Contact control automatic stabilization 1

[AC1999]

How to stabilize an unstable quasi static problem that includes contact. I am modelling a stick slip frictional behavior using FRIC user subroutine. The problem encounters instability while slipping, it has large number nodes taking part in the contact. Later, I have applied * CONTACT CONTROL, STABILISE which takes care of the convergence of this unstable part but as the surfaces comes to sticking after slip, it is expected that now the model is more stable and time increments will increase but it does not happen so. The time increments are still of the order of 1e-5 which makes the simulation take longer time and huge disk space. I have also tried *STATIC, STABILISE, but with several trial and errors I am not able to get the desired solution. In the later case solution obtained is more deviating it does not capture the slip properly and distorts the solution. Anyone with expertise in contact control automatic stabilization kindly help how to increase the time incrementation or convergence in the later part of the simulation.

Regards
Anulekha

 

[FEA way]

If possible, it’s better to avoid stabilization, especially the step level one. It often affects the results (you should always check the stabilization energy because of that). Instead, you could try with dynamic implicit quasi-static step. Also, always enable unsymmetric solver when modeling contact with friction.

 

[AC1999]

Thank you for responding.

 

[IceBreakerSours]

I think things like stabilization, springs, etc. are fine to use in some cases but clever as they might be, in general, they are numerical tricks. I prefer taking inertia into account and avoiding the instabilities altogether. Friction does result in an unsymmetric tangent stiffness matrix so an unsymmetric solver would be worth giving a shot but, since you take a computational hit, the RoI depends on the problem.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[AC1999]

The problem is I have a quasi static fluid diffusion regime before the sudden slip of the surfaces (like nucleation) takes place and after the slip comes to arrest there is again quasi static sticking phase. I have tried defining different steps *soils followed by *dynamic,implicit and then *soils. But in the dynamic implicit step after the sudden slip the kinetic energy becomes very high which gives NAN results in the following step.

 

[IceBreakerSours]

If I understand your description correctly, you have to let the dynamics stabilize/relax before turning off the acceleration abruptly.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[AC1999]

If I do the simulation in one SOILS step the problem statement is this:
The simulation first involves a quasi static regime, where the contact is at rest (stick). The solver takes large time increment size for this part (which I have limited by prescribing 1000 seconds as the maximum increment size). The initial time increment is also 1000 seconds.

Now, when the contact surfaces starts to slip, there occurs dynamic rupture, some chattering in the contact or interaction between the fault happens. Which is taken care by applying *CONTACT CONTROL, STABILIZE. In this phase the increment size gets reduced by the solver to 1e-5 to 1e-3 seconds. (I have prescribed a minimum time increment size for the step as 1e-10 sec).

Now once the rupture is arrested. The contacting surfaces again comes to sticking. The process returns to a quasi static state. I want the solver to take again higher time increment size (want it to go back to 1000 seconds). But it is not happening. It still is stuck to 1e-4 to 1e-3 seconds. It undergoes thousands of increments, which makes the computation time very very large.

Now I also tried to do this with three different steps (like I have written above), soils step for the first quasi static part, dynamic implicit for the rupture part and a static step for the next quasi static part. But this way also it did not work.

 

[AC1999]

667 SEVERE DISCONTINUITIES OCCURRED DURING THIS ITERATION.
667 POINTS CHANGED FROM CLOSED TO OPEN
I think this is causing the problem, after the slip, which makes average forces NaN

 

[IceBreakerSours]

Your method of diagnosis is incorrect. We have to keep the mechanics front and center, make sure the numerical methods approximate the mechanics as best as possible, and identify/isolate where/why/how/.. the numerics are breaking, and then come up with fixes. Your current method will lead you to fixes like minimum time increment size of 1e-10 sec - which almost never works.

It sounds like once the rupture has arrested (whatever that means), the bodies must have some inertia to keep moving. You could plot the displacement/velocities/accelerations for some nodes to be sure. Now, you may be introducing an abrupt deceleration by switching to a different solver which will cause a lot of eigenvalues to get excited and result in small time steps. If that is correct, then you need an intermediate step in which the transients settle down before you switch the solver.

Take that with a pinch of salt because I do not understand the physics you are trying to model.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[FEA way]

What type of contact are you using ? Can you share the keywords defining both contact interaction and contact property ?