OneScience/DeepCFD

DeepCFD

Model Overview

DeepCFD is a U-Net-based surrogate model developed by the German Aerospace Center for two-dimensional steady-state laminar flows. It uses geometric information to rapidly predict velocity and pressure fields, accelerating the evaluation of channel flows, flows around obstacles, and aerodynamic designs.

Paper: DeepCFD: Efficient Steady-State Laminar Flow Approximation with Deep Convolutional Neural Networks.

Model Description

DeepCFD uses a U-Net deep convolutional architecture that takes a signed distance field (SDF) and flow-region mask as inputs to rapidly predict velocity and pressure fields for two-dimensional, nonuniform, steady-state laminar flows.

Use Cases

Use Case	Description
Steady-state laminar flow prediction	Approximate velocity and pressure fields in two-dimensional, nonuniform, steady-state laminar flows
Channel-flow simulation	Predict flow distributions around randomly shaped obstacles in a channel
CFD surrogate modeling	Replace conventional CFD workflows such as OpenFOAM to improve inference efficiency
Aerodynamic and fluid-shape optimization	Evaluate large numbers of candidate geometries for low-speed laminar-flow design

Usage

1. OneCode

Use the online OneCode environment for an intelligent, one-click AI for Science (AI4S) programming experience:

Launch OneCode for one-click AI4S programming

2. Manual Setup

Hardware Requirements

• A GPU or DCU is recommended.
• A CPU can be used for import checks and small-scale pipeline validation, but full training and inference will be slow.
• DCU users must install DTK in advance. DTK 25.04.2 or later, or the OneScience-recommended version for the target cluster, is recommended.

Download the Model Package

modelscope download --model OneScience/DeepCFD --local_dir ./DeepCFD
cd DeepCFD

Set Up the Runtime Environment

DCU Environment

# Activate DTK and Conda first
conda create -n onescience311 python=3.11 -y
conda activate onescience311
# Installation with uv is also supported
pip install onescience[cfd-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai

GPU Environment

# Activate Conda first
conda create -n onescience311 python=3.11 -y libstdcxx-ng=12 libgcc-ng=12 gcc_linux-64=12 gxx_linux-64=12
conda activate onescience311
# Installation with uv is also supported
pip install onescience[cfd-gpu] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai

Training Data

The OneScience community provides the deepcfd dataset for training. Download it with the command below and verify that the data path in conf/config.yaml is configured correctly:

modelscope download --dataset OneScience/deepcfd --local_dir ./data

The project dataset can also be downloaded from Zenodo.

The dataset contains two files:

• dataX.pkl: Geometric input data for 981 channel-flow samples
• dataY.pkl: Ground-truth CFD solutions for the corresponding samples, computed with the simpleFOAM solver

Training

Single GPU:

• Training parameters are read from the root.datapipe, root.model, and root.training sections of config/config.yaml. The default configuration is intended for a minimal smoke test.
• For full training, set root.datapipe.source.data_dir to the actual DeepCFD dataset directory and increase the model size, batch size, and number of epochs as needed.

python scripts/train.py

Multiple GPUs:

torchrun --standalone --nproc_per_node=<num_GPUs> scripts/train.py

By default, training saves the best checkpoint to:

./weight/best_model.pt

Model Weights

This repository will provide pretrained DeepCFD weights in the weights/ directory. The weights will be uploaded soon.

Inference

python scripts/inference.py

Evaluation and Visualization

python scripts/result.py

Official OneScience Resources

Platform	OneScience 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

Citations and License

• Original DeepCFD paper: DeepCFD: Efficient Steady-State Laminar Flow Approximation with Deep Convolutional Neural Networks
• This repository retains source attribution and has been adapted for automated execution through OneScience and ModelScope.