I know the best way is to test but by any chance, can it be predicted how much inaccuracy will be introduced by hex and tet elements not sharing all the nodes of neighbor faces? 
Would it be below 5% or 10%? 
This only apply for case 2nd order or quadratic types of mixed element, since mid side nodes of diagonal unconnected. Not preferred by many, the problem can be eliminated by using 27node of hexahedral (if available) or some treatment by using MPC as Ansys capable.
It seems reliable enough if the mesh is in medium to fine setting. Interesting to know and reproduce the problem from C. Veyhl etal (2011) since volume fraction of filler element (tetrahedral & wedge) and plot of mesh is not reported.
For complex models of 3D parts, someone also possible to conduct convergences study compared to quadratic tetrahedral element by auto-meshing.
1 Like
I think the only way would be to know if the mesh has converged (not comparing both cases which I believe is impossible to predict) but if the mesh with the “gaps” has converged you could consider results are good. To do that you need to assess the stress discontinuity between the elements, the issue is that CCX only provide that results in the dat file and not in the frd file for postprocessing in CGX or prepomax, so is very difficult to do that; we hope in the future elemental results are available for plotting in the graphical postprocessors.
1 Like
Test with linear elements
Rather than quadratic hex20 and tet10, I used linear hex8 and tet4 elements. The PrePoMax is still grouping hex and tet elements separately into different parts/components. Maybe it’s expected, since neighbor hex and tet elements don’t share all the nodes of a face.
Indeed, this is similar as @JuanP74 notified. Seems the pre-post processor need to satisfy a neighbor element face have exactly the same number.
Even both face of element sharing common nodes, still detected as separation since one node is missing (linear element cases)
1 Like
The free nodes could be connected by an *EQUATION to the neighbor element. In cgx it could be done by an experienced user (commands seta, seti, setr (or the equivalent ‘q’ commands), send set abq areampc ).
3 Likes
thanks Klaus for the hints about areampc to generate automatically by CGX, interesting to try.