[A.Roy]

Hello SU2 community,


I am trying to simulate an empty wind tunnel geometry in 3D as preparation for in-tunnel airfoil CFD. As part of this I have a very coarse 2M cell mesh just to verify the setup; however, I am consistently getting divergence with Riemann inlet and outlet conditions as per the suggestions from the below threads.


Does SU2 implement any non-reflecting boundary condition?


Divergence occurs at CFL=0.8 (adapt = YES) around 938 iterations. I checked the last solution write before divergence and it appears a spot of low pressure / high velocity forms in the middle of the outlet plane; wondering what might be contributing to this - any ideas if it just poor mesh quality and / or sensitivity to separated flow, or something else like my boundary conditions or lack of upstream / downstream length?


Brief mesh quality metrics from SU2 listed below:

+--------------------------------------------------------------+
|           Mesh Quality Metric|        Minimum|        Maximum|
+--------------------------------------------------------------+
|    Orthogonality Angle (deg.)|        33.6256|        89.9995|
|     CV Face Area Aspect Ratio|        1.08326|        41330.2|
|           CV Sub-Volume Ratio|        1.00427|        146.656|
+--------------------------------------------------------------+

Config file, Mach contours on the outflow face, surface mesh on the outflow plane, and slice of volume mesh attached.


Thank you!

[pcg]

Do you really need non reflecting boundaries? Why not just use conventional inlets and outlets?

[A.Roy]

Hello Pedro,

I thought I may have required non-reflecting to try and minimize transient fluctuations upon startup (as it seems this is 1 of the 2 major issues that were giving me trouble), but it seems to work with regular inlet/outlet upon some tweaking, thank you.

If anyone else browsing this is running into similar trouble, some notes / comments:

Potential Issues / Fixes
- Almost always seems to be a culprit, but mesh quality especially near concave corners; turns out I had some fairly skewed elements in my near 90-degree corner of the flap cove of my 2D in-tunnel model
- For 2D startup with the in-tunnel model, and 3D startup regardless of empty tunnel or model, if the initial conditions are too far from the steady or quasi-steady state (and this may be even relatively minor variance in the case of sensitive in-tunnel conditions), this may cause divergence. I've found that I usually need to run about 250 - 500 iterations (about ~1.5-2 orders of magnitude of residual reduction) of Lax-Friedrich to get past initial transients interacting with potentially non-ideal cell quality. Word of caution: running too far with the 1st order scheme such as Lax-Friedrich may cause divergence issues or a bunch of delay after 2nd order restart as the solution corrects
- In some cases I was able to run Lax-Friedrich to overcome initial transients at something like CFL=8-10, but if I left that at 2nd order restart, the divergence re-appeared; I needed to switch to a lower CFL (1 or 2.5-5) sometimes when switching from Lax-Friedrich to Roe or JST_MAT