Dear LBM colleague,
I have been working recently on a case about office ventilation, which is similar to the following one:
http://www.flowkit.com/showcases/example-applications/98-heatconvection
except that my case is isothermal - no coupling of temperature, and the air outlet is under the inlet on the same wall. In my case there is no table in the box either. Re = 1078, built by the height of the inlet.
I modified the channel3d example in the showCases. Used 3d poiseuille profile for inlet, and zero gradient (ConstantDensity) for outlet, D3Q19, BGK… It worked out first with the following config:
(I work with dimentionless problem setting here.)
Velocity in lattice units: u=0.05
Reynolds number: Re=1078
Lattice resolution: N=6 // N of the inlet height
Relaxation frequency: omega=1.99667
Extent of the system: lx=14.2857 // dimentionless length built by the inlet height
Extent of the system: ly=11.9048
Extent of the system: lz=20
Grid spacing deltaX: dx=0.166667
Time step deltaT: dt=0.00833333
The result looks like this, colored for U_mag
https://lh4.googleusercontent.com/-B2_K09CkC3k/VMEOdkDQY_I/AAAAAAAABew/rtojSNZ4Y3k/s600/u024000.gif
https://picasaweb.google.com/107020002235760504567/Plb#6107178470747431922
I was quite happy with that, until I increased the resolution to 8 and 10, tried different configs until it gets stable. However the result looks quite different now:
Velocity in lattice units: u=0.05
Reynolds number: Re=1078
Lattice resolution: N=10
Relaxation frequency: omega=1.99445
Extent of the system: lx=14.2857
Extent of the system: ly=11.9048
Extent of the system: lz=20
Grid spacing deltaX: dx=0.1
Time step deltaT: dt=0.005
https://lh3.googleusercontent.com/-fu6Er9xcaTE/VMEOdgz7h7I/AAAAAAAABe4/Cb7svaJVtKc/s600/u174000.gif
https://picasaweb.google.com/107020002235760504567/Plb#6107178469877843890
The experiment data shows something more like the first picture with N=6, it should be more turbulence than the second one.
My question is, WHY do the results look so different?? How can I get more turbulent results with the N=10 configuration? I tried different dt values and it does not work.
====================================================
I also need to do another calculation with Re = 4312. I tried N=12, 14, 16, 18 with dt=0.002 ~ 0.010
It has get stable yet. Should I increase the resoluation, or decrease dt more?
Is it possible that for every Re & N there is a best dt and when dt gets bigger or smaller than the optimum, the simulation will get unstable (t_NaN get smaller)? (For me it does not make sense. I thought too small dt values for a fixed N does not matter, but my lots of palabos experiments show me the principle above.)
Thanks a lot for any help.
Best regards, Peter