Rapid Diagnosis of Induction Motor Electrical Faults using Convolutional Autoencoder Feature Extraction

Abstract

Electrical faults such as stator turns fault and broken rotor bars are among the frequently occurring failure modes in induction motors. This article presents a novel deep learning-based approach for the rapid diagnosis of these electrical faults within a short time window of 200 milliseconds. The extended Park's vector, calculated using three-phase supply currents, is chosen as the medium for fault detection. An unsupervised convolutional autoencoder is designed to detect features distinguishing healthy and faulty conditions. The developed features are supplied to a support vector machine to classify the fault conditions. The proposed approach is validated in a laboratory setup consisting of an inverter-fed induction motor operating under time-varying load and speed conditions with an accuracy > 95%.