Large Difference between Solution at Integration Points and Nodes

[Yannick]

Hello everybody.

we are using ccx to simulate centrifugal loads on a 3D body (multiple parts and materials, with contact). In order to see if we are doing it correctly we benchmark the solution against Ansys and we have noted some differences which we can not really explain.

The mesh used for our model consists primarily of C3D20 elements and a few C3D15 elements (in calculix). We try to have a similar mesh as in Ansys to have a somewhat meaningful comparison.

When evaluating the results in Calculix, a significant difference was observed between the results at the nodes (.frd) and those at the integration points (.dat). The analysis revealed that the maximum stress found at the nodes is approximately 10% higher than the maximum stress at the integration points (similar differences were observed for the strains).

In contrast, when we compare this in the Ansys solution we see virtually no differences in the maximum values between nodes and integration points.

Have you experienced similar behavior before?

Does ccx need a finer mesh (in comparison to Ansys) to have smaller / same divergence between nodes and integration points?

Thank you for your inputs

[Calc_em]

I don’t know how Ansys displays the stress contour plots, but CalculiX always shows averaged nodal stresses in .frd field output. So stresses are first calculated at the integration points (where they are most accurate), then extrapolated to nodes and averaged between elements sharing the same nodes. As you’ve already noticed, you can get the most accurate stress output (from integration points) using *EL PRINT, but it’s in a tabular format. Alternatively, you could use some scripts (such as this one) or request error measures (Zienkiewicz-Zhu stress ZZS or gradient error estimator ERR) to see how much of an error your mesh introduces.

[NorbertH]

You could also visualize the integration points in paraview for comparison. If that’s an option.