CFDBench
CFDBench is an evaluation dataset for engineering CFD scenarios, focusing on the most common industrial flow problems - pipe internal flow, airfoil external flow, and lid-driven cavity thermal convection. Compared with academic benchmarks (e.g. cylinder flow used by FNO), CFDBench's geometry and boundary conditions are closer to real engineering designs, making it an important benchmark for validating the engineering usefulness of surrogate models.
What's included
- Pipe internal flow: laminar and turbulent flow in circular, rectangular and varying-cross-section pipes
- External flow: steady and unsteady flow around cylinders, airfoils and square cylinders
- Lid-driven cavity: different Reynolds numbers
- Natural convection: heated cavity, buoyancy-driven thermal convection
- Multiphysics coupling: fluid-thermal conduction coupling (heat-sink problem)
Specs
- Format: VTK / CSV with mesh coordinates and field data
- Variables: velocity (u, v, w), pressure, temperature (some scenarios)
- Reynolds range: 100 ~ 10^6
- Resolution: 128x128, 256x256, 512x512
Use cases: Training and validating industrial CFD surrogates; comparing FNO / UNet / Transformer on engineering scenarios; fast design iteration (replacing full CFD solves); teaching the difference between engineering CFD and pure academic benchmarks.
Tags: CFDFluidThermal ConvectionInternal FlowExternal Flow
Related entries
- PDEBench · Large-scale PDE benchmark covering heat, advection-diffusion and fluid dynamics
- DrivAerNet++ · MIT's large-scale automotive aerodynamics dataset: 8,000 parametric car geometri
- AirfRANS · Parametric airfoil RANS flow-field dataset - compact and ML-pipeline friendly, t
- TU Delft Turbulent Wake Dataset · Classic experimental & simulation dataset of turbulent wakes/cavity flows from T
- U-bend Thermal-Fluid Dataset · 10,000 parametric 90 degree bend fluid-thermal coupling simulations across curva