Automotive Ahmed body
AhmedML
AhmedML provides 500 parametric Ahmed-body CFD cases and official train, validation and test splits for data-efficiency and out-of-distribution evaluation.
- Scale
- 500 geometries
- Method
- Hybrid RANS–LES
- Size
- About 2 TB
Computational fluid dynamics data catalogue
Open CFD datasets for engineering machine-learning research
The catalogue documents four simulation datasets covering automotive external aerodynamics and high-lift aircraft flows. Each record summarises the geometry, numerical method, available data products, licence and associated publication.
- 4datasets in the catalogue
- 2application domains
- Up to 500Mcells in an individual case
AutomotiveAhmed · Windsor · DrivAer
AerospaceHiLiftAeroML · 1,800 cases
Catalogue
Datasets and simulation scope
The collection ranges from parameterised bluff bodies to a complete high-lift aircraft configuration, with different numerical methods and data volumes.
Automotive 
Automotive Windsor body
WindsorML
WindsorML provides 355 Windsor-body WMLES cases and official train, validation and test splits for data-efficiency and out-of-distribution evaluation.
- Scale
- 355 geometries
- Method
- WMLES
- Size
- About 8 TB
Automotive DrivAer notchback
DrivAerML
DrivAerML is a 500-geometry road-car CFD dataset with official train, validation and test partitions for data-efficiency and out-of-distribution evaluation.
- Scale
- 500-geometry design space
- Method
- Scale-resolving CFD
- Size
- About 31 TB
Aerospace NASA CRM-HL
HiLiftAeroML
HiLiftAeroML comprises 1,800 WMLES cases of NASA CRM-HL aircraft variants for aerodynamic machine-learning research.
- Scale
- 1,800 cases
- Method
- Explicit WMLES
- Size
- About 66.9 TB stored
About the catalogue
Scope and documentation
This site collates dataset-level information needed to assess suitability for a research task. It complements, rather than replaces, the repository documentation and source publications linked from each record.
Data products
Records distinguish geometry, integrated coefficients, surface fields, volume fields and derived data products.
Methods and provenance
Simulation method, solver, publication, licence, contributors and documented limitations are reported together.
Access considerations
File-selection examples and dry-run commands are provided because the complete repositories range from approximately 2 TB to 66.9 TB.
Data access
Recommended sequence for initial access
Because the repositories are multi-terabyte, file lists and transfer sizes should normally be inspected before data are downloaded.
- 1
Select
Compare domain, geometry, numerical method and available data products.
- 2
Estimate
Use a selective Hugging Face command with
--dry-runenabled. - 3
Transfer
Retrieve the required tables, geometry or field data after reviewing the estimate.
Release record
Dataset releases and publications
Official WindsorML train, validation and test splits are now available. The release defines eight deterministic benchmark regimes: an approximately 80/10/10 random baseline, three nested data-efficiency subsets, and four out-of-distribution evaluations based on geometry, drag and image-derived wake structure. Download the JSON manifest, read the methodology, or browse the complete reproducibility files on Hugging Face. Partition sizes, provenance and a usage example are included in the WindsorML dataset record.
Official AhmedML train, validation and test splits are now available. The release defines eight deterministic benchmark regimes: a random baseline, three nested data-efficiency subsets, and four out-of-distribution evaluations based on geometry, drag and image-derived wake structure. Download the JSON manifest, read the methodology, or browse the complete reproducibility files on Hugging Face. Partition sizes and a usage example are included in the AhmedML dataset record.
Official DrivAerML train, validation and test splits are now available. The release provides eight deterministic benchmark regimes for a random public baseline, nested data-efficiency studies, geometry extrapolation, drag-regime extrapolation and rear-separation evaluation. Download the JSON manifest, read the methodology, or browse the complete reproducibility files on Hugging Face. The partition sizes and usage example are documented in the DrivAerML dataset record.
Catalogue maintenance
Corrections and related publications
Contact the maintainers to report a catalogue error, repository change, derived dataset or publication that uses these data.
Contact the maintainers