A Computationally Efficient Surrogate Model Based Robust Optimization for Permanent Magnet Synchronous Machines

One of the main obstacles to the robust optimization for permanent magnet machines is the high computational burden, which is mainly caused by the robustness evaluation process considering the manufacturing uncertainties. In this paper, a sequential sampling Kriging model is adopted to estimate the design robustness, and the problem of high-dimensional variables for the surrogate model due to uncertainties is avoided through adopting the worst-case based approach. To further reduce the number of finite element analyses (FEA) required for constructing a metamodel, a multi-points sequential sampling process with a two-step optimization oriented surrogate model updating algorithm is proposed. Compared with the FEA directly based robust optimization in previous work, similar Pareto Fronts are achieved by adopting the meta-model proposed algorithm but the overall run time reduced by 75%. In the end, an optimization problem for a 1.2 kW motor is considered by applying the proposed algorithm, and two prototypes with deliberately designed tolerances, imitating the worst-case scenarios in the real world, are manufactured. The worst-case torque ripple is reduced from 7.6% to 4.8% after optimizing, and this verifies the efficacy of the proposed algorithm.