Robustness criteria for concurrent evaluation of the impact of tolerances in multiobjective electric machine design optimization

This article is about a comparison of different measures for determining the robustness or reliability of electric machine designs in the presence of inevitable tolerances. The selected criteria shall be suitable for concurrent evaluation in the course of solving state-of-the-art large scale multi-objective optimization problems. In the past, besides particularly customized criteria, mainly gradient based measures, worst case information, or standard deviation based quantities were considered. In this work, the quantile measure is introduced for electric machine design optimization and compared with the existing solutions. The evaluation of a design's robustness is typically examined based on finite element simulations. As for most measures a significant number of parameter combinations and thus computations are required, a surrogate model assisted approach is presented to minimize computational effort and runtime. A test problem is defined and analyzed to illustrate the differences of selected robustness measures. Results reveal the importance of considering robustness in the optimization process. Moreover, a careful choice of appropriate measures has to be taken. Selected designs are compared and conclusions and an outlook on future activities are presented.