Research Topic:

Machine-learning reconstruction of moment equations for shock-dominated rarefied gas flows

Supervisor:

Prof. Manuel Torrilhon, RWTH

Thesis Abstract:

My doctoral research develops machine-learning-enhanced moment equations for shock-dominated rarefied gas flows. Higher-order moments and collisional production terms in strongly nonequilibrium regimes are remodeled using neural networks trained on high-fidelity kinetic data. By embedding these learned closures into numerical moment solvers, the approach indirectly reconstructs macroscopic flow quantities, including stress and heat flux, while improving the accuracy and applicability of moment methods for complex nonequilibrium flows.

Publications:

•Hang Song, Satyvir Singh, and Manuel Torrilhon. “Machine-learned R13moment closures for shock-dominated rarefied gas flows”. In: Computers & Fluids (2026). In press.

•Hang Song, Satyvir Singh, and Manuel Torrilhon. “Non-equilibrium flow simulations based on Grad-14 and Grad-17 moment equations for polytropic gases”. In: Physics of Fluids 37.3 (2025). doi: 10.1063/5.0257491.

•Hang Song, Satyvir Singh, and Manuel Torrilhon. “Numerical Study of Grad-14 and 17 Moment Equations for Rarefied Polyatomic Gases”. In: Rarefied Gas Dynamics: RGD 2024. Ed. by Martin Grabe, Georgii Oblapenko, and Manuel Torrilhon. Springer Aerospace Technology. Cham: Springer, 2026. doi: 10.1007/978-3-032-00094-1_36. url: doi.org 10.1007/978-3-032-00094-1_36.

Conference Presentations:

•Presentation at the 34th International Symposium on Rarefied Gas Dynamics (RGD34), Brisbane, Australia, July 2026.