EmmiAI/DrivAerML_subsampled_10x

---
license: cc-by-sa-4.0
task_categories:
- tabular-regression
tags:
- cfd
- computational-fluid-dynamics
- aerodynamics
- automotive
- drag-coefficient
- lift-coefficient
- physics-ml
- surrogate-modeling
pretty_name: "DrivAerML Subsampled 10x"
size_categories:
- n<1K
---

# DrivAerML Subsampled 10x

A subsampled and compressed version of the [DrivAerML](https://huggingface.co/datasets/neashton/drivaerml) dataset by Ashton et al. (2024), prepared for convenient use with ML frameworks for automotive aerodynamics tasks such as drag and lift coefficient prediction.

## Disclaimer

**This dataset is provided by [Emmi AI](https://www.emmi.ai) for convenience only, on an "as-is" basis, and without any warranty, express or implied.** Emmi AI does not own, and does not claim any ownership or rights over, the underlying data. All intellectual credit for the original CFD simulations, geometry generation, parametric morphing, and dataset design belongs exclusively to the original DrivAerML authors. Users should refer to the [original dataset](https://huggingface.co/datasets/neashton/drivaerml) and the accompanying paper for authoritative information.

## What Is This?

The original DrivAerML dataset is approximately 30 TB and contains 500 parametrically morphed variants of the DrivAer notchback vehicle with high-fidelity scale-resolving CFD (hybrid RANS–LES) simulation results. This version is a 10x subsampled derivative (~375 GB), intended to lower the barrier to entry for researchers and practitioners who want to experiment with aerodynamic ML models without downloading and processing the full dataset.

### What's Included

- Surface meshes (STL) for 500 parametric DrivAer variants
- Subsampled boundary fields (surface pressure and wall shear stress)
- Time-averaged aerodynamic force and moment coefficients (Cd, Cl, etc.)
- Geometric morphing parameters (16 parameters per variant)
- Per-variant reference quantities (frontal area, wheelbase)

### What Changed from the Original

- Surface and boundary field data subsampled by a factor of 10
- Volume field data (VTU) excluded to reduce storage
- Reorganized for direct use with the [Noether Framework](https://github.com/Emmi-AI/noether)

## Intended Use

This dataset is intended as a recipe and example for training and evaluating ML surrogate models for automotive aerodynamics. Typical tasks include:

- Scalar regression of drag (Cd) and lift (Cl) coefficients from geometry or geometric parameters
- Surface field prediction (pressure, wall shear stress)
- Surrogate modeling for rapid design-space exploration

For an end-to-end example using the Noether Framework, see the [aero_cfd recipes](https://github.com/Emmi-AI/noether/tree/main/recipes/aero_cfd).

## License

This dataset is distributed under the [Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)](https://creativecommons.org/licenses/by-sa/4.0/) license, consistent with the original DrivAerML dataset.

You are free to share and adapt this dataset for any purpose, including commercial use, provided you:
1. Give appropriate credit to the original authors (see Citation below).
2. Distribute any derivative works under the same CC BY-SA 4.0 license.

## Citation

If you use this dataset, please cite the original DrivAerML paper:

```bibtex
@article{ashton2024drivaer,
title = {{DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics}},
author = {Ashton, N. and Mockett, C. and Fuchs, M. and Fliessbach, L. and
Hetmann, H. and Knacke, T. and Schonwald, N. and Skaperdas, V. and
Fotiadis, G. and Walle, A. and Hupertz, B. and Maddix, D.},
year = {2024},
journal = {arXiv preprint},
url = {https://arxiv.org/abs/2408.11969}
}
```

## Acknowledgments

The original DrivAerML dataset was created by Neil Ashton and collaborators from UpstreamCFD, BETA-CAE Systems, Siemens Energy, Ford, and Amazon. It is hosted at [neashton/drivaerml](https://huggingface.co/datasets/neashton/drivaerml) and documented at [caemldatasets.org](https://caemldatasets.org). We thank the original authors for making this high-quality dataset publicly available under a permissive license.