Learning Forum

[stefan.krischer]

Dear all,

I am trying to perform a quite challenging transient simulation of inflowing air into a system using Ansys fluent. Due to the high speeds and large pressure gradients, I need a quite fine mesh on a quite large geometry. Therefore, to save computing power, I do the simulations in a 2D Axis symmetric setup. In some first tests, meshing and simulations on smaller and easier geometries was very successful, with element numbers in the order of 1 million. Now I am trying a lager geometry, but when using the meshing settings which were well proven for the easier geometry, I am getting the Error "A software execution error occurred inside the mesher. The process suffered an unhandled exception or ran out of usable memory." This always happens after multiple hours of meshing, when the progress indicator says it's "Converting triangles to quads" for the largest Volume in terms of number of Elements (it already displays that step for quite some time before showing that error).

I already tried different things:

1. Monitoring my used RAM. This never exceeds about 25% usage (I'm using quite a large memory machine due to the large number of cells).
2. Monitoring my used Disc-Space (as indicated in another thread) as well for the .ansys folder as for the Workbech_files folder. Both did not grow significantly during the process and were far away from disk limitations.
3. Reducing the number of Elements in that part. Then the meshing works, but my simulation diverges. Therefore, this is not an option.
4. In an official answer to how to response to this error, I found "Try different meshing methods. For example – try patch independent method instead of patch conforming tetra or Multizone instead of Sweep." but for the 2D meshing I do not find any of these options, and therefore assume they are not present in this case.

Does anyone have an idea what could cause the problem here, or how to fix this? That would be really helpful!

 

Here some more info on what I am doing and using:
I am using Meshing in the Fluent Workflow of Workbench 2024 R2 with a teaching license on an Alma Linux System with 48 Cores and 128GB RAM.
The system is in the order of multiple meters, the part where it fails to mesh has a length of about 3m and a height of about 2.5m. With a mesh of 2mm (using a sphere of influence) the meshing in that region was successful, but I need to refine it to 1mm (at least in some parts of the geometry). In the rest of the system I use a 5mm mesh.
For the rest, I am using the standard settings (No Adaptive Sizing, Growth Rate of 1.2, Mesh Defeaturing On, No Inflation, etc.) which are set when opening Ansys Meshing for a 2D Mesh from Workbench.

Here is a screenshot of the whole system I am trying to mesh:

This is not a very complex geometry and the selected (green) part is the one it is working on when crashing, as it needs a finer mesh than the rest of the system.

If you need any more information, I will very happily provide any.

Thank you in advance!

[Rob]

There's a surface facet limit to prevent silly cell counts in 3d.  If you have a 1mm cell in the green region, what is the (rough) cell count that you think you'll get? 

[stefan.krischer]

That will be in the region of 5-6Million cells. That is a lot, but not completely unreasonable for my understanding. Would that already reach that limit? And if yes, is there a possibility to reach such a cell count anyways?

[Rob]

The old limit was around 100k, that is likely to have changed but I've never seen it reported in the release notes. Try breaking the face down a bit more and start with the finer mesh region. Or use a coarser mesh and review adaption in the Fluent solver. 

[stefan.krischer]

Thank you a lot by now! Breaking the face down, helped in a way, that now a different message appears "The meshing process ran out of memory" although again it used only about 30% of the usable memory in the machine. Is there a build in memory limit in the software? Or is there another option to save memory?

[Rob]

You may still be hitting the limits I mentioned; you may have RAM free but Meshing may not be able to allocate that much.  Meshing was (and is) the Mechanical meshing tool, and was extended to CFD meshes in about R14. Whilst the mesh is good, the sizes we want has always caused some issues. Mechanical users often think a 1-2M cell model is huge, for CFD it's tiny by modern standards. 

[stefan.krischer]

Perfect, that actually worked! It still took several hours of meshing, but I'm just glad it worked. For reference and if anybody else get's this problem, I attached a Screenshot of the sliced up face which was then successfully meshed with 4.1 Million cells. Thank you a lot @Rob!