AirfRANS

The AirfRANS dataset (opens new window) was originally created for the publication AIRFRANS: High Fidelity Computational Fluid Dynamics Dataset for Approximating Reynolds-Averaged-Navier-Stokes Solutions (36th Conference on Neural Information Processing Systems (NeurIPS 2022)) (opens new window). It consists of 1000 computational fluid dynamics (CFD) simulations of steady-state aerodynamics over two dimensions (2D) airfoils in a subsonic flight regime, splitted in different tasks. More precisely, it contains numerical resolutions of the incompressible Reynolds-Averaged Navier–Stokes (RANS) equations over the NACA 4 and 5 digits series of airfoils and in a subsonic flight regime setup.

Regarding the raw data, each simulation is given as a point cloud defined as the nodes of the simulation mesh, that is to say a discretization of the 2D domain. Such a discretization is required to compute the ground truth solution through a classical CFD physical simulator.

Each point of a point cloud is described via 5 features:

• the inlet velocity (two components in meter per second)
• the distance to the airfoil (one component in meter)
• the normals (two components in meter, set to 0 if the point is not on the airfoil).

Each point is given a target of 5 components for the underyling regression task:

• the velocity (two components in meter per second)
• the pressure divided by the specific mass (one component in meter squared per second squared)
• the turbulent kinematic viscosity (one component in meter squared per second)
• a boolean is attached to each point to inform if this point lies on the airfoil or not.