Fast Prediction of Air-Gap Magnetic Flux Density and Electromagnetic Torque in Surface-Mounted Permanent Magnet Synchronous Motors

Accurate prediction of electromagnetic field distributions in Surface Mounted permanent-magnet synchronous motors (SPMSMs) is fundamental to achieving high-efficiency and high-performance designs. However, conventional finite element analysis (FEA) is computationally intensive and unsuitable for large-scale performance optimization. This paper proposes a fast prediction framework based on a conditional generative adversarial network (Pix2Pix) that integrates residual U-Net (ResUNet) and attention mechanisms to predict radial and tangential magnetic flux densities (Br and Bt) from motor geometry and current density maps. The proposed model introduces a four-channel input structure combining RGB geometry images with grayscale current-density maps to represent excitation conditions. A large-scale dataset comprising 16,650 paired samples across 31 temporal frames and multiple current levels was constructed, simulating realistic motor operation under dynamic excitation. Experimental results show that the proposed ResUNet-Attention Pix2Pix network achieves prediction accuracy exceeding 99% for Br and Bt distributions and torque estimation compared with FEA, while also demonstrating strong generalization ability across unseen SPMSM topologies (flat, radial, breadloaf). This approach provides a robust and precise framework for advanced electromagnetic characterization in SPMSM analysis, with cross-topology generalization validated across unseen SPMSM configurations.