Different behaviour of interFoam across versions

mrishi · June 8, 2022, 05:27

I have been observing strange artificial oscillations in my interFoam based simulations. So to isolate the issue I set up a cylinder domain half filled with water. No initial velocities anywhere, no inlets, only a pressure outlet at the top of cylinders and walls everywhere.

I expected this to remain stationary. But when I solved interFoam on this, I see that numerically induced velocities occur near the interface, which can be seen in the attached figure.

staticInterface.jpg

In the figure above: frame 1 is ESI solution (v2006), frame 2 (top right) is OF-9, frame 3 is OF-6 and the 4th quadrant shows that mesh is same along with a photo of the overall domain.

Most interestingly the oscillations has grid sensitive pattern, which is very strong in Foundation simulations (v6 and v9), and much better behaved in ESI version simulations (v2006). I though this has to do with compression term of interFoam. In the figures, vanLeer scheme with cAlpha=1 has been used. cAlpha is a suspected culprit for introducing oscillations but removing it fudges the interface, and even with it switched on, there is distinct difference between the two forks' behaviour, which is particularly bothersome. To state explicitly, the ESI version handles numerical oscillation much better than Foundation, with v9 performing the worst.
Since the cAlpha compression term is a feature of algebraic VOF, I also tried the alternative - geometric VOF. For ESI this was the interIsoFoam, where results were also similar to ESI interFoam (with different, small magnitude oscillations). On the other hand, the new geometric VOF scheme in v9 (MPLIC) has similarly poor behaviour as algebraic scheme above.

If there were to be motion in this mesh, the v9 interface will oscillate wildly over time, which may go unnoticed if the domain is not neatly ordered, as it is in this case. In the context of this motion, it is worth noting that the presence of turbulence increases these oscillations, but they do occur even without it, going from better to worse going from cAlpha 0 to cAlpha 1, and laminar to turbulence. The former can be somewhat tricky to test with the vanLeer scheme, since it is quite diffusive. So I checked with the CICSAM scheme by Ubbink and Isaa.

In general, OF-9 has by far the worst performance, even though the cases are set up identically, and I would like to understand why this is happening. Can we somehow improve the solution, or check for obvious mistakes, so as to get consistent solution from all these solvers, since this is a fairly simple case setup and a fairly standard solver?

Also may be worth noting that the solution does not converge to very low residuals (modeled using laminar flow), when using PISO and diverges if using PIMPLE loops. I checked this because maybe the insufficient convergence of single outer loop was causing spurious oscillations, but then I found that having multiple loops leads to divergence after a few timesteps.

Can someone kindly point me in the obvious direction about this?

vatavuk · June 10, 2022, 18:32

Dear mrishi,

This thread may contain information related to your problem:
InterFoam: Different results in OF5 and OF6

Observe that the time derivative calculation in ddtScheme is different in version 6. In version 7 the changes were reverted and it is the same as in version 5. This may explain the differences between versions 6 and 9, but I have no idea why v2006 performs better.

I hope this helps.

June 8, 2022, 05:27	Different behaviour of interFoam across versions	#1
mrishi Member Rishikesh Join Date: Apr 2016 Posts: 63 Rep Power: 9	I have been observing strange artificial oscillations in my interFoam based simulations. So to isolate the issue I set up a cylinder domain half filled with water. No initial velocities anywhere, no inlets, only a pressure outlet at the top of cylinders and walls everywhere. I expected this to remain stationary. But when I solved interFoam on this, I see that numerically induced velocities occur near the interface, which can be seen in the attached figure. staticInterface.jpg In the figure above: frame 1 is ESI solution (v2006), frame 2 (top right) is OF-9, frame 3 is OF-6 and the 4th quadrant shows that mesh is same along with a photo of the overall domain. Most interestingly the oscillations has grid sensitive pattern, which is very strong in Foundation simulations (v6 and v9), and much better behaved in ESI version simulations (v2006). I though this has to do with compression term of interFoam. In the figures, vanLeer scheme with cAlpha=1 has been used. cAlpha is a suspected culprit for introducing oscillations but removing it fudges the interface, and even with it switched on, there is distinct difference between the two forks' behaviour, which is particularly bothersome. To state explicitly, the ESI version handles numerical oscillation much better than Foundation, with v9 performing the worst. Since the cAlpha compression term is a feature of algebraic VOF, I also tried the alternative - geometric VOF. For ESI this was the interIsoFoam, where results were also similar to ESI interFoam (with different, small magnitude oscillations). On the other hand, the new geometric VOF scheme in v9 (MPLIC) has similarly poor behaviour as algebraic scheme above. If there were to be motion in this mesh, the v9 interface will oscillate wildly over time, which may go unnoticed if the domain is not neatly ordered, as it is in this case. In the context of this motion, it is worth noting that the presence of turbulence increases these oscillations, but they do occur even without it, going from better to worse going from cAlpha 0 to cAlpha 1, and laminar to turbulence. The former can be somewhat tricky to test with the vanLeer scheme, since it is quite diffusive. So I checked with the CICSAM scheme by Ubbink and Isaa. In general, OF-9 has by far the worst performance, even though the cases are set up identically, and I would like to understand why this is happening. Can we somehow improve the solution, or check for obvious mistakes, so as to get consistent solution from all these solvers, since this is a fairly simple case setup and a fairly standard solver? Also may be worth noting that the solution does not converge to very low residuals (modeled using laminar flow), when using PISO and diverges if using PIMPLE loops. I checked this because maybe the insufficient convergence of single outer loop was causing spurious oscillations, but then I found that having multiple loops leads to divergence after a few timesteps. Can someone kindly point me in the obvious direction about this? lourencosm likes this. Last edited by mrishi; June 8, 2022 at 17:36. Reason: additional clarification and elaboration.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
interFoam two-phase pipe flow air phase behaviour	katete	OpenFOAM Running, Solving & CFD	11	February 3, 2021 03:14
k-e & GAMG interFoam Schemitisation Stability Issue	JFM	OpenFOAM Running, Solving & CFD	3	December 1, 2015 05:58
strange pressure behaviour with symmetricPlane boudary condition - interFoam	duongquaphim	OpenFOAM Running, Solving & CFD	10	August 20, 2013 14:00
Strange boundary behaviour using interFoam	Andrea_85	OpenFOAM	11	January 22, 2013 15:09
divergence behaviour of interFoam solver for very fine mesh	purushotam85	OpenFOAM Running, Solving & CFD	0	June 23, 2010 18:57

June 10, 2022, 18:32		#2
vatavuk Senior Member Paulo Vatavuk Join Date: Mar 2009 Location: Campinas, Brasil Posts: 196 Rep Power: 17	Dear mrishi, This thread may contain information related to your problem: InterFoam: Different results in OF5 and OF6 Observe that the time derivative calculation in ddtScheme is different in version 6. In version 7 the changes were reverted and it is the same as in version 5. This may explain the differences between versions 6 and 9, but I have no idea why v2006 performs better. I hope this helps.