Emmi-Wing
收藏资源简介:
Emmi-Wing是由Emmi AI GmbH与林茨大学联合构建的高精度三维翼型气动数据集,专注于跨音速流场的高保真模拟。该数据集包含约3万组独特几何构型与入流参数组合的CFD仿真数据,涵盖体积场与表面场物理量,支持升阻力系数计算与帕累托前沿优化。通过参数化设计生成多样化翼型配置,采用OpenFOAM求解器进行稳态可压缩流模拟,为航空航天领域的数据驱动设计提供关键基准,有效解决三维跨音速流动中激波与涡旋现象的建模挑战。
Emmi-Wing is a high-fidelity 3D airfoil aerodynamic dataset jointly developed by Emmi AI GmbH and the University of Linz, focusing on high-fidelity simulations of transonic flow fields. This dataset contains approximately 30,000 sets of CFD simulation data for unique combinations of geometric configurations and inflow parameters, covering physical quantities in both volumetric and surface fields, and supports lift and drag coefficient calculation as well as Pareto front optimization. It generates diverse airfoil configurations through parametric design, and uses the OpenFOAM solver to perform steady-state compressible flow simulations. As a critical benchmark for data-driven design in the aerospace field, it effectively addresses the modeling challenges of shock wave and vortex phenomena in 3D transonic flows.
Emmi-Wing 数据集概述
数据集基本信息
- 许可证: CC-BY-NC-4.0
- 来源: NeurIPS 2025物理科学机器学习研讨会论文《Going with the Speed of Sound: Pushing Neural Surrogates into Highly-turbulent Transonic Regimes》
数据集内容
- 数据规模: 70个压缩案例文件(scans_reduced.zip)
- 评估模型: AB-UPT神经代理模型
- 完整数据集下载地址: https://data.emmi.ai/s/qTgKFQCRNnFTgXN
数据结构
每个案例目录结构(run_X)包含:
- PyTorch张量文件:
- design_parameters.pt
- surface_position.pt
- surface_pressure.pt
- surface_wall_shear_stress.pt
- volume_position.pt
- volume_pressure.pt
- volume_velocity.pt
- volume_vorticity.pt
- NumPy文件:
- cell_areas.npy
- cell_centers.npy
- cell_normals.npy
- 网格文件:
- wing.stl(翼面表面网格)
数据质量说明
- 质量控制: 使用最佳神经代理模型识别数值伪影
- 问题案例列表: erroneous_cases.npy文件包含潜在严重问题案例
- 使用说明: 训练过程中代理模型可平滑处理这些伪影,适合作为异常检测器
使用资源
- 代码示例: GitHub仓库提供数据加载和可视化脚本
引用信息
bibtex @inproceedings{ paischer2025going, title={Going with the Speed of Sound: Pushing Neural Surrogates into Transonic and Highly Turbulent Regimes}, author={Anonymous}, booktitle={Machine Learning and the Physical Sciences Workshop @ NeurIPS 2025}, year={2025}, url={https://openreview.net/forum?id=36Tpmdy1Cu} }




