SHIFT-SUV: High-Fidelity Computational Fluid Dynamics Dataset for SUV External Aerodynamics

We're excited to introduce the SHIFT-SUV dataset — a high-fidelity aerodynamic simulation dataset developed as part of the Luminary SHIFT Models initiative. This dataset enables the training and benchmarking of real-time physics AI models for automotive aerodynamics.

Website: shift.luminarycloud.com

Contact: [email protected]

Summary

Physics AI models can transform early stage automotive design by giving users real-time feedback on the physics-based performance implications of design decisions. However, the lack of high-quality training data has been a barrier to their development. Luminary SHIFT Models provide access to both high-quality datasets and pretrained models for a variety of applications and industries.

SHIFT-SUV is a massive step forward in this direction: purpose-built for high-fidelity aerodynamic inference, without requiring CFD expertise or meshing. Developed in collaboration with Honda and NVIDIA, SHIFT-SUV is based on thousands of parametrically morphed variants of the AeroSUV vehicle platform, developed by FKFS (Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart).

This dataset supports training surface-based or volume-based aerodynamic surrogate models, real-time inference systems, and exploring shape-performance correlations for automotive design.

Applications

• Rapid aerodynamic prototyping and shape optimization
• Research in aero-inference, point cloud learning, or physics-aware generative models
• Training and fine-tuning Physics AI models

Attribution

Please attribute FKFS for the original AeroSUV model, and Luminary Cloud for the SHIFT-SUV model and dataset.

An article is being prepared so users can cite this dateset - we will update this accordingly when available. Until then you can use this citation:

@misc{shift_suv_2025,
  author = {Luminary Cloud},
  title = {SHIFT-SUV: High-Fidelity Computational Fluid Dynamics Dataset for SUV External Aerodynamics},
  year = {2025},
  url = {https://huggingface.co/datasets/luminary-shift/SUV/}
}

Contents

This dataset contains the SHIFT-SUV dataset. We will continue to push newly computed samples to this repository periodically (approximately monthly), on our path towards 25K samples. The data generation and organization within the repository is described below.

Geometry Variation

A deformation cage approach was used to morph the discrete baseline AeroSUV model: The movement of the cage vertices were structured to mimic multiple different vehicle design parameters, informed by Honda Motors, common in early stage design. These parametric modifications are applied to multiple configurations of the baseline AeroSUV (e.g., estate vs. fast-back, smooth vs. detailed underbody) - these configuration options represent non-parametric parameters. Latin hypercube sampling is used to determine the specific values for the samples.

CFD Solver

All cases were run using the Luminary Cloud platform. The cases were simulated using simulation practices honed during our participation in AutoCFD4 and leveraged transient scale-resolving detached eddy simulation (DES):

Files

There following naming convention is used to describe the samples:

AeroSUV_<vehicle-back>_run_<index>_<simulation-type>

Legend

• vehicle-back: one of ["estate","fastback"], describing vehicle rear end design
• index: the sample index, which can be mapped to the parametric design values
• simulation-type: the solution method used for this sample. Currently the only value shared in this dataset is ["36606"], which corresponds to DDES simulations.

value	simulation type
36606	DES
64640	RANS

In each directory you will find the following files:

• merged_surfaces.stl: STL file with the vehicle's geometry
• merged_surfaces_filled.stl: STL file with the vehicle's geometry, with patches filling the holes at the tire/floor intersection (fully closing the vehicle)
• merged_surfaces.vtp: surface field solution file with pressure and wall shear-stress fields (both instantaneous and time-averaged) at the final iteration
• merged_volumes.vtu: volume field solution file with pressure, velocity, eddy viscosity, density, and temperature (both instantaneous and time-averaged) at the final iteration
• [DRAG, DRAG_COEFFICIENT, LIFT, LIFT_COEFFICIENT].csv: time histories of lift and drag during the transient simulation
• residuals.csv: time history of the compressible solver residuals
• residuals_and_forces.png: chart image displaying time-histories from above csv files
• viz: directory of images showing the geometry and flow fields

Downloading

You can use HuggingFace to gain access to the entire repository, but will require the associated TBs of storage available locally. Note you will need to have git lfs installed first, then run

git clone [email protected]:datasets/luminary-shift/SUV

If you will access only a subset of the data, or wish to interact in a staged manner, you can clone the repository where the LFS files are not checked out (simply pointers):

GIT_LFS_SKIP_SMUDGE=1 git clone [email protected]:datasets/luminary-shift/SUV .

# to ensure future `git pull` commands won't checkout full files, you'll want to ensure the skip is active in this repo
cd <path/to/repo>
git lfs install --skip-smudge --local

You can the pull down files you want to interact with in multiple ways:

# pull a specific file
git lfs pull --include="path/to/your/file"

# pull a directory
git lfs pull --include="path/to/file1,path/to/dir/*"

# pull, but exclude certain paths
git lfs pull --exclude="**/*.mp4"

and remove those files and reset them to pointers when done using them:

rm path/to/your/file
git checkout -- path/to/your/file

Credits

Timo Kuthada & FKFS

The AeroSUV model was developed by FKFS and provides the basis of SHIFT-SUV. The AeroSUV itself remains for non-commerical purposes only, the SHIFT-SUV dataset does not change this restriction.

Please attribute FKFS for the original AeroSUV model, and Luminary Cloud for the SHIFT-SUV model and dataset.

Fong Loon Pan & Honda

Guidance on geometry variants and industry relevant parameterization of the geometry was provided by Honda Motors.

NVIDIA

NVIDIA provided tools and infrastructure for model training via PhysicsNeMo and DoMINO. This is relevant if accessing the full dataset and pretrained models.

License

This dataset is distributed under the CC-BY-NC-4.0 license, which is also included in the dataset itself. By downloading the dataset you acknowledge the terms of this license.