Towards Efficient Simulations of Non-Equilibrium Chemistry in Hypersonic Flows: Neural Operator-Enhanced 1-D Shock Simulations

This work presents further advancements in designing a novel numerical framework to address the challenges in solving computationally intensive but physically accurate thermochem-ical non-equilibrium models for high-speed flows. The present study focuses on designing a systematic procedure to leverage the proven capabilities of neural operator-based surrogate modeling for chemistry in CFD environments. The proposed framework employs operator-splitting time integrators to decompose the governing equations into more tractable transport and reaction operators, each treated using its own optimal method. Here the integration of the stiff reaction operator, often accomplished via implicit methods, is replaced by a fast evaluation of a machine learning-based surrogate trained via zero-dimensional simulations. A specific neural operator architecture is used to construct the surrogate, with advantages in resolution invariance and generalizability compared to classical deep neural networks. A one-dimensional shock case scenario is used as a numerical experiment. Results demonstrate the effectiveness of the proposed framework, which is able to recover the correct solution with a relative error lower than 2% for quantities of interest, such as temperature and chemical composition.