Response Surface Model Evolution
Alexander I. J. Forrester, Neil W. Bressloff, Andy J. Keane · 16th AIAA Computational Fluid Dynamics Conference · 2003
Methods are presented for reducing time and effort when performing aerodynamic optimisation using response surface models. Significant time savings are made possible by monitoring the convergence of computational fluid dynamics simulations and omitting regions of poor designs. In so doing, optimal regions of the design space can be highlighted and surface refinement commenced early in the convergence of the design point set. A strategy employing surface updates with new data at points of maximum expected improvement is shown to perform more efficiently than reducing the design space to the region of the optimum. The response surface evolution methods are demonstrated through an example two parameter optimisation of a flap track fairing on a commercial airliner wing.