CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   ANSYS Meshing & Geometry (http://www.cfd-online.com/Forums/ansys-meshing/)
-   -   [ICEM] Quality reduced when smoothing (http://www.cfd-online.com/Forums/ansys-meshing/113404-quality-reduced-when-smoothing.html)

Bollonga February 19, 2013 08:17

Quality reduced when smoothing
 
2 Attachment(s)
Hi you all guys,

I've done a 2D hexa mesh with quality over 0.85 (see picture 1). Then I've converted it into unstructured, I've checked quality and it is obviously the same. But when I smooth it, dor 60 iterations forcing ortogonality in wall surfaces I get a way worse quality histogram (see picture 2).

The question is: Is it better to work with the non-smoothed mesh or does it worth it to use the smoothed with reduced quality?

Thanks!

stuart23 February 19, 2013 09:25

Hi Francisco,

I would go with the smoothed mesh. Even though the quality metric is lower, the mesh will probably have lower volume changes between adjacent cells. Rather than judging your mesh on the single quality metric, try plotting other metrics and seeing which ones increase and which ones decrease. You should consult the ICEM help file for definitions and details of each of the different metrics. I would recommend looking at aspect ratio and skewness and determinant for hexas plus Max angle and volume change.

Instead of using the unstructured smoother, try the Elliptical and Laplacian Premesh Smoothers, I find they work much better. These smoothers are both found under the orthogonality option of the Premesh Smoother and are applied before converting to unstructured mesh.

Good luck

Stu

energy382 February 19, 2013 14:07

I would suggest, NOT to use any of these smoother, as they all corrupt the mesh. They look pretty much better, but produce worse results from my experience. Better use the time to adjust the edges manually. For Cfx, aspect ratio doesn't matter, even 1000 is ok within boundary layer. But volume change is very important and max. dihedral angle as well (regarding hex mesh).

stuart23 February 19, 2013 18:43

Christoph, what version of ICEM are you using? In terms of "corrupt", do you mean negative volume elements?

I disagree that the Laplace and Elliptical Smoothers corrupt the mesh. Unfortunatly the smoothers are not perfect mesh fixing tools, and will create strange results if not given a good input. If there are already huge changes in volumes between neighboring cells, the Laplacian smoother will skew the mesh towards the larger cells. If some of the adjacent nodes are locked (for instance, a surface BC), the cells will distort and create lower quality or invalid mesh. The only way to get an orthogonal and conformal smooth mesh is to start with a good input mesh.

I think the *best* case to see the effect of smoothing is in the 3-edge intersection at the corner of an O-Grid. I would not solve on an O-Grid mesh without first smoothing it.

Having said that, sometimes the smoother will change mesh that you don't want changed. Where you have a large growth rate on cells near a boundary in order to capture the BL, but not waste too much mesh, the smoother will try to smooth the transition to be uniform.

All in all, I think most structured meshes should still be smoothed before solving.


Stu

BrolY February 20, 2013 04:02

I use smoother only with tetra mesh.

But the mesh is made from a blocking, I only move vertices, but never smooth the mesh.

Bollonga February 20, 2013 04:32

2 Attachment(s)
Isn't it important to have ortogonal edges coming out from surfaces with wall boundary condition to catch the boundary layer?
Hoy could I move all that amount of points/ edges without smoothing the mesh?
I attach pictures of my smooth and non-smooth meshes. It's a 2D hexa-mesh with an o-grid around wall surfaces.

BrolY February 20, 2013 04:36

When I wrote move nodes, I was talking of the vertices of the blocking, not the nodes of the mesh, sorry for the misunderstanding.

Most of the times, If you increase the orthogonality, or another quality criteria, it would decrease another quality criteria. So I focuse on determinant, min angle and volume mesh.

Far February 20, 2013 05:36

I find the smoother quit unfriendly for blocking, the reason may less knowledge in this area. However if you compare the smoother of ICEM to gridpro you will notice the significant difference in quality of mesh.

In my point of view smoother should make the mesh smoother, orthogonal with less volume change and at the same time does not the change the other metrics.

I usually concentrate on min angle and quality in first try and I achieve this through blocking fine tuning through vertex movement.

PS : Simon told me that they are working on smoother actively and hired one developer from turbogrid team.

energy382 February 20, 2013 05:38

wrong thread :)

energy382 February 20, 2013 06:06

Quote:

Originally Posted by stuart23 (Post 408857)
Christoph, what version of ICEM are you using?

v 14.5

Quote:

Originally Posted by stuart23 (Post 408857)
In terms of "corrupt", do you mean negative volume elements?

yes, but also distored elements, displaced boundary layer....Even if I freeze parts, boundary layers, keep periodic parts, the smoother often doesn't follow these criterias

Quote:

Originally Posted by stuart23 (Post 408857)
The only way to get an orthogonal and conformal smooth mesh is to start with a good input mesh.

Exactely that's the point! smoother makes a good mesh even better, but can't handle "acceptable" meshes.


Quote:

Originally Posted by stuart23 (Post 408857)
All in all, I think most structured meshes should still be smoothed before solving.

I was talking about unstructed meshes, as I always convert my blocking into unstructured mesh. Having said that, I have to admit, that I've never tried the structured smoother. But I think they're a new feature in v 14.5?!


All times are GMT -4. The time now is 00:14.