Uncertainty Estimation in the Aerodynamic Database for the Reusability Flight Experiment

The following thesis investigates the application of Bayesian inference for modeling aerodynamic uncertainties in the context of Reusable Launch Vehicle (RLV). The analysis is based on the Reusability Flight Experiment (ReFEx), which Aerodynamic Database (AEDB) is composed of aerodynamic force and moment coefficients, gathered from various CFD simulations and wind tunnel experiments, across a wide range of flight conditions and vehicle configurations. The primary objective is to develop Bayesian models capable of capturing uncertainty and providing reliable predictions for unobserved flight regimes. For this purpose two classes of Gaussian Process (GP) models are explored: standard GPs, which offer high fidelity at the cost of computational scalability, and sparse GPs, which approximate the full posterior distribution while being more computationally efficient for large datasets. Model performance is assessed using metrics such as the Root Mean Squared Error (RMSE) across multiple training and prediction schemes. In particular, sparse GP models are compared to full GPs by measuring the similarity of their posterior distributions using distance metrics such as Wasserstein and Energy distances. The results suggest that sparse GP models can effectively learn aerodynamic behavior and quantify predictive uncertainty, supporting fast, data-driven aerodynamic modeling for early-stage RLV design and the development of their Guidance, Navigation & Control (GNC) systems.