Cloud-Edge Federated Incremental Learning Framework for PMSM Efficiency Optimization with Lightweight CNN-LSTM Models and OTA Differential Deployment

Yilin Yao; Wenchao Zhang; Mengtong Li

doi:10.63313/JCSFT.9063

Authors

Yilin Yao International Business School, Henan University, Henan, China Author
Wenchao Zhang Trine University, Allen Park, MI, USA Author
Mengtong Li School of Engineering and Applied Science, Columbia University in the City of New York, New York, NY, USA Author

DOI:

https://doi.org/10.63313/JCSFT.9063

Keywords:

PMSM, Deep Learning, CNN-LSTM, Federated Learning, Incremental Training, Cloud-Edge Computing, OTA Deployment, Energy Efficiency Optimization

Abstract

Permanent magnet synchronous motors (PMSMs) are critical components in industrial automation and energy-efficient drive systems, where dynamic operating conditions, load variations, and equipment aging can significantly affect efficiency. Traditional fixed-parameter controllers or static models often fail to maintain optimal performance under these time-varying conditions. Leveraging advances in cloud computing and distributed learning, this study proposes a cloud-edge federated incremental learning framework specifically designed for PMSM efficiency optimization. The framework integrates lightweight CNN-LSTM models deployed on resource-constrained motor controllers for real-time inference with a cloud-based model repository that performs incremental training and federated aggregation, enabling continuous adaptation to diverse operational scenarios. To ensure safe and efficient large-scale deployment, a secure over-the-air (OTA) differential update mechanism with version management and gray-scale rollout is implemented, minimizing communication overhead while supporting heterogeneous terminal devices. Experimental evaluations conducted on a large-scale PMSM dataset collected from 50 industrial motors show that the proposed framework achieves the best overall performance among all compared models. The Lightweight CNN-LSTM with Incremental Federated Learning and OTA Differential Deployment attains a mean absolute error of 0.412, an RMSE of 0.611, and an R^2 value of 0.973, outperforming both non-federated and conventional baseline models. Furthermore, the proposed architecture reduces parameter size to 109 K and achieves the lowest edge-side inference latency of 8.9 ms, demonstrating its suitability for large-scale deployment on resource-constrained PMSM controllers. These results indicate that the framework not only enables lifelong learning and self-adaptive optimization of PMSM control but also provides a scalable and reliable solution for distributed industrial applications, bridging cloud-edge collaboration, federated learning, and real-time embedded inference.

References

[1] Holtz J. Sensorless control of induction motor drives[J]. Proceedings of the IEEE, 2002, 90(8): 1359-1394.

[2] Morimoto S, Tong Y, Takeda Y, et al. Loss minimization control of permanent magnet synchronous motor drives[J]. IEEE Transactions on industrial electronics, 2002, 41(5): 511-517.

[3] Ismail M M, Xu W, Ge J, et al. Adaptive linear predictive model of an improved predictive control of permanent magnet synchronous motor over different speed regions[J]. IEEE Transactions on Power Electronics, 2022, 37(12): 15338-15355.

[4] Quiroga J, Cartes D A, Edrington C S, et al. Neural network based fault detection of PMSM stator winding short under load fluctuation[C]//2008 13th International Power Electronics and Motion Control Conference. IEEE, 2008: 793-798.

[5] Jin L, Wang F, Yang Q. Performance analysis and optimization of permanent magnet synchronous motor based on deep learning[C]//2017 20th International Conference on Electrical Machines and Systems (ICEMS). IEEE, 2017: 1-5.

[6] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural computation, 1997, 9(8): 1735-1780.

[7] Eang C, Lee S. Predictive maintenance and fault detection for motor drive control systems in industrial robots using CNN-RNN-based observers[J]. Sensors, 2024, 25(1): 25.

[8] Liang Z, Wei W, Zhang K, et al. Research on multi-hop inference optimization of llm based on mquake framework[J]. arXiv preprint arXiv:2509.04770, 2025.

[9] Chen Z, Liu B. Continual learning and catastrophic forgetting[M]//Lifelong Machine Learning. Cham: Springer International Publishing, 2022: 55-75.

[10] McMahan B, Moore E, Ramage D, et al. Communication-efficient learning of deep networks from decentralized data[C]//Artificial intelligence and statistics. PMLR, 2017: 1273-1282.

[11] Zhang W, Li X, Ma H, et al. Federated learning for machinery fault diagnosis with dynamic validation and self-supervision[J]. Knowledge-Based Systems, 2021, 213: 106679.

[12] El Jaouhari S, Bouvet E. Secure firmware Over-The-Air updates for IoT: Survey, challenges, and discussions[J]. Internet of Things, 2022, 18: 100508.

[13] Mehar M, Waghole A, Bharti A, et al. Over the Air (OTA) Update System–A Systematic[J].

[14] Chen Z, Larsson E G, Fischione C, et al. Over-the-air computation for distributed systems: Something old and something new[J]. IEEE Network, 2023, 37(5): 240-246.

[15] Ning Z, Zeng H, Tian Z. Research on data-driven energy efficiency optimisation algorithm for air compressors[C]//Third International Conference on Advanced Materials and Equipment Manufacturing (AMEM 2024). SPIE, 2025, 13691: 1068-1075.

[16] Li T, Li H, Zhou Y. E-commerce Sentiment Analysis Using Fine-tuned LLaMA3 Models: A QLoRA-based Approach[J]. Journal of Technology Innovation and Engineering, 2025, 1(4).

[17] Fan P, Li H, Hu M. Profit-Oriented Production and Pricing Optimization for Manufacturing Enterprises Using Proximal Policy Optimization[J]. Economics and Management Innovation, 2026, 3(2): 8-17.