Code and Dataset Accompanying "Physics-Based Machine Learning Closures and Wall Models for Hypersonic Transition–Continuum Boundary Layer Predictions"

This dataset contains code and numerical data data accompanying a publication that proposes and demonstrates two methods, based on neural networks embedded in the Navier-Stokes equations, to enhance the accuracy of predictions of hypersonic transition-continuum boundary layers. The first method augments the continuum viscosity and thermal conductivity with anisotropic neural networks with appropriate constraints on realizability and invariance properties. The second method replaces the conventional slip-wall and temperature-jump boundary conditions, based on an assumed Maxwellian particle velocity distribution function, with neural network-modeled distribution functions, which much more closely approximate the bimodal (nonequilibrium) velocity distributions observed at these conditions. We train both sets of models using adjoint-based optimization targeting macroscopic properties from direct simulation Monte Carlo (DSMC) solutions of the Boltzmann equation. All code, trained models, and input files, such that the training runs can be reproduced, will be published in .py format. The dataset is published for reproducibility and to allow for future comparison and study by other investigators. Under the CC BY 4.0 license, any reuse of this dataset should cite the original accompanying publication.