[sina_mech]

Hi,

I'm trying to simulate the water flow in a huge tank (diameter: 20m) with an inlet of 5m*5m (of course turbulence!) . So should I even consider prism layer for such a flow?

I'm ok with having prism layers, but I realized that it makes the convergence much worse (specially in multiphase with mixture model).


PS: considering that the boundary layer for such a flow would be something around 10cm, or even less.

Thanks

[LuckyTran]

If you have a no-slip boundary condition, then I highly recommend it. If however you have a slip condition, then no.

The biggest thing about prism layers is that wall adjacent cells are parallel to the wall and this helps prevent really stupid things from happening (like velocity vectors that go into the wall).

Convergence problems with prism layers is most likely mesh quality issue. Automated meshers tend to have difficulty with prism layer generation. But you have no excuse for going into a simulation with a bad mesh, unless your geometry is really really complicated! If you want good results, make a nice mesh.

[sina_mech]

Originally Posted by LuckyTran

If you have a no-slip boundary condition, then I highly recommend it. If however you have a slip condition, then no.

The biggest thing about prism layers is that wall adjacent cells are parallel to the wall and this helps prevent really stupid things from happening (like velocity vectors that go into the wall).

Convergence problems with prism layers is most likely mesh quality issue. Automated meshers tend to have difficulty with prism layer generation. But you have no excuse for going into a simulation with a bad mesh, unless your geometry is really really complicated! If you want good results, make a nice mesh.

Thank you for your reply. It's a no-slip boundary condition. but having y+ = 30, the cell next to the wall is as small as 7 mm. Considering the dimensions I mentioned, if I want to have a smooth transition of cell sizes (from walls to the mid-fluid regions), I will end up having +60M cells, which is massive!

BTW, let me explain how I mesh, maybe you could kindly recommend a better way:
using ICEM, tetrahedral everywhere. One layer of prism layer with thickness of ~15cm. Splitting that single layer to 12 layers with size ratio of 1.1 or 1.2

PS: Unfortunately I'm not allowed to share the geometry.

[Kushal Puri]

Y+ = 30 m good, can proceed with standard k-epsilon turbulence model