A Multi-Fidelity Bayesian Optimization Framework for Hypersonic Vehicle Design

We propose a multi-fidelity computational framework for the preliminary design of hypersonic waveriders. The framework implements a two-fidelity Bayesian Optimization strategy merging cheap inverse design methods with expensive computational fluid dynamics models. Namely, the Osculating Cone method is employed to generate low-fidelity data, whereas the compressible Euler model with anisotropic mesh adaptation is used to generate high-fidelity data. A novel parametrization is proposed to enable direct control on the geometry, ensuring straightforward compliance between different methods, and seeking a reliable and flexible approach to the design of the vehicle. The optimization framework is tested on two use cases.