Surrogate thermochemical kinetics for nonequilibrium hypersonic flows

Conference item

Nonequilibrium thermochemical processes such as vibrational excitation, chemical dissociation, and ionization are critical to accurately characterize hypersonic flows. The numerical modelling of such mechanisms is computationally demanding due to the onerous number of degrees of freedom and kinetic parameters coupled with wide flow spatio-temporal scales, resulting in highly non-linear and stiff chemical kinetics. To accelerate computing of reacting flow simulations without losing fidelity, data-driven strategies, such as Principal Component Analysis (PCA) and physics-based models, such as Computational Singular Perturbation (CSP), are typically implemented. In the present work, a novel reduced-order model (ROM) framework is proposed to combine the two strategies into a surrogate model. First, it is shown that the equations governing the evolution in space and time of a small disturbance around an equilibrium state can be formulated as a generalised eigenvalue problem. The solution to this problem can be used to represent the evolution of disturbances relaxing towards equilibrium downstream of an initial perturbation, such as a shock. The eigenvectors of the system distinguish between groups of slow non-equilibrium processes and fast, near-equilibrium processes. The calculation of the system eigen-basis, however, is the most computationally expensive operation of the numerical algorithm. This paper proposes the substitution of this operation with non-linear regression fitting, namely Gaussian Process Regression (GPR), to map the system eigenvalues and eigenvectors. To reduce the training cost of GPR, a truncated basis of independent variables is identified using PCA. The method is applied to a two-temperature plasma evolving downstream of a shock, modelled using Park's two-temperature model with 11 species for air. The feasibility and fidelity of the novel ROM framework are presented, demonstrating the viable application of computational learning methods for eigen-basis prediction.

Authors: Rapisarda, C; Clarke, J; McGilvray, M; Di Mare, L

• Publication status: Published
• Peer review status: Reviewed (other)
• Publisher: American Institute of Aeronautics and Astronautics
• Host title: AIAA SCITECH 2025 Forum
• Article number: AIAA 2025-2343
• Publication date: 2025-01-03
• Acceptance date: 2024-08-26
• Event title: 2025 AIAA Science and Technology Forum and Exposition (AIAA SciTech Forum)
• Event location: Orlando, FL, USA
• Event website: https://aiaa.org/scitech/utility/past-forums/#2025
• Event start date: 2025-01-06
• Event end date: 2025-01-10
• DOI: 10.2514/6.2025-2343
• EISBN: 9781624107238
• Language: English
• Keywords: nonequilibrium flows; reduced order modelling; Gaussian process regression; perturbation analysis; intrinsic low dimensional manifold; numerical algorithms; computing; reacting flow; Rankine Hugoniot jump conditions; FORTRAN