| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "sc": "https://schema.org/", |
| "cr": "http://mlcommons.org/croissant/", |
| "ml": "http://mlcommons.org/schema/", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "prov": "http://www.w3.org/ns/prov#", |
| "dct": "http://purl.org/dc/terms/", |
| "data": { |
| "@id": "cr:data", |
| "@type": "@json" |
| }, |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "field": "cr:field", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "format": "cr:format", |
| "includes": "ml:includes", |
| "containedIn": "ml:containedIn", |
| "recordSet": "cr:recordSet", |
| "source": "cr:source", |
| "extract": "cr:extract" |
| }, |
| "@type": "sc:Dataset", |
| "name": "TurtLES", |
| "description": "Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates", |
| "url": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "license": "https://creativecommons.org/licenses/by/4.0/", |
| "conformsTo": "http://mlcommons.org/croissant/1.1", |
| "citation": "Please cite TurtLES when using this dataset.", |
| "datePublished": "2025-01-01", |
| "version": "2.0.0", |
| "distribution": [ |
| { |
| "@type": "sc:FileSet", |
| "@id": "train-coords-files", |
| "name": "train-coords-files", |
| "description": "Coordinate files (x, y, z) for each simulation case in the training split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/train/*/*_coords.h5", |
| "encodingFormat": "application/x-hdf5" |
| }, |
| { |
| "@type": "sc:FileSet", |
| "@id": "val-coords-files", |
| "name": "val-coords-files", |
| "description": "Coordinate files (x, y, z) for each simulation case in the validation split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/val/*/*_coords.h5", |
| "encodingFormat": "application/x-hdf5" |
| }, |
| { |
| "@type": "sc:FileSet", |
| "@id": "test-coords-files", |
| "name": "test-coords-files", |
| "description": "Coordinate files (x, y, z) for each simulation case in the test split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/test/*/*_coords.h5", |
| "encodingFormat": "application/x-hdf5" |
| }, |
| { |
| "@type": "sc:FileSet", |
| "@id": "train-solution-files", |
| "name": "train-solution-files", |
| "description": "Solution snapshot files containing fluid variables (p, rho, u, v, w) grouped by timestep batches, for the training split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/train/*/*_b*.h5", |
| "encodingFormat": "application/x-hdf5" |
| }, |
| { |
| "@type": "sc:FileSet", |
| "@id": "val-solution-files", |
| "name": "val-solution-files", |
| "description": "Solution snapshot files containing fluid variables (p, rho, u, v, w) grouped by timestep batches, for the validation split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/val/*/*_b*.h5", |
| "encodingFormat": "application/x-hdf5" |
| }, |
| { |
| "@type": "sc:FileSet", |
| "@id": "test-solution-files", |
| "name": "test-solution-files", |
| "description": "Solution snapshot files containing fluid variables (p, rho, u, v, w) grouped by timestep batches, for the test split.", |
| "containedIn": "https://huggingface.co/datasets/turtlesprojectml/turtles", |
| "includes": "data/test/*/*_b*.h5", |
| "encodingFormat": "application/x-hdf5" |
| } |
| ], |
| "recordSet": [ |
| { |
| "@type": "cr:RecordSet", |
| "@id": "coordinates", |
| "name": "coordinates", |
| "description": "3D spatial coordinates of the unstructured mesh nodes, shared across all timesteps for a given simulation case.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "coordinates/x", |
| "name": "x", |
| "description": "x-coordinate of each mesh node.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-coords-files" }, |
| "extract": { "cr:hdf5Path": "/x" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "coordinates/y", |
| "name": "y", |
| "description": "y-coordinate of each mesh node.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-coords-files" }, |
| "extract": { "cr:hdf5Path": "/y" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "coordinates/z", |
| "name": "z", |
| "description": "z-coordinate of each mesh node.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-coords-files" }, |
| "extract": { "cr:hdf5Path": "/z" } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "solution-snapshots", |
| "name": "solution-snapshots", |
| "description": "Per-timestep fluid simulation snapshots. Each file contains multiple timestep groups (sol_t0, sol_t1, ...) each holding five 1D arrays (p, rho, u, v, w) with one value per mesh node. The number of mesh nodes is fixed within a case and ranges from about 2.9 to 10.5 million across cases.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "solution-snapshots/p", |
| "name": "p", |
| "description": "Static pressure at each mesh node for a given timestep.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-solution-files" }, |
| "extract": { "cr:hdf5Path": "/sol_t{t}/p" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "solution-snapshots/rho", |
| "name": "rho", |
| "description": "Fluid density at each mesh node for a given timestep.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-solution-files" }, |
| "extract": { "cr:hdf5Path": "/sol_t{t}/rho" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "solution-snapshots/u", |
| "name": "u", |
| "description": "x-component of velocity at each mesh node for a given timestep.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-solution-files" }, |
| "extract": { "cr:hdf5Path": "/sol_t{t}/u" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "solution-snapshots/v", |
| "name": "v", |
| "description": "y-component of velocity at each mesh node for a given timestep.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-solution-files" }, |
| "extract": { "cr:hdf5Path": "/sol_t{t}/v" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "solution-snapshots/w", |
| "name": "w", |
| "description": "z-component of velocity at each mesh node for a given timestep.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileSet": { "@id": "train-solution-files" }, |
| "extract": { "cr:hdf5Path": "/sol_t{t}/w" } |
| } |
| } |
| ] |
| } |
| ], |
| "rai:dataLimitations": "TurtLES covers turbulent wake flows at specific Reynolds number ranges and bluff-body geometries within predefined simulation bounds. It does not include wall-bounded turbulence, as the turbulent structures arise solely in the wake at the simulated Reynolds numbers. Generalisation beyond the covered parameter space (Reynolds numbers, angles of attack, and geometry configurations) cannot be guaranteed. The dataset is not recommended for use cases requiring wall-bounded turbulence, highly compressible flows, or multiphase flow phenomena.", |
| "rai:dataBiases": "All samples are generated from a controlled parameter sweep within predefined bounds of Reynolds number, angle of attack, and geometry. The distribution of simulation parameters is determined by the experimental design rather than real-world occurrence frequencies, which may not reflect the distribution of conditions encountered in practical engineering applications. No demographic or human-related biases are present, as the dataset is purely physical simulation data.", |
| "rai:personalSensitiveInformation": "None. TurtLES contains only fluid dynamics simulation data (velocity components, pressure, density, and spatial coordinates). No personal, demographic, or otherwise sensitive information about individuals is present.", |
| "rai:dataUseCases": "TurtLES is designed to benchmark and train neural surrogate models for turbulent 3D PDE systems. Established use cases include: (1) benchmarking neural PDE surrogates on 3D turbulent flows, (2) training data-driven models as replacements or accelerators for high-fidelity CFD solvers. Use cases for which validity has not been established include real-time flight control, safety-critical structural analysis, or any application requiring certified physical accuracy beyond the dataset's parameter bounds.", |
| "rai:dataSocialImpact": "Positive impacts include accelerating CFD research by providing a large-scale benchmark for neural surrogates, reducing the computational cost of high-fidelity simulations, and enabling ML-based engineering and environmental simulation workflows. Potential risks are limited given the purely scientific nature of the data; however, surrogate models trained on this dataset should not be deployed in safety-critical systems without independent validation against certified solvers.", |
| "rai:hasSyntheticData": true, |
| "prov:wasDerivedFrom": [], |
| "prov:wasGeneratedBy": [ |
| { |
| "@type": "prov:Activity", |
| "name": "CFD Simulation with PyFR", |
| "description": "All fluid dynamics data was generated using PyFR (https://www.pyfr.org), an open-source, high-order accurate, GPU-accelerated solver for the compressible Navier-Stokes equations. Simulations were run across a parameter sweep of Reynolds numbers, angles of attack, and bluff-body geometry configurations. No data annotation or labelling was performed; all fields (p, rho, u, v, w, x, y, z) are direct outputs of the solver on a case-specific unstructured mesh with about 2.9 to 10.5 million nodes. No human annotators were involved. Data was partitioned into train, validation, and test splits by simulation case." |
| } |
| ] |
| } |
|
|
