Uncertainty Estimation in the Aerodynamic Database of ReFEx using Bayesian Inference

Abstract

The Reusability Flight Experiment (ReFEx) is a vertical take-off horizontal landing demonstrator developed by the German Aerospace Center (DLR) to investigate technologies for aerodynamically controlled Reusable Launch Vehicles (RLVs). A key requirement for such missions is an Aerodynamic Database (AEDB) that reliably predicts aerodynamic forces and moments while quantifying the associated uncertainties, which are critical for mission and control design. Conventional AEDB generation from Computational Fluid Dynamics (CFD) and Wind Tunnel Test data relies heavily on heuristic interpolation and expert judgment, making uncertainty quantification subjective and difficult to reproduce. This study applies Bayesian inference, using Gaussian Process (GP) regression, to model aerodynamic coefficients of ReFEx and to provide principled uncertainty estimates. Both full GP models and sparse GP approximations were investigated on a representative AEDB subset covering critical phases of the re-entry trajectory. Results show that full GP models reproduce aerodynamic behavior with high accuracy and robust uncertainty bounds, while sparse GP models closely approximate their posteriors at significantly reduced computational cost. These findings demonstrate that Bayesian GP models enable objective, data-driven AEDB model generation, and extend previous applications on the vertical take-off vertical landing demonstrator CALLISTO to the vertical take-off horizontal landing configuration of ReFEx.