LINEAR TET ELEMENTS 2

[feadude]

Do u know why when plotting contour stresse of linear tet elements (which I usually avoid except for first run or displacements only) there are several color bands (stress levels)across the element.
I thought linear displacement interpolation should give constant stress across the element because stress is the derivative of the displacement?
Thank You

 

[johnhors]

feadude

Are you plotting average nodal stresses? If so then a stress variation across an element face is quite normal, so try plotting the un-averaged element stresses instead.

 

[feadude]

I plotted unaveraged element stresses, stress across the element is constant. But I still do not understand why stress variation for averaged nodal stresses for a constant stress element is quite normal.
Thank you

 

[johnhors]

The average nodal stress is simply the average of all the element stresses that have that node as part of their topology. Then since each node of a linear tet4 element will be used by different elements the average stress at each node will almost certainly be different. The rendering routine that plots the contours knows nothing about linear or parabolic elements, it just sees three different stress levels at each corner of the element face, hence stress contours across the face will be drawn.

 

[feadude]

johnhors:
"The rendering routine that plots the contours knows nothing about linear or parabolic elements" makes it clear.
I do not think they taught us the rendering routine in FEA courses.
Thank you.

 

[jhardy1]

feadude,

A basic course on FEA should have covered the theory of FEA, but may not have gone far into the "art" and practice of effective FEA.

Modern post-processors will give you a great deal of control over contour plotting etc. We usually like to see a node-averaged stress plot for final results, because it looks more reasonable than an un-averaged plot. (In general, we expect stresses to vary gradually across a structure, not to jump in discrete increments.)

However, looking at the un-averaged stresses plots is usually a very important part of the verification process. If element stress differences are excessive between adjacent elements, this can be a sign that your mesh is too coarse, or possibly you have some modelling errors. If the stress differences are OK, then you may prefer to revert to averaged stress plots to produce better-looking graphics for final reporting etc.

 

[ParabolicTet]

correct me if i'm wrong, but I would not use linear tets for any first-run or displacement only analysis. i think just 100 parabolic tet elements will give you much more accurate results than 1000 linear tets.

http://web.peoriadesignweb.com/eos

 

[johnhors]

No Parabolic Tet, you are not wrong! You should avoid using linear tets with a barge pole, as all literature about them clearly states.

You could perhaps only justify their use in an exceptionally large complex model (which cannot be easily simplified) where the degrees of freedom in a parabolic tet model would be beyond your computer resources.

 

[corus]

Even johnhors exception case would be invalid. The only time that linear tet elements could be used is in a thermal analysis. If your model is too big then you would have to resort to sub-modelling parts of interest and using a coarse mesh for the global model.

corus

 

[tobbe]

I would certainly use linear tets for a first-run, since my pre-processor (and virtually any other) easily converts them to parabolic, in order to save analysis time if needed. My definition of "first-run" is a test-run where I check that the model is possible to analyze and have a reasonable behaviour. Not all of my analysis-models pass the entire analysis the first time I run them.

Regards.

 

[feadude]

tobbe:
I agree. By first run I did not mean that I will use the results. I meant that I will check the model for singularities, etc with linear tet before I convert them to parabolic tet. This saves meshing/solve time.

 

[corus]

Using linear elements will not ensure that a future parabolic element will work. You can find that some elements created are very thin and the parabolic elements can produce mid-side nodes that make the element have a negative volume. Besides that, you'll always get someone who doesn't know otherwise saying that the results you've got will do as they need the answers, or they need a pretty picture. Best to do it right in the first place.

corus