Load steps and contact convergence

[JonathanHodgson]

Hi all,

I'm currently attempting a fairly meaty (read: probably overly-ambitious) contact analysis, and trying to make it run as efficiently as possible!

In particular, I'm wondering whether adding some user-defined load steps ('time steps') could help convergence by allowing the model to 'settle' quickly at, say, 0.1 or 0.01 of full load, and thus solve more easily when the full load is subsequently applied; or is it quickest to just leave it to its own devices, going straight to load step 1?

Has anyone run any tests? Does anyone know whether the convergence from a previous step is carried forwards to the next, or indeed whether it helps if so?

Thanks!

[rrabe]

Hello

I had mixed experiences with defined steps and contact (not only). Especially the cases that you see that Simulate starts to cut steps up to ridiculous increments until finally fails.

In some cases defining more steps can even help you to finish a simulation or even make it quicker - in opposite to the case explained above.  But there is no general rule. It depend on the model.

Anyway I don´t Simulate goes directly to the final step. Would be very hard to obtain convergence. 

Other trick is to bypass a step that you see that is causing problems. Ex. step 0.2 and 0.4 goes fast but 0.3 is a lot slower (for no reason). So you let it out.

When working with contact I notice that the "light" initial contacts takes much more time than when the system os more preloaded. So I start already with a larger step and not a light one.

Finally - but I have not much experience - is to increase the number of defined steps but let it all in automatic. I used this for shock analysis and not for contact.

Regarding if the number of steps has influence?  I would say yes - but sometimes this is not noticeable and you will need only create more simulation time.

A famous work around due to contact bug in Simulate 3 (not sure if was 3 the 1st version with finite friction) was to use a huge amount of steps to have the right result. 

[JonathanHodgson]

Thanks for the comments.

I have in the past seen Mechanica creating intermediate loadsteps to help convergence; it hasn't been reporting that this time (so I can only assume that it goes straight to load increment 1.0 after solving 0.0).

I've now tried a number of different runs, although none were like-for-like and I've only tried very light initial loads, which you say can often take longer to converge. Nonetheless, the evidence is that there is no benefit (on a run which otherwise works) in adding intermediate steps - it's quicker to leave it on Automatic and let it go straight to the answer.

Cheers,

Jonathan