A One-Dimensional Exact Optimization Framework for Permanent Magnet Synchronous Machine Design

초록

Permanent magnet synchronous machines (PMSMs) exhibit complex and non-differentiable characteristics that hinder the application of derivative-based optimization methods, leading to the widespread use of metaheuristic algorithms. However, these methods face critical limitations, such as the lack of optimality guarantees and inconsistent results due to their stochastic nature. This study introduces a novel one-dimensional (1-D) exact optimization framework that eliminates reliance on metaheuristics. A minimal multilayer perceptron (MLP) training phase accurately predicts split and aspect ratios, enabling a structured optimization process where air gap diameter becomes the sole design variable. The proposed framework consists of two sequential stages: stator design using predicted ratios and rotor design via the Levenberg-Marquardt method. This approach transforms the PMSM design process into a deterministic optimization problem, guaranteeing consistent and repeatable outcomes. The results derived from the proposed method are validated through comparison with finite element (FE) and experimental results.

키워드