[tomaspzapico]

Hello,

I am trying to run a turbulent NACA 0012 simulation with SA and JST. I am using a CFL of 5 but it took a large and not reasonable number of iterations for it to converge. I tried to accelerate the convergence by using the CFL_ADAPT parameter, but when I monitor it I see that, even though it is supposed to follow the equation mentioned in this post CFL Adapt parameter, it just multiplies the old CFL by the factor I put in the configuration file each iteration until it reaches the maximum CFL (50, because with higher values the solution diverges). Nevertheless, this did not help much as it only reduced the number of iterations from 22000 to 20000 more or less.

I have also tried running the simulation for a number of iterations and then restarting it, but nothing seems to work. Additionally, I have reduced the convergence criterion from 0.01 d.c. to 1 d.c., but it provided a solution with a very high error when compared to other solvers and experimental data.

Could anyone help me please? You will find the mesh and configuration files in the following link: https://github.com/tomaspzapico/SU2.git

[bigfootedrockmidget]

Hi,
You have to make the repository public :-)


We have many naca0012 tests in the testcases section of the su2 repository. Most of them work pretty well and converge within 1000 iterations. On pure triangular meshes or pure quadrilateral meshes, multigrid also works pretty well. Maybe you can check what the difference is in setup between what is in the repository and what you have?

If in the end it is the mesh, then it would be good to check what makes your mesh different from other meshes.

[tomaspzapico]

I have already made it public.

[bigfootedrockmidget]

Just switch off the MUSCL scheme for the turbulence equations. And this case also benefits from multigrid, if you lower the relaxation factors a bit.

[tomaspzapico]

Thanks for your response. I had also tried switching off the MUSCL reconstruction for the turbulence equations, but the results for the CL and CD were further away from the experimental data I am using. I will give the multigrid a look in order to accelerate the convergence.

[bigfootedrockmidget]

muscl for turbulence equations is not very robust at the moment. If the results are still far away, you might need a finer mesh.

[tomaspzapico]

I mean that with MUSCL on for the turbulence equations I got more accurate results than without MUSCL