Robust Optimal Design for Worst-Case Tolerances

Abstract

Engineering models can and should be used to understand the effects of variability on a design. When variability is ignored, brittle designs can result that will not function properly or that will fail in service. By contrast, robust designs function properly even when subjected to off-nominal conditions. There is a need for better analytical tools to help engineers develop robust designs. In this paper we present a new approach for developing designs that are robust to variability induced by worst-case tolerances. An advantage of this approach is that tolerances may be placed on any or all model inputs, whether design variables or parameters. The method adapts nonlinear programming techniques in order to determine how a design should be modified to account for variability. We tested the method under relatively severe conditions on 13 problems, with excellent results. Using this approach, a designer can account for the effects of worst-case tolerances, making it possible to build robustness into an engineering design.