Meta-Learning for Dynamic Stability Analysis of Atmospheric Entry Vehicles

Entry, Descent, and Landing (EDL) is a critical stage of space missions, where atmospheric entry vehicles experience oscillations due to aerodynamic forces. Uncontrolled oscillations can lead to catastrophic outcomes, such as failed parachute deployment or tumbling. Hence, it is essential to understand the vehicle's dynamic stability characteristics during the design phase. However, identifying these stability coefficients with associated uncertainties in predictions is challenging due to the complex physics involved. This study proposes a meta-learning approach to estimate dynamic stability coefficients, integrating numerical and experimental datasets to enhance prediction accuracy. Our method involves: (i) reconstructing trajectories from sparse Ballistic Range (BR) data, (ii) fusing Free-Flight Computational Fluid Dynamics (FF-CFD) and BR data using meta-learning, (iii) estimating stability coefficients with the Markov Chain Monte Carlo (MCMC) method, and (iv) introducing a data source bias to ensure robustness. This approach reduces uncertainties and improves model generalization across different flow conditions, addressing limitations in both experimental and computational techniques. The results promise a more accurate, cost-effective method for dynamic stability analysis, contributing to a comprehensive aerodynamic database for atmospheric entry vehicles.