Migration to CS 3

[Luca]

Dear all,

I'm trying to move to Code_Saturne 3. The installation worked fine but i'm not able to run simulations. I changed the path to the new installation directory but still there are some problems:

Parallel Code_Saturne with partitioning in 2 sub-domains

Code_Saturne is running
***********************

Working directory (to be periodically cleaned) :
/home/luca/tmp_Saturne/RUSSKY.CASE1.11241822

Kernel version: /opt/code_saturne-3.0.0-betaR4215
Preprocessor: /opt/code_saturne-3.0.0-betaR4215
./runcase: line 285: /opt/code_saturne-3.0.0-betaR4215/share/ncs/runcase_mpi_env: No such file or directory
Usage: /opt/code_saturne-3.0.0-betaR4215/bin/code_saturne <topic>

Topics:
help
autovnv
compile
config
create
gui
info
run
salome

Options:
-h, --help show this help message and exit

I guess that the problem is in the runcase file but I cannot find out how to fix it.
Does anybady have a fully warking example for CS 3.0? The Tjunction given in the examples folder doesn't seam to be complete...Or maybe I'm missing something...

Thanks
Luca

[Yvan Fournier]

Hello,

The tutorials for version 3.0 are probably not up to date (I beleive they were updated for a recent training sessions, but they might not have been commited yet).

Sorry about that, but the simplest is just to create a case using "code_saturne create" (add "--help" for options), and work from that.

The tutorials will be fixed by final release (expected January 2013).

Regards,

Yvan

[Luca]

Thanks you again Yvan!
now it's working!
Luca

[Luca]

Dear all,

I keep having severe problems with CS 3. I attach a screenshot of the pressure field around a bridge deck. The case is extremely simple: inlet, outlet, wall at the deck and symmetry in all other BC. The case is essentially 2d because only one cell (extruded from 2d) is present in the thickness. I have quite many doubts:

1) The mesh I used originally had hexa cells at the boundary domain that have been transformed splitting them. Is it normal? Is it only graphical? This happened also in the boundary layer close to the deck.

2) Why do I get this non physical pressure field?

3) Did anything change with respect to version 2.2 in the wall treatment? I cannot get the same results...the flow is unnaturally attached (it was not in 2.2). I'm using a k-omega sst with 1 scale wall function and y+Max=5.
I read that the RANS model in CS are all high Reynolds number models. Can I integrate through the boundary layer?
Is the two-scales wall treatment a possible approach for massively separated flows? What kind of y+ should i seek for?

Many many thanks for the support!

Luca

[Yvan Fournier]

Hello,

Are you sure you are using the same mesh ? You could get this sort of issue with a mesh with different element types and a different scaling.

Otherwise, could you post the "listing" files you obtain with versions 2.2 and 3.0-beta, for comparison ?

Regards,

Yvan

[Luca]

Dear Yvan,

the mesh I used before didn't have the structured part at the boundary. As explained in the file attached I solved the problem using Extrapolation instead of Neumann to calculate pressure gradients at boundaries.
I attach a file where you can find some details regarding the difficulties.

Sorry to abuse your kindness!

Thanks,
Luca

[Yvan Fournier]

Hello,

Does the "pressure waves" problem dissapear when you switch from extrapolation of pressure gradient on domain boundary by Neumann 1st order ?

Also, you mesh seems to contain tetrahedra only (judging by the histogram of the number of interior faces per cell, which maxes at 4). As your computation seems to be basically 2D, extruding a 2D mesh (producing prisms) would help. Also, the structured part you add at the inlet would truly be boundary aligned.

Regarding turbulent models, Martin has made some changes to defaults and fixed some bugs. So this could explain some things, but the mesh also has a major importance. I would suggest checking results with different turbulence models and possibly scalable wall functions, but I'll let Martin or Jacques give you recommendations, as they are more knowledgeable than me on that count.

Also, there are both high and low-Reynolds numbers in Code_Saturne. For low-Reynolds models, an y+ value close to 1 is recommended. For high-Reynolds models, y+ closer to 10-30 is better, but here again, I'll let specialists confirm this.

If your data is not confidential, you may also post it here.

Regards,

Yvan

[Luca]

Dear Yvan,

I confirm that the pressure waves disappear is I use EXTRAPOLATION instead of NEUMANN.

I totally agree with the horrible mesh alignement, the problem is that when i built the mesh i extruded the quad obtaining hexas. If I visualize the results in paraview the outer skin is still correct (quads) but inside the quads are splitted! I cannot really understand how this happens. This is done automatically at some point (i will check Salome) but the question is, shouldn't it change the mesh topology making the postprocessor fail?

I will ask if I can send the data!

Thanks a lot!

[Luca]

Dear Yvan,

I cannot send exactly the same case but I send you another that before worked fine with CS 2.2.0. I rebuilt it for CS 3Beta and I can observe the same problems that i described yesterday.

There is the case, the case used with CS 2.2.0, some images and a README with some comments.

Many thanks for your help! I hope it will be useful!

Luca

[Yvan Fournier]

Hello,

I have not looked at your other case yet, i am having second thoughts regarding your mesh: according to the "listing" files, you have at most 4 interior faces per cell. This would happen either with a tetrahedral mesh, or with hexahedra extruded on just 1 layer (as you seem to have 155230 boundary faces vs. 76370 cells, I would assume the latter). So your mesh is probably actually OK, and you only have a visualization problem. You may check this also with the "preprocessor.log" files.

I'll check with other colleagues regarding the turbulence issues.

Regards,

Yvan