Tetrahedron - hexahedron

Was wondering if someone could bring light on a small problem.
I'm trying to compare results from 2 different mesh for the same system. But the same problem with the same BC does not pass in hexahedron mesh, while it passes in tetrahedron.
Is there any kind of special condition that has to be declared for hexahedron elements ?

I'm using two different studies with different files and directories. If it is needed I can put the geometry too as an attached image, but it is rather a simple device.

Thanks in advance.

Well, I was expecting something not application related, something more general about the use of the type of elements of the mesh.

I'm trying to describe the problem with what I have at hand (not on linux for the moment).

Here it is, the system in 2D view with the boundary conditions (a fixed surface and a symmetry surface) and the applied load (a pressure perpendicular to the plane of the screen). It should capture a torsional movement of the plate.

i82.servimg.com/u/f82/12/42/63/52/microm11.png

The geometry is 3D (simply extruded the 2D view by a small thickness).
Boundary conditions are imposed for the surface Fixed (group of faces) for DX, DY, DZ and symmetry conditions for the group Symmetry (DZ=0).
The load is applied along z-axis, either with PRES_REP or FORCE_FACE on a surface defined with grey in the image.

The thing is, using a mesh with tetrahedral elements everything worked well, got the expected deformed shape and even the expected deformation values. Moreover, I checked the documentation and hexahedral elements seem to have an improved accuracy so I thought I'll give it a try.
I kept the same geometry (the groups are defined already in the geometry so the mesh type does not interfere with their creation), only changed the mesh element type. It doesn't work.
The message I get is for a node "Pivot presque nul ... matrice non factorisable" and I get a node number and DY direction. The node does not appear in any of the groups on which BC were imposed. If it counts, I can put an image of where it is found.

I've tried a finer mesh, thinking that maybe the number of nodes wasn't high enough on the surface of fixed conditions. Still the same error.

I tried to move the mesh about Y axis, not to have the torsion axis aligned on z axis and I got the same type of message but now for an another node and the DX axis.

I even tried applying only the fixed surface and the applied force, I still got the same error.

Don't hesitate to ask me further details and I appreciate much your help.

Post edited by: Boloni, at: 2009/10/15 15:20<br /><br />Post edited by: Boloni, at: 2009/10/15 15:23

It seams to be a meshing error specific to the problem. Make sure your mesh is continuous. If you only apply the fixed BC does it converge? The other option is that you matrix could be sparse. Try using a pivoting solver such as mumps.

Try to see Advanced Mesh Info for the mesh and check for leftover tetrahedra or or just try Clear Mesh Data and Compute again. Could also try to merge nodes if some are unattached. Just to check if it is something with the mesh. With respect to pressure application it could be a missing MODI_MAILLAGE.

Also some elements generated with Salomé need conversion to work in ASTER. This is the case for 2nd order formulations of some elements. See u4.23.02.pdf - but still the answers are application related. Sorry.

Only thing I can come up with which is not application specific is that to me hexahedra may not be the geometric most obvious choice to model a thin plate as the aspect of the generated faces could be unfortunate.<br /><br />Post edited by: Torben, at: 2009/12/18 18:16