CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Siemens (
-   -   Mesh Refinement (

JY August 11, 2002 19:17

Mesh Refinement
I am really frustrated when dealing with mesh refinement with PRO-STAR:

For a big model (with multiple blocks or even multiple areas in which the mesh was generated in different orders), the default order of I, J and K for mesh refinement always get massed and confused! This always results in a mesh refinement in a dimension other than you expected. You know, you don't always want to refine the mesh with the same ratio in all three dimensions.

For example, you want to double the mesh in the direction normal to the solid wall, so you choose a layer (or several layers) of cells as a cell set parallel to the solid wall. But be careful, if your cell set cover different blocks or even different areas where you generated the original cell with the reflection function or something else, the underlying orders of I, J and K within these cells might be inconsistent! If this is the case, and you choose "1 by 1 by 2" (or what ever different times in the three dimensions), you get mesh density somewhere doubled in Z-direction, but probably somewhere else doubled in Y-direction!

This is so annoying as you don't always remember how you generate the original coarse mesh.

Another annoying thing: If you did the mesh refinement from the command line or the Cell Tool bar, it is not always an easy task to see immediately the refinement effect when you have many cells. If done from STAR GUI panel, it is very convenient for you to see the effect immediately. However, from the GUI panel, you can only refine the mesh with the same refinement ratio in all three dimensions, i.e., IxJxK = 2x2x2 or 3x3x3 or what ever.

I spent several hours for a simple case but I had to resume the model to start from the previous step, again and again!

Any experienced user please light me up -- any trick to reduce the effort in mesh generation?

Thanks a lot,


steve August 12, 2002 11:04

Re: Mesh Refinement
Use the CDIRECTION and RESTRUCTURE commands first to consistently orient all the cells that you want to subdivide. Then they will all be subdivided the same way.

Jiaying Xu September 4, 2002 06:24

Re: Mesh Refinement
Hello, Steve,

Thanks for your response. I tried but had other problems up.

What I did are:

1) define a cell set in which structred mesh exist;

2) use CDIRECTION to define the I,J,K directions of a base cell;

3) use RESTRUCTURE to reorder all vertices and cells in the set.

4) merge and compress cells and vertices.

I was warned that some faces would be destroyed during the step 3, but I still answered yes. I finally found that some disconnectivities occured after I did a "CSET ALL $ CPLOT". Obviously some cells were distorted.

Did I do something wrong?

Thanks for you further time.


Jiaying Xu September 4, 2002 06:46

Re: Mesh Refinement
Another question related:

What if I use tetrahedral cells? I mean, do we still have I,J,K directions in this case? What are they if yes?

If I use hexahedral cells, the saved .cel file has the format of "CNumber N1 N2 N3 N4 N5 N6 N7 N8 CType MType" each line. But what if I use tetrahedral cells all over the flow domain? Still use this kind of eight-node format?

I know the user guide says a tetrahedral cell can be described as an eight-node N1-N2-N3-N3-N4-N4-N4-N4 sequence. But this by default suggests the I,J,K directions. I am wondering what will happen if I refine the mesh in this case.

Thanks for your explanations.


steve September 4, 2002 09:13

Re: Mesh Refinement
Sorry - I have no idea what you did wrong. Perhaps you have non-hex cells mixed in with everything else. You can't have a structured mesh with unstructured (ie non-hex) cells.

steve September 4, 2002 09:29

Re: Mesh Refinement
You can't divide tets into anything else other than NxNxN where N is an even number. I don't think it is physically possible to take a tet and divide it 2x1x3. If you want to work with structured meshes, you have to stick to a structure, and tets are definitely not structured. The numbering convention does suggest that there is still an I,J,K direction, but its not useful for refinement. If you want to divide any type of cell, look at the CMREFINE command instead of CREFINE. It limits you to equal divisions, but works on everything.


Jiaying Xu September 4, 2002 10:09

Re: Mesh Refinement
Yes, all the cells in my case are hexehedral, i.e., every call consists of eight different nodes.

It is true that the overall mesh occupying the whole flow domain is unstructured -- they are made up of two different sub-domains, within each sub-domain, the mesh is however fully structured, see the schematic from

I did things (CDIRECTION + RESTRUCTURE) within each structured sub-domain separately. Was I correct?

Thanks a lot for your time.


Jiaying Xu September 19, 2002 13:37

Re: Mesh Refinement
Any exprienced person help me on this problem, please? Thansk a lot! Jiaying

All times are GMT -4. The time now is 05:16.