Abstract:To address the precision limitation of traditional thermal models and the computational inefficiency of electromagnetic-thermal coupling finite element models, this work proposes a multidimensional correction-based electromagnetic-thermal coupling design optimization method for electric machines. Firstly, this method corrects the sensitive parameters in the analytical lumped parameters thermal network model (LPNM) based on prior information from historical experimental data to ensure the accuracy for thermal performance analysis. Subsequently, to enhance computational efficiency, a large number of low-fidelity (LF) samples are obtained using the corrected thermal model in a serial electromagnetic-thermal field calculation manner, complemented by a small number of high-fidelity (HF) samples obtained from the electromagnetic-thermal coupling simulation model. Then, by employing a transfer learning technique, the data information from the low- to high-fidelity sample sets is comprehensively utilized to train a multi-fidelity (MF) surrogate model for the final optimization. The proposed method applied to a high-torque-density aerospace propulsion permanent magnet synchronous motor (PMSM) with short-term strong overload capability, it not only significantly improves the accuracy of LPNM, but also reduces the training sample acquisition time by over 50% under the same model accuracy. Finally, a prototype is tested to validate the effectiveness of the proposed method.

Multi-Objective Optimization of Electromagneticthermal Coupling Performance for High-torquedensity PMSM Based on Multi-dimensional Correction Techniques