How do you think the robustness of SU2?
 

Hi, guys!

I have used SU2 for almost one year, but it't still no easy for me to make a converge computation. For example, I want to compute the drag and lift of the M6 wing , which is a testcase along with SU2 , when I try a new grid (structure grid) and use MultiGrid it just diverge.

So how do you think about SU2, is it a robust tool for CFD ? Is it easy or hard for you to make a converge computation with SU2? Which tool do you use for mesh generation and which kind of mesh do you use, structure or unstructure?

Best regards,

Jason Shih

Hi Jason,

This is a very interesting an useful post. Sorry for my delay getting back to you.

Based on my personal experience what makes a CFD code more or less robust is not written in articles or papers, and it is mainly based on user experience and ad hoc tricks. SU2 1.0 started with the basic algorithms, and step by step, the SU2 developers have been improving its robustness by adding limits and meaningful checks to avoid the divergence when the grids are not optimal.

The current version has, in my opinion, a good level of robustness (any help is very welcome!) if we compare the code with non-commercial codes. Commercial codes are in other category... they are very robust but you really don´t know what do you have inside and if they are trading robustness with accuracy. With SU2 you are 100% sure about what kind of algorithm are you running.

A different chapter is the Multigrid. MG is a great algorithm, when it works, but... massive parallelization, RANS and unstructured grids degrade its performance an affects the code robustness. In fact, MG was created to solve some intrinsic problems of the explicit schemes... and keeping in mind that 99% of the time we use implicit integration my recommendation is to not use the MG.

I hope this reply helps.

Best,

Francisco Palacios

Thank you for your reply, Francisco.
Now I'm doing some wing optimization, but I can not get the converge adjoint solution. I have turned down the CFL numer and turned off MG, but they didn't work, so may it be the problem of the mesh? By the way, I'm not quite clear about the requirement for the mesh when using SU2.
As you said, in 99% of the time we use implicit integration we should not use the MG. I found that there is a option "MG_ADJFLOW", should I set it to "NO"?
The cfg file is attached below, is that any option in the cfg file should be change?
Thanks!