data_driven_rans:

Repository on Master's thesis "Machine Learning Augmented Turbulence Modelling for Reynolds Stress Clorsure Problem"

Introduction

This repository contains a pytorch based implementation of the Tensor Basis Neural Network proposed by Ling et al. [1]. While the network architeckture is the same, the feateure set was extended as in Wu et al. [2], [3].

What the code in this repository can do:

• Read in RANS data from OpenFOAM, extract scalar invariants, and compute tensor basis
• Read in preprocessed DNS/LES data stored in .th files and interpolate them onto RANS grid
• Store both RANS and DNS/LES data so it can be accessed for training
• Select data and train the TBNN to find a mapping from invariant input features from RANS to labels (anisotropy tensor) from DNS/LES
• Store trained model so it can be accessed for prediction
• Create OpenFOAM file for the anisotropy tensor to be used as source term for baseline RANS equation
• Compute barycentric map coordinates [4] and visualize anisotropy tensor predictions

Dependencies

The following python packages are used and can be installed by executing the following commands:

pip3 install numpy pandas matplotlib scikit-learn torch torchvision scipy seaborn

Citation

@article{riccius2023physics,
  title={Physics-Informed Tensor Basis Neural Network for Turbulence Closure Modeling},
  author={Riccius, Leon and Agrawal, Atul and Koutsourelakis, Phaedon-Stelios},
  journal={arXiv preprint arXiv:2311.14576},
  year={2023}
}