shear stress is checkboarding

[romoli]

Hello,

I am using CS 4.2.1 for a computation at Re 1e7.
My mesh is block structured, very good quality (it's not mine, so it is not self satisfaction ).
Y+ average is 0.75 so i use k-w without wall functions.

Flow is incompressible, no thermic, with objective to compute drag on body.
I can observe that the wall shear stress (field Shear_Stress) has strange behavior (checkboarding effect, see attached picture).

I would like to know if any of CS users or developpers have already faced this problem and what could be the cause of this patterns ?
Could it be a post-processing problem or bug ? (My overall forces (pressure+shear) on the wall are correct regarding to other CFD codes)

Thanks for any help.

[Yvan Fournier]

Hello,

Which field are you visualizing ? There was a bug in the z-component of the boundary stresses postprocessing which might explain your issue, and was fixed a few months ago, so the fix is available in the latest 4.0, 4.2, and 4.3 series (4.2.1 is quite old, but 4.2.2 and 4.2.3 contain the fix).

Best regards,

Yvan

[romoli]

Hello,

Which field are you visualizing ?

I am visualizing the field Shear_Stress defined in the output for boundaries of type wall

a bug in the z-component of the boundary stresses

I was thinking of that, but the x and y component are also checkerboarding...

but 4.2.2 and 4.2.3 contain the fix

So far what i've done :
I made a restart of my 4.2.1 computation with v5.0 beta version and with just one iteration to see if it was only the post-processing writing that was messed up. But the problem is still there.

I queued a full computation with v5.0beta for the week end to see if results are different.

Best Regards,

[Yvan Fournier]

Hello,

Do you observe a similar effect on other variables ? Checkerboarding may happen with very small CFL numbers.

Otherwise how big is you mesh ? If it is small and non-confidential, we could take a look at your case. Otherwise, I'll need to check how this is computed, to provide you options for partial computation of the stress to see which term is causing the issue.

Regards,

Yvan

[romoli]

Yvan,

Pressure field looks good, Yplus shows the same checkerboarding effect than wall shear.
I made an additionnal test by switching from SIMPLEC to SIMPLE algorithm for p-v coupling.
The results are showing the same aspect for wall shear stress. No improvement so far.

If you have possibilities for helping further, I can share the model but it is quite large (3Gb for uncompressed mesh file). In that case It would be nesessary that we continue in private mode. Is it possible ?

Thank you again for your help.

[Yvan Fournier]

Hello,

Could you try also using a "cell data to point data" filter ? This is generally useful for tetrahedral meshes, where we can observe "false checkerboarding" due to the cell centers being a a different distance from the wall for neighboring cells, but this should net be the case according to the way you mesh was built... Unless your y+ values are out of the valid range for your turbulence model.

Which turbulence model do you use, and what values of y+ do you obtain ? Do you use scalable wall functions ?

With version 5.0-beta, quality criteria also allow visualizing the boundary cell thickness, which is also interesting (it checkerboarding appears at that level, the workaround for tetrahedral meshes described above is useful).

We can take this offline and I can check on your mesh as a last resort, but if we find a simpler solution first, that would save time...

Regards,

Yvan

[romoli]

Hello Yvan,

Could you try also using a "cell data to point data" filter ?

Yes, it slightly improves, but for pure Hex structured mesh, i didn't expected to need this. And still with this filter, results does not look as physical as it should be (see attach picture comp.png with comparison for "cell data to point data" filter for cs (left) and other code (right)

Which turbulence model do you use, Do you use scalable wall functions ?

k-w SST, without wall function, but i have same effect with a restart using EBRSM model

what values of y+ do you obtain ?

typically around 1 where checkerboarding is observed (see attached picture)

visualizing the boundary cell thickness

see attached picture, no checkerboarding seen at this level



Regards,

[Yvan Fournier]

Hello,

I'll try to check if I reproduce this issue on a smaller mesh (a U-bend should have some similarly-shaped cells), but If you want to send me your own mesh and setup (send a link to a download via private messaging or to saturne-support at edf.fr), and we'll take a look at this.

Best regards,

Yvan