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] Low quality Tet prism mesh (http://www.cfd-online.com/Forums/ansys-meshing/85150-low-quality-tet-prism-mesh.html)

saisanthoshm88 February 18, 2011 06:37

Low quality Tet prism mesh
 
I was meshing a tray in a cooking oven. The tray is enclosed in a volume and then meshed. The quality of the Tet prism mesh is too bad.

The following is the URL link for a ppt that contains the images of the tray geometry and the cut plane for the tet prism mesh in different views:


PPT link:
http://www.2shared.com/document/yngE...ay_images.html
The following are the URL links to the geometry, mesh and project files:
Geometry link: http://www.2shared.com/file/XpDKNDNz/Tray1.html
Mesh link: http://www.2shared.com/file/cOZ-pgUJ/Tray1.html
Project file: http://www.2shared.com/file/6X8ZHvly/Tray1.html

Can some one please look into the ppt and suggest upon the probable ways of improving the mesh quality.


Any help is greatly appreciated.


Thanks in advance.

PSYMN February 18, 2011 11:05

Prism suggestions...
 
The most likely potential problem I see is the Ortho Weight set to 1.0... This parameter tries to make a balance between orthogonality and quality or smooth transition.

If your model were just flat, then orthogonality (mesh aligned with surface normal) wouldn't need to compromise. But when you have an inside or outside corner, that corner must split the angle between the adjacent surface normals (~45 degrees in your model). This ratio determines how quickly the neighboring elements must become vertical. Setting it to 1 forces verticality very quickly without regard to quality. Try 0.5.

Before generating prism, run cycles of smoothing with laplace alternating on and off (10 Laplace, 10 without, 10 Laplace, 10 without). Laplace is great for improving the transition of the surface triangles, but it lowers individual quality. By iterating and smoothing without Laplace, the quality is improved again and transition is not lost. This creates the best starting point for Prism.

This next suggestion may sound strange, but try it anyway. Your model is pretty simple. I think you are over smoothing it during the prism run. Try a few changes.

First, try reducing mas directional smoothing back down to 5 or less, and First layer smoothing down to 1.

You may also want to try setting all the smoothing to 0. Prism will grow quickly because it is not taking time for smoothing.

Sometimes I just turn on a single first layer smoothing step and some volume smoothing steps, but 0 for everything else.

Anyway, try those suggestions and see where it goes.

saisanthoshm88 February 21, 2011 08:31

Thanks for the reply Simon. Iíve tried the approaches suggested but couldnít get any improvement in the tet-prism mesh.

Cycles of Laplace smoothing were run.


The ortho weight was reduced to 0.5 and the Fillet ratio was taken as 0.2.

The following smoothing steps were adapted:

Surface smoothing steps = 0.

Volume smoothing steps = 5.

Directional smoothing steps = 3.

First layer smoothing steps = 1.

Iíve also tried it by setting all the smoothing to zero but in spite of this there is no improvement in the mesh quality.

PSYMN February 21, 2011 09:24

Hands on help...
 
All right, I'll download your geometry and see if I can give it a try in the next few days.

saisanthoshm88 February 21, 2011 12:24

Many Thanks Simon.

saisanthoshm88 March 2, 2011 03:26

Help requested
 
Hi Simon,
In this mean while I tried to build the mesh with Quick Delaunay and Smooth (Advanced Front) so that there can be a better volume transition than the robust octree mesh.

But I had a bad experience in either case. The Quick Delaunay algorithm resulted in a very low quality mesh and I get the following error with the Smooth (Advanced Front) algorithm:

ERROR: Canít build closed shells with single edges.
status 1
Application did not finish correctly.
Error: afmesh exited with a non- zero status.

In both the cases I tried to generate a volume mesh from an existing surface mesh.

I donít see such problems with the robust octree.

Please do look into the geometry whenever you find some time.

Thanks in advance.

saisanthoshm88 March 3, 2011 03:58

I used the color by count option for the curves to figure out the problem but I couldn't infer anything appreciable from that. Can you please help me out by looking into the geometry if you find some time.

I'm extremely sorry if I seem to be disturbing you having known that you may be busy handling a lot of issues there.

saisanthoshm88 March 9, 2011 08:21

Well Simon I found that the problem with the error in advance front algorithm can be resolved by fixing the single edges issue in the geometry by using either the Use Local Tolerance (or) Single curve cleanup options in the build topology set up.

But one inference is that when compared to the quality of the mesh obtained with the robust octree approach , the meshes with the advancing front (or) delaunay are not that good

The way I used those algorithms was to first generate a mesh with robust octree and then delete the volume elements to use the retained surface mesh as an input either to delaunay (or) advancing front approaches

Any way the main problem with the prism mesh still persists and I'm unable to fix it indeed

PSYMN March 9, 2011 12:57

Try smoothing...
 
Yes, the top down methods require you to have a "good" envelope. The Advancing front method (from CFX) is the most picky... The Delaunay TGlib AF method is much more robust and much faster...

The reason the octree method produces better quality is the smoothing.

Try smoothing the output from the Advancing front or delaunay methods. The quality will improve significantly.

saisanthoshm88 March 9, 2011 13:54

Thanks for the help Simon

I didn't get any error out of the existence of single edges while using the top down octree method but I had an error while using the bottom up smoothing (Advanced front) method in ICEM saying

ERROR: Canít build closed shells with single edges.
status 1
Application did not finish correctly.
Error: afmesh exited with a non- zero status.

This error was later resolved by fixing the single edges in the geometry

So I had a problem in that regard with the bottom up approach but not the top down approach

PSYMN March 9, 2011 18:51

The top down doesn't mind it... But the bottom up requires a nice envelope...

saisanthoshm88 March 20, 2011 03:06

Hi Simon, The problem with the prisms was resolved when the prism layers were redistributed with an initial height of 0.1

PSYMN March 24, 2011 21:41

Oh, were you getting the inverted distribution from the "Split Prisms" function? Yes, the redistribute command is the work around.

Please Note: That 13.0 defect has already been fixed for 14.0 and is included in the preview2 release that CADFem has access to.

saisanthoshm88 March 25, 2011 00:14

Thanks for the reply Simon , can you please explain about the inverted distribution with the split prisms option in ICEM and then were you referring to the split face ( blocking) defect ( the work around suggested in the past was to reset the " index control" ) in ICEM 13.0

And can you please tell me the recommended ICEM CFD mesh quality range if i'm using CFX
( for a tet prism mesh and a Hex mesh)

PSYMN March 25, 2011 18:32

Quality...
 
If you haven't seen the inverted issue, then no worries... It happens only in 13.0 for redistribute prism, but is already fixed... In the mean time, the work around is to redistribute using the "Move => Redistribute Prisms" command. There isn't any more to tell.

Min tetra quality should probably be above 0.2, but if you have some below, it is probably fine, particularly in a non critical area. CFX is pretty robust, you could try as low as 0.05.

The Prism metric is really harsh in ICEM CFD. If you can get above 0.01, it will probably run fine in CFX... You could even go lower. (Maybe others can comment on their worst mesh that converged)...

Hexa, you are usually looking at min angle and aspect ratio... We shoot for an angle over 18, but will take anything over 9.

saisanthoshm88 March 26, 2011 03:53

Many Thanks for all the worthy replies Simon

njenkins360 August 8, 2011 17:55

Simon, you're amazing. I've read like 30 of your posts and learned so much.

Thanks.


All times are GMT -4. The time now is 02:53.