RAndomly-SEEDed super-resolution GAN (RaSeedGAN)

This repository covers the Python implementation of a generative adversarial network (GAN) for estimating high-resolution field quantities from random sparse sensors without needing any full-resolution field for training.

The proposed network has been testes with four different cases. The associated raw data will be completed in the coming weeks. The datasets are:

• Fluidic Pinball: direct-numerical simulation (DNS) data generated from Deng et al. (2020). (Available)
• Turbulent Channel Flow: DNS data from a turbulent channel flow with friction Reynolds number available at Johns Hopkins Turbulence Database. (To appear soon)
• Turbulent Boundary Layer Flow: experimental data of a turbulent boundary layer with friction Reynolds number acquired in the water-tunnel facility at Universidad Carlos III de Madrid. (To appear soon)
• Sea Surface Temperature: experimental data of the global sea surface temperature from January 2000 to December 2019, downloaded from NOAA. (To appear soon)

Installation

Use the package manager pip3 to install the required dependencies. Python 3.8 is required.

pip install -r requirements.txt

Usage

To generate the tfrecord files, execute:

python run_generate_tfrecords.py -c pinball -u 4 -n 010

To run the training procedure, execute:

python run_training.py --case pinball --upsampling 4 --model_name architecture01-noise-010 --noise 10 --learning_rate 1e-4

To compute the prediction of the testing dataset, execute:

python run_compute_predictions.py -c pinball -u 4 -m architecture01-noise-010 -n 10 -l 1e-4

On a system with one GPU available, fluidic pinball case with upsampling factor takes approximately 100 seconds to run a sinfle epoch.

Publication

This repository has been used for the following scientific publication:

• Güemes, A., Sanmiguel Vila, C., & Discetti, S. (2022). Super-resolution GANs of randomly-seeded fields. arXiv preprint arXiv:2202.11701.

Authorship

This repository has been developed in the Experimental Aerodynamics and Propulsion group at Universidad Carloss III de Madrid. The following researchers are acknowledged for their contributions:

• Alejandro Güemes
• Stefano Discetti
• Carlos Sanmiguel Vila

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

Funding

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 949085).

License

Creative Commons Attribution 4.0 International. See LICENSE.md for further details.