C3D10 vs C3D10M ?

[labosonic]

I'm seeking for any piece of information about the accuracy of both elements in abaqus static analysis.
Especially :
Which one is "better" (in terms of Stress and Plastic Strain) in a simple analysis i-e without contact ?
Which one is "better" (in terms of Stress and Plastic Strain) in a contact analysis ?
Why are they sometimes such differences in results between both modelling strategies ?
Is the approach using mixed elements in analysis correct ? (i mean C3D10M near contact zones and C3D10 elements elsewhere)
If so : is there a rule to know how many C3D10M elements must be used to have correct results in such approach (in terms of numbers of elements to put or in terms of physical data to check)?

 

[DickButkus]

1. Without contact I would expect you to get identical results with both elements. The "M" is more expensive, that is, requires more memory to run.

2. Never ever use C3D10 in contact. Always use C3D10M in contact.

You can mix C3D10M and C3D10 elements - but I would not trust the stresses at the interface. The load transfer should be correct. I think it would be best to avoid this because the interface may not be obvious.

Always use C3D10M for contact problems. If there is no contact, then use C3D10 because they are faster.

 

[brad]

Additional note--C3D10M are also more accurate for high-strain plasticity (but for relatively low strain--ballpark < 5%--both should give similar results for no contact). This is not commonly mentioned, but certainly used to be in the User's Manual (and probably still is).
Brad

 

[labosonic]

The problem is that in my curent version of Abaqus (6.3-1) and in the other one in an use (6.2-7) : those are differences (only a little : less than 5%) but there is a difference - and my fellows disturb me with such a small difference 8-( ).
I cannot use a modelling strategy with only using C3D10M because I would reach limits of my computer.
So I have to build in methods with mixed elements C3D10M near a contact and C3D10 further.
Have anyone any advice about the further notion in terms of number of elements (1,2 or more C3D10M before meshing with C3D10) ? Have anyone tested a solution with a physical data to check at the interface ?