Strange numerical divergence in FSI example

[kvslab]

Strange numerical divergence in FSI example

[keiyamamo]

Dear @zhangmuElias , thank you for sharing it with us and good to know it worked for you! Please feel free to close the issue.

[zhangmuElias]

ok, dear @keiyamamo , thanks for the discussion with me!

[keiyamamo]

Hi @zhangmuElias
you wrote that the left end of plate is fixed, but i can't seem to find such a boundary condition. In fact, I don't think I can find any bc for the plate.

[zhangmuElias]

Dear @keiyamamo, thanks for replying. In fact I still used the name barwall for the left circle bc of the plate, which has the id of 15.
Do you think it's enough?

[zhangmuElias]

Dear all:

To update, I tested other mesh models for the case file I attached above.
It is surprising that the Laplace mesh model with volume_change, constant and small_constant successfully simulates this case with water as the fluid medium. The strange velocity surge near the head and tail of the plate disappears.
Therefore at least there are no missing settings in my attached case file above.

I guess when the fluid viscosity changes from 1 [Pa.s] to 1e-3 [Pa.s], the diffusion effect decreases which might make the solver numerically unstable. The Laplace mesh model may serve itself as an extra diffusion term to stabilise the solver.

I also tried different parameters of alpha_u in Biharmonic model from 0.01 to 0.1 to 1 to 10, but all of them failed with the same problem I reported above.
There is a possibility Biharmonic model may have some difficulties in dealing with low-viscosity problems, I need to check it from articles. I am not sure about the exact reason. Did you try using the Biharmonic model in fluids like air or water before? It might help if others can share experiences.

Figure: the result of the reduced scaled case with water as fluid (Re=200)

Best regards
ZHANG

[keiyamamo]

Hi @zhangmuElias

Good to know it worked. I don’t have much experience with biharmonic model, thanks for sharing your experience with us.

[zhangmuElias]

Dear @keiyamamo, then in case of fsi problem, which mesh model do you use? Although the Laplace model can eliminate the velocity and pressure surge problem in the low viscosity situation, I also found that whenever the displacement is a little bit too large, the Newton solver will diverge. So this modification of biharmonic to Laplace can only partly solve this problem.

[keiyamamo]

I use Laplace model and it works fine for us because the displacement is not so large. One thing that I noticed now is that your mesh size changes dramatically from boundary layer to adjacent cells. Additionally, I’m not sure if it’s a good idea to have boundary layer around the moving object since boundary layer cells have skewed shape, potentially causing “crash” of cells when deformed.

[zhangmuElias]

Dear @keiyamamo, thanks for noticing me about the mesh.
I also tested the original benchmark case TF_fsi inside the file folder problems. It also diverged at time t=3.7s, and I noticed that it was the time when the plate started to vibrate. But, as shown in the article "Slyngstad, Andreas Strøm. Verification and Validation of a Monolithic Fluid-Structure Interaction Solver in FEniCS. A comparison of mesh lifting operators. MS thesis. 2017." the laplace model can at least handle the fsi2 and fsi3 benchmark case. I am not sure why the TF_fsi using laplace diverged in my computer, have you tried that?

Figure: The result of TF_fsi case at the last time step

[keiyamamo]

Hi @zhangmuElias
Unfortunately, no. I have only tried biharmonic type for TF_fsi problem.

[zhangmuElias]

Dear @keiyamamo , ok I see, thanks!

[zhangmuElias]

Dear @keiyamamo , in these days' tests, I found the parameter of the biharmonic model $\alpha_U$ should be reduced for low-viscosity fluid. In my case, when $\alpha_U$ equals 1e-8, the problem I reported above can be solved.