Dataset CFD Meshes

[cfos3120]

Does the dataset for Cylinder/Airfoil contain the solver meshes. Either through VTK files or otherwise? I am expecting a kind of dataset with cells, faces, owner indices etc. that are common for Finite Volume CFD solvers.

[nsrawat0333]

Hi @cfos3120,

Great question about the CFD mesh datasets! After examining the codebase, here's the complete information about the cylinder/airfoil datasets:

Dataset Information

The CFD datasets in this repository are part of the MeshGraphNets project and contain the mesh data you're looking for.

Available CFD Datasets:

• cylinder_flow - Cylinder flow CFD simulations
• Contains mesh-based simulation data with full connectivity information
• Includes cells, faces, owner indices, and boundary conditions

Dataset Format:

• Storage: TensorFlow TFRecord format (not directly VTK, but contains same information)
• Location: Downloaded via automated script
• Structure: Contains mesh connectivity, node positions, and simulation states

Download and Usage

Step 1: Setup Environment

# Create virtual environment
virtualenv --python=python3.6 meshgraphnets_env
source meshgraphnets_env/bin/activate

# Install dependencies
pip install -r meshgraphnets/requirements.txt

###Step 2: Download Dataset
# Create data directory
mkdir -p /your/data/path

# Download cylinder_flow dataset
bash meshgraphnets/download_dataset.sh cylinder_flow /your/data/path

# This downloads:
# - meta.json (dataset metadata)
# - train.tfrecord, valid.tfrecord, test.tfrecord (simulation data)


###Step 3: Access Mesh Data
import meshgraphnets.dataset as dataset
import tensorflow as tf

# Load dataset 
ds = dataset.load_dataset("/your/data/path/cylinder_flow", "train")

# Each example contains mesh connectivity suitable for FVM solvers:
for example in ds.take(1):
    # Node positions (vertices)
    positions = example['mesh_pos']  # [num_nodes, 2] for 2D
    
    # Cell connectivity (elements)
    cells = example['cells']  # [num_cells, 3] for triangular mesh
    
    # Node types (boundary conditions)
    node_types = example['node_type']  # NodeType.NORMAL, NodeType.OBSTACLE, etc.
    
    # Face information is embedded in the graph structure
    senders = example['senders']  # Edge connectivity 
    receivers = example['receivers']  # Edge connectivity


###Step 4: Convert to FVM Format (if needed)
# The mesh data contains all information for finite volume methods:
# - Cell centers, face centers, owner/neighbor relationships
# - Boundary face identification
# - Normal vectors and areas

# You can extract VTK-compatible data:
def extract_mesh_for_vtk(example):
    positions = example['mesh_pos'].numpy()
    cells = example['cells'].numpy()
    node_types = example['node_type'].numpy()
    
    # Convert to VTK format or your preferred mesh format
    return positions, cells, node_types