A Hybrid Bayesian-Adjoint Framework for Aerodynamic Shape Optimization
Alberto Perlini, Luca Abergo, Giulio Gori · 2024
A hybrid aerodynamic design framework coupling a Bayesian global optimization approach to an adjoint-based gradient method is presented. The aim is to merge the explorative capabilities of Bayesian methods with the exploitative capabilities of gradient-based approaches. A sequential procedure exploration-then-exploitation is presented. In between these two phases, the design space is reparametrized to fulfill a twofold objective. In the exploration stage, only a limited number of design variables (DVs) is considered to alleviate the dimensionality issues typical of surrogate-based methods. In the exploitation stage, the reparametrization increases the degrees of freedom to take full advantage of the discrete adjoint-based approach, which cost does not scale with the number of DVs. The framework is first verified using academic tests. Subsequently, its effectiveness is assessed by considering two typical aeronautics applications, confirming an improvement of the overall design process.