Hybrid Bayesian–Adjoint Framework for Aerodynamic Shape Optimization

A hybrid framework for aerodynamic shape optimization is presented. The framework implements a sequential procedure, exploration-then-exploitation, coupling Bayesian global optimization with an adjoint-based gradient method. Gradient-based methods efficiently refine a design but are inherently sensitive to the initial point and often converge to a local minimum. To address this limitation, the Bayesian phase first explores the design space globally. A well-performing design is identified and serves as the initial point for the subsequent gradient-based phase, improving the chances of reaching a better optimum. A probabilistic metric governs the transition between global and local optimization, ensuring efficiency. This approach is particularly suited for multimodal aerodynamic design problems. Here, it is employed to design representative aeronautical use cases, namely, airfoils and wings.