Thermal stresses Abaqus

Hi all,

I have a 20 node hexahedral "Thermomechanical" element in our inhouse software which i am trying to validate against Abaqus .

The problem is, the displacements, nodal forces match exactly ( Just compared the max min values in the legends! and checked a few random nodes).

The Abaqus input file is attached:

The problem is as follows:

One single element is chosen with a linear elastic material model (Some parameters are chosen in an unphysical manner but here the idea is

to just validate the implementation):

In Z direction, the bc used in abaqus is encastre. Then there are two prescribed temperature fields:

Initial step = 0 on the entire element (all nodes)

年代tep 1 0.005 on all nodes

YOungs modulus E = 2.1e11, Poissions ratio nu = 0.31

and heat expansion coefficient alpha = 1.0 (unphysical) No heat conduction!

Essentially a thermal expansion problem.

The element type is the quadratic 20 node with full integration!

As mentioned earlier, the results of my code and abaqus match exactly on the nodal displacements and reaction forces. But I am trying

to compare the stresses at Nodes (Least square fit). What appears is that the normal stresses at the corner nodes seem to match well but the

mid-side nodes, differ by a factor of atleast 1.5! (Any).The Shear stresses are exactly the same!!!

In order to get to the source of the problem, i first compared a simple uniaxial tensile test against abaqus. There everything matched very well. All stresses and strains, displacements, Reaction forces.

Therefore there seems to be no problem with the implementation of the shape functions or any of its derivatives.

We use the same quadratic shape functions for the temperature field as for the displacement field. Now since the thermal strains only affect the diagonal stress components,

the matching of shear stresses is no surprise. But the difference in normal stresses suggests that Abaqus is doing something else.

I am not sure why because the displacements and reaction forces are the same? Any help is appreciated.

One educated guess would be that abaqus uses linear shape functions to interpolate temperature distribution. But then the question is which is correct
and why? With such large difference in stresses, which one is more reliable?

Thanks,

MS