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] width of prism-layers in unstructured grids (http://www.cfd-online.com/Forums/ansys-meshing/97802-width-prism-layers-unstructured-grids.html)

Peter2010 February 25, 2012 15:46

width of prism-layers in unstructured grids
 
Hello. I am quite new in CFD and I'm going to mesh a radial flow pump with an unstructured grid. I have a general question concerning the creation of prism layers.
So I am wondering how large the whole prism-area should be to get a good mesh? (I'm using SST-Model with automatic wall function, so I found out that the boundary layer should be resolved with about 5...10 prism-layers and the value of y+ should be about 30 in my case.) But is it worse to make the whole prism-area much wider than the boundary layer? I think that the prism elements in general have a better quality than tet-elements... so why not creating as much prisms that reach pretty far into the inner flow? Is there a limit, or a general recommendation, for the width of the prism layers? Should they only be used for the region close to the wall? If true, why is it better to use tet-elements from a certain distance from the wall?

Thanks very mush for your help!

PSYMN February 26, 2012 21:15

In many situations prism quality can be difficult to maintain as you add more and more layers... Tetras tend to fill any volume without suffering from quality issues... Tetras are also "omni-directional"...

The point of prisms is to have an-isotripic mesh... A mesh with more nodes in one direction, particularly a direction where you expect to see a higher result gradient, without also adding too many elements in the less interesting direction...

But, in answer to your question, if the prism quality is good and you limit the aspect ratio so it doesn't go to far the other way (tall skinny prisms), you can fill your volume with prisms and be very happy. This is the idea behind many swept meshing schemes that will sweep triangles from one end of the model to the other...

Peter2010 February 28, 2012 12:10

Okay, thank you for the answer. I think I still have further questions:
So if I understand you correctly, prisms should better be used if the flow is anisotropic (Close to the wall)? But in regions where the flow is more isotropic (in the inner flow) tetras are better because of their “omni-directional”-shape, which results in better quality?
Which criteria is suitable best to determine the prism and tet – quality in ICEM? Could you give me a recommendation for this?
If I set the prism-height in ICEM to “0”, it is computed automatically… What does the resulting height of the prism-layer that ICEM generates depend on?

Thanks again!

PSYMN February 28, 2012 23:15

There are a variety of quality metrics that people use... It really depends on your solver. One thing to note, if you run the "Quality" metric it really is very harsh on prisms... The "Quality" metric divides each prism into 3 tetras and gives you the worst quality of the three. You can imagine that a perfectly good prism (eqilateral base, orthogonal sides, but thin) would give a very poor quality if broken into tetras and evaluated... So don't worry so much if your prism quality looks much lower than your tetra quality... Tetra should be over 0.15 or so, but prism should be over 0.0001. Actually many solvers can handle prisms (and tetras) worse than that. Look in the help for more details.

When smoothing tetra prism mesh, we usually "freeze" the prisms while smoothing the tetras.

As for setting the initial height to zero (allowing it to "float"), it looks at the base triangle area and the number of layers and growth ratio and back calculates the initial height so that the last prism layer height has a volume similar to the adjacent tetra... Of course, there is some fudging done to keep heights similar to their neighbors also... This smoother volume transition is easier for the solver to handle...

Best regards,

Simon

Peter2010 February 29, 2012 04:43

I looked in the ICEM-help manual, but the definition of the criteria „quality“ seems to be different from your explanation: It is said that for prism elements the quality is calculated as the minimum of determinant and warpage. I can’t find a description of warpage… is it the same as “Max Warpgls”?
I am now confused, what is really done to determine the „quality“ of a prism-element.

My procedure to create the prism layers was always first creating one layer with “zero”-initial height. Afterwards I splitted this layer. If ICEM calculates the height of the initial layer in that way that the volume is similar to the adjacent tetra: after splitting the prism the volume transition can’t fit any more…?
So is the combination of setting the initial height to zero and splitting the Prism afterwards not advisable?

Thank you,

Peter

Peter2010 March 7, 2012 05:29

Is there anyone else who can tell me how the quality of a prism element is calculated in ICEM when using the “Quality”-metric? Does it calculate the aspect ratio of the 3 internal tetras, like Simon told me… or is it calculated by determinant and warpage? If so, how can I imagine these criteria “determinant and warpage”?
Thanks very much!

PSYMN March 7, 2012 10:09

The Help says the minimum of determinant and warpage... (so I will need to go back to development and check what I was told previously... Maybe we are both right and it is divided into tetras and then the minimum of determinant and warpage is calculated on the tetras... )

For more on determinant and warpage, you can start in the help, but then you may also need to google to get more background info on the terms used in the help...

We really should have pictures to help with these definitions...

Peter2010 March 8, 2012 09:32

Okay, thank you. So I’ll go on googling to find out more about determinant and warpage…
Could you tell me something concerning my question about setting the initial prism height to zero:
My procedure to create the prism layers was always first creating one layer with “zero”-initial height. Afterwards I splitted this layer for example into 10 sublayers. If ICEM calculates the height of the initial layer in that way that the volume is similar to the adjacent tetra: after splitting the prism the volume transition can’t fit any more…?
So is the combination of setting the initial height to zero and splitting the Prism afterwards not advisable?
Thanks again!

cfd seeker March 9, 2012 01:44

Peter i guess and i hope so i have understood your problem of creating just one layer and splitting it....After splitting your only prism layer into 10 layers then go to Move Nodes>> Redistribute Prisms and it will ask you about the redistribution of prism layers based on "Fixed Growth Factor" or "Fixed Initial height"...here come your choice if you want to adjust prisms so that to achieve certain initial height for controlling wall y+ values or you can also adjust prisms based on fixed growth factor, in this case initial height will be adjusted by the ICEM....In either case your last prisms will be in continuity with the adjacent tetra...i hope it will help you..thanks
Regards

Peter2010 March 9, 2012 03:31

Thanks for your answer, cfd seeker! But what you mentioned in your last sentence is exactly the point:
At first I generate 1 global layer with initial height = 0. So ICEM generates it that way that its height fits to the adjacent tetra.
Then I split this prism layer into 10 sublayers. No matter if I choose "Fixed Growth Factor" or "Fixed Initial height" while redistributing... afterwards the the inner prism layer is certainly much lower and so doesn't fit to the adjacent tetra any more...

cfd seeker March 9, 2012 11:54

You are always welcome Peter.
What I got from your post is that redistribution is also not helping your cause isn't it or its otherwise?? If redistribution is not helping you in getting the continuity then I will suggest you to create more than 1 prism layer at first, may be 3 or 5 and then split and redistribute. Try this also.
Regards

PSYMN March 9, 2012 17:34

Setting the initial height to zero works best if you are actually generating the number of layers you intend to keep...

If you only generate 1 layer, then that layer will be setup to be approximately as thick as the adjacent tetra volume. If you split that, you will end up with a smaller prism element and a larger jump to the tetra. I do not recommend it.

If you want to try the split and redistribute method (to save time), then you should probably calculate how thick the total hight should be and set it up for that...

Personally, I don't like the 1 layer and split. If I am in a rush, I usually do at least 3 layers and then split/redistribute. Prism has a lot of smart things that it does to improve quality as it grows subsequent layers. if you just grow one layer, you are not using most of those things. You might as well grow all your layers and just turn off all the smoothing.

katarin February 25, 2013 07:02

wedged prism layers 3D
 
Quote:

Originally Posted by PSYMN (Post 346378)
In many situations prism quality can be difficult to maintain as you add more and more layers... Tetras tend to fill any volume without suffering from quality issues... Tetras are also "omni-directional"...

The point of prisms is to have an-isotripic mesh... A mesh with more nodes in one direction, particularly a direction where you expect to see a higher result gradient, without also adding too many elements in the less interesting direction...

But, in answer to your question, if the prism quality is good and you limit the aspect ratio so it doesn't go to far the other way (tall skinny prisms), you can fill your volume with prisms and be very happy. This is the idea behind many swept meshing schemes that will sweep triangles from one end of the model to the other...

Hello, I also want to make o triagle prism mesh with layers , trying to solve a 3D Dam break problem. It seems like it cannot be done. Is there a way, besides ICEM, through the FLUENT mesher environment that I could do it??? Please help


All times are GMT -4. The time now is 02:52.