[Libmesh-users] Remeshing during calculation with tetgen

[robert <rob...@un...>]

ok, thanks for your reply.
In general, do you have a better suggestion for a mesh generator? I need 
to predefine my geometry and subdomains in any CAD program, export it 
and mesh it. Then, during the calculation re-meshing would be great when 
the domains get deformed and the elements badly shaped.

So far I have tried the following: During the calculation the mesh is 
deformed. Then I write it to a file, re-mesh it manually with tetgen, 
import the new mesh to libmesh (new EquationSystems object), manually 
project the old solution on the new equationsystems with 
system.point_value (...) . However, point_value works with point_locator 
and for meshes with several million elements this takes too long. 
Strictly speaking, I start a new calculation with new mesh and 
equationsystems objects and use the old solution as starting conditions.

As I said, this is not very efficient and quite limited for application.

Thanks, Robert


On 06/07/2012 06:45 PM, John Peterson wrote:
> On Thu, Jun 7, 2012 at 10:23 AM, robert<rob...@un...>  wrote:
>> I think the problem is really my computer - i installed paraview on a better
>> machine and it works with the same file.
>>
>>
>> I'd just like to ask another question which might not be interesting enough
>> for the list:
>> can I use the class TetGenMeshInterface to re-construct my mesh when it
>> becomes badly shaped due to deformation?
> Probably not without some additional development and an improved
> understanding (on my part) of what Tetgen is actually capable of.
>
> If the geometry is relatively simple (no holes) it should in principle
> be possible to throw away all the elements and re-tetrahedralize
> through the TetGenMeshInterface::triangulate_conformingDelaunayMesh()
> interface.
>
> I will say that I have not found tetgen to be a particularly robust
> mesh generator in the past, however.  For example, I had to make the
> miscellaneous_ex6 geometry very simple for Tetgen to successfully
> generate a Mesh (and it behaved differently on Mac vs. Linux).
>
>> What happens with the
>> regions/subdomains and the EquationSystems object then.
> Well, throwing away all the elements as I suggested would obviously
> lose the subdomain information, so that would be trouble.
>
> Assuming you had a list of tet faces defining the subdomain
> boundaries,  it might be possible to perform a constrained Delaunay
> tetredralization in Tetgen, though I don't know how to do it and our
> interface doesn't provide the capability.
>
> The EquationSystems object *should* be able to re-initialize itself
> after significant changes to the Mesh... it is designed to do this for
> arbitrary refinement/coarsening, so it may just "work".
>
> A long-term goal of mine is to interface with a better tetrahedral
> mesh generation library in libmesh.
>