[Jiricbeng]

Hello,

I created mesh of a complicated geometry in Ansys meshing. It consists of more than 10 millions elements. Then in CFX I use mesh tranformation (rotation) to create full domain. So the full domain consists of more than 100 millions elements.

1) Once I drag the mesh into the CFX, a .cmdb mesh format is created. Then after the mesh is copied by the rotation, the periodic boundaries are glued together (they are interior) which is fine.

2) If I export the mesh in .msh format and then this .msh is imported into CFX, the periodic boundaries are not glued and the volumes are isolated. I tried to change the settings of mesh tolerance and double nodes removal etc and nothing works.

The reason why I want to use .msh format is that .msh does not include all faces, it includes only my name selections. Because when using .cmdb mesh in CFX, the calculation does not run due to memory issues. I read that it could be caused by the high number of faces and using .msh could be the workaround. But I cannot glue the segments.

[ghorrocks]

Maybe try defining named selections for the other surface groups as well to reduce the number of surfaces?

Have you tried adjusting the duplicate node tolerances?

If nothing works then you might join them up with GGIs.

Finally - have you shown that you actually need a mesh this big? A mesh sensitivity study will show you how much mesh you really need.

[Jiricbeng]

Well, that is true, maybe I could try to create name selection for the rest of the domain, but it will be a hard task as the number of surfaces is really high to select.

I tried to adjust to duplicate node tolerances but it did not work.

I need to avoid GGIs in this case as it would deteriorate the solution (I am sure it would not converge at all and memory requirements would be even higher). Yes, I need this size of mesh, otherwise divergence occurs. It would be better to have even finer mesh.

[ghorrocks]

it will be a hard task as the number of surfaces is really high to select.

Then select all, then unselect the surfaces which are not part of it.

need to avoid GGIs in this case as it would deteriorate the solution (I am sure it would not converge at all and memory requirements would be even higher).

How would this deteriorate the solution? Why would not you get convergence?

Yes, I need this size of mesh, otherwise divergence occurs. It would be better to have even finer mesh.

This comment suggests there is a more fundamental problem. Finer meshes are normally harder to converge, not easier. So if this is not the case for you then it is likely something is wrong. Is your mesh quality acceptable? Have you got some unusual physics in the model?

[Jiricbeng]

My existing name selection consists also of many faces, if there was possibility to subtract name selection, it would be the easisest way.

In my case GGI interface would be in the place where I need to evaluate forces and torques. I think it is generally known that it is not a recommended to use an interfaces in places where you need to evaluate values very sensitive to the complicated flow conditions. Moreover there are enormous changes in the size of the domain from a few um to cm. The experience is the less interfaces the better.

Since there are such big changes in the geometry you must respect the smallest volumes in order not to have abrupt changes in element sizes. I simulate flow in the bearing based on full geometry, without any simplification.

Moreover I found another thing - the success of the mesh gluing depends on the history of mesh transformation in given CFX file.

[ghorrocks]

If a face is selected and you click on it, it unselects it. If a face is not selected and you click on it, it selects it. So unselecting is easy.

[Jiricbeng]

I know how to select/unselect faces . However the calculation already proceeds as .cmdb due to small mesh reduction.