[Sereff]

Greetings everyone,

I am trying to use Wall Modeled LES (Smagorinsky, damping function included) to simulate a boundary flow and I noticed some wierd behavior at the first few cells near the wall. I have attached the plots I processed from my results. They are time and space averages from a full-developed boundary flow. The geometry of my case are:

> 6282m*1000m*3141m (x,y,z)
> resolution: 64*64*64 (x,y,z)
> cyclic BCs on x and z direction
> y is the vertical direction and ground is with a surface roughness of 0.1m

I have the following BCs on y direction for U, p and nut

U:   upperAtm:   zeroGradient;
     ground:     noSlip;

p:   upperAtm:   zeroGradient;
     ground:     zeroGradient;

nut: upperAtm:   zeroGradient;
     ground:     nutURoughWallFunction;
                   roughnessConstant  0.5;
                   roughnessHeight    3.0; // Ks = (20~30)*y0
                   roughnessFactor    1.0;

I found a documentation explainning how the wall functions work in OpenFOAM here, and it explains how OpenFOAM calculate the velocity of the first cell center by modifying nut of the first cell center. However it didn't explain the weird behavior I marked with red on the pics. As can be seen from nut.png, there is an unexplainable drop on nut on the second cells off the wall. And from the Usemi.png, to me the trend of hand-drawn red line is more realistic, in case of I didn't get the entries on wall functions right. In an ideal world the velocity profile should almost overlaps the log-law line.

Can anyone give me some hints? Any suggestion is deeply appreciated.

kind regards,

[HPE]

Hi,

- Is it possible for you plot "y+" instead of "y/H" in the vertical direction, please?
- Could you please compute min/max y1+ values for the walls?
- Have you seen the kink in your visualisations of the field? May be it is just a problem in sampling?

Hope it helps.

[Sereff]

Hi HPE,

Thank you for your reply. To answer your question, I have checked all my plotting and averaging scripts. They are working properly. And the y1+ is somewhere around 420k.

1. The zig-zag shape for nut plot would go away once I switched to vanDriest for delta in turbulenceProperties, However I doubt this is the correct way to do so since van Driest damping function is irrelevant for wall-modelled LES.

2. Regarding the mismatch of velocity profile at near wall region, it seems to be a very well known issue for LES of high-Reynolds-number wall-bounded flows. Here is an article that describe the overshoot problem. I am still struggling resolving it for standard Smagorinsky model. The article claims that DES models are able to avoid the overshoot to some extend however DES is not implemented for DPMFoam.


kind regards,

[HPE]

Hi,

- The y1+ is extremely high. It must be at least on the order of O(100), preferably ~30 for a zero pressure gradient boundary. I bet whatever deem problematic stems from this.
- van Driest is a damping function of nut towards walls - hence relevant. It seems that whatever you are not able to do with wall modelling can be performed by the analytical damping detached from numerical predictions of the flow.

Hope this helps.

[Sereff]

Hi,

Thank you again for you pathient reply and kind help.

I can't say anything about your first comment for now regarding the y+ issue until I change my mesh setups. And now you mention that, the 'zeroGradient' BC for pressure at a such high y+ DOES seem unphysical. I will try to do some more search on this issue.

As for you second comment, I think I didn't put my words wisely in my last comment. From what I read in Mason & Thomson's paper in 1992, and also from my supervisor, the van Driest damping function is not the correct one to utilize in a wall-modeled LES.

Finally, I am sorry this mght be stupid, but I don't quite get what do you mean by " the analytical damping detached from numerical predictions of the flow". Would you mind to elabrate a bit more?


Kind regards,

[HPE]

Hi,

- " the analytical damping detached from numerical predictions of the flow". Sorry my bad.
- What I meant in a very horrible way was that van Driest function is an semi-empirical function calibrated with canonical flows, and its behaviour is independent of the computed flow variables within your simulation. For van Driest function, the only important independent variable is the normal distance to a wall boundary, hence however your simulation is bad away from the walls, it will smoothly damp the free-wall quantities towards the wall - hence expected or plausible wall flow predictions from your simulation (these remarks are speculations since I don't know your setup details).
- Your supervisor is also right - but sometimes these heuristic/empirical function work like charm. There has been updates to van Driest function as well, e.g. Piomelli's damping function. Have a go, if need be.
- For y1+, what I meant was that the wall-normal height of the first grid in your simulation, 420 thousand, is very very high. May be it was just a typo that you wrote 420k instead of 420 above. But if 420k is what you get - you need to refine your grid towards the wall for your wall function operates within its design limits.

Hope these help.

[Sereff]

Hi HPE,

Thank you for your advise, and I'm here to kill more of your time :P

I have tried to make y+ in the range of few hunered but it seems that it didn't really help. The velocity profile maintained the same trend at the lower 20% of the domain. I kepted tweaking the coefficient of Mason&Thomson Dmaping function but it seems it can only do so much (see attactchment MT_UsemiPlus.png).

Now I am moving to dynLagrangian model, which is a variation of dynamic Smagorinsky model (lagrangian averaged scale dependent model). The problem is there is no clear tutorial or documentation on how to implement this model. Here Cs = flm/fmm dynamically and I need to initialise flm and fmm (initial values and boundary conditions). I did the following

flm
    internalField  uniform 1e-6;
    upperAtm       zeroGradient;
    ground         fixedValue uniform 0;

fmm
    internalField  uniform 1e-4;
    upperAtm       zeroGradient;
    ground         zeroGradient;

And I made it run for a very long time to stablize, I then tooke the average flm and fmm and plot the Cs = flm/fmm ad attached (Cs.png). Tbh, it looks horrible. Do you have any experience on this particular model?


kind regards,