[lorenzo95]

Hi all!

I am simulating the flow over a 3D wing at low Re (290000) and I found that using Roe (2 order) or JST completely changes values of CD. Both methods give good results of CL instead.

In particular, JST matches pretty well with experimental results, but the simulation seldom converges (sometimes residuals remain stable and aerodynamic coefficents are almost constant and fine).

On the other hand, Roe is much more stable and faster, always converges but completely misses CD (overestimate of 5x the correct value). In particular, looking at the breakdown_forces file it is the drag related to pressure which is so overestimated.

Did somebody face the same problems or has any hints on what the problem(s) could be?

I attach the cfg with JST and Roe.

Thanks in advance for any help

[aero_amit]

Hi,
I saw your cfg files -
1. Mach number is less than 0.1, use AUSMPLUSUP or SLAU instead of ROE or JST.
2. CFL is set to high value 400, start with lower CFL, say 20 and ramp up later.
3. Add option -
USE_ACCURATE_FLUX_JACOBIANS =YES option in cfg file
4. Is flow really fully turbulent (You have mentioned Re=2.9e5)??
5. Also make sure you have proper mesh confirming required y+

[lorenzo95]

Thanks for your kind reply amit,

the flow is not expected to be entirely turbulent, I included the transition BC model to deal with the low Re.
The mesh has y+<1 all over the surface.
That CFL value was optimed for Roe (after a considerable effort in tuning) I forgot to change it.
I'll try with your suggestions and see if there is some improvement!

A further question: the option for accurate flux jacobian is appliable both to SLAU and SLAU2? I don't know this method and on the SU2 page (https://su2code.github.io/docs_v7/Convective-Schemes/) for sure it applies to SLAU2 (same for AUSMPLUSUP/AUSMPLUSUP2)

Thanks again!

[aero_amit]

Yes, option for accurate flux Jacobian is applicable to SLAU/SLAU2 and AUSMPLUSUP/AUSMPLUSUP2

[lorenzo95]

Thank you! I will study these schemes.

I tried AUSMPLUSUP and it seems now I have non-convergence issues.
The CFL I tried started from 10 and ramps up very mildly with FACTOR_UP = 1.01. Linear Solver Iterations and LS Error are as in the posted config. Apart from these parameters, have you some advice to make it converge?

[aero_amit]

I see...
While doing some test runs back then for pull request by pcg (hybrid Jacobian AUSMPLUSUP / SLAU), I observed some peculier behavior -
AUSMPLUSUP/SLAU (2 as well) will have restricted allowable CFL as you go for lower Mach numbers with option USE_ACCURATE_FLUX_JACOBIANS=YES.
So you may have to restrict the CFL to some limit only (find by numerical experiment). Don't go for very high CFL (It may not converge or even may diverge)
Now coming to another choice for USE_ACCURATE_FLUX_JACOBIANS = NO (basically inconsistent Roe Jacobian) for AUSMPLUSUP / UP2, allowable CFL was very small and its hopeless but surprisingly for SLAU / SLAU2, setting it to NO allowed much higher CFL than YES (consistent numerical Jacobian).
Basically I am just sharing my experience (also it was EULER case) .....
So... for AUSMPLUSUP / UP2, limit CFL with settings you have.
I suggest you to give a try to SLAU2 (or SLAU) with USE_ACCURATE_FLUX_JACOBIANS = NO option (try ramping to higher CFL.
Let me know the outcome....)

On another note, there is incompressible solver as well in SU2 (I don't have any experience with that)

[lorenzo95]

I tried your suggestions..

Unfortunately both SLAU with USE_ACCURATE_FLUX_JACOBIANS = NO and high CFL (10 to 100) and AUSMPLUSUP with USE_ACCURATE_FLUX_JACOBIANS = YES and low CFL (10) do not converge (and residuals stay very high).

Until now, the best shot has been JST. I think I'll work on it to make it converge.
I also saw there are low Mach Roe schemes.
Otherwise I'll move on to the incompressible solver...

Thanks a lot for your advices!

[aero_amit]

Did you give a try to VENKATAKRISHNAN_WANG limiter or VAN_ALBADA_EDGE limiter??

[lorenzo95]

Not yet..
I run many simulations varying the convective scheme and CFL. I'm still searching for a robust approach but generally JST is the only one which gives nearly the correct results without diverging. On the other hand, it is really slow to reach convergence!