CFD Online Discussion Forums

CFD Online Discussion Forums (
-   ANSYS Meshing & Geometry (
-   -   [ICEM] Hexa mesh of eliptic bottom (

kutis August 25, 2011 10:33

Hexa mesh of eliptic bottom
2 Attachment(s)

I have very difficult geometry for me -- eliptic bottom with cylindrical connections (see figures), There is many cylindrical connections (cca 750). I'd like to mesh it with hex but I have problem to create blocking in ICEM CFD. Please could you suggest me best blocking strategy for this geometry? Can I use some pattern for cylindrical connections, or should I use copy of blocks or some other combination of blocking strategies?

Attachment 8952

Attachment 8953

PSYMN August 26, 2011 09:18

Blocking approach...
If you are meshing this with ICEM CFD Hexa Blocking, the strategy is very straight forward, but will be fairly tedious, so you should probably start with the smallest section of the model you can... It looks like you can split this in half again and go for quarter symmetry?

The basic strategy is to start with an Ogrid (or for this half symmetry model, a quarter Ogrid). You do this by selecting the first block and then putting two faces, one on the top and the other on the symmetry plane.

The resulting topology will have a box in the middle of the geometry surrounded by 4 Ogrid blocks.

Split one of the radial Ogrid edges for the inside of the cylinders and then again for the outside. Associate edges to curves, verts to points...

Then start splitting the blocking the other way to capture the circles... You are just splitting to put each cylinder into a box. Associate edges to curves.

Use the index control to adjust 03 until you only see the blocks in the gap... Delete (not permanently, just send to VORFN) all the blocks between the boxes around the cylinders.

Next you have a choice... How do you want the boundary layer to go thru the tubes? You can have one continuous boundary layer thru the whole model, or you can treat each tube as its own separate nozzle with its own boundary layer...

I will assume the later...

For this to work best, you want the corners of the Ogrid to be away from the geometry (not projected to a circle that will give bad quality in the corner). So, you will need to split the radial Ogrid again on either side of the tubes so you have an extra block to work with on each side... The block on the inside of the flow should be about as long as it is wide for best mesh quality, the block on the outside should be about as long as you expect the jet of fluid to be (for best resolution).

Then use index control again (to make things easier to see and select) to reduce to the three Ogrid indecies that contain your tubes and the small blocks on either side. Using the Ogrid command, select all of the appropriate blocks. 3 blocks for each tube. Apply... You will get a nice Ogrid (all at once) in each tube and extending out each side.

Then adjust vert locations, edge distributions, etc.

PSYMN August 26, 2011 09:21

Other approaches...
If you have used hexa blocking, this model will be straightforward, but tedious, and you will get a gorgeous mesh. On the other hand, if you have never used it, or simply don't want such a manual/interactive approach, perhaps try another method such as BFCart or the new CutCell Mesher in TGrid/ANSYS Meshing.

Another option would be the octree tetra with conversion from tet to hex (looks like hexa core mesh).

Best regards,


kutis August 26, 2011 16:20

Thank you for your reply Simon.

The half of eliptic bottom is the smallest symmetric part. So I can't use smaller part of the bottom. And the number of cylindrical connections (tubes) is different in each layer of connections.
I think about the problem in the similar way as you wrote. I split the main block into 4 ogrid blocks and 1 central block, and divided each ogrid block into 3 blocks. Then I started spliting ogrid blocks into individual layers, each layer represents one layer of cylindrical conections. Then I started to divide the layers (made from ogrid blocks) into blocks, which should represent cylindrical conections. But how should I associate edges to curves? There are only curves which represents cylindrical connections, there is no curves in outer or inner surfaces to associate edges of layer blocks. When I try to associate vertex to surfaces, the blocks became very deformed.

In your suggestion, you are writing: "Then start splitting the blocking the other way to capture the circles..." Do you mean divide all blocks or only the middle blocks where are tubes?

Thank you also for idea about BFCart mesh. I will try it. Is it suitable for CFX?

Thank you once more.


PSYMN August 26, 2011 21:37

Yes, BFCart is suitable if you keep it uniform (no hanging nodes for CFX). The Octree tetra followed by 12 tet to 1 hex may also work well for you.

I usually just split the blocks right through rather than worry about splitting individual blocks. The implied splits go thru anyway, and things are just easier to control if you the real splits go thru. Besides, the splits on one side work for the other side...

As for associations, you associate to curves when you have them... IF you don't have curves for a particular edge, just leave it alone. It will default to surface projection... I guess it does help to move it to the correct surface.

If you have images, i can help more.

kutis August 27, 2011 01:56

Thank you Simon for your reply.

The problem with the model is also that the individual layers have different number of tubes, but the diameter of tubes is still the same. The number of layers is 22. So for example, at the top layer (layer 1), there is 67 tubes and somewhere at the bottom (for example layer 18) there is only 30 tubes. So, should I divide all layer blocks into the same number of blocks according the highest number of tubes (model A) or each layer block should contain the number of blocks equal to number of tubes in the layer (model B)?

Model A: in all layer blocks would be 67 blocks, in layer 1 - each block would contain also tube, in other layers - not all blocks would contain tubes, some blocks would be without tubes, but the mesh density in tubes across layers would be the same and the mesh density of blocks without tubes I think would be deformed because also the blocks would be deformed due to streching of blocks with tubes.
Model B: There would not be any blocks without tubes, but the mesh density of tubes would be different at the top and the bottom of the model. The mesh density of tubes at the bottom of the model would be much higher than at the top model.
So which spliting of layer blocks is better from mesh quality view point?

I will also send some pictures of model but my connection with computer in work failed, I am going to work at tuesday and I will send the pictures.

Thank you once more Simon


PSYMN August 27, 2011 10:44

I would split to capture the top row, then try to re use those splits as often as you can on the way down.

You can decide if you want to combine two blocks (double density) into some of the tubes as you head down, or if you want to squeeze the higher density around the tubes... If you are more interested in the flow in the tubes, you might focus your mesh there. On the other hand, there is something to be said for keeping it even. I would probably plan on controlling the final tube refinement with an ogrid, and since I wasn't worried about that, I would just go for minimum block distortion. Some tubes would have two blocks, the spaces next to them may also have two or more blocks. Just keep the lines as straight as you can.

When you put that Ogrid thru all the tubes at the end, the Ogrid refinement will dominate the node count in those tubes. Also, you can apply those Ogrids only to the tubes, so the mesh density in blocks without tubes will not be anywhere close to the mesh density in the blocks with tubes...

Have fun.

kutis August 27, 2011 13:54

Thank you Simon for all your suggestions.

I'll try all your suggestions on Tuesday when I'll come to work.
When I worked with the model on friday, the projection of vertex of blocks to eliptic surfaces (outer, inner and both intersurfaces) seems very problematic. The curves witch can be used to associate with block edges are only on symmetric plane, on top plane and on tubes diameters. There are no curves between layers of tubes. When I moved the vertex to eliptic surfaces, the blocks became very distorted, they didn't have regular shape. Maybe the geometry spliting in Desing Modeler into individual layers could create the curves between the tube layers, which I could use in association and vertex projections. Or is it possible to create such curves directly in ICEM CFD without geometry manipulation in other CAD system?

Thanks once more Simon


PSYMN August 28, 2011 15:59

Geometry creation...
Sure, I am not sure what curve you think you need, but you can always create these in the CAD system or in DM or in ICEM CFD.

If you just mean you are missing curves along sharp features (essential for crisp hexa mesh), then just run build topology with filter curves and points on... It will give you all the curves you need.

In addition, I sometimes find it helpful to have a curve down the center of some fillets, This sort of iso-parametric curve is easy to create in ICEM CFD.

There are a number of other point or curve creation tools under the geometry tab.

Best regards,


All times are GMT -4. The time now is 13:23.