Bayesian Models for Uncertainty Estimation in Aerodynamic Databases of Reusable Launch Vehicles

Abstract

The definition of Aerodynamic Databases (AEDBs) is an important yet very complex and labor-intensive task during the design of new aerospace vehicles. This is particularly true for Reusable Launch Vehicles (RLVs), as it has been observed during the development of CALLISTO, a demonstrator for a Vertical-Takeoff Vertical-Landing (VTVL) first stage which is jointly developed, manufactured and tested by DLR, JAXA and CNES. In this paper, we present an Inference-based methodology to define various types of Bayesian models exemplarily for a subset of CALLISTO's AEDB to assess their usability and prediction qualities. First, a short introduction to the underlying aerodynamic dataset will be given which has been aggregated from various Computational Fluid Dynamics (CFD) and Wind Tunnel Test (WTT) campaigns. Then, the different Bayesian models will be defined and their inference results compared against each other, according to common error metrics. It will be shown that, within the limits and assumptions of this study, several types of Bayesian AEDB models provide better accuracy in the prediction of uncertain aerodynamic coefficients compared to classical expert-fitted models for the given CALLISTO dataset. Generally, it can be concluded that Bayesian models are not only a promising new method for the definition of AEDBs, but could also find many potential applications in other engineering domains.