ashiq24/FSI-pde-dataset

Multi-Physics Fluid-Solid Interaction Dataset

• How to Download ?

  from huggingface_hub import snapshot_download
  
  folder_path = snapshot_download(
      repo_id="ashiq24/FSI-pde-dataset",
      repo_type="dataset",
      allow_patterns=["fsi-data/*"]
  )
  

Dataset Description: Fluid-Solid Interaction Simulations (fsi-data)

This dataset contains simulations of fluid dynamics (using the Navier-Stokes equations) and elastic wave equations. The primary focus is on simulating the flow of water around an elastic rod.

Simulation Variables

For each time step (t), the simulation records the following variables:

1. Velocity (v)

   • A 2D vector with components:
     • vx: velocity in the x-direction
     • vy: velocity in the y-direction

2. Pressure (P)

   • Represents the pressure field at time t

3. Displacement (d)

   • A 2D vector representing the displacement of the elastic object and changes in mesh location:
     • dx: displacement in the x-direction
     • dy: displacement in the y-direction

Mesh Representation

The simulation uses a 2D mesh to represent the spatial domain:

• Initial Mesh (M0)

  • Represented as a matrix with N rows and 2 columns
  • Each row contains (x, y) coordinates of a mesh point
  • Example:

    x1, y1
    x2, y2
    ...
    xN, yN
    

• Time-dependent Mesh (Mt)

  • The mesh updates over time based on displacement
  • Calculated as: Mt = M0 + dt, where dt is the displacement at time t. That means the displacement at each time step also encodes the shift in the mesh from the initial Mesh M0.

Dataset Description: Computational Fluid Dynamics (cfd-data)

This dataset follows a similar structure to the fsi-data but focuses solely on simulating water movement around a rigid object. Key differences include:

• Only the Navier-Stokes equation is simulated
• The displacement field is zero (no movement of the rigid body)

Loading Dataset

FsiDataReader: Fluid-Solid Interaction Data Loader

The FsiDataReader class provides an interface to load and process simulation data for fluid-solid interaction. It handles structured datasets containing velocity, pressure, and displacement fields at different time steps.

Key Features

• Loads simulation data for various viscosity (mu) values.
• Supports flexible filtering by inlet boundary conditions (x1 and x2).
• Reads and processes both HDF5 and text-based dataset formats.
• Provides a PyTorch DataLoader for easy batch processing.

Usage Example

from fsi_data_reader import FsiDataReader

# Initialize the dataset loader
data = FsiDataReader('./fsi-data/', mu=['1.0'], in_lets_x1=['0.0'])

# Access the mesh structure
mesh = data.input_mesh
print(mesh.shape)

# Get a PyTorch DataLoader
data_loader = data.get_loader(batch_size=1, shuffle=False)

Constructor Parameters

• location (str): Path to the dataset directory.
• mu (list): List of viscosity values (mu) to load. Must be one of ['0.1', '0.01', '0.5', '5', '1.0', '10.0']. The value 0.5 should not be mixed with other mu values.
• in_lets_x1 (list, optional): Inlet boundary condition values for x1. Allowed values: ['-4.0', '-2.0', '0.0', '2.0', '4.0', '6.0'].
• in_lets_x2 (list, optional): Inlet boundary condition values for x2. Allowed values: ['-4.0', '-2.0', '0.0', '2.0', '4.0', '6.0'].

---

Methods

get_data(mu, x1, x2)

Loads simulation data for a specific mu, x1, and x2 configuration.

get_data_txt(mu, x1, x2)

Loads simulation data from text files instead of HDF5 format.

get_loader(batch_size, shuffle=True)

Returns a PyTorch DataLoader with batched velocity, pressure, and displacement data.

Visualization

Code of visulization is provided in plotting.py and data_vis.ipynb

To Cite

@article{rahman2024pretraining,
  title={Pretraining Codomain Attention Neural Operators for Solving Multiphysics PDEs},
  author={Rahman, Md Ashiqur and George, Robert Joseph and Elleithy, Mogab and Leibovici, Daniel and Li, Zongyi and Bonev, Boris and White, Colin and Berner, Julius and Yeh, Raymond A and Kossaifi, Jean and Azizzadenesheli, Kamyar and Anandkumar, Anima},
  journal={Advances in Neural Information Processing Systems},
  volume={37}
  year={2024}
}