Dear colleagues,
The short version of my question is: At what circumstance can an Abaqus simulation yield kinetic energy far greater than all other energy components? I can understand kinetic energy increase due to external work or a release of strain energy. But in those cases, the kinetic energy that results will be a fraction of the external work or strain energy. However, I am seeing kinetic energy far greater than external work or strain energy and cannot fathom how this can be possible.
Details for those that are interested:
I am trying to run a static simulation involving contact. As I encounter convergence issue, I also attempted implicit dynamic model with quasi-static setting. However, the model still fail to converge. During investigation, I found out that at the abort of the simulation, there is a kinetic energy (ALLKE) spike several orders of magnitude larger than most other energy components, taking up 99% of the total energy (ETOTAL).
This is quite surprising as right before this spike the energy breakdown in this model was reasonable. With internal energy (ALLIE), contact friction dissipation (ALLFD) and contact elastic strain energy being the major component of total energy, balancing with the total external work (ALLWK). Thus, I am trying to look into the mystery behind the spurious kinetic energy.
I understand I have much to learn about the software and the numerical methods and would appreciate it if someone can name a few key terms that I can look up or direct me to some related documentations or examples.
Much thanks in advance.
Hi Yunpei Yang,
It is not strange that unsuccessful attempt at current increment causes unbalanced energies. Much more important question: what is the nature of your convergence issue - only numerical or physical?
most likely the spike in ke is due to intermittent impact calculation (e.g. inc =1e-6*previous inc size).... meaning contact conditions are changing...try increasing loading duration (change initial inc, max inc, n total time = 10*previously used values, update amp as well)
btw what was the reason behind convergence issues with static procedure???
Hi Victor J. Аdlucky ,
Thanks for the suggestion. After some more testing with static, looks like you are right. In some cases there's a static stabilization energy spike (which likely relates to the kinetic energy spike if the simulation is dynamic). In some other cases, the energy spike is not there but the simulation would still abort. In the cases without the stabilization energy spike, I may encounter a sudden drop in some of the energy components. The only difference in these different tests were changes in time increment sizes.
Given the inconsistency of the issue (sometime an energy spike and some other time an energy drop), and how sudden these spikes are. I am inclined to believe that my model has an issue that is more numerical in nature. However, I do suspect that the numerical "instability" is also related to the physics of the model. After all, the model always fail at a specific stage of deformation.
I am currently working through a list of things that may help with this issue. Looks like using dynamic instead of static as well as using stabilization will not suffice. The next I will probably try is to use explicit. I am a bit skeptic though because I feel it is still quite possible for explicit to be affected by this numerical issue. If you can think of other things I can try please let me know.
Thank you
My advice is to try contact damping: https://abaqus-docs.mit.edu/2017/English/SIMACAEITNRefMap/simaitn-c-contactdamping.htm
Also I recommend the next book:
Hi Yunpei Yang.
I suggest you check the mesh quality particularly the Jacobian ratio values. Make sure it is more than 0.3. Hope this helps.
Hi Mir Aamir Abbas ,
Thanks for the suggestion. All elements that belongs to deformable geometries are square shaped and does not distort heavily during simulation.
Though not the issue in this case, I agree it is a recommended practice in almost all FEM applications.
Hi Nils Wagner ,
Thank you very much for the offer. Unfortunately my model involves a UGENS subroutine which I do not have the liberty to share at the moment. However, if I manage to recreate a similar case using Abaqus built in material models I will let you know.
- Badges
- Science topic