A simple lattice-Boltzmann code for 2D flow resolutions.
This LBM code includes:
- D2Q9 lattice
- TRT collision operator
- Zou-He on all boundary conditions
- Drag/lift computation using interpolated bounce-back
- Core routines are deferred to Numba
Cases are described in the lbm/src/app/ repository. To run a simulation, adjust the parameters in the related python file, then run python3 start.py <app_name>. A results folder will be generated in ./results/ with the current date and time. If you wish to add a new application, you must create a new app, and register it in the factory, located in lbm/src/app/app.py. Below are some examples and benchmarks that were ran with the code. The related cases are available in the repository. The computational times are obtained on a standard laptop.
A simple driven cavity in unit square. Below are the computed time-domain velocity norms at Re=100 (left) and Re=1000 (right).
A comparison of u = f(y) and v = f(x) at the center of the domain with reference data from "U. Ghia, K. N. Ghia, C. T. Shin, High-Re solutions for incompressible flow using Navier-Stokes equations and multigrid method".
The Turek cylinder benchmark CFD case is described in "Schafer, M., Turek, S. Benchmark Computations of Laminar Flow Around a Cylinder". Here, we explore the accuracy of the drag and lift computation (using IBB). Note that for the 2D-2 case, the values correspond to the max drag and lift.
ny |
2D-1 (Re=20) Cd, Cl, CPU | 2D-2 (Re=100) Cd, Cl, CPU | |
|---|---|---|---|
| Turek | --- | 5.5800 - 0.0107 - N/A | 3.2300 - 1.0000 - N/A |
| lbm | 100 | 5.7111 - 0.0285 - 236 s. | 3.5409 - 1.0696 - 518 s. |
| lbm | 200 | 5.6250 - 0.0212 - 1367 s. | 3.2959 - 1.0253 - 2762 s. |
Below is a video of the 2D-2 case:
An array of square obstacles at Re=2000, with ny=200. This computation took approx 20 minutes on my laptop, although the accuracy here is questionable.
Two steps in a channel at Re=500, with ny=150. This computation took approx 15 minutes on my laptop.
