Awesome
taichi_LBM3D
Documentation
https://yjhp1016.github.io/taichi_LBM3D/
Background
Taichi_LBM3D is a 3D lattice Boltzmann solver with Multi-Relaxation-Time collision scheme and sparse storage structure implemented using Taichi programming language, which is designed for porous medium flow simulation. Taking advantage of Taichi's computing structure, Taichi_LBM3D can be employed on shared-memory multi-core CPUs or massively parallel GPUs (OpenGL and CUDA). The code achieved 900 MLUPS in a 3D lid-driven test case using a A100 NVIDIA GPU. The code is only around 400 lines, highly extensible and very intuitive to understand.
Installation
This solver is developed using Taichi programming language (a python embedded programming language), install Taichi is required, by python3 -m pip install taichi
.
Pyevtk is required for export simualtion result for visualization in Paraview, install Pyevtk by pip install pyevtk
Usage
There are several place for users to modify to fit their problems:
set computing backend
First the computing backend should be specified by ti.init(arch=ti.cpu)
using parallel CPU backend, or by ti.init(arch=ti.gpu)
to use OpenGL or CUDA(is available) as computing backend
set input geometry
LBM uses uniform mesh, the geometry is import as a ASCII file with 0 and 1, where 0 represent fluid point and 1 represent solid point. They are stored in format:
for k in range(nz)
for j in range(ny)
for i in range(nx)
geometry[i,j,k]
You can specify the input file at:
solid_np = init_geo('./img_ftb131.txt')
For two phase solver, a two phase distribution input file is also requred. This file is composed of -1 and 1 representing phase 1 and 2 respectively
set geometry size
Set geometry input file size here: nx,ny,nz = 131,131,131
set external force
Set expernal force applied on the fluid here: fx,fy,fz = 0.0e-6,0.0,0.0
set boundary conditions
There are three boundary conditions used in this code: Periodic boundary condition, fix pressure boundary condition, and fix velocity boundary condition
We use the left side of X direction as an example: bc_x_left, rho_bcxl, vx_bcxl, vy_bcxl, vz_bcxl = 1, 1.0, 0.0e-5, 0.0, 0.0
set boundary condition type in bc_x_left
; 0=periodic boundary condition, 1 = fix pressure boundary condition, 2 = fix velocity boundary condition
if bc_x_left == 1
is select, then the desired pressure on the left side of X direction need to be given in rho_bcxl
if bc_x_left == 2
is select, then the desired velocity on the left side of X direction need to be given in vx_bcxl, vy_bcxl, vz_bcxl
The same rules applied to the other five sides
set viscosity
Viscosity is set in niu = 0.1
for single phase solver
niu_l = 0.05
niu_g = 0.2
for two phase solver, niu_l for liquid phase, niu_g for phase 2
Additional parameters for two phase solver
- Contact angle of the solid surface can be specified in
psi_solid = 0.7
this value is the cosine of the desired contact angle, so the value is between -1 and 1 - Interfical tension of two phases is set in
CapA = 0.005
- Boundary condition for the phase setting:
bc_psi_x_left, psi_x_left = 1, -1.0
bc_psi_x_left = 0 for periodic boundary for the phase field, 1 = constant phase field value boundary. If bc_psi_x_left is set as 1, then the next parameter is desired constant phase for this boundary: psi_x_left should be set as -1.0 or 1.0 for phase 1 or phase 2 respectively.
All the quantities are in lattice units
Examples (Direct Numerical Simulation)
Flow over a vehicle: inertia dominated
Single phase flow in a sandstone (Sandstone geometry is build from Micro-CT images at 7.5 microns): viscous dominated
Urban air flow: inertia dominated
Two Phase flow: oil (non-wetting phase) into a ketton carbonate rock saturated with water (wetting phase): capillary dominated
Authors
Jianhui Yang @yjhp1016 Liang Yang @ly16302
License
MIT
Citations
If you use Taichi-LBM3D in your research, please cite the corresponding paper:
@Article{fluids7080270,
AUTHOR = {Yang, Jianhui and Xu, Yi and Yang, Liang},
TITLE = {Taichi-LBM3D: A Single-Phase and Multiphase Lattice Boltzmann Solver on Cross-Platform Multicore CPU/GPUs},
JOURNAL = {Fluids},
VOLUME = {7},
YEAR = {2022},
NUMBER = {8},
ARTICLE-NUMBER = {270},
URL = {https://www.mdpi.com/2311-5521/7/8/270},
ISSN = {2311-5521},
DOI = {10.3390/fluids7080270}
}
Acknowledgement
We acknowledge the support from ARCHER2 eCSE Programme.
To do list:
- wrap functions into class to faciliate direct use from users
- Modify single phase gray scale and two phase data structure to improve computing efficiency (single phase code is done)
- Packing through pip
- Installation on ARCHER2