Reasonable Design Space Approach to Response Surface Approximation

Abstract

Response surface approximations have become very popular as a tool for multidisciplinary optimization. However, the accuracy of these models is a major concern. One method to increase the accuracy of response surface approximations is to construct them only for the portion of the design space that represents reasonable designs. Our paper describes the methodology underlying this reasonable design space approach. This methodology is illustrated using two examples involving multidisciplinary optimization of a high-speed civil transport model. In these examples, low-fidelity analyses are used to estimate the boundary of the reasonable design space. Within the reasonable design space, response surface models are constructed, based on results from higher-fidelity analyses. The accuracy of the response surface models created with and without using the reasonable design space approach is compared, and the advantages of employing response surface models created via the reasonable design space approach are demonstrated.