Current and Stray Flux Sensors for Anomaly Detection in PMSM Drive Based on Gradient Boosting Machine

Detecting inter-turn short-circuit faults (ITSC) and local demagnetization (LD) in permanent magnet synchronous motors (PMSMs) at low severity is very challenging during dynamic operations. This paper proposes a learning based classifier for detecting abnormality in a PMSM caused by ITSC and LD using external stray flux and motor current sensors. Features of the classification are extracted based on power density estimation to enhance the accuracy of the learning model. Within this framework, 4 supervised learning methods are tested: Adaptive Boosting (AdaBoost), Light Gradient Boosting Machine (Light-GBM), Categorical Boosting (CatBoost) and Extreme Gradient Boosting (XGBoost). All algorithms are trained on datasets from one operational profile and then tested on other operational profiles. The proposed scheme is validated using data from an in-house experimental setup.