## Biasing in Unstructured Meshes

Moderator: GiD Team

kilgalli
Posts: 2
Joined: Thu Jun 25, 2015 6:44 pm

### Biasing in Unstructured Meshes

Hi,

I am currently evaluating GiD to potentially replace our current meshing software. My company develops and maintains an electromagnetic solver that has a range of meshing challenges to optimize the problem.

In the example posted an element size range has been specified, a size gradient, and a geometrically adaptive function for all surfaces. The function places a bias away from lines. Does GiD have any adaptive features that could reproduce this or is there a work around to get close? The closest I have gotten was by building new geometry and setting sizes on each surface. My attempt with preferences is also posted. Another way to apply this type of thing after mesh generation would be somthing like refine the mesh X times with Y bias around Z line.
Objective mesh
CubitSquareOnly.jpg (265.35 KiB) Viewed 2451 times

GiD Attempt
Attempt.png (343.27 KiB) Viewed 2451 times

Thank you,
Sean Kilgallin
Engineer
IERUS Technologies Inc
escolano
Posts: 1625
Joined: Thu Sep 18, 2014 3:31 pm

### Re: Biasing in Unstructured Meshes

With GiD you could assign sizes only to geometrical entities, then is not possible to localize a finer mesh in some inner part of a surface.
To do it is necessary to modify the geometry, adding more entities to assign sizes.

for example to have a mesh similar to your picture in a 1x1 square surface I had to split the surface, and assign size to the points, lines and or surfaces.

e.g spliting in two parts the square, and assigning size 0.015 to the inner lines and points, and general size=0.035
surface_splitted.png (8.09 KiB) Viewed 2449 times
mesh
surface_splitted_mesh.png (52.86 KiB) Viewed 2449 times

or splitting in three parta, and assigngin size 0.015 to the inner surface, lines and points, and general size=0.035
surface_splitted_3.png (8.13 KiB) Viewed 2449 times
surface_splitted_3_mesh.png (54.65 KiB) Viewed 2449 times
kilgalli
Posts: 2
Joined: Thu Jun 25, 2015 6:44 pm

### Re: Biasing in Unstructured Meshes

Thank you for the quick reply! Can you speak to any developments that might solve a problem of this nature without manipulating underlying geometry?
abelcs
Posts: 141
Joined: Mon Oct 13, 2014 9:14 am

### Re: Biasing in Unstructured Meshes

Dear collegue,

unfortunately, it seams that the only way to deal with this situation is the way Enrique has explained: to introduce new geometrical entities in the model, onto which you can assign different sizes.

Actually there is another way to assign sizes independently on the geometrical entities, but it is not so easy to use. It is using a background mesh to indicate the desired sizes in the different parts of the domain. This is typically used in optimization loops, where the mesh of the previous step can set the desired sizes for the new iterations based in some error measurement. If you are interested in this option we can provide you with more details.

It is planned to introduce entities in GiD just to provide with mesh sizes information, but it is not foreseen in short term.

Regards,

ABEL
escolano
Posts: 1625
Joined: Thu Sep 18, 2014 3:31 pm

### Re: Biasing in Unstructured Meshes

About the approach or 'refine the mesh' it is possible to do it, selecting the triangles to be splitted:
Mesh->Edit mesh->Split elements->Triangle - Triangle
and select the triangles to be splitted
surface_mesh.png (41.07 KiB) Viewed 2442 times
elements selected
surface_mesh_selection.png (44.58 KiB) Viewed 2442 times

mesh after split selected triangles
surface_mesh_splitted.png (49.19 KiB) Viewed 2442 times
laplacian smoothing of the final mesh
surface_mesh_splitted_smoothed.png (49.02 KiB) Viewed 2442 times

This woks, and also the conditions and materials of the original triangles are inherit by the new triangles, but direct modification of the mesh is not the recommended way to work with GiD (for example the manual selection of the elements can't be exactly repeated).
The normal way is to apply everything to the geometry and then generate the mesh.