Inter-turn fault detection in induction motors using deep neural networks and signal processing methods

Abstract

This study aims to enhance the effectiveness and reliability of stator winding fault detection in squirrel-cage induction machines. It specifically investigates the influence of signal preprocessing methods, including the Fast Fourier Transform and the Wavelet Transform, on the extraction of fault-related information and the subsequent performance of Deep Neural Networks (DNNs) in classifying inter-turn faults. The study uses a quantitative mixed-method approach, combining field-circuit modeling with experimental validation. A field-circuit model was developed to simulate both healthy and faulty motor conditions and generate training data. Diagnostic features derived from measured phase current waveforms were processed and used as inputs to a deep convolutional neural network. The framework was rigorously tested using laboratory experiments to verify the classification of inter-turn short-circuits. The results reveal that appropriate signal preprocessing enhances feature representation quality, indicating that the choice of transform method directly impacts diagnostic accuracy. These findings provide evidence that the developed framework achieves higher precision, sensitivity and robustness, confirming its capability to detect early-stage multi-phase inter-turn short-circuits effectively. This research offers a novel perspective on noninvasive diagnostics by integrating advanced signal processing with deep learning classifiers. The study provides valuable insights into optimizing input data for DNN fault detection, contributing to the advancement of reliable condition monitoring systems for industrial induction motors.