Time step dependent mesh artefacts

[mattfalc1997]

Hi,

I have been running pipe flow simulations to better understand the causes of mesh artefacts that can occur in LES simulations. Similar to mesh artefacts previously posted on this forum, this is a time-step issue, with the implication being that this is related to the Rhie-Chow interpolation, as previously discussed. An image of two cases, which are the same but with the time step of the second case 10 times lower than the first, is shown below.

I have also attached the setup file for each case for context.

Large time step:

solu_lsq_lbe.xml

(8.22 KiB) Downloaded 126 times

Small time step:

solu_lsq_lbe_short.xml

(8.22 KiB) Downloaded 120 times

My main question is on the implementation of the Rhie-Chow interpolation and to get a better understanding of why time-step causes such a big problem. I have attached a paper which discusses the unsteadiness in the Rhie-Chow interpolation. Is the artefact issue when using small time steps related to eq. (55) in this paper and as . If so, is there a way of remedying the issue using eq. (56) which the paper states should be independent of .

https://www.sciencedirect.com/science/a ... 9118305539

If there are any questions, let me know!

Thanks in advance,
Matthew

[Yvan Fournier]

Hello,

I am not familiar with the details of the Rhie & Chow implementation, though if you want to kook at it, you can search for/grep the "arak" variable in the source code (in the current dev branch, this is in src/base/cs_pressure_correction.c and cs_velocity_pressure.c).

If you are interested in more experimentatrion, the Rhie&Chow filter may be turned off completely (search for "arak" in these examples https://www.code-saturne.org/documentat ... eters.html), but this leads to checkerboarding in many cases.

I know a colleague tested many things a few years ago to reduce this time step dependency, with no satisfying solution, but I was not aware of the article you linked to. I'll suggest to colleagues who are a bit more on the numerical side of the code to take a look at it. Otherwise, CDO solvers should avoid this issue as they are no collocated, but the work on that aspect is still in progress...

Best regards,

Yvan