Aerodynamics CFD · PLAID-datasets
AirfRANS_original
Graph-structured CFD dataset of RANS simulations over NACA airfoils for physics learning and graph neural networks.
What AirfRANS_original is
AirfRANS_original is an aerodynamics computational fluid dynamics (CFD) dataset containing Reynolds-averaged Navier–Stokes (RANS) numerical simulations around NACA airfoil geometries. It is formulated to benchmark graph-structured machine learning models, geometry learning, and physics-informed learning approaches on fluid flow problems.
This distribution reformulates the original AirfRANS simulation data using the PLAID (Physics Machine Learning Datasets) library and container format. The release supports reproducible evaluation on physics simulation tasks, including benchmark workflows aligned with the ML4PhySim challenge.
How it was made
The dataset is stored in Parquet tables within a single split named all_samples, where each entry is a serialized instance of a PLAID sample container. Each sample bundles the numerical simulation field arrays together with the underlying spatial discretisation.
Mesh structures within the samples adhere to the CFD General Notation System (CGNS) standard. These can be extracted directly and converted into Muscat mesh data structures via standard CGNS bridge utilities for geometric processing and graph extraction.
Key facts
- Publisher
- PLAID-datasets
- Category
- Aerodynamics CFD
- Task categories
- Graph machine learning, physics learning, geometry learning
- Data format
- Parquet containing PLAID sample containers and CGNS meshes
- Dataset variant
- AirfRANS_original (paired with AirfRANS_remeshed)
- Related benchmark
- ML4PhySim Challenge
What you can use it for
- Graph Neural Network Surrogates
- Benchmarking graph neural networks on unstructured fluid dynamics meshes to predict physical flow fields over airfoils.
- Physics-Informed Learning
- Evaluating surrogate models and physics-informed loss formulations against Reynolds-averaged Navier–Stokes aerodynamics solutions.
- ML4PhySim Challenge Participation
- Training and cross-validating competitive physics simulation models using the specified ML4PhySim_Challenge_train split.
Explore the data
The schema, real rows and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from PLAID-datasets/AirfRANS_original at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Provides standardised CFD mesh representations using the widely supported CGNS convention.
- Integrates directly with the PLAID data model and Muscat library for fluid flow feature inspection.
- Includes defined sub-split metadata matching the ML4PhySim competition benchmark.
Watch out for
- Requires domain-specific libraries including PLAID and Muscat to deserialize and manipulate sample objects.
- Aerodynamic coverage is confined to 2D NACA airfoil profiles under RANS flow physics.
Questions
Can I look inside AirfRANS_original without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser and a SQL filter. Nothing is downloaded to your machine.
What licence is AirfRANS_original under?
Its authors declare odbl. Read the terms on the original page before reusing the data.
What mesh standard does the dataset use?
Meshes inside the PLAID sample containers follow the CGNS (CFD General Notation System) standard and can be converted to Muscat mesh objects.
How are samples structured in the dataset tables?
Each row contains a pickled PLAID sample object that encapsulates the CGNS mesh data and the associated physical CFD fields.
What is the relationship between AirfRANS_original and AirfRANS_remeshed?
They are the two official PLAID variants of the AirfRANS benchmark, with AirfRANS_original retaining the original mesh discretisations of the simulations.
How do you access the ML4PhySim challenge training data?
The split information is contained in the dataset description metadata, pointing to the specific sample indices comprising the ML4PhySim_Challenge_train subset.
How to cite it
arXiv:2212.07564
Written from AirfRANS_original's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.