[Vino]

Dear Foamers,

I am trying to simulate square duct problem (Re=3500) using cyclic inlet outlet with fvOptions(momentumsource:meanVelocityForce). I am using OpenFOAM 4.1 with simpleFoam with k-epsilon model.

I am not getting convergence for pressure & cross stream velocities. However, I am getting fully developed velocity profile in stream wise direction. Also, gradP term reaches steady state.

Could you please give me your suggestions to improve the convergence?

The case file that I used is available in the following link:

https://github.com/vino-123/square_duct_cyclic_inout

The following are the convergence I got with above case settings.

Convergence:
https://drive.google.com/open?id=0B1...2hnUkVjQkZtVjQ

gradP:
https://drive.google.com/open?id=0B1...3ZLeDhieGd2ZEU

Ubar:
https://drive.google.com/open?id=0B1...ktVM3NmOExNcU0


Thank you very much for your help.

[Tobi]

As you showed in your pictures the case should be converged. There are cases in which the pressure cannot go below a certain value. For me it seems fine. However I have not checked your.

[Vino]

Dear Tobi,

Thank you very much for your reply.

If I run the same case with inlet outlet BCs, instead of cyclic fvOptions(momentumsource:meanVelocityForce), the pressure converges to 1e-4 easily. Also, the cross stream velocity components (v,w) reach 1e-6.

But for the benchmark case I am working on, I need to run the case using cyclic conditions and I need to compare the cross-stream velocity components (secondary flow). Unfortunately, the cross stream velocity components also does not reach a good convergence, as a result the contours are oscillatory.

I would like to know whether there is any better way to achieve convergence using cyclic conditions. Also, I am worried whether there may be a possibility of bug in this particular implementation.

Thank you very much for your time.

Regards,
Vino.

[Tobi]

Depend on your case setup, if you are simulating some turbulence related stuff, it is obvious that cyclic can give you higher residuals based on the fact that you map the inlet to the outlet. In fact, if you set defined boundaries, the whole linear system is moved to a certain direction.
I am not an expert in turbulence but don't you have to specify a transform mode in the boundary type (polyMesh/boundary) for cyclic as well as the matchTolerance? I guess default values will be set but you might check it out. In addition I would set your residual control for your LS to 1e-12 or something like that and the relTol for U/k/epsilon to 0.01. You can also check if the SIMPLEC converge better.

For the scheme of U I would prefer the linearUpwindV scheme (V - vector). If your mesh is orthogonal then you can use the related schemes for the laplacian too and thus the snGrad should not be needed to be corrected. In addition - based on your topic I would use the FDIC preconditioner for the pressure and the stabilized PBiCG for the non-sym Matrix-Systems. In addition a correction of the pressure could help in order to fulfill the mass conservation in each pseudo time-step. However, if a steady-state is reached, this is not really dramatically but could influence your system at the beginning. If you induce errors at the start, they should be transported within your domain (with cyclic, you do not have any limitation of the solution).

But again, I was not checking your geometry and set-up accurately.

[Vino]

Dear Tobi,

Thank you very much for your detailed reply. I will tryout the suggestions given by you & will post the update. thanks.

Regards,
Vino.

[HoneyBadger]

Originally Posted by Vino

Dear Tobi,

Thank you very much for your detailed reply. I will tryout the suggestions given by you & will post the update. thanks.

Regards,
Vino.

Hi, Vino,



Any update? I've got the same problem as yours. Wondering is there any update after this? I found simplefoam converges really slow using this method.

[Vino]

Dear Kari,


I tried suggestions given by Tobi(in the previous reply), and I did not find much improvement. But the same setting without cyclic boundary (using inlet-outlet) showed a better convergence.



I was more interested on the cross-stream velocity components. I found that the cross-steam velocities were not symmetric. So, later I dropped the test case.


Also, I am not sure whether any problem with my mesh file or cyclic boundary condition setting.



Kindly post it, if you are able to improve the results.



thanks.

Originally Posted by HoneyBadger

Hi, Vino,



Any update? I've got the same problem as yours. Wondering is there any update after this? I found simplefoam converges really slow using this method.

[HoneyBadger]

Originally Posted by Vino

Dear Kari,


I tried suggestions given by Tobi(in the previous reply), and I did not find much improvement. But the same setting without cyclic boundary (using inlet-outlet) showed a better convergence.



I was more interested on the cross-stream velocity components. I found that the cross-steam velocities were not symmetric. So, later I dropped the test case.


Also, I am not sure whether any problem with my mesh file or cyclic boundary condition setting.



Kindly post it, if you are able to improve the results.



thanks.

I’ve tried for a long while as well. Didn’t manage to find a reasonable approach