Hexahedric on round profiles
- Alessandro
- Topic Author
- Offline
- Premium Member
-
- Posts: 108
- Thank you received: 0
to try it, I made a simple benchmark on a beam subjected to a simple bending, with different meshes but same boundary conditions:
1) Linear tetra mesh
2) Quadratic tetra mesh
3) Linear hexa mesh
4) Quadratic hexa mesh
In the attached image you can see the displacement.
Results surpised me a lot, now I don't know what to do with my analysis made using tetra elements.
My problem is that every time that I try to mesh a non-prismatic component (I mean, a part with one radius, a circle, etcetc) trying to use hexa elementes, the software fails to mesh.
It is my fault? what can I do to solve? [file]<br /><br />Post edited by: Alessandro, at: 2009/01/23 19:39
- Alessandro
- Topic Author
- Offline
- Premium Member
-
- Posts: 108
- Thank you received: 0
Attachment bending-a9ba5556f8f924f0e25de1547ca5363c.pdf not found
- Joël Cugnoni
-
- Offline
- Moderator
-
the difference that you observe between quadratic tetra (2nd in the list, sure?, U=~3.8) and hexa (3rd & 4th, U=~3.3) seems really big to me..
For sure , you should completely forget about linear tets for stress analysis... but usually quadratic tetrahedral meshes give reasonable predictions of the stiffness at least!! (but not very good for local stress & plasticity).
So 2 questions:
1. Element size: in slender structure (bending dominates), you should have at least 4 elements in the thickness.. maybe more with tetra. What is you mesh density here (nb elem along each direction)? Did you try to look inside the volume mesh (clipping plane) to see if the internal elements are not "too big" compared to the surface? Convergence (evolution of error with mesh size) is slower with tetrahedral mesh than hexa; so try to play with mesh size to see what happens... Could you post a picture of your meshes?
2. How do you apply your load? with Force_Face? or Pres_Rep?
Using Force_Nodale can be dangerous when changing mesh as the number of nodes can change and the total force will then be wrong (number of node * applied nodal force!)
Joël Cugnoni - a.k.a admin
www.caelinux.com
- Alessandro
- Topic Author
- Offline
- Premium Member
-
- Posts: 108
- Thank you received: 0
Hi,
the difference that you observe between quadratic tetra (2nd in the list, sure?, U=~3.8) and hexa (3rd & 4th, U=~3.3) seems really big to me..
For sure , you should completely forget about linear tets for stress analysis... but usually quadratic tetrahedral meshes give reasonable predictions of the stiffness at least!! (but not very good for local stress & plasticity).
So 2 questions:
1. Element size: in slender structure (bending dominates), you should have at least 4 elements in the thickness.. maybe more with tetra. What is you mesh density here (nb elem along each direction)? Did you try to look inside the volume mesh (clipping plane) to see if the internal elements are not "too big" compared to the surface? Convergence (evolution of error with mesh size) is slower with tetrahedral mesh than hexa; so try to play with mesh size to see what happens... Could you post a picture of your meshes?
2. How do you apply your load? with Force_Face? or Pres_Rep?
Using Force_Nodale can be dangerous when changing mesh as the number of nodes can change and the total force will then be wrong (number of node * applied nodal force!)
Dear Admin,
I answer to your question:
1) I attach the whole folder of this case. It seems to me that the mesh is not so coarse, but I'm still a beginner in this filed.
2) Yes, I use FORCE_FACE on the top face of the beam (20 MPa in the X direction).
To be honest, my answer was related to the possibility to mesh with hesaedric elements some special solids, like a cylinder (when I try to mesh a cylinder using "automatic hexaedrization", the resultant mesh is open and Salome gives me an error).
Thank you for your attention.<br /><br />Post edited by: Alessandro, at: 2009/01/23 21:38
- Alessandro
- Topic Author
- Offline
- Premium Member
-
- Posts: 108
- Thank you received: 0
Attachment tetra_esa_beam-37a3bdd5a4268f0b94ba464539b182ba.gz not found
- Alessandro
- Topic Author
- Offline
- Premium Member
-
- Posts: 108
- Thank you received: 0
Hi,
So 2 questions:
1. Element size: in slender structure (bending dominates), you should have at least 4 elements in the thickness..
You are right: making a finer mesh things are ok. Difference between a quadratic thetra mesh and a quadratic hexa mesh is less then 1% for this particular case.
Thank you for the hint.<br /><br />Post edited by: Alessandro, at: 2009/01/23 22:17
