A Screening Method for Customizing Designs Around Non-Convergent Regions of Design Spaces

Abstract

In engineering design, Response Surface Methodology is often used to explore a design space and to generate a meta-model of the space. Designs of Experiments are used in this process to obtain the most behavioral information from the space with the fewest number of runs. Often, the designer discovers that the analysis program cannot converge to an answer for one or more regions of the design space. Traditionally, the designer gets around this problem by shrinking the variable ranges so that there are no regions of non-convergence. Instead of this process, this paper suggests that it is better to treat the non-convergent points as a region of infeasibility. Extra runs can be made to determine the border of this infeasible region, and then a custom Design of Experiments can be constructed to explore the design space while excluding the non-convergent region. This process still generates a valid meta-model yet retains more of the design space than the simple reduction of variable ranges.