Fast prediction of chemically reactive rarefied hypersonic flows using boundary condition-based machine learning algorithm

A machine learning (ML) framework for complete analysis of high-altitude, rarefied hypersonic reentry flows has been developed in the present work. The ML framework comprises the boundary condition-based machine learning (BCML) algorithm for predicting flow fields and a deep neural network-based model (ChemDNN) for computing the mole fractions of various species around arbitrarily shaped reentry geometries. The BCML model is a physics-guided neural network trained using flow field data generated from the in-house direct simulation Monte Carlo (DSMC) code for a wide range of hypersonic flow conditions, spanning Mach numbers from 8 to 35. The BCML model predicts flow properties within the domain by incorporating the effects of surrounding boundary conditions. Results demonstrate that the new high-altitude DSMC-based BCML model achieves good agreement with the DSMC predictions while requiring only a fraction of the computational cost. Furthermore, the BCML framework accurately captures key flow features such as shock waves, including stand-off distances and shock strengths. Additionally, the ChemDNN model effectively predicts the mole fractions of various species for the five-species air chemistry model, enabling efficient modeling of nonequilibrium high-enthalpy reacting flows. A separate deep neural network model is also reported, which can estimate the coefficients of heat flux and pressure on the surface of the reentry geometry, which are critical inputs for the design of thermal protection systems in hypersonic vehicles.