Different result with linear and quadratic mesh

Marcelo wrote:

Hi Alessandro

Thanks for the attencion. I´ve already seen your topic, but in your case the error is 12% and I foud more than 7x in the results of stress.

Anyway should I use only quadratic on structural problems?
If i could get a smaller mesh the results would be more similar?

Had a quick look at you mesh, and yes, you should refine it and/or use quadratic mesh.

Hi

Concerning the choice of element / mesh refinement, here is a summary of my recommendations:

for 3D stress analysis in linear statics without contact:
- USE QUADRATIC ELEMENTS BY DEFAULT: they will give you systematically better results than linear elements.
- COMPLETELY FORGET ABOUT THE LINEAR TETRA, they are always too stiff and give BAD stress predictions, their convergence rate is also very low (error does not change much if you refine the mesh).
- if you NEED to use linear elements: only use hexa or prisms. When bending is significant, put at least 2-4 linear elements through the thickness. To avoid "locking": try to keep the element aspect ratio between 1 and 5 OR use reduced elements



for 3D stress analysis in non-linear statics & contact:

- in large strain analysis (plasticity), the enhanced accuracy of quadratic elements is usually less significant than in linear statics. The non-linear geometric effects and integration of local constitutive behaviour are more important. In this sense tetra are not the best choice in this case (strange plastic strain localization patterns...) My advice is to try several meshes and different element types (and exact/reduced integration). For highly non-linear studies, the best choice usually is a nice structured hexahedric mesh (linear or quadratic depending on the case).

- with contact, quadratic elements tend to have large contact pressure oscillation and may induce solver convergence problems. I would recommend to use a fine linear hexahedric (or at least prisms) mesh in this case.

for 2D stress analysis:
- in linear statics, prefer quadratic elements
- for large strain / plasticity, try both and look which one is best
(there are no really pathologic elements in 2D)

I hope that it will help you in your analysis

Joël Cugnoni