[NablaDyn]

Hi all,


I searched the forum but did not find any solution or thread regarding my issue:


I use pretty good quality and smooth (pure) hex meshes in OF. Now, I want to refine the mesh using refineMesh and refineHexMesh on well-defined cellSets but neither utility manages to generate a valid mesh. After refinement all (I'm not quite sure but it seems like that) faces of the refined cells are highly non-orthogonal, rendering the result rubbish. The mesh looks appropriate, though.


So here's my question: Has anyone encountered the same and maybe found a remedy for this or does someone know what is going wrong here? I can provide checkMesh results before and after as well as some pictures (will do later).


Regards and thanks in advance,


Martin

[Fauster]

Originally Posted by NablaDyn

Hi all,


I searched the forum but did not find any solution or thread regarding my issue:


I use pretty good quality and smooth (pure) hex meshes in OF. Now, I want to refine the mesh using refineMesh and refineHexMesh on well-defined cellSets but neither utility manages to generate a valid mesh. After refinement all (I'm not quite sure but it seems like that) faces of the refined cells are highly non-orthogonal, rendering the result rubbish. The mesh looks appropriate, though.


So here's my question: Has anyone encountered the same and maybe found a remedy for this or does someone know what is going wrong here? I can provide checkMesh results before and after as well as some pictures (will do later).


Regards and thanks in advance,


Martin

Dear Martin,

I am facing exactly the same problem. refineMesh creates non-orthogonal cells when used multiple time on the same cell set. I tried to play with tolerance in controlDict without success. I thought it was due to reconstructParMesh that I used before but it's not the case. I tried the option use geometry or use hex without effect. I also tried to use refineMesh in parallel or in serial (no change !).

F

[NablaDyn]

Hi Paul,


well, yet I did not solve my issue but, as my computational results do not seem to be very sensitive to the determined nonorthogonality, I assume it only is a matter of how nonorthogonality is computed for the newly introduced cells containing hanging nodes.



I suppose, by splitting cells in octree fashion, the locations of the new cell centroids cause high nonorthogonality where there was none before. I further assume that - given OF can handle hanging nodes by means of flow computations - this nonorthogonality might not pose a 'real' problem but rather the incapability of the mesh checks to account for properly account for hanging nodes.


Please, anyone, correct me if I'm wrong...

[choch]

Hello Paul and Martin,

I am facing the same error after iterating refineMesh on the same cell set. I tried to increase writePrecision in controlDict with no effect.

Did you ever find a solution to this problem ?

Regards,

C.

[NablaDyn]

Hello Charlie,


indeed, I reviewed this specific problem recently. I'm pretty confident, that I have now understood what's the point after reading Juretic's thesis ("Error Analysis in Finite Volume CFD") and having a look at how cfMesh handles "split hexahedra", i.e. hexahedral cells that have been split in such a way that they now have a hanging node on at least one of its six cell faces:


Consider two adjacent hex cells, whereby one is decomposed in octree fashion into, say, 2 cells. As OpenFOAM treats any cell generically as arbitrarily polyhedral cell, it does not account in any special way for hanging nodes. Thus, non-orthogonality arises or increases locally as soon as such a non-symmetric (only one cell of a pair) is refined (non-ortghogonality is defined as the angle between the normal vector of the cell-connecting face and the vector connecting the cell centres of two neighbouring cells. The introduced non-orthogonality is thus a "real" numerical concern which reduces accuracy and might result in convergence degradation.

I hope this helps but much more I hope I'm right in my interpretation. Anyone correct me please if I'm wrong.

[choch]

Thank you very much for your answer Martin.

You are totally right. I was using cells with high aspect ratio in one direction. By refining the cells in that direction I was creating non-orthogonal cells.

By lowering the aspect ratio in the zone of refinement this issue was solved.

C.

[Carlo_P]

I'm having a similar problem using improveMeshQuality.
I used it, but in the end, I have a worst mesh.
You can see the report of checkMesh:


before:

Max cell openness = 7.35622e-16 OK.
Max aspect ratio = 29.7317 OK.
Minimum face area = 3.92408e-13. Maximum face area = 1.47856e-07. Face area magnitudes OK.
Min volume = 1.60163e-18. Max volume = 3.43724e-11. Total volume = 3.3578e-06. Cell volumes OK.
Mesh non-orthogonality Max: 84.9321 average: 12.8083
*Number of severely non-orthogonal (> 70 degrees) faces: 309.
Non-orthogonality check OK.
<<Writing 309 non-orthogonal faces to set nonOrthoFaces
Face pyramids OK.
***Max skewness = 4.04614, 2 highly skew faces detected which may impair the quality of the results
<<Writing 2 skew faces to set skewFaces
Coupled point location match (average 0) OK.

Failed 1 mesh checks.


after:


Checking patch topology for multiply connected surfaces...
Patch Faces Points Surface topology
Inlet_up 4820 5250 ok (non-closed singly connected)
Inlet_sx 17145 18233 ok (non-closed singly connected)
Outlet 1743 1893 ok (non-closed singly connected)
Wall 291922 301087 ok (non-closed singly connected)

Checking geometry...

Max cell openness = 4.85654e-16 OK.
Max aspect ratio = 41.771 OK.
Minimum face area = 1.25284e-12. Maximum face area = 1.65683e-07. Face area magnitudes OK.
Min volume = 1.20562e-18. Max volume = 3.30891e-11. Total volume = 3.3578e-06. Cell volumes OK.
Mesh non-orthogonality Max: 79.9838 average: 18.2811
*Number of severely non-orthogonal (> 70 degrees) faces: 797.
Non-orthogonality check OK.
<<Writing 797 non-orthogonal faces to set nonOrthoFaces
Face pyramids OK.
***Max skewness = 7.31313, 6 highly skew faces detected which may impair the quality of the results
<<Writing 6 skew faces to set skewFaces
Coupled point location match (average 0) OK.

Failed 1 mesh checks.

End


What do you think about?
Thanks a lot,
Carlo

[NablaDyn]

1. Your problem is not related to "mesh refinement", so I'd advice you to open a new thread, e.g., concering the improveMeshQuality utility (I suppose it belongs to cfMesh)


2. If you have a mesh that bad, an implicit improvement technique will surely not "improve" your mesh. I'm afraid you need to review your first mesh entirely. I suspect you don't capture geometric features well enough achieve a decent quality in that region (or even globally). But this is all speculation. Without seeing your mesh I cannot make any reliable statement.

[thanosniotis]

Hello community,

I have been investigating the same condition two weeks now. I perform wave propagation simulation so i would like to refine the mesh to specific regions close to the water free surface. When i refine one cellSet at the three dimensions it creates non orthogonal mesh; but when i refine it only to Z anisotropicaly the refined cells are purely hexahedral. If anybody has an idea regarding this issue please let us know.

Best regards,
Thanos