Learning Fluid-Structure Interaction Dynamics with Physics-Informed Neural Networks and Immersed Boundary Methods
Source code of Learning Fluid-Structure Interaction Dynamics with Physics-Informed Neural Networks and Immersed Boundary Methods described in the paper: Learning Fluid-Structure Interaction Dynamics with Physics-Informed Neural Networks and Immersed Boundary Methods.
Anaconda/Miniconda (recommended) or any other Python environment.
.
├── README.md
├── result
│ └── fsi
├── src
│ ├── data
│ │ ├── IBM_data_loader.py
│ ├── nn
│ │ ├── bspline.py
│ │ ├── kan2.py
│ │ ├── nn_functions.py
│ │ ├── pde.py
│ │ ├── tanh.py
│ │ ├── tanh2.py
│ ├── notebook
│ │ ├── M1_fsi_spectral_with_adaptive_methods.ipynb
│ │ ├── M2_fsi_spectral_adaptive_methods.ipynb
│ │ └── contour_plot_all_models_.ipynb
│ ├── trainer
│ │ ├── m1_trainer.py
│ │ └── m2_trainer.py
│ └── utils
│ ├── colors.py
│ ├── fsi_visualization.py
│ ├── line_plot2.py
│ ├── logger.py
│ ├── plot_losses.py
│ ├── combine_fluid_csv_files.py
│ ├── combine_solid_csv_files.py
│ ├── draw_contour_plts.py
│ ├── line_plot2.py
│ ├── logger.py
│ ├── plot_losses.py
│ ├── plotting_irregular_2D_interface.py
│ ├── plotting_regular_2D.py
│ ├── plotting_regular_2D_time_seqeunce.py
│ ├── printing.py
│ └── utils.pyThe code is tested in Ubuntu 20.04 LTS, using Nvidia A100 GPU.
conda env create -f environment.yml
conda activate pinn_fsi_ibm
# Check if PyTorch and CUDA available
python -m src.utils.check_torch
Version 2.4.0
CUDA: True
CUDA Version: 12.4
NCCL Version: (2, 20, 5)To train models run the following commands.
# fsi
python -m src.trainer.m1_trainer
We provided all pre-trained models and training loss log history. The notebooks can be run independently of training models.
Test models
- fsi:
fsi_test_model.ipynb
Plot loss history and test results
- fsi contour plot of test and error:
contour_plot_all_models_.ipynb
MIT LICENSE
If you find this work useful, we would appreciate it if you could consider citing it:
@article{farea2025learning,
title={Learning Fluid-Structure Interaction Dynamics with Physics-Informed Neural Networks and Immersed Boundary Methods},
author={Farea, Afrah and Khan, Saiful and Daryani, Reza and Ersan, Emre Cenk and Celebi, Mustafa Serdar},
journal={arXiv preprint arXiv:2505.18565},
year={2025}
}