Robust multi-fidelity aerodynamic design optimization using surrogate models

Abstract

We present a novel framework for robust aerodynamic design optimization with respect to CFD modeling errors using multi-fidelity simulations, a multi-objective optimization algorithm, and a surrogate model employed to map the error landscape across the design space. We use the low- and high-fidelity CFD model divergence as a proxy for simulation risk, which is simultaneously optimized along with a measure of performance. Instead of generating high-fidelity simulations directly, we employ a Sparse Pseudo-input Gaussian Process surrogate modeling algorithm to predict the divergence. We apply this approach to a simple diffuser design problem, coupled with a multi-objective Tabu Search optimization algorithm, which shows encouraging results. We are able to generate a range of Pareto optimal design, which display a trade-off between aerodynamic performance and simulation risk. This approach is applicable to more general problems and would be of interest in an industrial design setting.