I. Introduction
In designing practical electric machinery, it is fairly important to consider not only the performance but also the robustness with respect to the deviation of design variables. Recently, various robust optimization methods which can evaluate the robustness are proposed [1], [2]. In general, however, these methods require large computational costs because of additional field analyses for the evaluation of the robustness in the perturbation domain. In order to reduce these costs, the method which evaluates the robustness indirectly by considering the effect of the noise instead of additional field analyses is already proposed [3], [4].