[alainislas]

Hello everyone,



I am meshing a 20L Siwek (Sphere) Apparatus. I blocked all fluid region with hexa elements and already associated all the edges that are close to the nozzle. The edges that stick to the sphere surface are catched up automatically, but somehow when checking my mesh by quality i got some elements with "volume-orientation" problem.


I already tried to associate the contiguos edges of these bad elements (colored in red) to the sphere surface, but still have this error.


I would appreciate any suggestion to help me fix this problem.

[bluebase]

Hi Alain,


you seem to have a similar problem as with blocking a planar circle with quads. A single structured block in a circle cannot yield good quality in the corners of the block. That's why we have to use an o-grid strategy to overcome this issue. It seems in the locations with bad elements in your sphere, you also have block corners which are degenerated. Take focus on the corner of the block with the worst elements, 5 out of 6 corners have to conform to the sphere's surface. Since the curvature of your sphere is small, that element is almost flat.

The plainly speaking, your need an ogrid in the second direction.

However, this seems to be not trivial in your case. You blocking seems somewhat convoluted. You could try to extruded a thin o-grid all around from the sphere's surface into the domain. This might make your blocking even more convoluted though. If you haven't permanently deleted the blocks covering the probe, you might need to include blocks from VORFN as well in order to let the ogrid project into the the probes base, and to avoid introducing an additional layer on the probe.

You might even need to rethink your essential blocking strategy. In my opinion, the blocking for the probe needs to fit into the blocking for the sphere. I have the impression, you started the other way around?


Best,
Sebastian

[alainislas]

Dear Sebastian,

Thank you very much for your reply. Since my geometry is a nozzle contained within a spheric vessel, my strategy was to block and delimit all fluid regions contained within the sphere and the walls of the nozzle. In fact, I believe I started the way around as you suggest. I didn't block any solid region at all, what I did was to start blocking the fluid from the nozzle in the direction out to the walls.

I am attaching an image with cut planes in the XY and YZ planes respectively. As you indicate I need O-grids for the blocks where the bad elements are present. However I already tried to use the O-grids (not shown in this pic), but the error persists, and the number of bad elements increases. I also tried the orthogonal smooth option, but it doesn't improve much.

At this point, I am not sure if my strategy was correct or I can still fix this blocking,

[bluebase]

Originally Posted by alainislas

However I already tried to use the O-grids (not shown in this pic), but the error persists, and the number of bad elements increases. I also tried the orthogonal smooth option, but it doesn't improve much.
At this point, I am not sure if my strategy was correct or I can still fix this blocking,

I understand that introducing another o-layer will likely seem to introduce more bad elements in the beginning. But i would expect it would be possible to fix such issues with shifting vertices in order to make the involved blocks "more square" again. Of course this will be a somewhat laborious task...

It seems difficult to find a specific solution via a few messages.
May I suggest you to take a step back instead, and train on a simplified geometry? Create a new project, insert a sphere, and try to block it perfectly with o-grids running in all directions inside of the sphere. Basically you select the initial single block, and no faces when using the o-grid feature to create the o-grid all around. then move the vertices to the spheres surface.

I imagine training on a sphere could help visualizing what the geometry needs.
This sphere-only blocking is what you need to find in your complex blocking as underlying structure.

Right now, i see in your xz-plane screenshots(labeled xy) that you have something more like a c-grid - an o-grid which is open to the top of the sphere. This will inherently lead to "bad corners". The yz plane shows the same issue.
I see the following options to continue:

• Start over. will cost much time, but you have more experience now; will likely lead to best mesh quality
• Introduce an additionally, circumferential o-grid, will likely lead to workable mesh quality since there will not be degenerated corners left; tedious, somewhat error prone
• make the problematic blocks unstructured, least amount of work, likely yield workable mesh, but might not suit your requirements anymore regarding structured only mesh.
• If above is an option you could extend the idea of a hybrid mesh, by making most of the domain structured, leaving a hole where the nozzle is. Then deal with the boundary layer region with high quality prism mesh, and have unstructured regions to patch the regions together

Regarding smoothing, this usually is the very last step in meshing. In my (limited) experience, the smoother likely enhances problems in the mesh, and will only improve a mesh if its structure is good by itself.

[alainislas]

Dear Sebastian

Thank you so much for your kind comments. As suggested I spent some time practicing with a simple O-grid contained into a perfect sphere. This, in fact, helped me so much to improve my blocking strategy.
This time, I started the way back by first introducing a perfect O-grid into the sphere which I was able to perfectly section and to associate to the top part of the sphere and to an extruded bounding box at the bottom, which covers the nozzle (see pic).

Up to now, I believe it would be good to try a hybrid HEX-TETRA mesh, but I can't make it work yet. For this, I saw it is necessary to create a floating surface (in my case the cyan surface).

I already meshed the bottom lid (red bowl-shaped geometry) with tetras, and then I tried the Edit Mesh -> Merge Nodes -> Merge Meshes option. Although It works, some degenerated elements are introduced, and many errors appear in the log, like :

* unable to include quad 14612 (241820 401229 241823 401226)
* I am trying to include edge 231577 402144 but vertex 231577 isn't even in the region

This gives me a global bad quality. I would heartedly appreciate any suggestions.