OpenRadioss does not converge with increasing spatial resolution for a cantilever beam simulation about openradioss

[qiangRadioss]

Hello Fonotec,

thanks for this interesting studies.
I guess the problem is that finer meshes excite higher frequency modes and need to be stabilized more numerically.
I think increasing dn (in /PROP/SHELL) with finer mesh could improve the convergence issues, you could take a trying.
Anyway, the default value of dn has not been designed for very finer mesh, i think it's even questionable for the modeling with shell element with high ratio of thickness/mesh_size (e.g. >10).
B.T.W. some of you rad files use npt=1 (at least in the file of 0.625mm) instead of npt=5 (or 0), npt=1 will have only membrane stiffness, i guess that's just a typo error.

hope this helps.

Qiang.

[Fonotec]

Hello Qiang,

Thanks for your reply! I will follow your suggestion and try slightly higher values of a factor 2, 4 and 8 for the shell numerical damping (dn). I guess this will indeed improve the numerical stability but I think this will not let the solution converge to the correct solution. I think it will cause the amplitude to be damped more, so the solution for different resolutions will not be converged, they will diverge a bit more.

You have a valid point that the thickness/mesh_size might be a bit questionable for modeling with shell elements at the highest resolutions. Parallel to this I will investigate whether this indeed is causing problems. For this I propose that I will do a similar set of cantilever beam simulations over 2^6 orders of resolution but for which the thickness/mesh_size does not exceed a factor of a few at the highest resolution.

Yes, indeed npt=1 is a typo in some rad files that was a parameter I changed in an other set of tests that I was performing.

I will come back to you once I performed the tests.

Regards,
Folkert

[Fonotec]

Hello Qiang,

I tried to use a larger dn to improve the convergence, I increased dn for my highest resolution simulation and find that this does not seem to influence the solution visibly. I tried besides the fiducial value of 0.015 (for QEPH shells) 3 values that are 2, 4 and 8 times larger. In the figure below you can see that the result is (almost) identical:

cantileverbeamtest_100percent.pdf

I double checked that the parameter indeed is read in correctly, I can see in the .out file that the value of dn is set to the value that I want.

Regards,
Folkert

[qiangRadioss]

Hi Folkert,

I've tested the case of 0.3125mm, i got the following :

the increasing factor is about 3.33, there is still the losing of amplitude, but it does be improved.
the ratio thickness/mesh_size of this case is 8, and for your finest mesh case, the ratio is 16.

Regards

Qiang.

[Fonotec]

Hi Qiang,

I agree with what you are finding, this indeed reduces the decay in the amplitude also for me for this simulation.

I performed a similar simulation such that the thickness is always smaller than the width of the shell elements and in this case I do not seem to get any decay. So I think that the main reason indeed is that the shell element becomes too thick. See the figure below, which is a cantiliver beam of 4000 mm x 1000 mm x 8 mm at different resolution.

cantileverbeamtest_100percent.pdf

Overall, I think this implies that the shell should be thin enough for the standard OpenRadioss values to work. For me this is a satisfactory conclusion and for me the issue could be closed.

Regards,
Folkert

[qiangRadioss]

close the issue as mentioned Folkert in his last comment.