Distributed Edge-Fog-Cloud Architecture for Intelligent Fault Diagnosis of Permanent Magnet Synchronous Motors Using Temporal Fusion Transformers and Federated Learning

Permanent Magnet Synchronous Motors (PMSMs) are widely deployed in industrial manufacturing, where continuous high‑precision operation demands reliable and real‑time fault monitoring. However, traditional centralized diagnostic architectures struggle with increasing data volume, communication latency, and heterogeneous operational conditions across distributed PMSM installations. To address these challenges, this study proposes a distributed edge–fog–cloud intelligent diagnostic framework tailored for PMSM fault detection and predictive maintenance. The framework integrates on‑device lightweight sequence models for real‑time inference, fog‑level stream aggregation for regional anomaly assessment, and a cloud‑based Temporal Fusion Transformer (TFT) for long‑horizon multivariate forecasting. A federated learning orchestration mechanism is further employed to enable privacy‑preserving cross‑device model adaptation, reducing communication overhead while supporting large‑scale industrial deployment. Experimental results on a PMSM condition monitoring dataset show that the proposed distributed TFT-FL system improves fault prediction accuracy by 5.9% over the LSTM baseline and achieves significantly lower prediction error (MAE 0.118, RMSE 0.183). The results confirm the effectiveness and scalability of the distributed architecture for next‑generation industrial PMSM health monitoring and predictive maintenance.