Skip to content

Code associated with the paper "RANS wake surrogate: Impact of Physics Information in Neural Networks"

License

Notifications You must be signed in to change notification settings

jenspeterschoeler/RANS-Wake-PINN

Repository files navigation

DOI DOI

RANS Wake PINN

This repository is a companion piece for the article "insert article link when published". It includes the code necessary to train the models described in the paper. PINN concept


Setup and running the code

The code has been developed and run with an Anaconda environment sml. To install said environment: Clone the repository, navigate to the project directory and run the following line:

conda env create -f environment.yml

To run the code activate the environment and run the mainscript run_PINN_experiment.py.

conda activate sml
python3 run_PINN_experiment.py

🚨 The code was developed with the Jax/Flax libraries. At the begining of the project Jax only supported Linux, since support has been added for Windows and Mac. However, the code has only been verfied to run on Linux. For detailed installation instructions visit the Jax documentation.

💡 To configure hyperparameters during training the Hydra framework has been used. The predefined cases that was used in the paper are included and can be uncommented at the top of the run_PINN_experiment.py script.

Results

The results can be found in the Results folder. It contains data from the training runs described in the paper and post-processing notebooks.


Available data

The data used in the training of the models is available in the data folder (RANS_1wt_irot_v2.nc). The dataset was originally generated with the DTU in-house CFD solver Ellipsys using the PyWakeEllipsys framework. The data shared in this repo is post-processed into a cylindrical axisymmetric dataset as illustrated below: Data

Citing

If you use this repository in your scientific work please consider citing us:

Paper

@article{Schoeler2024,
    doi = {10.1088/1742-6596/2767/9/092033},
    url = {https://dx.doi.org/10.1088/1742-6596/2767/9/092033},
    year = {2024},
    month = {jun},
    publisher = {IOP Publishing},
    volume = {2767},
    number = {9},
    pages = {092033},
    author = {J. P. Sch{\o}ler and N. Rosi and J. Quick and R. Riva and S. J. Andersen and J. P. Murcia Leon and M. P. van der Laan and P.-E. Réthoré},
    title = {RANS wake surrogate: Impact of Physics Information in Neural Networks},
    journal = {Journal of Physics: Conference Series},
}

Code

@misc{Schoeler2024code,
    author       = {J. P. Sch{\o}ler and N. Rosi and J. Quick and R. Riva and S. J. Andersen and J. P. Murcia Leon and M. P. van der Laan and P.-E. Réthoré},
    title        = {RANS Wake PINN},
    month        = jun,
    year         = 2024,
    publisher    = {Zenodo},
    doi          = {10.5281/zenodo.10846076},
    url          = {https://doi.org/10.5281/zenodo.10846076}
}

About

Code associated with the paper "RANS wake surrogate: Impact of Physics Information in Neural Networks"

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published