Transfer Learning-Assisted Analytical Quasi-3D Surrogate Modeling for Electromagnetic Performance Prediction of AFPMs Considering Eccentricity

Axial flux permanent magnet motors (AFPMs) are prone to static and dynamic eccentricities due to assembly and manufacturing tolerances, which result in nonuniform air gaps. Such eccentricities distort the air gap flux density and consequently affect electromagnetic performance, including back EMF and cogging torque. In addition, tolerance analysis and robust design under eccentric conditions require repeated performance evaluations. However, high-fidelity 3D FEA entails a very high computational cost. This paper proposes a surrogate modeling framework that predicts electromagnetic performance using a quasi-3D analytical method considering eccentricity and calibrates the analytical predictions to approach the 3D FEA results by applying transfer learning. The proposed approach improves the computational efficiency of AFPM tolerance analysis under eccentric conditions and provides a practical electromagnetic performance prediction model applicable in the design stage.