Optimal design of hollow conductor for high‐speed synchronous motor exploiting adaptive‐sampling radial basis function algorithm

Ruiye Li,Peng Cheng,Hai Lan

DOI: https://doi.org/10.1049/elp2.12509

2024-10-12

IET Electric Power Applications

Abstract:This study introduces an adaptive sampling radial basis function (ASRBF)‐based multi‐objective optimisation design to efficiently optimise hollow conductors in high‐speed motors. By progressively populating samples and applying a probabilistic improvement approach, ASRBF reduces the number of function calls and avoids unnecessary explorations, achieving a 15% reduction in AC losses and enhancing heat dissipation and slot fullness with significantly lower optimisation costs. As aircraft electrification advances, permanent magnet synchronous motors (PMSMs) require higher power density and efficiency, but optimisation is hindered by high computational costs and resource consumption. To address this, the paper proposes a multi‐objective optimisation method based on adaptive sampling radial basis function (ASRBF). The ASRBF algorithm adaptively adds sample points by estimating expected improvements at prediction points, enabling the surrogate model to rapidly approximate the global optimum while significantly reducing function evaluations. It integrates optimisation objectives and constraints using probabilistic improvement techniques, enhancing robustness and convergence speed by avoiding excessive exploration of invalid regions. Mathematical test functions validate ASRBF's excellent performance in handling complex objective domains. Applied to high‐speed PMSM with hollow conductors, it aims to minimise AC losses while maximising slot fill factor and heat dissipation, resulting in a 15% reduction in losses and an increase in conductor heat dissipation area and slot fill factor, at one‐thousandth of the cost of the full factorial optimisation method. The ASRBF algorithm efficiently constructs surrogate models for multi‐dimensional, multi‐objective, non‐linear, and constrained problems, providing a powerful tool for comprehensive performance optimisation of complex systems such as motors.

engineering, electrical & electronic