Surrogate-Based Modeling of Induction Machines to Reduce the Computational Burden of Robust Multi-Objective Optimization

One of the main obstacles to robust design optimization of Induction Machines (IM) is the high computational burden, which is mainly due to time-intensive nonlinear finite element (FE) simulations. The large multivariable design space of electric machine optimization typically requires running thousands of simulations taking many hours, if not days which is quite prohibitive. To overcome this bottleneck, hybridization of FE-based optimization with approximate models can lead to expedite the process. This paper is focused on accelerating the typical FE-based optimization scenarios by implementing and systematically studying an ensemble of surrogate models of IMs in terms of computational burden and performance. In this regard, after adopting the most significant surrogate model, a multi-points sequential sampling process with a two-step surrogate-based optimization approach is developed. Compared with direct FE-based robust optimization, competitive results are achieved by adopting the proposed hybrid surrogate-based approach and the overall runtime is reduced by 69%. Furthermore, as a case study, an optimization problem for an 11-kW IM is considered by applying the typical FE-based optimization task followed by the proposed hybrid technique. Hence, the achievable speed improvements, as well as further possible enhancing means are discussed. The detailed comparison of the presented surrogate models makes a comprehensive source for engineers and designers to follow.