Thermal Fault Analysis and Optimisation of Induction Motors Using Current and Temperature Signals with Artificial Intelligence

Abstract

Induction motors are widely used in industry, yet thermal stresses caused by electrical and mechanical losses reduce their reliability, efficiency, and service life. This study presents an autoencoder-based, non-intrusive fault-detection framework that utilises synchronised current and bearing-housing temperature signals. Data quality measures, min-max normalisation, and sliding-window segmentation support robust time-series learning of normal operating patterns. Implemented in TensorFlow/Keras and trained over 12 epochs using the Adam optimizer with L2 regularisation, the model achieved stable convergence by the sixth epoch, with training loss at 0.0345 MSE and validation loss at 0.0265 MSE. Anomalies were detected through reconstruction errors, applying a 97th percentile threshold of 3.5288 to distinguish normal behaviour from fault conditions. Performance metrics demonstrate strong diagnostic ability, including a True Positive Rate of 94.5%, False Positive Rate of 4.3%, Precision of 95.1%, Recall of 94.5%, and an F1-score of 94.8%. The results confirm reliable detection of current and temperature deviations, enabling early fault intervention, reducing unplanned downtime, and improving energy efficiency. The approach generalises well across various operating conditions and offers a scalable foundation for intelligent condition monitoring of industrial motor fleets.