"nfaced" with periodic boundaries

Questions and remarks about code_saturne usage
Forum rules
Please read the forum usage recommendations before posting.
knewlands

"nfaced" with periodic boundaries

Post by knewlands »

Hello,

I'm using C_S version 2.0.5 and, whilst I've never had problems with periodic boundary conditions in the past, in my latest run on a basic annular configuration I obtained "nfaced" values in the output files. There is no mention of the "nfaced" cells in the pre-processing stage when I check the mesh and I have identified that this only happens when I apply periodicity between the inlet and outlet. The "nfaced" cells seem to correspond to the extremities of the domain in proximity of the inlet and outlet boundaries.

The command I use has worked in the past, but in any case this is what I add to the runcase file: "--perio --group INLET OUTLET --trans 0 0 0.45". I have also attached the output from the mesh_check.

Can anyone suggest what the problem might be and how I might fix it?

Thank you,

Kristin
Attachments
test_mesh.log
(17.54 KiB) Downloaded 253 times
Brian Angel

Re: "nfaced" with periodic boundaries

Post by Brian Angel »

Hello Kristin,

I've just tried a simple test case using V3.0 with the dimensions given in your test_mesh.log file. CS V3.0 does not seem to have a problem. Can you send me the output files of your calculation so that I can have a look.

Regards,

Brian Angel.
knewlands

Re: "nfaced" with periodic boundaries

Post by knewlands »

Hello Brian,

Thank you for your reply. The output files are rather large, so I'm not sure I can send them through this forum. Essentially, in the output text files for the volume, I have a list of values for hexa8, until approx. the last 4000 values, which are preceded by the word "nfaced". I checked the surface output file and strangely it lists about 100 tri3 values before the remaining correct quad4. The grid was generated as a block-structured mesh in ICEM with hexahedral cells and there should not be triangular faces. When I do not apply periodicity, all output values are for quad4 and hexa8.

When I open the EnSight case file in ParaView, it looks as though the outer wall boundary condition doesn't extend all the way to the inlet, as it should. Instead, what should be the outer wall just downstream of the inlet looks like internal flow.

I'm not sure if this information helps in diagnosing the problem. Please let me know if there are any files that would help specifically and perhaps where I could upload them.

Thank you,

Kristin
Brian Angel

Re: "nfaced" with periodic boundaries

Post by Brian Angel »

Hello Kristin,

Can you please check the output in the file "joining.case" which provides information for the periodic condition. Can you also post this file as well as some images of your volume mesh (where the tri3 faces appear in ParaView and the Inlet and Outlet surfaces) on the forum so that I can download them and have a look here.

I would also like to confirm that, for the groups Inlet and Outlet, you are not assigning boundary condition values to these groups (e.g. a velocity) but you are using, for example, a source term to drive the flow through the flow domain from inlet to outlet (e.g. a prescribed pressure drop).

Best regards,

Brian Angel.
knewlands

Re: "nfaced" with periodic boundaries

Post by knewlands »

Hello Brian,

With regards to the boundary conditions, I apply a pressure drop to drive the flow and I've never had a problem with this method in the past. I'm not sure what the issue is this time, but the tria3 and nfaces values don't appear without periodicity.

I can't find a joining.case file, but I've attached the output log of the mesh_check with periodicity applied (tria3 faces are mentioned). I've also attached images of what the mesh should look like (preprocessor_mesh as well as an image of the inlet superimposed on the outlet, as generetad in ICEM) and what the output inlet and outlet faces actually look like. I cannot see the tria3 faces in ParaView, but judging by the coordinates in the output files they should be along the outer wall at the inlet (axial location ~ -0.25e-09 where the minimum z-coord should be 0). I haven't attached the output files as the surface one is 4MB and the volume one is 87MB in size.

Does anything in the above suggest where I could've made a mistake?

Kindest regards,

Kristin
Attachments
perio_problem_mesh.tar.gz
(1.83 MiB) Downloaded 220 times
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: "nfaced" with periodic boundaries

Post by Yvan Fournier »

Hello,

You images are difficult to compare, as one seem to be a volume mesh, the other a surface mesh, and some polyhedra don't appear in the inlet image (either a bug in ParaView, or a too old version).

First, what is exactly the issue ? does your computation fail to run ? or are you simply surprised that the periodic joining is non-conforming when it should be conforming ?

It seems as though the periodic faces do not overlap perfectly, a few INLET faces near the wall overlap a bit more than the matching OUTLET faces. Did you experiment with periodic joining parameters ? (increasing/decreasing the fraction, and possibly using a stricter "plane" parameter).

Regards,

Yvan
knewlands

Re: "nfaced" with periodic boundaries

Post by knewlands »

Hello,

Thank you both for your help. I must apologise as I really should've taken a closer look at the original mesh. The calculation did run, but I couldn't understand why I had non-conforming faces. The problem has been solved by fixing the original mesh :oops:

I'm also having an issue of a different kind with a much larger mesh (~ 23M cells). I have deliberate non-conforming faces in this mesh, but when I try to check the mesh I obtain an error in ECS execution saying that it cannot allocate the memory, even though I run it on a high memory node on the cluster. I've attached the .log file and I wonder if you could suggest what I'm doing wrong?

Thank you,

Kristin
Attachments
mesh_check.log
(13.95 KiB) Downloaded 234 times
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: "nfaced" with periodic boundaries

Post by Yvan Fournier »

Hello,

Did you configure/compile the preprocessor with the "--enable-long-int" option ? Not sure it will be enough, but it might help, if the issue is due to 32-bit integer counters not being large enough for some sizes (though I'm not sure: these types of issues should occur more around 50 to 80 million cells than at 27 million).

Regards,

Yvan
knewlands

Re: "nfaced" with periodic boundaries

Post by knewlands »

Hello Yvan,

Whilst I installed the third party libraries manually, I used the installer and the setup file to install Code_Saturne. As a result, I didn't add specific options when configuring/compiling the preprocessor. Forgive the basic question, but how would I go about adding the "--enable-long-int" option now that I have the code installed?

The current installation has worked well up to 15 million cells and this is the first time I've tried to use a mesh of 23 million elements with non-conforming faces. Is it sufficient to use the "code_saturne check_mesh -m meshname.cgns" command or should I be specifying further options?

Thank you for your help,

Kristin
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: "nfaced" with periodic boundaries

Post by Yvan Fournier »

Hello,

to add --enable-long-int, if you still have the build directory from the installation, "cd" to the preprocesser (ecs) build directory, and check the top of the "config.log" file. Copy-paste the logged "configure" command and its options to a terminal (in the same directory), add --enable-long-int, then run "make && make install".

I don't remember which partitioner you are using. Metis 4 tends to fail around 20 million cells (YMMV depending on what cell type is dominant). Metis 5 or Scotch should be fine with 23 million cells.

Regards,

Yvan
Post Reply