[semions]

Hi,

I'm trying to simulate blood flowing inside an irregularly shaped 3D vessel with one inlet and one outlet. The inlet is not normal to x, y or z. For now, I'm running steady state simulations.

I'm using SU2 v 7.0.2 and I started by editing the cfg file from a 2D channel example. In my case, I'm prescribing the an inlet velocity profile from a file.

I'm reaching convergence in a few iterations, but I'm not convinced by the results. In particular, I don't understand:
- why the INC_VELOCITY_INIT value will affect the computed velocity field so much,
- how to properly set the CFL number,
- why the velocity profile doesn't seem to fully develop to a parabolic-like profile,
- if there is a way to strongly impose the inlet boundary condition.


Many thanks.

[pcg]

You have INNER_ITER set to 20, your case cannot possibly be converged.
I guess that is why you see a dependence on initial conditions and no parabolic profile.

Increase INNER_ITER to a large number, say 10000, and monitor the residuals and some post processing metric like drag. For a simple problem you want at least 6 orders of magnitude drop in residuals and the value of drag to be the same for a significant number of iterations.

CFL for steady state problems acts as a form of under relaxation. For a laminar problem you should be able to use a large value 1000 or more.
With very low values your case will never converge.
With very large values the solver will become unstable.

[semions]

Thanks for the advice.

I increased INNER_ITER and I tried different combinations of CFL number and linear solvers.
I can't make my simulation converge. The residuals start from -6 and if CFL > 10, the residuals increase and the simulation eventually diverges; if CFD < 10, the residuals won't go below the initial orders of magnitude even after a few thousand iterations.

I wonder if there's something that I'm missing, since I've run this kind of simulation many time with openfoam and it converges in a few minutes.

Thanks again.

[akoodlyr]

Hello,



If possible could you share the mesh file?



Also, can you try the simulation by setting the INC_NONDIM to INITIAL_VALUES instead of DIMENSIONAL and see if you notice a difference? I know in theory there shouldn't be, but I have noticed some weird convergence behavior when using dimensional simulations with very large values for density or velocity.


Cheers

[semions]

Unfortunately, my mesh is too large (approx. 98 MB) and CFD online won't let me upload such a file.

[akoodlyr]

Ok, no problem.

Did you have any progress by changing the non-dimensionalisation?

[semions]

No, no noticeable change in the residuals.