An Efficient Serial-Loop Strategy for Reliability-Based Robust Optimization of Electromagnetic Design Problems

An efficient reliability-based robust design optimization (RBRDO) method is proposed to substantially improve computational efficiency without sacrificing numerical accuracy when applied to electromagnetic design problems in the presence of uncertainties. Unlike the conventional parallel-loop optimization program architecture, robust optimization and reliability assessment are totally decoupled from each other during the overall design process. To ensure prescribed design feasibility, constraint boundaries in an original RBRDO formulation are first shifted to feasible directions based on the reliability information obtained at a previous design cycle. Then, the first two statistical moments, mean and variance, of a quality loss function is estimated by the univariate dimension reduction method. After all, the proposed serial-loop methodology employs sequential cycles of equivalent robust design optimization with shifting constraint boundaries. A simple mathematical model and a blushless direct current motor design problem are provided to demonstrate the effectiveness of the proposed method.