Machine Learning-Based Bearing Fault Detection in VFD-Driven Induction Motors for Predictive Maintenance

Bearing faults are among the most prevalent causes of failure in induction motors, often leading to unplanned downtime and maintenance costs. Traditional fault detection techniques such as vibration analysis or acoustic emission methods require intrusive sensor placement and are often sensitive to environmental noise, limiting their effectiveness in industrial environments. This study presents a cost-effective, non-intrusive approach for early detection and classification of bearing faults in a Variable Frequency Drive (VFD)-powered induction motor using phase current signals and machine learning (ML) techniques. A 200 W asynchronous motor with 6202 bearings was operated under variable-speed conditions and current signals from each phase were acquired using current transformers (CTs), filtered, and digitized via an Arduino-based signal conditioning setup. A dataset comprising 60 K plus samples for each fault type—including healthy bearings (HB), ball fault (BF) and race fault (RF)—was constructed and pre-processed using MATLAB. Multiple classifiers were trained and evaluated, including Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Linear Discriminant Analysis (LDA) and Gaussian Naıve Bayes (GNB). Among these, the tuned SVM model achieved superior performance, with an accuracy, precision, F1 score respectively 97.3%, 97.4% and 97.3%. These results confirm the feasibility of using stator current signals for robust and real-time bearing fault diagnosis under VFD excitation, offering a scalable and reliable predictive maintenance solution for industrial settings.