Loïc BREVAULT, Mathieu BALESDENT
DOI Number: N/A
Conference number: HiSST-2025-182
Simulating the trajectory of hypersonic vehicles using high-fidelity aeropropulsive performance models is computationally intensive due to the need for thousands of evaluations during numerical ordinary differential equations integration. To address this, an active learning surrogate-based trajectory simulation strategy is proposed that enables accurate hypersonic vehicle trajectory computation while significantly reducing computational cost. The method builds Gaussian process surrogates using a tailored design of experiments and employs an adaptive enrichment process guided by trajectory simulation to selectively add informative samples. By leveraging prediction uncertainty from the surrogates, the approach identifies and evaluates new points with the high-fidelity aeropropulsive performance model only where needed, ensuring accurate trajectory integration with a reduced number of high-fidelity calls. The proposed methodology is demonstrated on a representative hypersonic vehicle ascent trajectory. Results show that the approach achieves trajectory accuracy comparable of the high-fidelity reference while reducing computational costs by more than an order of magnitude compared to direct integration and
offline surrogate-based strategies. This work highlights the potential of inline, active learning surrogate modeling for enabling high-fidelity trajectory simulation of hypersonic vehicles in computationally constrained contexts.