[zloty001]

Hello,


I am beginner in opensource CFD and it is my first - non tutorial - SU2 simulation and here is short description of my case issue.


I have some trouble with coverage compressible simulation over airfoil (M=0.85). I ran my calculation and residuals are not "going down" so I stopped SU2 after 1000 iterations. To compare in Fluent it took me about 300 iterations to converge solution. Moreover comparing to Fluent solutions, the SU2 solution has a hudge difference (Cd in Fluent: 0.0643, Cd in SU2: 0.13). I don't know why the difference is so huge and how to make my results more accurate.


My config file is based on turbulent NACA 0012 tutorial although I made some changes, but in my opinion it shouldn't be a problem. Nevertheless, I attach my config, maybe I am missing something. I would appreciate if someone could see into it. I used Multigrid as it should make calculation converge faster and I used linear solver as well.

My second though "why can't it converge' was that it is due to mesh. I made structured mesh in ICEM and in my opinion it is fine. I would be grateful if someone could look into it. Moreover, I am thinking whether triangle mesh would be better to calculate for SU2?


Below I attach only residuals plot.
Here you can see config file, mesh and results (SU2 and Fluent to compare). https://app.box.com/s/hqseqg9qds2l0h8ivz5asziuurv7qhq0

Maybe one more sentence, why I am using SU2. This is because I am doing CRM wing optimization project on my University. I am going to optimize 6 section of CRM wing and compare "2D wing optimization" (through airfoils optimization) to the whole 3D wing optimization.


Thanks in advance for your help !

[zloty001]

In addition to post above, I am attaching some mesh screenshots.

[talbring]

Hi zloty001,

I had a look at your case. I modified some config options and it converges now better (https://gist.github.com/talbring/b16...b0ea3bc7b59494). See the attached picture.

However, this seems to be a tough case due to shocks on the upper and lower surface. Plus there seems to be a shock-induced separation on the lower surface.

Still, I cannot tell you why the values for C_D are different from Fluent (I get 0.105 now).

Regards,
Tim

[zloty001]

Hi Tim,

thank you for your help. I see that main change in config was this line:

REF_DIMENSIONALIZATION= FREESTREAM_PRESS_EQ_ONE

and adding adaptive CFL number up to 100.

Non-dimensionalization of the Navier-Stokes Equation helped in converge and to be honest I've missed this parameter and I assume that I wouldn't pay attention to it, so once again thank you, I am glad you helped me. Problem with converge is solved .


However, I looked deeper into the differences between Fluent and SU2. I thought it could be helpful to run some validation case, so I used NACA0012 validation case. It gave an interesting results which I attach below.
In both cases in Fluent and SU2 I used the same BC and method: Re=9.0e6, M=0.82, alpha= -0.14, RANS, SA.
You can see, that SU2 provides more reliable results of pressure distribution than Fluent.

So returning to my case and the differences in calculation, I believe Cd calculated by SU2 is more reliable than Fluent. The difference in Cd between SU2 and Fluent is quite significant - around 30%.

Validation data: https://ntrs.nasa.gov/archive/nasa/c...9810014503.pdf

Regards,
Zloty