Abstract
This paper discusses the effectiveness of cumulative function approximation for improving the computational cost of robust optimal design. Since robust optimal design constructs intermediate model to evaluate the robustness of a design within distribution region, it requires much computational cost for executing sampling to capture the variation of the objective function and constraints within the distribution region. Cumulative function approximation, which is one of response surface methods, initializes approximations of functions and improves them through iteration of optimization over the approximated functions and adaptive addition of sample points based on the optimization result. Its effectiveness is much more outstanding for robust optimal design than for nominal optimal design because of substituting executing sampling over approximated functions instead of executing system analysis. After describing the concept for integrating the cumulative function approximation with robust optimal design, this paper shows its effectiveness through its approximation to various intermediate models of robust optimal design.
