The dependability of three-phase induction motors is vital in industrial applications, requiring accurate and robust fault detection strategies. This paper introduces an automated diagnostic framework for two critical electrical faults: Open Circuit and Unbalanced Voltage faults. Three machine learning classifiers: Artificial Neural Networks, Support Vector Machines, and K-Nearest Neighbours were employed to develop and evaluate the models. A comprehensive dataset of approximately 100,000 samples was collected under different load scenarios (Full, Half, and No load), incorporating important features such as stator currents, rotor speed, and electromagnetic torque. Hyperparameter tuning was performed using Grid Search to enhance model generalization and optimize performance. Standard evaluation metrics, including the confusion matrix, ROC curve, F1-score, precision, and recall, confirmed highly reliable results. All models achieved 100% classification accuracy on both training and unseen test datasets across all load conditions. These outcomes demonstrate the effectiveness of machine learning in building load-independent fault diagnosis systems, offering significant potential for predictive maintenance in industrial environments. |
