OneScience-Sugon/cylinder_flow

Cylinder Flow

Dataset Description

The Cylinder Flow dataset is sourced from DeepMind's MeshGraphNets benchmark and describes flow around a cylinder (vortex shedding) on a two-dimensional unstructured triangular mesh. The data was generated using COMSOL simulations. Each trajectory contains 600 time steps with a time-step size of dt=0.01 and records the fluid velocity and pressure fields.

Paper: Learning Mesh-Based Simulation with Graph Networks

Supported Tasks

Scenario	Description
Flow field time-series prediction	Predict subsequent velocity and pressure fields from historical mesh states.
Vortex shedding simulation	Learn the evolution of transient flow around a cylinder.
Graph neural network simulation	Provide training and evaluation data for mesh-based models such as MeshGraphNet.
Mesh-based physics modeling	Learn physical relationships between nodes and edges on an unstructured triangular mesh.

Dataset Format and Structure

The data uses the TFRecord format and is divided into training, validation, and test sets:

data/cylinder_flow/
  train.tfrecord
  valid.tfrecord
  test.tfrecord
  meta.json
  stats/

Each record corresponds to an unstructured mesh trajectory. The main fields are as follows:

Field	shape	dtype	Description
cells	[1, -1, 3]	int32	Node indices of triangular cells.
mesh_pos	[1, -1, 2]	float32	Two-dimensional coordinates of mesh nodes.
node_type	[1, -1, 1]	int32	Node types, such as interior, boundary, inlet, and outlet.
velocity	[600, -1, 2]	float32	Two-dimensional velocity fields over 600 time steps.
pressure	[600, -1, 1]	float32	Pressure fields over 600 time steps.

stats/ stores normalization statistics for edge features, node velocities, velocity differences, and pressure.

How to Use the Dataset

This dataset is compatible with the OneScience-Sugon/MeshGraphNet model. Download the dataset and model:

hf download --dataset OneScience-Sugon/cylinder_flow --local-dir ./cylinder_flow
hf download --model OneScience-Sugon/MeshGraphNet --local-dir ./MeshGraphNet

The validation script included with the dataset can be used to check the directory structure, metadata, and the first TFRecord sample:

cd cylinder_flow
python scripts/validate_cylinder_flow_dataset.py

Append the --verify-sha256 option to perform full SHA256 verification.

Official OneScience Information

Platform	OneScience Main Repository	Skills Repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Citation and License

• Original Cylinder Flow paper: Learning Mesh-Based Simulation with Graph Networks
• This repository retains information about the data source and organizes the data for automated runs on OneScience ModelScope. Before public distribution or redistribution, confirm the licensing requirements of the upstream project.