[abonfi]

I have tried to run TestCasesv3.2.0/rans/naca0012 using triangular grids, rather than quadrilateral ones.
The triangular grids have been obtained by cutting into triangles the structured meshes available on the NASA website.
The three coarsest meshes are available here: https://dl.dropboxusercontent.com/u/60787983/meshes.zip
Using the same configuration file (except for the names given to the boundary patches) that runs successfully with the quad mesh, SU2 3.2 generates NaNs after a few non-linear iterations.

Aldo

[fpalacios]

Interesting question... I think you should play with the value of the limiter a small value ~0.01 could be a good idea. And, obviously, reduce the CFL and disconnect the MG.

This is a good question, we will probably chat about this in our next YouTube video.

Cheers,
Francisco

[akun646]

Hello,

I have tried to run the triangular mesh (the medium one from the website), with both the VENKATAKRISHNAN and BARTH_JESPERSEN limiters. It did work, but the best residual I've got was only about 10E-3 and the results are still quite different from those of the quad mesh. So, still an open question!!

Cheers!!!

Facundo

[fpalacios]

Please find in this link the reply to this post.

"This week in SU2": Wednesdays at 10:00am (PDT) on YouTube.

Best,
The SU2 team

[abonfi]

Thank you for the video reply to my post.

Here

https://dl.dropboxusercontent.com/u/...3/naca0012.pdf

you can find a short document that reports numerical experiments with
the http://turbmodels.larc.nasa.gov/naca0012_val.html testcase, at 10 degrees AOA.

Figs. 5.7, 5.8 and 5.8 show comparisons between my own code, SU2 and the CFL3D data available on the NASA website.

Please note that the quad grids available in ver. 3.2.0 of the SU2 testcases have not been labelled correctly:
mesh_NACA0012_turb_897x257.su2 has NELEM=57344, so it has been derived from the 449x129 structured mesh, not the finest one.

I agree that as long as you can generate a quadrilateral boundary layer mesh,
that is probably preferable than using a triangular one.
I am not using unstructured FV schemes, but Residual Distribution (or Fluctuation Splitting, similar
to SUPG-FEM on iso P1 elements) ones, which require simplicial elements (triangles and tets).

Concerning testing SU2 on triangular boundary layer meshes: I am now using the
grid (and testcase) shown in Fig. 5.10(a) of the aforementioned document.
It has a "nice" triangular boundary layer mesh and is truly unstructured elsewhere.

Re-starting the 2nd order limited calculation from the 1st order one (as you suggested) is ok,
but iterative convergence of the 2nd order calculation remains problematic.

Regards,
Aldo