[Gmsh] Generating mixed element mesh

Trevor Strickler trevor.strickler at gmail.com
Tue Jul 9 17:49:58 CEST 2013


If you need a structured mesh region to be ALL structured inside and still
interface to an unstructured region, you can do the following:

If your structured region is extruded, surround your source surface with
another concentric surface that you will extrude alongside the central
surface...the outer surface will create a volume around the sides of your
structured surface that can be your interface to the unstructured regions.
You will then need to extrude quadtri regions from the top and bottom
surfaces of these extrusions to close those two ends.

If your structured region is transfinite, you can, with some work, create a
layer of transfinite regions around the entire original region, even fixing
the gaps at sharp edges.

Quadtri also works with extruded boundary layers as well.


Now, with that all said:


Admittedly, some current QuadTri features are inconvenient.  As of last
week, I have started changing how it works.

My 3 main changes:

1. There will be no more 'double layer' option for meshing the extruded
quad to triangle interfaces. That is being replaced by a method that works
just like the transfinite quadtri interfaces that use a body centered
vertex to divide hexahedra and prisms.  Only ONE layer is required.

2. Source surfaces will be able to extrude around in loop onto themselves
and quadtri will still work on the sides.

3. (This one is hard) I am changing the boundary layer quadtri method to
allow the outer quadtri layer to be locally scaled to give higher quality
elements in the transition to unstructured regions. I did this for cases
where source elements vary in size across a source surface and one value
for the height of the quadtri layer will not give high quality transition
elements everywhere (even using vector views to define the extrusion
lengths cannot adjust the last layer independently and locally). (Note to
Christophe, this requires adding just one more 'float' variable to class
xyzv in smooth_data normals...I hope that is not too bad).

 I am half finished with these changes, so hopefully soon I can submit
these soon.  If anyone needs the features before they go into the nightly
snapshot, I will be able to send a source patch you may use to build your
own version with my changes (I will announce when they are finished and
debugged).

Cheers,

Trevor Strickler
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.geuz.org/pipermail/gmsh/attachments/20130709/996f16b3/attachment.html>