CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Prism in ICEM (

Eric October 15, 2006 07:07

Prism in ICEM
Dear Friends,

I use ICEM to generate Tetra mesh, but after Prism, the mesh quality is so low. What Should I do ?

Thank you Eric

Guest October 15, 2006 13:10

Re: Prism in ICEM
There are several ways you can get rid of this problem. You need to follow some guidelines like smoothing the tetra mesh by the Laplace method before generating prisms. You might be having stuck tets as well, watch out for those in sharp regions of your geometry. Lastly, the way you are generating prisms....

Chebeba October 16, 2006 15:55

Re: Prism in ICEM
In my experience the only way is to work on the original triangle surface mesh to maximize the prism meshers possibilities to generate good mesh.

Running in order 1:Surface 2:Prism 3:Volume improves end quality (compared to 1:Surface 2:Volume 3:Prism), but the prism mesher tends to fail more often running that way. YMMW.

Smoothing after prism mesher only messes things up terribly.

myron October 17, 2006 08:03

Re: Prism in ICEM
You can do smoothing - if you do it correctly. First smooth everything but Prisms - set Prisms-6 to FREEZE. Smooth everything else as much as possible. Then at the end set Prism-6 to smooth also - but set Up To Quality to 0.1 or so. This will only move a bit of the prisms to clean the last few tets in the interior.

TB October 17, 2006 22:17

Re: Prism in ICEM
I used to have this problem in ICEM too. It ended up with some pyramid elements at the corner fillet, and the quality is extremely low. Smoothing mesh doesn't help at all. I did get over it sometimes by specifying only the height ratio and number of layers and then letting ICEM to decide the initial height & total thickeness based on adjacent tet cells. However, this seemed to work only when my tet quality > 0.5. As you reducing the tet size, the approach will fail especially if you have dirty geometry with lots of sliver surfaces.

I'm pretty aware of the approach suggested by Chebeba & myron. However, these will not guarantee 100% problem-free. I still think ICEM need to improve its prism algorithm to make it more reliable in the future.

Michael Bo October 18, 2006 03:41

Re: Prism in ICEM
I often use a different approach when creating prisms in my existing volume mesh. I only create one layer of prisms setting the "Initial Height" to the size of my "Global Element Scale Factor" or higher and the "Prism Height Limit Factor" to 0.95. After running the prisms mesher I smooth everything but "Penta_6" which are "freezed". Finally I split the prisms to 5 to 10 Number of layers using a prism ratio of 1.2. That gives you a good prism mesh...

Michael Bo October 18, 2006 03:48

Re: Prism in ICEM
Here's how it looks

Guest October 19, 2006 06:17

Re: Prism in ICEM
Yeah you are right and you can get away with it but what about when you have a complex geomtery like a Formula 1 car?

Chebeba October 19, 2006 13:57

Re: Prism in ICEM
Hey that's a pretty neat trick!

I shall have to try it sometime next time...

TB October 20, 2006 00:33

Re: Prism in ICEM
I occasionally used this approach as well, but it doesn't guarantee 100% working under all conditions. I still haven't find a method that will work 100% without giving some low quality mesh with pyramid elements. If anyone has a perfect solution, I'll grateful if he/she can share the knowledge.

Robin October 20, 2006 12:03

Re: Prism in ICEM
Pyramids will be necessary wherever there are exposed quad faces, which you will have if a neighboring prism layer has a different number of layers. Pyramids aren't all that bad. What is worse is to have the first layer of prisms varying greatly in height.


TB October 21, 2006 01:08

Re: Prism in ICEM
I agree with you. However, whenever I get pyramid elements, the mesh quality goes really bad (e.g. more than 100 elements have quality < 0.1). Smoothing doesn't help, and solver sometimes has convergence difficulty with this type of mesh compared to those without pyramid elements.

I did try on a very simple rectangular flow domain before. Pyramid elements start to appear at the corner when tet is refining to a certain value, and the mesh quality is very low for such case. Can't explain why. Of course you will find it okay if you just try those tutorial examples.

Robin October 22, 2006 15:27

Re: Prism in ICEM
Don't worry about "Quality" in ICEM CFD, it really has little to do with what the solver requires. About the closest thing in ICEM CFD to a quality measure for the CFX solver is the Dihedral Face Angle, which you should try and keep above 20 degrees.

A bad element here and there probably won't be a problem.

Regards, Robin

TB October 23, 2006 20:58

Re: Prism in ICEM
That is a very general comment. It's really problem dependent. Sometimes the bad elements will get you into trouble.... As CFD guys used to say: Gabbage in, gabbage out....

Robin October 24, 2006 08:19

Re: Prism in ICEM
Good point, but "Quality" in ICEM CFD is not related to the production of errors, which is what really matters. Sure, bad elements may get you into trouble, but what you do to avoid them may also give you grief.

Allowing the first node height to change across a surface results in high aspect ratio elements that are not parallel to the wall. This significantly increases error. Having a few "bad" pyramids away from the wall is probably less harmful.


All times are GMT -4. The time now is 20:21.