CFDBench: lid-driven cavity flow

Goal

Test generalization across boundary conditions and physical properties in a canonical incompressible CFD problem.

Data

navier datasets download cfdbench --local-dir ./data/cfdbench --pattern "cavity/**"

Recommended models

FNO, U-NO, Transolver, PIBERT, DeepONet, and U-Net.

Metrics

Velocity and pressure MSE, relative L2, divergence RMS, recirculation-center location, wall shear, and matched-error latency.

Study design

Use separate BC, property, and combined OOD splits. Report whether the data are raw or interpolated to 64×64.

References

Luo, Chen & Zhang, CFDBench: A Large-Scale Benchmark for Machine Learning Methods in Fluid Dynamics, 2023.