## Multi-phase flow simulation 2D - Level Set Method

Elmer cases by the users for the users
dmitry
Posts: 52
Joined: 09 Dec 2014, 16:52
Antispam: Yes
Location: Russia

### Multi-phase flow simulation 2D - Level Set Method

Hi dear Elmer users!
I want to share a non-GUI case of 2D multi-phase flow simulation by the means of Level Set Method. Falling water column simulated here.
Case is based on "levelset3" test. Also present post touch on a matter of following topic (Density ratio of phases 1/1000):
viewtopic.php?f=3&t=1755&start=0&hilit= ... 2ca04b5969
So, solution for high density ratio was obtained by using direct umfpack solver for linear system and dense mesh, especially near the walls.
I've performed simulation on both coarse and fine mesh.
Transport properties was smoothed and assigned as in "levelset3" test by hyperbolic tangent function.

Coarse mesh:

We see, phase interface quite smeared. During simulation appears serious instabilities in pressure and velocity.

Fine mesh:

Interfase calculated more acuratly.

The case is placed at my google drive:

PS: I will really appreciate in advise on improvement of performance. Calculation for fine mesh was about 3 hours. Also any comments are welcomed!
Regards, Dmitry!
Last edited by dmitry on 11 Jan 2017, 20:36, edited 1 time in total.

raback
Posts: 3453
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

### Re: Multi-phase flow simulation 2D - Level Set Method

Hi Dmitry

I think levelset method is nice when the interface is rather smooth between two rather similar substances. However, I haven't had very good experiences with liquid/air type of interfaces. If the surface tension is not very large the interface quickly becomes irregular and difficult to track. I was also lucky that Lagrangian approach have suited many of our problems so I didn't really try to exploit all the tricks of the trade.

I think one problem may be the timescales. Elmer historically solves the Navier-Stokes equations in a monolithic and fully generic fashion. This is nice on the elliptic regime of the equation but I would guess that the time-scales in the free surface problem here dominates and therefore we end up solving the N-S equation (and the levelset equation) a huge number of times. The monolithic machinery becomes unnecessary expensive in the parabolic regime where the mass matrix dominates. Codes like OpenFOAM exploit this in pressure and velocity correction schemes which are much cheaper. Hence I think you can obtain better solution with same resources using InterFOAM solver, for example, than the current Elmer version of levelset method.

I would think that the levelset solution of Elmer could be made much better but it would require evaluating critically the steps taken in the simple test cases.

-Peter

dmitry
Posts: 52
Joined: 09 Dec 2014, 16:52
Antispam: Yes
Location: Russia