An Efficient Multi-Objective Bayesian Optimization Approach for the Automated Analytical Design of Switched Reluctance Machines

The design and optimization of electrical machines are often formulated as multi-physics and multi-objective problems that are complex and computationally intensive. The Bayesian optimization framework has been proven to be a powerful optimization tool that is able to take advantage of the information from prior optimization history and make an optimal search more efficient. Specifically, Bayesian optimization algorithms have been widely applied in optimization problems, especially in the field of machine learning and computer vision where evaluations of the objective functions are costly or the derivatives are often not accessible. Because of their similarities to the electrical machine design and optimization problems, this paper employs an expected-improvement-based Bayesian optimization approach for the automated analytical design of switched reluctance machines, and the expected improvement is selected as the acquisition function to balance exploration and exploitation throughout the optimization process. The optimization results demonstrate that compared to the popular and benchmark NSGA-II algorithm, the proposed Bayesian optimization method is more effective and efficient in terms of finding design candidates of better qualities that dominate the solutions generated by NSGA-II in a small number of evaluations.