[Geraud]

Hello everyone,

I'm currently working on slosh damping in a cylindrical tank. I try to set a two-ways FSI model to study the impact of a vibrating latex membrane located on the bottom of the tank on free-surface waves.

To start progressively, I am trying to run a simulation without any forced motion for the membrane. I want to check the deformation of this 1mm thick membrane due to the water weight.

The pressure profile monitored is more than strange and the simulation crashed after 50 iterations with "Overflow". So I guess it's a convergence problem.

[PICTURE 1]

I was thinking the problem comes from the first timestep where the solver evaluate the pressure at 12bar (it should be around 1atm).

On the first loop of my simulation I get some 'F' for the linear solution. So I tried to focus on this first loop and try to minimize the linear solution residuals.
I changed the global number of cells of my mesh from 50.000 to 1.000.000 elements. For each mesh, I change the timestep between 1e-3s and 1e-7s.
-> I obtain a minimum for the linear solution residual around: Timestep=1e-5s and Number of Elements=200000 cells. But those residuals for momentum equations are still in 1e+1.

The plot of the monitored absolute pressure is different but don't match with expectations.

[PICTURE 2]

Am I doing all wrong? Any ideas or methods to make this simulation converge? Is it possible that a simulation cannot converge regardless of the mesh size and the timestep? I'm new in CFD and are learning on the job so I may be doing stupid things!

(I already start reading the Roache's book on CFD to learn more about "sensitivity analysis" but if there is any other reference that could help, let me know).

Thank you,
Geraud.

[ghorrocks]

If you define a motion (ie no FSI) does it converge?

You will be able to get this simulation to work, but you should get individual bits of it working before you combine it all together. For instance, do a transient simulation first with no FSI and mesh motion, then add mesh motion with a prescribed motion (no FSI), then add the FSI.

[Geraud]

Thanks for answering,

Actually, I already ran a CFD analysis without FSI.

In this analysis, I use mesh motion to shake the tank and create free-surface waves: I get pretty good results, moreover, that converges easily for a very large range of timestep and mesh size. It's why I am really surprised about the instability of the FSI simulation.

At the beginning, I was worried about the membrane which is really fine and flexible but with a timestep smaller than 1e-4s, the structural part converges in one or two iterations without any problem. The fluid part, however, is getting me a very hard time, what is strange because the geometry is so simple.

Geraud.

[Azy]

Hello,

I am facing the same problem, I went through all the method that comes to my mind. Could you tell me whether you fixed your problem or not? and how?

Thanks

Originally Posted by Geraud

Thanks for answering,

Actually, I already ran a CFD analysis without FSI.

In this analysis, I use mesh motion to shake the tank and create free-surface waves: I get pretty good results, moreover, that converges easily for a very large range of timestep and mesh size. It's why I am really surprised about the instability of the FSI simulation.

At the beginning, I was worried about the membrane which is really fine and flexible but with a timestep smaller than 1e-4s, the structural part converges in one or two iterations without any problem. The fluid part, however, is getting me a very hard time, what is strange because the geometry is so simple.

Geraud.