[hcl734]

Hi all,

I am simulating a backward facing step with heat transfer at the lower wall after the step using Nusselt-number as benchmark.
With two relatively coarse meshes I obtain results close to experimental data.

Using a third mesh which is heavily refined to the walls I get bad results overestimating Nusselt-number especially in the region of reattaching flow.

I think it could be a problem of the near-wall-treatment in kOmegaSST-model I am using, since for Mesh 3 yPlus seriously drops below 1.
BUT I thought using nutLowReWallFunction & kLowReWallFunction (+omegaWallFunction + alphatJayatillekeWallFunction)
OF would automatically switch to LowRe-mode when the mesh get's to fine at the wall?
Is that not implemented in OF?

[hcl734]

solver: buoyantBoussinesqSimpleFoam

geometry




yPlus-values with N3 being the finest grid





Mesh criteria

[hcl734]

Is OpenFOAM sensitive to aspect ratio?
I learned it is with a limiting value of five but then I read in some sources that it isn't as long as checkMesh doesn't fail.

[ssss]

I am not quite sure, but I think that the current implementation of kOmegaSST is based on a High-Re implementation. I propose you to use this komega model instead

https://www.google.fr/url?sa=t&sourc...wr89UzCCo2wesA

Or try using kOmegaSSTSAS which seems to be low-re

Be careful always with the boundary conditions and wall functions with the different turbulence models in OpenFOAM. If you want to fully resolve the boundary layer try not using wall functions and see if it works better

[hcl734]

Thank you so much for this advice.

But from my understanding it is a "pure" LowRe-model thus always simulating the near wall region, no matter which yPlus-value present.
In Fluent there is a scalable wall function which uses wall function if yPlus is big enough and switches to LowRe (simulation) if yPlus is below 1.
Which is quite useful for "grid independent" solutions.
Is something like this implemented?

[ssss]

Originally Posted by hcl734

Thank you so much for this advice.

But from my understanding it is a "pure" LowRe-model thus always simulating the near wall region, no matter which yPlus-value present.
In Fluent there is a scalable wall function which uses wall function if yPlus is big enough and switches to LowRe (simulation) if yPlus is below 1.
Which is quite useful for "grid independent" solutions.
Is something like this implemented?

The fluent k-w-SST implementation has indeed extra terms to ensure that the model can operate at low-re and high-re mode. I think that the OpenFOAM's implementation of the k-w-SST lacks of this terms, and it always operates at high-re mode. Using the low-re wall functions can mitigate the effect of using a high-re model in a mesh with y+<30 but it is not the optimal solution.

That's why I proposed you to use the k-w-SST model in the link of my last post. It has de extra fluent terms and it can operate at low and high-re mode

[jherb]

Perhaps this helps:
http://www.cfd-online.com/Forums/ope...ng-fluent.html

[hcl734]

Ok, so I used the low Re-model but it still gives the same results.

I used the following boundary conditions

alphat:
fixedValue 1e-12;

nut:
fixedValue 1e-12;

k:
fixedValue 1e-12;

omega:
omegaWallFunction;

as recommended in the mentioned thread
http://www.cfd-online.com/Forums/ope...ng-fluent.html

boundary conditions for walls are:
U=(0,0,0), p=zeroGradient, k=1e-12, omega=omegaWallFunction, nut=calculated (or 0)

Should I use no WallFunctions at all?
Then I would put omega to a near zero value as well?

This makes the results even worse, so does anybody has an idea what could cause those problems, if the wall functions are not the matter?

I uploaded an example of the calculations using the finest mesh here (including latest time step result):

https://owncloud.tu-berlin.de/public...fe3a71201fbfa5

[hcl734]

Using only omegaWallFunction




Using no wall function at all

[jherb]

You also might find these threads helpful:
http://www.cfd-online.com/Forums/ope...estigated.html
http://www.cfd-online.com/Forums/ope...ons-2-3-a.html

[hcl734]

First of all thank you for your suggestions.

What I got from the threads you pointed me to is:

kOmegaSST_lowRe with omegaWallFunction

or without wall functions at all using Wilcox BC for omega

I did both and the results still deteriorate with increasing mesh density, since I am not using wall functions here I assume it is a general mesh problem. Or we simply shouldn't use kOmegaSST for yPlus lower 1.

Could you answer the question for me how sensitive OF is to mesh parameters?
Since I got quite high aspect ratios and I am not sure whether this could be the main problem.
checkMesh tells everything is ok and I imported the meshes from Ansys where they work just fine.