Interpolation Issue?

[MaddogMcCree]

Hey there,

I´m researching composite structures with Abaqus and
I`ve written a script to evaluate failure criteria - i.e.
efforts of the structure. This is done by accessing the
odb-file and writing a new FieldOutput to it (Hashin- and Puck-Criterion).
So far everything worked well. But though the criteria
are defined only for positive values, the elemental progression shows negative values in the contour plot.

As the listet effort values for the integration points
within the report file are all positive, I guess this is
an issue of internal Abaqus interpolation.

Element used is C3D20 - full integration ->27 integration points. No single value of the failure criteria within the rpt.-file is negative. I´m really confused.

 

[MaddogMcCree]

What I forgot to mention:

The model is build from several solid-layers of
CFRP with different orientations.

Having only one orientation, there´s no such
issue with negative values in the contour plot.

 

[Elabbasi]

This is probably a standard interpolation issue. You can get negative extrapolated values even if all integration point values are positive. It could even happen if you use lower order elements such as C3D8.

Consider a simple 1D example with 2 Gaussian integration points. If one integration point has a failure criterion value of 0.0 and the other a value of A, the extrapolated values at the two end points are -0.366A and 1.366A.

Nagi Elabbasi
Veryst Engineering

 

[MaddogMcCree]

Thanks for your response.
It s just as I supposed.

But how can I overcome it? The values of the contour plot are definitely not the right ones.
I thought of deleting all layers except one to have the interpolation performed on only one layer without disturbing high differences in the values.

Abaqus right now is a real pain in the a.. .

 

[mrgoldthorpe]

I'm going back a few years, probably to ABAQUS/Post, but there used to be a contour plot option that didn't smooth results to the nodes but plotted results element-wise. This avoided displaying the erroneous plots described by MMC. I've forgotten the word ascribed to it.

Is this feature available in CAE?

 

[mrgoldthorpe]

I've attached an old example of what I mean. I was doing the same kind of thing as MMC - working out a creep damage parameter that is quite sensitive to the input stresses and strains. As I recall, plotting in the usual (default) manner shows unreasonable results, so we displayed this damage parameter smoothed over each element.

I still can't remember the term for this kind of plot though.

 

[mrgoldthorpe]

It's "Quilted". From 6.8 CAE User Guide:

26.4.1 Choosing line-, banded-, or quilt-type contours

.......

You can choose Quilt-type contours only for element field output variables. Variable values are extrapolated to element faces on the surface of your model, with no averaging between elements. Quilt contours are useful for evaluating results on an element-by-element basis. Only one color per element is plotted for axisymmetric elements with asymmetric deformation (CAXA or SAXA).

Banded-type contours are the default...

Figure 26–3 shows the appearance of line-, banded-, and quilt-type contour plots...

 

[Elabbasi]

Good point mrgoldthorpe.

A quilt contour plot can do it. It produces one color per element face. A 2D problem with only positive/zero values at integration points should never result in a negative value in the quilt plot.

In 3D however, there is no guarantee! According to the manual, the plotted parameter is still extrapolated from the integration points to the faces (and each face gets a different contour value). It is much less likely to see a negative value compared to the default banded contour plot, but not impossible.

Nagi Elabbasi
Veryst Engineering

 

[MaddogMcCree]

Thank you for all your help.
I think the quiltet plot option is a real good idea.
For the beginning this one will satisfy me.
Yet there still remain some negative values.
The resulting contour-plot with the quiltet option is much more reasonable now.
The best way maybe is a results-mapping, as I suggested before. With a separation of the layers and also a separate postprocessing of the fieldOutput on one layer at a time.

Best regards,
MaddogMcCree