Informed Active Learning for Hypersonic Vehicle Co-Design
Name
JackmanPhantomPaper.pdf
Size
478.62 KB
Format
Adobe PDF
Checksum (MD5)
c092af3074fffb88c7b6b559f87a55f9
Author(s)
Jackman, Ethan
Crouse, James
Date Issued
August 20, 2026
Abstract
Hypersonic flight o ffers c ompelling c apabilities for a wide range of applications, including responsive space access, long-range atmospheric transport, and reusable launch systems. Conceptual design of hypersonic vehicles often consist of highdimensional design spaces in which high-fidelity c omputation is prohibitively expensive, motivating the use of surrogate models refined u nder a l imited c omputational b udget. Conventional active learning refines s uch s urrogates b y s ampling w here predictive uncertainty is greatest, potentially allocating resources towards non-competitive regions of the design space. This paper proposes a composite, non-intrusive active learning criterion that allocates high-fidelity e valuations b y j ointly w eighting surrogate uncertainty, proximity to relevant trajectories, and the geometric promise of each candidate. Applied to a parameterized hypersonic vehicle with geometry and trajectory optimized jointly, the criterion reduces the uncertainty in lift coefficient a long the optimal trajectory by roughly 70%, whereas a pure uncertaintybased baseline yields negligible improvement under the same computational budget.
Subjects
active learning
surrogate modeling
Bayesian optimization
MIT Department
Lincoln Laboratory
Terms of Use
All rights reserved
Persistent DSpace Link
