Real-Time Estimation of Inter-Turn Fault in Inverter-Fed Three-Phase Induction Motor Using ML-based Vibration Analysis

This paper presents a novel approach to diagnose the inter-turn short-circuit (ITSC) faults in three-phase induction motors using machine learning. Induction motors possess many attractive advantages such as high efficiency, cost-effectiveness, and high reliability. However, with their widespread applications, real-time online fault detection and maintenance play a vital role in providing continuous operation. This paper focuses on developing a real-time inter-turn fault detection of a three-phase induction motor using machine learning. The proposed method detects a minor ITSC fault in the starting stage itself. Therefore, the faults can be removed, or the faulty motors can be replaced with healthy ones to provide continuous operation. Also, using machine learning the severity of faults can be classified as taking adequate maintenance action. The ML-based fault detection method uses a convolutional neural network (CNN) and an Extreme Learning Machine (ELM) for training the diagnosis model with experimental data. The features required for training the model are selected from the faulty inter-turn model and vibration analysis of the three-phase induction motor. The proposed ML-based fault detection gives an accuracy of about 99%.