[Scigraphica-devel] Re: [gtkextra] finite element meshes

[Eric M. M. <emo...@be...>]

Adrian Feiguin wrote:

> I'm working on a way to generate arbitrary meshes for finite elements.
> What I need to know is: 
> 
>  Is there a standardt format/representation for this kind of meshes. I
> mean, if you needed to define one in a worksheet in scigraphica, how would
> you do it, considering that you need to determine edges and holes.
> 
>  First, of course, you need a set of x,y,"intensity" points/nodes. Then
> you need to add constraints for the edges and the holes. If you have
> experience with programs that do that, can you give me some feedback,
> please? 



Adrian,

Not sure about standardized formats, but please continue to try to find 
one before embarking on creating your own!  I have done some FE 
application usage, and at one point looked at what (free-as-in-beer) 
tools were available for download and use on Linux.  There are many that 
do parts of the job, but few that interoperate.


Second, regarding your comment on x,y, intensity points:  If you use a 
point-based FE method, then edges and holes are very hard.  I.e. if a 
point represents a given location in space, and heat flows along lines 
to the other points surrounding it, then it is easy to solve heat flow 
but hard to represent edges and holes in an accurate way.

Instead, the FE codes I have used were all area/volume based.  I.e., the 
  characteristics of the material were not associated with nodes at all, 
but with the elements defined by the triangles.  Nodes are typically 
just location and identifier.  The elements, whether triangles, quads, 
or a 3d type, store and ordered list of nodes defining them, and the 
identifiers of the elements with which they share edges or faces (2 or 
3D).  They also hold whatever physical properties are associated with 
the material.  E.g., a node will have a temperature, but not a thermal 
conductivity, because triangles that share the node may be of different 
materials.

So, a mesh definition for solving FE is significantly more elaborate 
than just creating a surface for point plotting.

HTH,

Eric