In designing electrical machines, multiple performance objectives must be considered, often requiring time-consuming evaluations with stochastic algorithms. Bayesian Optimization (BO) provides an efficient alternative, particularly for costly objective evaluations, using probabilistic surrogate models based on Gaussian Process Regression (GPR) for high accuracy. Despite its potential, BO is rarely used in electric machine design. This study explores its application for optimizing a reluctance synchronous machine with 14 design variables. Compared to the common NSGA-II, BO achieves significantly better outcomes in much less time.

Access the paper here.