DrivAerRim

DrivAerRim: A full vehicle dataset with rotating wheels for data-driven aerodynamic design of wheel rims

DrivAerRim is a full vehicle computational fluid dynamics (CFD) dataset designed to isolate how rim geometry affects rotating wheel flow, component forces, and complete vehicle aerodynamics. It contains steady Reynolds-averaged Navier-Stokes (RANS) results for all 904 rim geometries from the DeepWheel dataset, installed on a fixed, full scale DrivAer estateback.

Only the rim geometry is intentionally varied between cases. The vehicle body, closed cooling configuration, smooth underbody, deformed Rain tyres, wheel positions, open road domain, boundary conditions, meshing strategy, solver settings, and export definitions are held constant. The same rim design is installed at all four wheel positions within each case.

The release contains geometry, integrated aerodynamic coefficients, accumulated drag, complete vehicle and wheel component surface fields, prescribed flow slices, local wheel volumes, a vehicle wake volume, a fluid volume surrounding the complete vehicle, preview images, and drag convergence histories. It is intended for controlled aerodynamic comparisons and for training and evaluating geometric deep learning surrogate models.

Contact

Chao Xia
Department of Mechanical Engineering, Chalmers University of Technology
Email: chao.xia@chalmers.se

Key features

• A fixed, full scale DrivAer estateback with a closed cooling configuration and smooth underbody.
• Deformed Rain tyres with longitudinal grooves, ground contact patches, and sidewall bulging.
• Rim rotation represented by cylindrical moving reference frame (MRF) regions, with rotating wall boundary conditions on the tyres.
• Complete vehicle and wheel component aerodynamic coefficients and streamwise accumulated drag traces.
• Complete vehicle and component surface pressure and wall shear stress fields.
• Velocity, pressure, and total turbulent kinetic energy on 128 prescribed slices and in six fluid volumes per case.

CFD configuration

Item	Setting
CFD software	Simcenter STAR-CCM+ 2506
Governing equations	Three-dimensional steady incompressible RANS
Turbulence model	Realizable $k$-$\varepsilon$ two-layer model
Wall treatment	Two-layer all-$y^+$ wall treatment
Spatial discretisation	Second order for the flow equations and turbulence convection
Flow solver	Coupled implicit solver
Inlet speed	140 km/h
Inlet turbulence intensity	0.1%
Ground	Fully moving no-slip road at the inlet speed
Rim rotation	Cylindrical MRF region around each rim
Tyre rotation	Rotating wall tangential velocity

Repository structure

The repository uses the following file patterns:

DrivAerRim/
├── Dataset_Splits/
│   ├── validation_ids.txt
│   ├── test_ids.txt
│   └── case_XX_train_NNN/train_ids.txt
├── NumData/
│   ├── Force_Coefficients_Combined.csv
│   └── Cd_Accumulated_Combined.csv
├── Geometry/
│   ├── Car_Body/Car_Body.stl
│   ├── Wheel_FL/Wheel_FL_XXXX.stl
│   ├── Wheel_RL/Wheel_RL_XXXX.stl
│   ├── WheelHouse_FL/WheelHouse_FL.stl
│   ├── WheelHouse_RL/WheelHouse_RL.stl
│   ├── WheelSupport_FL/WheelSupport_FL.stl
│   └── WheelSupport_RL/WheelSupport_RL.stl
└── RimXXXX/
    ├── Surfaces/
    │   ├── FullCar/FullCar_XXXX.vtp
    │   ├── Wheel_FL/Wheel_FL_XXXX.vtp
    │   ├── Wheel_RL/Wheel_RL_XXXX.vtp
    │   ├── WheelHouse_FL/WheelHouse_FL_XXXX.vtp
    │   ├── WheelHouse_RL/WheelHouse_RL_XXXX.vtp
    │   ├── WheelSupport_FL/WheelSupport_FL_XXXX.vtp
    │   └── WheelSupport_RL/WheelSupport_RL_XXXX.vtp
    ├── Slices/
    │   ├── X/X_<position>/X_<position>_XXXX.vtp
    │   ├── Y/Y_<position>/Y_<position>_XXXX.vtp
    │   └── Z/Z_<position>/Z_<position>_XXXX.vtp
    ├── Volumes/
    │   ├── Wheel_FL_region/Wheel_FL_region_XXXX.vtu
    │   ├── Wheel_FL_volume/Wheel_FL_volume_XXXX.vtu
    │   ├── Wheel_RL_region/Wheel_RL_region_XXXX.vtu
    │   ├── Wheel_RL_volume/Wheel_RL_volume_XXXX.vtu
    │   ├── Wake_volume/Wake_volume_XXXX.vtu
    │   └── FullCar_volume/FullCar_volume_XXXX.vtu
    └── Pictures/
        ├── <preview_name>.webp
        └── Cd_steady_history.png

<position> denotes the slice coordinate in millimetres, and <preview_name> denotes the descriptive name of a preview image.

RimXXXX identifies a rim configuration and its case directory, with XXXX running from 0001 to 0904. The suffixes FL and RL denote front left and rear left. The right side component geometries are mirrored counterparts of those on the left. With geometrically symmetric configurations and zero yaw, the released component geometries and local spatial records focus on the front left and rear left regions. The complete vehicle surface, wake volume, and full vehicle volume include both sides.

The repository contains approximately 9.32 TB of files (8.48 TiB). Every case provides 128 slice files, 7 surface files, and 6 volume files, together with at least 32 WebP previews and Pictures/Cd_steady_history.png. The volume records total 5,424 files, including 904 FullCar_volume files.

Geometry and numerical summaries

Geometry/ contains 904 wheel assembly STL files for each of the front left and rear left positions. The vehicle body, two wheelhouses, and two wheel supports are shared across cases and each stored once. The STL geometry is separate from the exported CFD surface meshes.

NumData/Force_Coefficients_Combined.csv contains one row per case, with the integer Rim identifier mapping to RimXXXX, for example 1 to Rim0001. Columns contain complete vehicle drag, lift, front and rear axle lift, and side force coefficients, together with wheel component coefficients. NumData/Cd_Accumulated_Combined.csv contains the streamwise coordinate X in metres and one accumulated drag coefficient column for each case, from Rim0001 to Rim0904.

Spatial fields

Surface and slice files use VTK PolyData (.vtp), and volume files use VTK Unstructured Grid (.vtu). Surface paths follow RimXXXX/Surfaces/<surface>/<surface>_XXXX.vtp. The 128 slices comprise 80 planes normal to x, 30 normal to y, and 18 normal to z. Slice directory and file names give the plane position in millimetres, for example Rim0001/Slices/X/X_-100/X_-100_0001.vtp.

Records	Field	Quantity	Units
Surfaces	MeanPressure	Static pressure	Pa
Surfaces	MeanWSS	Wall shear stress vector	Pa
Slices and Volumes	MeanVelocity	Velocity vector	m/s
Slices and Volumes	MeanPressure	Static pressure	Pa
Slices and Volumes	MeanTotalTKE	Total turbulent kinetic energy	m²/s²

The Mean prefix denotes averaging over the final 500 steady solver iterations, not a physical time average of an unsteady simulation.

The six volume records comprise the front left and rear left MRF regions (Wheel_FL_region, Wheel_RL_region), larger local wheel regions (Wheel_FL_volume, Wheel_RL_volume), the vehicle wake (Wake_volume), and the fluid region surrounding the complete vehicle (FullCar_volume).

Volume	x range (m)	y range (m)	z range (m)
FullCar_volume	-1.5 to 6.5	-1.5 to 1.5	-0.3 to 1.5
Wake_volume	3.2 to 5.7	-1.25 to 1.25	-0.3 to 1.25

Each case includes RimXXXX/Volumes/FullCar_volume/FullCar_volume_XXXX.vtu, averaging approximately 5.9 GiB (6.3 GB) per file. It contains the same three flow fields as the other volume records. Unlike Surfaces/FullCar, which describes the vehicle surface, Volumes/FullCar_volume contains the surrounding fluid mesh and flow fields.

Intended uses

DrivAerRim can support:

• analysis of how rim geometry changes complete vehicle and component aerodynamics;
• development and evaluation of geometric deep learning surrogate models for aerodynamic coefficients and surface or flow fields.

Surrogate-model dataset splits

The root-level Dataset_Splits/ directory contains the fixed case IDs used for the GAOT-3D surrogate-model example. Each header-free text file contains one front-left wheel sample ID per line in the form Wheel_FL_XXXX, where XXXX maps to the corresponding RimXXXX case.

The shared validation and test sets contain 92 cases each. Six nested training sets contain 720, 576, 288, 144, 72 and 36 cases, corresponding to 100%, 80%, 40%, 20%, 10% and 5% of the 720-case training pool. The 720-case training set, validation set and test set are mutually exclusive and together partition all 904 cases. See Dataset_Splits/README.md for the directory layout and exact file counts.

Download

The complete download comprises approximately 9.32 TB of files (8.48 TiB). Targeted downloads are strongly recommended.

Install the Hugging Face client:

pip install -U huggingface_hub

Download the predefined surrogate-model splits:

hf download BeyondXia1212/DrivAerRim \
  --repo-type dataset \
  --include "Dataset_Splits/**" \
  --local-dir DrivAerRim

Download the two numerical summaries:

hf download BeyondXia1212/DrivAerRim \
  --repo-type dataset \
  --include "NumData/**" \
  --local-dir DrivAerRim

Download spatial fields and pictures for one case. The shared and case-specific STL files are stored separately under Geometry/:

hf download BeyondXia1212/DrivAerRim \
  --repo-type dataset \
  --include "Rim0001/**" \
  --local-dir DrivAerRim

Download only the full vehicle volume for one case:

hf download BeyondXia1212/DrivAerRim \
  Rim0001/Volumes/FullCar_volume/FullCar_volume_0001.vtu \
  --repo-type dataset \
  --local-dir DrivAerRim

Python example for the numerical summaries and one case's spatial fields and pictures:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="BeyondXia1212/DrivAerRim",
    repo_type="dataset",
    allow_patterns=["NumData/**", "Rim0001/**"],
    local_dir="DrivAerRim",
)

The CSV files can be read with standard tabular data tools. The .vtp and .vtu files can be inspected with ParaView or other software supporting the Visualization Toolkit (VTK).

Code

The released case-preparation, Python batch controller, HPC job, and aerodynamic analysis scripts are available in GitHub: DrivAerRim/Codes. The code documentation describes configuration and use. The GitHub project page provides illustrated dataset overviews.

Citation

The accompanying manuscript is in preparation. Until a formal publication identifier is available, please cite the dataset as:

@dataset{xia2026drivaerrim,
  author    = {Xia, Chao and Lewerth, Carl and Zeng, Xi and Vdovin, Alexey and Sebben, Simone and Jia, Qing and Yang, Zhigang},
  title     = {DrivAerRim: A full vehicle dataset with rotating wheels for data-driven aerodynamic design of wheel rims},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/BeyondXia1212/DrivAerRim}
}

The source rim collection should also be cited:

Yoo, S. and Kang, N. DeepWheel: Generating a 3D Synthetic Wheel Dataset for Design and Performance Evaluation. Journal of Mechanical Design 148, 051702 (2026). https://doi.org/10.1115/1.4069899

Please also cite the accompanying DrivAerRim article once its formal publication identifier becomes available, while retaining the dataset and source geometry citations.

Credits

The dataset was created by researchers at Chalmers University of Technology and Tongji University. The authors acknowledge Ford of Europe for access to the DrivAer model and Volvo Cars for tyre, rim, and wind tunnel validation data, technical discussions, and advice. We further thank Soyoung Yoo and Namwoo Kang for their helpful correspondence and clarification regarding the use of the DeepWheel dataset in this work. This work was funded by the Transport Area of Advance Seed Project 2026 at Chalmers University of Technology. Computational resources were provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), funded by the Swedish Research Council.

License and provenance

DrivAerRim is released under the Creative Commons Attribution-NonCommercial 4.0 International licence.

The rim geometries incorporated into DrivAerRim were adapted from the third-party DeepWheel dataset, created by Soyoung Yoo and Namwoo Kang and distributed under CC BY-NC 4.0. The source STL meshes were repaired, scaled, and integrated into the DrivAer wheel assemblies for the present CFD simulations. The non-commercial restriction is retained in this derived release in accordance with the upstream licence. Users should attribute both DrivAerRim and DeepWheel and indicate any further modifications when redistributing the material.