Quantifying stator winding fault severity in induction motors: A machine learning approach with spectral feature extraction

Induction motors (IMs) constitute a fundamental component of industrial infrastructure, accounting for approximately 60%–70% of global industrial energy consumption and driving critical manufacturing processes. The proliferation of electrical faults in electric motors, particularly in their incipient stages, presents significant challenges to operational reliability and energy efficiency, necessitating advanced fault diagnostic methodologies. The challenge lies in detecting the incipient fault in its early stages, as there is minimal damage and it may not be easily detectable. This paper introduces a novel machine learning-based framework for quantifying stator winding fault (SWF) severity in IMs through multi-domain signal analysis. The proposed methodology implements a hybrid feature extraction approach, integrating Fourier and wavelet transform coefficients derived from three-phase current signals to capture both frequency-domain characteristics and localized time–frequency patterns indicative of electrical degradation, along with statistical features. The experimental validation incorporates a diverse range of machine learning architectures, spanning traditional algorithmic approaches such as Support Vector Machines (SVMs) and advanced ensemble methodologies including Histogram Gradient Boosting (HGBM) and XGBoost (XGB). The proposed framework demonstrates superior fault detection capabilities, achieving classification accuracy exceeding 95% across multiple evaluation metrics including precision (0.9952), recall (0.9949), and F1-score (0.9952), which is an extremely high score for a multiclassification model of four classes in signals. Furthermore, SHapley Additive exPlanation (SHAP) is used to recognize the important features that led to the success of the models. The obtained results establish the efficacy of machine learning techniques in facilitating early fault detection and enabling condition-based maintenance strategies, thereby enhancing operational reliability in industrial applications. To promote reproducibility and facilitate further research in this domain, the complete implementation framework, including source code, is made publicly available in https://github.com/rehaan-hussain/Quantifying-Electrical-Severity-in-Induction-Motors-A-Machine-Learning-Approach.