hacktoberfest-pinns

PINNs on Openfoam for hacktoberfest 2025.

🏍️ Motorbike PINN (Physics-Informed Neural Network)

A physics-informed neural network (PINN) surrogate model trained on the OpenFOAM motorBikeSteady case.
The network learns the steady-state incompressible Navier–Stokes solution and predicts velocity and pressure fields (u, v, w, p) directly from spatial coordinates.

---

Quick Start

Clone the repository and run the prediction script using your trained model:

# 1. Clone the repository
git clone https://github.com/<your-username>/motorbike-pinn.git

# 2. Train your model. Here, I used Nvidia B200 as part of the GPU Mode IRL 2025 San Francisco Hackathon.
# Will create the models/motorbike_pinn.pt for weights and step 3
cd src
python3 src/openfoam_pinn.py --vtk /data/VTK/motorBikeSteady_500.vtk

# 3. Run inference on the motorbike case
python3 predict.py --vtk /data/VTK/motorBikeSteady_500.vtk \
                   --weights /models/motorbike_pinn.pt \
                   --out-csv /data/predictions/pred_points.csv \
                   --out-vtk /data/predictions/pred_points.vtp

---

Arguments

Flag	Description
--vtk	Path to the input VTK file exported from OpenFOAM (foamToVTK -latestTime).
--weights	Path to the trained PINN model weights (.pt file).
--out-csv	(Optional) CSV export of predictions for analysis.
--out-vtk	(Optional) VTK/ParaView file containing predicted fields (U_pred, p_pred).

---

Output

After running the command, the results will be stored in:

data/predictions/
├── pred_points.csv   # Numerical results (x, y, z, Ux_pred, Uy_pred, Uz_pred, p_pred)
└── pred_points.vtp   # Visualization file for ParaView

Open the .vtp file in ParaView to visualize the predicted velocity and pressure fields.

---

Requirements

Install dependencies:

pip install -r requirements.txt

---

Repository Structure

motorbike-pinn/
├─ data/
│  └─ VTK/                      # OpenFOAM exported data
├─ models/
│  └─ motorbike_pinn_weights.pt         # trained PINN weights
├─ src/
│  ├─ openfoam_pinn.py         # training script
│  └─ predictopenfoam_pinn.py                # inference script
└─ requirements.txt

---

Credits

• OpenFOAM motorBike case (steady, simpleFoam)
• PyTorch, PyVista, MeshIO, HacktoberFest
• Developed for research on Physics-Informed Neural Networks (PINNs) for CFD surrogate modeling.

License

This project is licensed under the MIT License