[sasanghomi]

Hi

I am trying to model the fluid flow inside a vortex tube.

I applied
Total Pressure @ Inlet
Static Pressure @ Cold Outlet
Static Pressure @ Hot Outlet

You can see the geometry in the attached pic.
My problem is that it is too sensitive to the mesh.
I started with 2 million grid cells and now 8 million mesh (for the attached sector). The cold temperature does not stop changing.

Does anybody have the same experience?
Regarding the turbulence model, I am using K-epsilon RNG. if you have any suggestions about turbulence options, please let me know.

[ghorrocks]

A separate point:
You will not be able to get any k-eps family turbulence model to work on this. You have a slight chance that SST with the curvature correction model will work, but you more likely will need RSM or an LES model for this.

Your question:
If the value does not stop changing then you do not have an accurate solution to it yet. You might need to refine the mesh further, but you also might need to do other things (eg tighter convergence tolerance, switch to transient simulation, different turbulence model etc). You need to look at the results and work out whether it is mesh resolution or something else causing the problem.

Finally:
A CFD with 8 million elements is not a very big mesh. It is common to require meshes of this size or larger for mesh independence. Sometimes much larger

[sasanghomi]

Thanks Glenn for your response.
I would be thankful if you could let me ask you another question.
Do you have any ideas about the Mesh? I can see most articles have used hexahedral grids. I suppose in this case the flow direction is not necessarily perpendicular to hexahedral mesh. So, do you think that tetrahedral cells reduce the accuracy in this case?

[ghorrocks]

Dissipation is lower for a flow which is aligned to the grid compared to one flowing at an angle to the grid. So if the flow is aligned with the grid with a hex mesh then you will get a bit lower dissipation. If the flow is not aligned to the grid you will get about the same dissipation between hex and tet meshes.

Also, hex meshes form polygons around the nodes with less faces than tets. This leads to reduced memory consumption. It is quite a big difference, about a factor of 2 or 3. But if you have heaps of memory this is not important. But if you are running out of memory you will run a bigger simulation with a hex mesh than a tet mesh.

But at the end of the day these are just general guidelines. Try a hex and tet mesh in your case and see if it makes any difference for you.