کنفرانس بین المللی ماشین و محرکه های الکتریکی

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

Few-Shot Learning for Precise Excitation-Winding Fault Localization in Resolvers

نویسندگان :

Ali Pour Ghoraba¹ Zahra Nasiri-Gheidari²

1- ُSharif University of Technology 2- ُSharif University of Technology

کلمات کلیدی :

Resolver fault localization,،Hybrid model,،Few-shot learning,،Excitation-ITSC,،Metric learning

چکیده :

Resolvers play a critical role in rotor-angle sensing for high-performance electric drives, where even small measurement deviations can destabilize control loops and reduce system reliability. Among various failure modes, excitation-winding interturn short-circuit (ITSC) faults are particularly challenging to detect and localize due to their subtle electromagnetic signatures and strong dependence on operating conditions. This paper proposes a physics-informed, few-shot learning framework for precise localization of excitation-winding ITSC faults using minimal labeled data. A validated hybrid resolver model—96% faster than full finite-element analysis and within 13% of its accuracy—serves as a high-fidelity data engine to synthesize diverse, labeled fault samples under varying severities and operating conditions. These synthetic signals train a Siamese metric-learning network that operates directly on raw sine–cosine resolver voltages, learning a robust similarity space that generalizes to unseen fault locations with only a few exemplars per class. Experimental and simulation results demonstrate that the proposed approach achieves accurate, noise-tolerant fault localization while dramatically reducing data and computational requirements. By uniting model-based synthesis with few-shot inference, the framework enables scalable, data-efficient, and field-deployable diagnostics for resolver-equipped electric drives.