Use of Data Fusion of AI/ML and High-Fidelity Physics Simulations to Predict High-Mach Number Atmospheric Re-Entry

Atmospheric re-entry presents several challenges, ranging from the management of the extremely high heat-fluxes to communication blackout and vehicle stability. Since it is virtually impossible to reproduce the exact re-entry flight conditions in ground facilities, and flight testing is extremely expensive, high-fidelity, physics-based computer simulations become a critical asset for the proper characterization and understanding of the high-enthalpy, high-Mach number environment the vehicle experiences. In particular, high-fidelity Computational Fluid Dynamics (CFD) simulations have been instrumental in improving the design and proper operation of the vehicle. However, CFD simulations tend to be very expensive and if the characterization along an entire flight trajectory is needed, the costs can be prohibitive. In this context, the implementation of Artificial Intelligence/Machine Learning (AI/ML) methods to augment and extend the data obtained through CFD has shown great promise. In the present study, an AI/ML algorithm is trained on a limited set of existing high-fidelity CFD simulations at specific altitudes and Mach numbers along a trajectory. The AI/ML algorithm is then tasked to predict the vehicle state at additional points along the trajectory without repeating the more expensive high-fidelity simulations. These AI/ML-predicted solutions are benchmarked for validation purposes against test data and/or additional simulations. The key advantage of this methodology is the high speed and low computational cost at which reliable predictions of the vehicle state can be obtained once the AI/ML algorithm has been trained and properly validated. We show the application of the combined modeling workflow described above for the re-entry flight experiment of a space capsule, Japan Space Agency’s OREX, for the prediction of a range of key measurements, such as stagnation heat flux and electron concentration around the capsule. This methodology builds on the previous validation of the accuracy of the simulation results as presented before in this forum. Recommendations are suggested regarding future work and the development of best practices for the use of AI/ML methods to augment CFD results.