Levelset method

[Roland]

Hi Peter,
Yes I agree completely with you and this is why I continue to struggle (!!) with the Lagrangian method and the transient phase change solver(say TPC solver). But, as I told it already, I only get a vertical move of the triple point (which is the working principle of the TPC solver, according to the doc and to your former explanations) when it should also move horizontally for increasing the solid crystal diameter. How could this be done? Have you progressed on your side with your last Cz model? Please keep me informed since I am waiting for your help without I cannot go further for the moment.
Roland

[raback]

Hi Roland,

This work on phase surface + free surface is fairly old stuff even in Elmer. Would you have any references on how this is dealt in other codes today?

I think we would need a completely new free surface solver operating on normal-tangential coordinate system. It should include two parts. One where the shape is given by melt rate and on the other by free surface of the melt. Even there I can see a challenge that for the triple point the normal is not uniquely defined...

-Peter

[Roland]

Hi
After our last discussion with you, Peter,about the Eulerian levelset method, and after non succeeding with the Lagrangian method (with the Transient Phase Change Solver and the triple point modeling problem), I switched my growth model to levelset after looking at the very interesting phaseChangeB test example which you mentioned.
The model works globally well (see the General_plot result) in the first seconds: the physical values are close from real ones with a small pulling velocity (0.15 mm/s). The thick white line is the melting isotherm (hence the phase change interface) and the black arrows show the flow.
Everything seems very nice (the crystal is growing and there is no more triple point problem !) except that after a few seconds (see the t=0s, t=5s, t=10s plots in tis post and in the next) the mesh becomes unstable in the circular corner as you can see it in the t=10s plot.
All this seems very promising (and exciting!) since it looks close to the experimental behaviour (like for instance the phase change interface which is nicely curved, the liquid flow stirring, etc…).
This is why I would like to ask for some help about this only remaining problem of the strange mesh behaviour in the corner of the grown crystal. Has somebody an idea of how this could be fixed? I also attach the corresponding zipped GUI project.
Many thanks in advance for any help!
(The t=5s and t=10s plots are in the following post)

[Roland]

t=5s and t=10s plots.
Roland

[kevinarden]

Just an initial glance would indicate a lot more mesh density would help. It appears to be very much to coarse for the deformation that is happening.

[Roland]

Hi Kevin,
Thanks for your reply have you are completely right!
Now that the model works globally nicely, I will try different things and especially a finner mesh.
We keep us informed
Roland

[raback]

Hi Roland,

Nice progress!

Temperature is very kind as it allows to be used as levelset function eliminating a lot of the machinery.

You can also make the body freeze if is cold

Code: Select all

Body Force
  Velocity 1 = 0.0
  Velocity 1 Condition = Variable Temperature
     Real
       0           1.0
       1687.0   0.0 
     End 

And similarly for 2nd component. This enforces velocity to zero when temperature is below melting.

Probably you current method with viscosity is smoother but this could be more liked by the iterative linear solvers, if there are issues.

-Peter

[Roland]

Hi Peter,
Thank you for your good idea about the forcing of the velocity which I will try (but just, when Temperature is below melting how have the velocity be equal to the calculated Navier Stokes stirring velocity ?).
As you saw the model works now nicely except concerning this problem of meshes in the corner of the growing crystal which become crazy and overlapps after a few seconds. According to Kevin's advice I tried it with a finner quad mesh but the problem remains the same. See the attached plot (at 30s).
The trick could be to have an adaptive mesh or a remeshing when the meshes strech too much and overlapp.
Do you think that this could be possible?
Anyway the model works much much better than in Lagrangian for the phase change interface and the triple point modeling, and this is why I would like to absolutly fix this only remaining meshing problem! Please keep me informed if you (or anybody) has some ideas about this.
Thanks in advance for your precious help!
Roland

[Roland]

Hi,
As explained in the former post, I am still wondering, in the presented crystal growth model, how this mesh excessive streching anf then overlapping occuring after a few seconds can be avoided. Has somebody an idea of how this can be fixed?
Any help is wellcome....
Thanks in advance for some advices about that!
Roland

[kevinarden]

There is adaptive meshing available and I believe it can be controlled by mesh distortion settings, or other parameters.