Artefacts

Questions and remarks about code_saturne usage
Forum rules
Please read the forum usage recommendations before posting.
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi,

The slope test for the velocity with blending factor 0.98 (2% upwind) has not been working as well. There are still mesh artefacts in the pipe flow case (please find the attached file).
Attachments
artefacts.png
Yvan Fournier
Posts: 4153
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

I showed your case to Sofiane today, and he beleives that the issue is due to a too strong chenge in cell size in the junction of you O-grid mesh. Zooming on the mesh, you can see that the cells are close to 10 times less thick just ouside the center "square". So using a less refined mesh in that area, to ensure a smoother cell size transition should help. Again, this is due to the fact that filtering is done only by the cell size, as we have o explicit filter (which would be useful but not trivial to implement).

So keeping the same refinement on the outer boundary (wall) of your mesh but thicker cells at the junction with the central "square") should help.

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi Yvan,

I have modified the mesh and trying to make it uniform in all the direction (please find the attached file). However, the solution still have artefacts. please have a look and give your view on this.
Attachments
velocity_magnitude.png
(144.08 KiB) Not downloaded yet
new_mesh_05_05_2022.PNG
Yvan Fournier
Posts: 4153
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hi,

Yes, I see that each time you improve the mesh, the artifacts diminish, but do not dissapear. The behavior confirms Sofiane's analysis, which is (summarily) that changes in cells size lead to changes in filter size, influencing the solution, but besides implementing an explicit filter (which would probably be very useful, at least as an option, but not necessarily a trivial undertaking), I do not have any "miracle" solution. If you do want to invest time in developing an explicit filter, I can try to assist, but I will also need to involve colleagues, who would probably thing this is a good idea in principal but might have other more urgent matters to handle, so I cannot guarantee anything, though I could at least get suggestions from them (and going this way depends on how much of your own time you can invest here).

Also, reading your previous posts, it is not clear to me if you compared Smagorinsky, 'Dynamic Smagorinsky, and WALE submodels. The filtering in dynamic Smagorinsky is a bit more involved than that in standard Smagorinsky (I am not familiar with WALE, but it may also be interesting), so comparing artefacts (and their amplitude) with these models my be interesting.

Also, I would need to check with our turbulence specialists (Sofiane or Jean-François), but i think Re 4000 is on the low-end of the range of Reynolds numbers we handle with LES. If I recall correctly, most of our experiments are closer to the 1e4 - 1e5 range than the 1e6 we can have in actual flows (with pressurized water), but I am not sure 4000 is so common. It would be interesting to completely cut off the turbulent viscosity model (not necessarily a good idea in terms of best practices/model validity, but at least an interesting test to understand/analyse the cause of this behavior).

I am aware of computations or liquid metal flows with code_saturne (including the "classical" wire-wrapped SFR fuel pins), but thee go back a while, and I was not directly involved, so I do not know if any special behavior was observed.

That's about all I can come up with for now.

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi Yvan,

The artifacts have been reduced by decreasing the value of pressure residual with the iterative scheme of gradient reconstruction. however, in our model (E-scape model) we have a combination of structured and unstructured mesh (tetrahedral mesh) and the iterative scheme is not working for the unstructured mesh region. The simulation has diverged in a very early stage.
Give your view on this.
Yvan Fournier
Posts: 4153
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

The hybrid (default) least squares face values estimation + Gauss reconstruction scheme should procide results similar to rhe iterative scheme without convergence issues (for the gradient itself). Witj an extended neighborhood (for the first part of the algorithm) you may ger slightly resume results.

You may still get divergence as of the complete time scheme, in which case a small portion of upwind and an extended neighborhood may help. More sweeps may help, up to a point (you already have more sweeps by default in LES than in RANS).

Besides that, I am out of ideas (the CDO version of LES should be much more robust relative to tetrahedral but is not quite ready yet).

Best regards,

Yvan
Post Reply