Hexa Mesh Smoothing
I have built a 3D hexa grid for a simulation using the CFX 12 solver and the simulation does not converge due to bad quality elements (low minimum angle of approx. 6°) in two certain regions of the mesh, where there is a tangential connection in the geometry, as you can see in the pictures attached.
As one can see in the quality histogram in one of the attached pics, there are other regions of very low min angle (as low as 1.3°) volumes, which result from resolved block refinements, but these cells are otherwise rectangular and no high residuals appear there, so I'm not concerned about these.
So what could I do about it?
I have attempted to smooth the mesh in order to improve the mesh quality in the region of concern (using a subset / extra part) , trying both the "smooth mesh globally" and the "smooth hexahedral mesh orthogonal" options. While the "smooth mesh globally" option had very little to no effect on the mesh, "smooth hexa mesh orthogonal" always resulted in negative volumes being produced by the smoother. I increased the stability factor to 4.0 for surface and 8.0 for volumes and checked the "fix orientation" option, but neither helped. I have also tried both the laplace smoother and the orthogonal smoother option. Laplace gives a lot less -ve volumes, but still several (around 10 ... 100) that cannot be fixed afterwards.
After all, the smoother routines do not improve the minimum angle of the volumes of concern but instead produced additional low quality and inverted volumes.
My question: Am I doing something wrong? Or can smoothing the mesh simply not be expected to solve my problem?
Has anyone any suggestions how to otherwise solve the problem?
I have alredy slightly modified the geometry, so that the joint is not actually tangential anymore but in fact there is an angle of approx. 6.3°
I could, of course, drive this a bit further, but I'd rather not violate the geometry too much.
Besides, I would like too smooth the mesh after all, because it currently looks very "patchworked".
- The flow is assumed to be laminar and it's just single phase air
- The geometry and mesh are symmetrical with respect to the x-y-plane. And no, modelling half the thing and using a symmetry plane in the middle is not an option because the flow is driven by an oscillating membrane which performs a non-symmetrical oscillation. ;-)
I couldn't attach the project files here because the geometry file is too large even compressed. I could e-mail it to anyone who is interested, though.
So, I hope that I provided all necessary information. If anything is unclear, please don't hesitate to ask.
Any help is greatly appreciated.
Some more pics ...
... displaying the blocks there.
project files ...
... in bits :D and pieces.
For anyone who might be interested to take a look at it.
zipped it and split the archive into four files with 7zip.
! remove .txt ending !
To get to my current state of the mesh, just convert the pre mesh to unstruct, then resolve refinements with "allow unstable patterns" option.
This is the sort of thing that it is good to send to tech support... Even if someone here can help you get by, it helps to make them aware of these smoothing issues for next time.
My suggestion is to alter your geometry or blocking topology... Even if the smoother were not causing trouble else where, it probably couldn't do much to help in this critical area due to your other constraints. If you can't alter the geometry any further, then perhaps you (or someone online) can think of a better blocking.
It is interesting, but I can't take this on right now. Ping me again later if no one else steps up. ;)
I'll try to change the geometry a little more and add some curves to better control the block edges. Maybe I can reduce the problem to only the volumes on the surface.
I can't really think of a better way to do the blocking, but if someone has an idea, please tell me.
I'll let you know when I find a way to resolve this.
Simon, if you could find some time to look further into this, that would be really great.
I have a general question that came to my mind when I read this topic. What is the minimum acceptable quality of hexa mesh?
What is your solver.
That question is very solver dependent...
Check your solver manual.
As a basic rule, a quality of better than 0.1 is good for most. But some also have specifics for min angle (>18 degrees is usually good enough), or determinant or aspect ratio...
|All times are GMT -4. The time now is 14:55.|