M{\'e}thodologie de dimensionnement par optimisation bay{\'e}sienne d'une machine synchro-r{\'e}luctante assist{\'e}e d'aimants permanents

TL;DR

This paper compares fixed and Bayesian optimization methods to enhance the performance of a permanent magnet-assisted synchronous reluctance machine, demonstrating Bayesian methods' efficiency and direct simulation-based solutions.

Contribution

It introduces Bayesian optimization approaches for machine design, improving performance without additional verification compared to fixed model methods.

Findings

Bayesian methods outperform fixed models in performance.

Bayesian approaches require the same computational effort as fixed models.

Solutions are directly based on finite element simulations without verification need.

Abstract

In this article, three optimization approaches are exploited to improve the performance of a permanent magnet-assisted synchronous reluctance machine: a first optimization using fixed substitution models and two Bayesian optimization approaches based on adaptive substitution models. The results show that Bayesian approaches lead to machines with better performance using the same computation time (same number of finite element simulations). Unlike optimization methodologies based on fixed substitution models, Bayesian approaches provide solutions directly based on finite element simulations and, therefore, do not require verification.