I am trying to utilize the block solver to couple the Helmholtz and StressSolver for an FSI problem as demonstrated in the group of tests/ShoeboxFSI group of test cases.
I have two questions.
This test case was introduced as strong FSI coupling, does this mean two-way coupling(or fully coupling)?
Is there a reason why convergence is not possible(or too many iterations needed) at some frequencies?
FSI coupling
-
- Site Admin
- Posts: 4823
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: FSI coupling
Yes, the test cases correspond to strong coupling.
You could solve the system as a monolithic one. Probably the block preconditioning does not work so well near resonance.
-Peter
You could solve the system as a monolithic one. Probably the block preconditioning does not work so well near resonance.
-Peter
Re: FSI coupling
Thank you Peter.
But I don't know what you mean by "strong coupling" and "the system as a monolithic one".
Some definitions, the way I understand them:
Single-way coupling : The structure is deformed, and the deformation is applied to the fluid as boundary condition. No iterations.
...versus:
Two-way coupling : The deformation of the structure perturbs the fluid, and the fluid perturbs back the structure, and so on until convergence.
Fully coupled: All the unknowns (pressures and displacements) are solve at the same time in the same matrix
...versus:
Segregated: Treats each physics sequentially, and iterates until convergence.
So, by "strong coupling" do you mean "Two-way coupling" ??
Thanks again
But I don't know what you mean by "strong coupling" and "the system as a monolithic one".
Some definitions, the way I understand them:
Single-way coupling : The structure is deformed, and the deformation is applied to the fluid as boundary condition. No iterations.
...versus:
Two-way coupling : The deformation of the structure perturbs the fluid, and the fluid perturbs back the structure, and so on until convergence.
Fully coupled: All the unknowns (pressures and displacements) are solve at the same time in the same matrix
...versus:
Segregated: Treats each physics sequentially, and iterates until convergence.
So, by "strong coupling" do you mean "Two-way coupling" ??
Thanks again
-
- Site Admin
- Posts: 4823
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: FSI coupling
Hi,
Sorry for being ambiguous. These have two-way coupling that is basically solved from the same matrix equation except the matrix is treated as a block matrix and block matrix preconditioning is used whereas the outer iteration minimizes the residual of the full system using GCR. There is a possibility to solve the system also as fully monolithic one which will allow the use of direct solvers but consumes more memory and has worse scalability.
-Peter
Sorry for being ambiguous. These have two-way coupling that is basically solved from the same matrix equation except the matrix is treated as a block matrix and block matrix preconditioning is used whereas the outer iteration minimizes the residual of the full system using GCR. There is a possibility to solve the system also as fully monolithic one which will allow the use of direct solvers but consumes more memory and has worse scalability.
-Peter
Re: FSI coupling
Thank you peter , it's clear now.
I'm comparing the result of Comsol and Elmer, but it's not perfect yet.
How can I fix my sif file to get better results?
I'm comparing the result of Comsol and Elmer, but it's not perfect yet.
How can I fix my sif file to get better results?
- Attachments
-
- case.sif
- (7.8 KiB) Downloaded 47 times
Re: FSI coupling
I'm testing the B.SIF file, Do you think the results will be different from A.SIF file?
B takes too long to converge, so I think it will take too long to compare with A.
B takes too long to converge, so I think it will take too long to compare with A.
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: FSI coupling
Better results most likely will come with mesh changes not sif changes. Many structural codes automatically use higher order elements even if you don't create them. Commercial codes put a lot of development in their element formulations and they keep them proprietary. Use a slightly finer mesh on the elmer problem and see if the answer changes. If it does than the mesh is not converged. If you do this on tests/ShoeboxFSI you will see that the results change.
Re: FSI coupling
Thank you for your advice. I'll try it.