Learning Forum

[zoelle.wong]

Hello! I'm resolving a mesh that has a refinement on the nm scale. My y+ =1 gives a minimum cell height of 2e-7 m. Consequently, my mesh contains cell volumes beyond machine double precision. Does Fluent normalize the cell volume in the back-end or is there a geometric pre-conditioner for users to scale their set up? My thougt process is to scale my set-up, but I was curious if Fluent already does this in the background. Thank you!

[Rob]

The solver theory is covered in the manual, but even with some very small cells I've never seen any issues with the flow field. However, if you're resolving the near wall to that level of resolution I'd also question how confident you were about the surface roughness and surface faceting if it's curved. 

[zoelle.wong]

I'm pretty sure my mesh follows best practice (aspect ratio < 500, good orthognality etc), but I see that the minimum cell volume is negative. I also get a long warning message saying about left-handed faces vs right handed faces. Does this mean the cell volume is an artifiact of orientation or cell skewness? Alternatively, could it be that the cell volume is too small given that the maximum and minimum cell volumes are close to machine precision? I tried to fix the mesh via ANSYS Fluent (this normally works), console outputs what I see below. Thank you so much in advance! 

[zoelle.wong]

I went back to my mesh and I found that my normals weren't properly oriented; I think that's the root of the problem.

I also have a quick follow-up question: Does Fluent perform well when simulations are done across different nodes? Other researchers at my institution have told me that they see the best performance on 1 node with multiple cores; is that generally true? Thanks again! 

[Rob]

Yes, the problem there is cell definition. Re the aspect ratio, I assume there is no separation in regions with a value that high? If there is you'll want to aim for a much lower value. 

Yes, no, maybe. It depends.  Fluent performs very well in parallel, but.... If the cell count per partition is too low, or there is a lot of delay on the network/bus memory stuff (I cause work for IT, I don't solve it) you may find a model slows with more cores. The other thing to avoid are the virtual & efficiency cores on newer Windows based systems, the former means Fluent is fighting over one real core, and the latter should really only be used to run YouTube or the like while waiting for meshing to complete.