Online Characteristics Prediction for Permanent Magnet Traction Motors Based on Electromagnetic-Thermal Coupling Reduced-Order Model

To support a high-fidelity digital twin for permanent magnet motors, this paper presents a fast coupled electromagnetic-thermal modeling framework, which integrates a temperature-aware electromagnetic reduced-order model (ROM) and a sixth-order lumped parameter thermal network (LPTN). The electromagnetic ROM is constructed using proper orthogonal decomposition (POD) and neural networks (NN), with a key contribution being an integration of transfer learning and Kriging interpolation that significantly reduces the training data required to generalize the electromagnetic ROM across temperature variations. Thermal parameters of the LTPN are efficiently identified via particle swarm optimization (PSO) under dynamic multi-condition operating profiles. Experimental validation shows that the proposed framework performs coupled field simulation within 0.8 seconds, achieving high accuracy—electromagnetic performance errors remain below 4%, and critical temperature predictions deviate by less than 5°C from measured values. Owing to its computational efficiency and lightweight architecture, the model is readily deployable on edge devices, demonstrating strong potential for industrial digital twin applications such as predictive maintenance and real-time online monitoring.