Fluent strategy for hexa mesh
 

Hi,

I've did some calculations with tetra mesh (+ prism for boundary layer) and now switched to hexa mesh (+ prism for BL).

My strategy was:
Start with 1st order scheme
Decrease under-relaxation factors to about 1/3 of default values
Turbulence model: k-e-realizable
Ideal gas, other properties are piece-wise linear
BC: Mass-flow-inlet and pressure outlet

Then I start with mass flow of 1/10 of final mass flow for 250 iterations. Increase to 1/2 (250 iteration) and finally "final mass flow".

For the tetra mesh cases I could switch to 2nd order and everthing works fine.
For the hexa mesh cases (cell number twice as large as for tetra case) I get divergence.

Are there any special things with heay mesh? I tried to switch to QUICK ... but again I got divergence.

Any advice?
Thanks in advance

What do you mean by Hexa + Prism for BL. When you have a hexa mesh the quad surface grows directly to a hex mexh without prisms, your BL can be captured by biasing your element size near the walls.

Regards
Luke

The hexa part of my mesh (it is actually a octree mesh) is not alighted with the BL. So I have prisms at the Boundary than a triangular transition zone, before the main part of the mesh is heaxeder-style.

Try to change the underrelaxation factors!

"Decrease under-relaxation factors to about 1/3 of default values"

You think I should decrease the udf even more? 1/3 of default is quite less .... :confused:

transition zones can introduce many many errors. mesh quality is also very hard to maintain. you might have non-orthogonal or very highly skewed cells that are causing divergence.

i wouldn't reduce the urf anymore than 1/3.

if you are using Fluent 14, there is an option of higher order relaxation, it is an under relaxation on your higher order terms so you do not need to manually switch between 1st order, 2nd order, or quick schemes, it will just use an under relaxation factor. But I would definitely focus on getting a converged 1st order solution first.

Hi LuckyTran,

thanks for your answer.
I also think, that the grid quality might be the reason. I tryed the same mesh with CFX12.1 and I got a solution. But Fluent12 does diverge .... So it seems that fluent is more demanding a better grid ....

The fundamental methodology of all the codes are very similar. If you look into it there are not many differences between the commercial codes in how the equations are solved if you choose the same options but they often have different default options. More often differences in user input and these defaults lead to more differences in how a simulation behaves than actual differences in the code. Of course, if you choose different options, differences in results should be expected. It's up to you on how you want to proceed, if you want to figure out why Fluent in particular is diverging, i.e. what Fluent is doing differently.

The test with CFX was just, because some other people did it with CFX and got results. SO I wanted to check if CFX would work for me as well (same BC, all settings as close as possible to the fluent ones) ....
But there has to be a way to get results with fluent .... Nevertheless fluent requires "more work" to get convergence, compared to CFX ... but that’s just my experience.
I just wanted to check if I should use a different strategy for the different kind of mesh.