Uncertain Reduced Order Model Predictions of an Unsteady Field
Author(s)
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
The design of blunt-body entry vehicles balances atmospheric heating and drag to ensure payload safety during entry, descent, and landing. Their blunt shape creates turbulent, recirculating wakes, leading to uncertain flight paths and challenging mitigation strategies. Traditionally, uncertainties are managed using conservative scalars and multipliers, resulting in over-engineered designs with reduced payload capacity and less accurate landings. While advances in computational fluid dynamics (CFD) enable high-fidelity analysis, the cost of extensive simulations remains prohibitive. Surrogate models, such as reduced-order models (ROMs), offer a faster alternative but must address the uncertainty introduced by unsteady aerodynamic training data. This paper presents a methodology to capture, encode, and propagate uncertainty in unsteady high-dimensional fields using parametric ROMs. The approach employs a Lorenz model to emulate unsteady fields and applies replication sampling to capture full-order model (FOM) nonparametric uncertainty. Proper orthogonal decomposition (POD) reduces dimensionality, and sparse Kriging regression predicts latent space mean and variance. A linear covariance back-mapping technique is applied to propagate uncertainty from the latent coordinates to integrated scalar coefficients. Results demonstrate the ROM accurately predicts scalar uncertainties consistent with FOM validation, supporting future application to more complex problems, such as entry vehicles in unsteady free-flight simulations.
Sponsor
Grant GR00019617
Date
2025-01
Extent
Resource Type
Text
Resource Subtype
Post-print
Rights Statement
Unless otherwise noted, all materials are protected under U.S. Copyright Law and all rights are reserved