Changing Total Time changes the result?

[sanju261991]

I am simulating a ball bouncing using Abaqus/Explicit. I have noticed that if the total time of the simulation is changed theresults also change, which I think does not make sense as the whole setup is the same except the total time.

Here is the plot of Kinetic Energy vs Time for 3 different total times.

After I zoomin, I would expect the plots to overlap, but I find that the plot do not overlap.

I am using automatic timesteping for the simulation. Also the 'Initial time increment' is the same for all the cases.

Can anyone explain why this happens?

Thanks

 

[gravityandinertia]

I'm going to assume it's related to the automatic time stepping. Whatever it is using for the 0.0002 and 0.0006 values is probably a similar time step. The 0.0004 maybe somehow triggering a different automatic time step. The thing to always remember about FEA is that almost every parameter (time, mesh size, force distribution, force vs. time application, stress-strain curve) in real life is a continuous system, yet we are representing them with a number of discrete points, therefore every parameter must be tested for convergence to find when we have a value small enough to represent the "infinitesimal" well.

This convergence process makes non-linear problems and dynamics problems much more time consuming than your basic linear static analysis, but it's work that needs to be done nonetheless if you want accurate results. It's usually when this process doesn't take place that you start to get people that say FEA can't accurately predict real world values. In nearly all cases it can, however that doesn't mean that it is always worth the effort. Sometimes doing so means spending more money and time than a physical test would take.

I got a little off topic there, but hopefully you understand what I'm saying.

 

[IRstuff]

Time steps are often based on the total time of the simulation. There's an implicit assumption that you wouldn't want to start a 1 hr simulation with picosecond time steps.

TTFN
I can do absolutely anything. I'm an expert!
faq731-376 forum1529

 

[Mustaine3]

I'm quite sure, that the issue is related to output. The output times are not equal and with that high frequent behavior the curves do not match. When you apply a filter the get rid of the noise, the issue would disappear.

 

[sanju261991]

Thank you all for your replies.

Okay, I understand that the automatic time step might result to different time increments for different total times based on various factors with the ultimate aim to reach convergence.

Do a similar optimization also take place when a user-defined time increment is used? I guessed in this case a constant time increment would be used throughout the simulation. I tried it with fixed time increments and got a similar graph.

Or can I just assume it as some numreical error?


@Mustaine3 -- I am a little unclear about your point. How do I apply the filter? Can I apply it using Abaqus post processing? or using python, MATLAB?

 

[IRstuff]

There's a concept of quasi-linearization in solving certain systems of equations. Since a bouncing ball is a non-linear system, one can imagine that smaller timesteps would result in a better piecewise linear solution.

TTFN
I can do absolutely anything. I'm an expert!
faq731-376 forum1529

 

[Mustaine3]

@sanju
Numerical errors are always there, but usually the error/difference is quite small. So it depends on the scale you are looking at. Very small differences can also come from different versions, OS, domain decomposition, ...

Filters are available during the generation of output and also in postprocessing, when you operate on xy data.


@IRstuff
I think I know what you mean, but the step time should have no influence in the stable time increment.
At longer runtime the round off errors might become an issue. Running the job in double precision can help here.

 

[IceBreakerSours]

.. automatic time step might result to different time increments for different total times based on various factors with the ultimate aim to reach convergence.

There are no convergence checks in Abaqus/Explicit.

*********************************************************
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