The SANS equations are used to reduce a turbulent flow presenting an homogeneous direction into a 2-D system, effectively cutting the computational cost of a simulation by orders of magnitude. This is accomplished by including additional terms in the 2-D momentum equations which account for the 3-D turbulence mixing effects. The additional unclosed terms are modelled here using a convolutional neural network (CNN). Check this preprint for more info, or our journal publication:
- Font, B., Weymouth, G.D., Nguyen, V.-T. & Tutty, O.R. (2021) Deep learning the spanwise-averaged Navier-Stokes equations. Journal of Computational Physics, 2021, 434(10):110199. doi:10.1016/j.jcp.2021.110199
The use of a virtual environment is recommended for the installation of the package. Here we will use a conda environment based on Python 3.6. To create a new virtual environment:
conda create --name my_env python=3.6
Now you can install this package into the my_env environment using pip:
git clone https://github.com/b-fg/sanspy
cd sanspy
source activate my_env
pip install -e ~/sanspy
The environment can be deactivated running conda deactivate. In order to be able to perform modification on the package without the need of reinstalling the -e (editable) argument is used. This provides the source path of the package to the conda environment so any modifications on the source code (the folder you have downloaded and installed from) is immediately available with no need to re-install.
To use the package just activate the conda environment in your terminal with source activate my_env or load it into your preferred IDE. In your Python script, you should now be able to import the modules with:
from sanspy.model import Model
Scripts examples on training and testing are provided in the examples/ folder.
If there is a GPU in the current machine, this can be used to train or test the model (see in examples/).
In this sense, install the keras-gpu and package from conda, which will also install the tensorflow and tensorflow-gpu libraries as the backend:
conda install -c anaconda keras-gpu
Finally, there is a dependency to my postproc package, which you can install from this repository.
This is used for reading the dataset, but you can just use your own read/write routines.