Rapid Optimization Framework for Axial Flux Permanent Magnet Motors Using Initial-State FEA and Surrogate Modeling

Abstract

The optimal design of axial flux permanent magnet synchronous motors (AFPMMs) relies on three-dimensional finite element analysis (3D FEA) to accurately evaluate nonlinear electromagnetic characteristics. However, the high computational cost of 3D FEA limits its applicability to large-scale design optimization. To address this issue, this paper proposes a computationally efficient optimization framework that integrates initial-state finite element analysis (ISF), light gradient boosting machine (LightGBM) surrogate modeling, and genetic algorithm (GA) optimization. ISF evaluates only the initial rotor position, significantly reducing the computational effort required for electromagnetic analysis while maintaining acceptable accuracy. The generated dataset is used to train surrogate models that predict motor performance, thereby replacing repeated FEA evaluations during optimization. The proposed framework is validated through comparison with full periodic FEA and experimental measurements obtained from a prototype motor. Results show that ISF reduces computation time by more than 90% while maintaining a torque error within 1%. In addition, the surrogate models achieve high prediction accuracy with R2 values exceeding 0.98. The optimized design improves average torque and power density by 1.67% and 1.09%, respectively, while maintaining magnetic robustness under severe operating conditions. These results demonstrate that the proposed framework provides an effective and computationally efficient approach for AFPMM optimization.