetry in the three-phase stator currents with slight variations in magnitude at the incipient level of fault. Among the evaluated models, CatBoost performed better with an accuracy of 85.93%.