Cokriging for Robust Design Optimization
Andy J. Keane · AIAA Journal · 2012
This paper addresses the problem of robust design optimization. Such formulations are inevitably multi-objective because the designerwants goodperformance andalso small variations in that performance. The desire for processes that produce robust designs stems from the observation that if only nominal performance is considered during design optimization, sensitive designs often result, and these commonly fail to meet objectives when the inevitable uncertainties of manufacture, operating conditions, and degradation in operation are considered. It is assumed that design is carried out using analysis codes that are expensive to run. Because of this and the need for themultiple calls associatedwithMonteCarlomethods, use ismade of surrogate-based optimization tools to speedup the search.Here, themethodology of cokriging is examined to allow results fromusing differing numbers ofMonteCarlo samples to be simply combined. The primary aim was avoid always having to use large numbers of samples in the Monte Carlo assessment of design robustness. The application of these methods is illustrated by considering a gas-turbine compressor blade optimization, inwhich a range of shape errors are considered that simulate foreign object damage, erosion damage, and manufacturing errors. Consideration is also given to variation in operating conditions.