Precise Temperature Prediction for Magnetically Suspended Permanent Magnet Synchronous Motor Based on Few‐Shot Transfer Learning

With increasing speed of high‐speed magnetically suspended permanent magnet synchronous motors (MSPMSMs), losses rise significantly, intensifying the risk of thermal runaway. Accurate temperature prediction is therefore essential. Conventional methods rely on extensive experimental data, which are unavailable in the early stage where only low‐speed, light‐load tests can be conducted. To address this limitation, a few‐shot transfer learning method is proposed in this paper to predict full‐speed, full‐load temperatures using limited experimental data. An accurate reduced‐order lumped‐parameter thermal network (LPTN) model was identified for an existing prototype MSPMSM to generate a large current–speed–temperature dataset, serving as the source domain for model training. The target MSPMSM was then fine‐tuned with its low‐speed, light‐load temperature data samples, enabling transfer of the source model to the target model. Experimental validation confirms that the proposed method accurately predicts MSPMSM temperatures under full‐speed, full‐load conditions, with errors less than 5%. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.