ScholarWorks@성균관대학교

초록

Permanent magnet assisted synchronous reluctance motors (PMa-SynRMs) are promising traction candidates due to their high efficiency, power density, and reduced reliance on rare-earth magnets; however, multi-objective design in large variable spaces demands extensive Finite Element Method (FEM) evaluations. We propose a dual-output surrogate framework that decouples feasibility classification from performance regression. A classifier first filters infeasible designs induced by geometric and physical constraints; the regressor then predicts key metrics (energy consumption, torque, torque ripple) only on feasible samples. This separation preserves physical validity near active constraints and improves sample efficiency in high-dimensional searches. The approach is validated on FEM-generated datasets of a traction-scale PMa-SynRM, where it reliably reconstructs the constrained Pareto structure-including the knee region relevant to trade-offs-while maintaining robust discrimination of infeasible designs. Overall, the framework reduces reliance on expensive FEM calls without sacrificing decision quality, providing a practical foundation for scalable, constraint-aware optimization of railway traction motors.

키워드