Multi-Objective Bayesian Optimization of Squirrel-Cage Induction Machine

In electrical engineering, a design of electrical machine using numerical methods, such as Finite element method, is a common practice. Electrical machines are complex multi-physical systems where for finding the optimal sets of designs solutions, called Pareto fronts, a very effective approach is to use multi-objective optimization. The most popular method for multi-objective optimization of machine design is the use of numerical optimization algorithms such as NSGA-II. However, due to the time-consuming nature of induction machines simulations, this approach is not very effective. This paper addresses this issue by proposing machine learning as a solution, specifically utilizing Multi-objective Bayesian optimization. This optimization method has been used in many industries as an efficient global optimization of the modeled system. By using the right acquisition function, the search space can be efficiently navigated to find the optimal candidates. Moreover, the optimization requires only a limited number of samples. The main aim of this paper is to present this method, which is demonstrated on the optimization of a 1.5 kW induction machine with time-consuming calculations. The machine optimization approach is not the main focus here, as this method can be effectively applied to any machine design or even any optimization approach. Furthermore, two possible approaches of machine optimization using this method are presented here.