Supervised Machine Learning for Inter-Turn Short-Circuit Detection in Field-Oriented Control Induction Motors

Abstract

Owing to the widespread use of induction motors, early detection of inter-turn short-circuit faults is essential for predictive maintenance and asset management. However, detecting incipient faults in motors operating under field-oriented control remains challenging, because fault signatures can be masked by the actions of the closed-loop control system. This paper presents a systematic evaluation of feature extraction, feature selection, and supervised machine-learning techniques for detecting incipient inter-turn short-circuit faults in field-oriented control induction motors. Multiple signal representations were investigated, and four feature selection methods, including a previously unreported cascaded correlated neural network-based strategy, were comparatively analyzed. The developed models were trained and validated using simulation data, and subsequently tested using experimental measurements, enabling the assessment of simulation-to-experiment generalization. In addition, the effects of the sampling frequency and signal observation window length on the diagnostic performance were investigated by considering the constraints of industrial signal-processing devices.