MULTI-FIDELITY GAUSSIAN PROCESS MODEL FOR CFD AND WIND TUNNEL DATA FUSION
Résumé
In aerodynamic design, both Computational Fluid Dynamics (CFD) simulations and wind tunnel (WT) experiments deliver datasets that are complementary in terms of uncertainties and density of information. In many situations it is desirable to use methods that combine all available information while accounting for the strengths of both sources. In this paper, we combine the sources in a multi-fidelity Gaussian process (GP) model to foretell the aerodynamic forces and moments. The resulting model accounts for input-dependent error measurements in both experimental and simulated data. To validate the model, we construct a proper multi-source aerodynamic database containing CFD and WT data. This database is based on the NASA Common Reference Model. The CFD simulations are based on Reynolds-averaged Navier-Stokes (RANS) equations. We demonstrate in numerical settings that the suggested multi-fidelity GP framework outperforms the single-fidelity one in terms of prediction accuracy at the highest level of fidelity (i.e. WT data). The resulting model also allows to reconstruct common aerodynamic profiles (e.g. representations of longitudinal forces) with uncertainties.
Origine : Fichiers produits par l'(les) auteur(s)