A New Robust Dominance Criterion for Multiobjective Optimization

This paper discusses robust design issues in the multiobjective design of electromagnetic devices. A Pareto front of the worst cases of the solution due to the perturbation of the design variables is used to describe the robustness of a design. A robust dominance (r-dominance) relationship between two solutions of a multiobjective problem is defined. To reduce the computational cost of evaluating the robustness, a local response surface model is built on the uncertainty set. This robust multiobjective design scheme was applied to the design of an interior permanent magnet machine, where a tradeoff exists between the average torque and the torque ripple as the rotor angle is varied. The results show that the proposed r-dominance, when embedded in a multiobjective optimization search method, is able to find robust solutions to multiobjective design problems.