Best way to handle hanging nodes in Fluent for LES?
 

Lets say I wanted to use a mesh with a topology like this for LES validations.
Attachment 27026

What would be the best way to handle the interfaces between the mesh regions with different resolutions?

1. Setting up non-conformal interfaces in Fluent (I know that this kind of interface does not preserve turbulent fluctuations and alters the instantaneous flow field even if the flow scales are well-resolved on both sides of the interface, so I only add this here for completeness)
2. Solving with hanging nodes. This means I load the coarsest mesh into Fluent and apply the mesh adaption near the boundary to get a mesh topology similar to the one in the picture.
3. Converting Cells with Hanging Nodes / Edges to Polyhedra via tui command "mesh → polyhedra → convert-hanging-nodes". This should convert the hexahedron cells adjacent to a region of smaller cells to polyhedron cells with 9 faces each (provided a size ratio of 2:1) and avoid hanging nodes.
   Unfortunately, the menu itempolyhedra is not available in my fluent version. Strange...

With the focus on LES or later DNS, which approach would be the best in terms of accuracy?
I would simply test both methods, but since this is not really part of my project I hope that someone else has some experience to share on this topic.
Maybe there are even more possibilities or number 2 and 3 on my list are treated in the same way by Fluent or they will not work the way I expect. Feel free to correct me.

Edit:
Apparenty, converting hanging node cells to polyhedra is only possible with a 3d-mesh.

Any thoughts?

As i know there is no possibility to convert hexa to polyhedra in fluent till now: perhaps i make here a mistake.

I think solution number will be the best to follow despite the non conservatism of the turbulent fluctuations through the cells.

By the way why not merging all the parts into multibody part and getting conformal mesh for such a geometry or are u posting there only an example.

Are you sure about this? The manual is quite clear about converting hanging nodes to polyhedral cells.

I guess you mean number 1?
In fact this approach had already been followed by some students under the supervision of one of my colleagues, and the results were not very promising to say the least.

One of the reasons is that we wanted to have nearly cubical-shaped cells in the whole region for comparison with a Lattice-Boltzmann method with grid refinement that works on a similar looking mesh, so the discretization would be similar.
Another reason is that the filter width of LES subgrid models like Smagorinsky is based on the cubic root of the cell volume, which at least to my understanding is only an appropriate choice if the cell is nearly cubical-shaped.

As far as i know, method 2 and 3 are not exactly the same (not anymore, at least) but i think the differences are not that big with respect to the main problem here, the fact that the resulting cells are not even closely to be cubical.

I have experience on method 2 for the turbulent channel flow at Re_tau = 590 and a dynamic smagorinsky model; i was unable to get meaningful results with number of grid points comparable to the ones from a structured grid. Actually i used a bounded central scheme, which might have had a strong impact with this specific topology, but i guess the pure central scheme could have created additional problems.

To be extremely sure, i would go number 3 (as 2 might have some approximations and is not really developed for LES). Of course, you MUST go 3D, there are no additional MEANINGFUL options for LES.