Systematic review of artificial intelligence for thermal and fault state estimation in electric motor drives

Abstract

Electric motor drives are increasingly used in electric vehicles, industrial automation, robotics, renewable energy systems, aerospace actuators, and smart manufacturing. Their reliable operation requires accurate fault diagnosis, robust thermal supervision, and timely health state estimation under variable operating conditions. Artificial intelligence and deep learning models have been widely applied to induction motor fault diagnosis, permanent magnet synchronous motor fault detection, demagnetization assessment, inverter fault diagnosis, anomaly detection, and temperature prediction. However, many studies still emphasize model accuracy while giving limited attention to dataset transparency, validation reliability, data leakage, operating-profile dependency, and real-time deployment constraints. This paper systematically reviews AI-enabled thermal and fault state estimation methods for electric motor drives, with emphasis on permanent magnet synchronous motors and induction motors. Studies are compared by motor platform, monitoring target, signal modality, model family, dataset source, validation design, leakage-control evidence, generalization setting, and implementation feasibility. Particular attention is given to train-test contamination mechanisms in correlated motor time series, including window overlap, preprocessing fitted before data separation, operating-profile overlap, and feature-selection leakage. The synthesis indicates that trustworthy motor monitoring depends less on isolated accuracy gains than on transparent datasets, group-aware validation, training-only preprocessing, evaluation under unseen conditions, physically credible modeling, and reporting of hardware-relevant metrics. The review concludes with a research agenda for integrated thermal-fault monitoring that combines robust validation, interpretable learning, compact models, and real-time motor-drive deployment.