Number of Elements through Wall Thickness

[JT100]

Could someone direct me to some literature that will describe how to determine the minimum number of elements that should be used through a wall thickness for different element types? Thanks

 

[mechfeeney]

You could perform this analysis on your own. I can't point you to any literature, but the general rule of thumb is 3. In all cases, however, it is important to run mesh convergence tests (which is a study showing that the results are no longer a function of the element size). In short, 3 is a safe starting number, but your results can only be validated with a mesh convergence.

 

[JT100]

Thank you, I just thought there may be a minimum required due to the maths involved, I was advised that for linear elements a minimum of 6 is required?

 

[corus]

It depends. If the wall sees only a direct load then one element would be sufficient. If it sees a transient thermal stress then you might need more than 6. Mesh convergence tests are one way, but you could also compare the results with hand calculations, where this is possible.

Tara

http://tinyurl.com/4ydjg7m

 

[rstupplebeen]

The rule of thumb that I was taught was 3 but recently I have heard 4.

My guess is that it is similar to camera's megapixels I remember reading consumer reports saying 2 megapixels was all that is needed.

Basically run a mesh sensitivity study is the answer. Depending on how similar your problems are you may get a feel for your analysis type and may not need to run these all the time. I typically run a sensitivity and then deploy this as a method for many runs with similar geometries or loads.

Rob Stupplebeen

https://sites.google.com/site/robertkstupplebeen/

 

[loki3000]

how do you do a mesh convergence test?
do you just refine to 2x elements and check stress on a number of points or is it more statistical (K-S test)?

 

[ESPcomposites]

You just refine the mesh (1,2,3.. to N elements through the thickness) and compare the results. The stress will reach an asymptotic value. Compare the results for difference mesh densities to the asymptotic value to gauge convergence.

For general purposes, this works fine. However, note that stresses do not converge "smooth", but rather in a sawtooth like fashion. This is in contrast to displacements, which do converge "smoothly". Just something to be aware of if the data doesn't show a nice smooth convergence curve.

It is not related to a statistical distribution.

Brian

http://www.espcomposites.com

 

[mechfeeney]

As a supplement to ESP's helpful comment, yes you can probe (query) one point or multiple points and watch the behavior.