کنفرانس بین المللی ماشین و محرکه های الکتریکی

صفحه اصلی / The 5th International Conference on Electrical Machines and Drives

Explainable Machine Learning for Multiclass Fault Diagnosis in Inverter-Driven Permanent Magnet Synchronous Motors

نویسندگان :

Hooman Aminzadeh Vahedi¹ Mahsa Panahi Azar²

1- دانشگاه تهران 2- دانشگاه تهران

کلمات کلیدی :

Permanent Magnet Synchronous Motor،Fault Diagnosis،Explainable Artificial Intelligence،Random Forest،Condition Monitoring

چکیده :

Despite their high efficiency and torque density driving widespread use in renewable energy and electric mobility, Permanent Magnet Synchronous Motors (PMSMs) can suffer significant performance and reliability degradation from inverter and machine faults. This paper employs an explainable machine learning framework for multiclass fault diagnosis in inverter-driven Permanent Magnet Synchronous Motors (PMSMs). A Random Forest classifier is paired with two model-agnostic explanations: Permutation Importance and Individual Conditional Expectation (ICE) and delivers both accuracy and transparency with its introduced pipeline. The method is evaluated on a public dataset of 10,892 samples spanning nine operating conditions (normal, open/short-circuit, and overheating) achieving robust classification and clear sensor-level insights. The trained model indicates that thermal features dominate discrimination, while electrical variables contribute less under the tested conditions. The approach yields actionable explanations suitable for real-time condition monitoring and maintenance of PMSM drives.