Hi everyone, I am trying to model the formation of shear band in sand specimen in ABAQUS/Standard. I used a user defined work hardening-softening model for sand. But the global equilibrium cannot be obtained after peak, so the solution diverges.
On the other hand, if I changed the consistent tangent matrix (its determinant seems to be negative when the solution diverges) to elastic matrix, the computation will complete, but I get a homogeneous deformation, thus no shear band occurs.
While in ABAQUS/Explicit, the computation will also complete, and a shear band will occur. But the stress-time curve in the softening regime will oscillate severely. And it is unacceptable in most times.
"But the global equilibrium cannot be obtained after peak"
This is expected if you have softening - d sigma/ d epsilon has to be negative in this case.
You may try to change to displacement control instead of load control, this usually works even with instabilities.
Hi Dr. Baeker, in my simulation I am using the displacement control. I applied nodal displacement at the nodes of top boundary to compress the soil specimen. So maybe this is not the reason for this problem... But it is true that d sigma/ d epsilon become negative in softening stage.
If you use displacement-control, there should be no problem at all beyond th epeak point since stresses can decrease in the structure. So the problem may be either with your material specification itself or with the model. Perhaps you can see where the problems occur - is it a convergence issue inside the UMAT (do you use a UMAT?) or of the external Newton iteration?
Yes I use a UMAT implemented by closest point projection method. The simulation is aborted due to nonconvergence of external Newton iteration, "too many attemps made for this increment'. The global equilibrium cannot be obtained for even extremely small increment.
It seems that the problem is happening like this. At softening stage, at first local convergence is obtained, but global equilirium is not. So another external iteration is performed by ABAQUS. While afer some iterations, the predicted strain will become extremely large for some elements, which resulted in the non-convergence of local problem (inside UMAT).
But it is strange that by using Elastic matrix as DDSDDE in ABAQUS, a whole stress-time curve can be obtained, although no shear banding occurs. See the Figs.
Is it possible that your model becomes inherently unstable at this point? Carefully check the damage and stress distribution in the last converged increment to find out what is happening in the model.
Hi Dr. Baeker, thank you for your patience! The strain and stress distribution is shown in Figures. It is happening like this: The stress and strain are the same for all elements before time=22.5s. Then, minor differences start to occur. And you can tell from the curves that the differences are becoming large. But the simulation diverged after only four or five increments from then on at time=27.5s.
These differences among elements are actually very small. And I have no idea why this resulted in the non-convergence. Could you help me check the figures and see if you can find out something? Thank you so much!
- Badges
- Science topic