Low Friction Force in Abaqus/Explicit 1

[SinaPeugeot]

Hi all,

We’re working on a large model with a complex contact condition; however, we’re getting some unreasonable results. We’ve pinned down the problem to a strange issue with the explicit solver. I’ve made a small model (*.inp files attached; figure shown below) which clearly shows the issue. The model consists of a series of blocks which all have contact (coef. Friction = 1.0) with each other. They are first vertically compressed (from above by ~900kN), and then the upper blocks are pushed sideways to slide over the lower blocks. Each small block is meshed with more than a dozen C3D8R element (mesh is not shown in the figure).

The expected outcome is that the blocks don’t move until the force is ~900kN (i.e., the frictional resistance = mu*N) and then sliding will initiate and the force would be constant at 900kN. This is exactly what happens when the implicit/quasi-static solver is used. However, when the exact same model is re-run using the explicit solver, the frictional resistance reduces by ~50%! Please see the figure below for a comparison of the results. I haven’t been able to identify the reason (I have looked at a slower application of the load; introducing contact damping; etc. but with no luck)

In our larger model, the difference between the theoretical friction force and that found from the explicit solver is even more drastic; but we must use the explicit solver for this analysis.

Would really appreciate any advice.

Thank you.

 

[FEA way]

Did you check all model energies ? It’s very important in analyses with the explicit solver. Especially take a look at ALLVD. Abaqus/Explicit adds some contact damping by default and it may cause various issues with the accuracy of results.

 

[IceBreakerSours]

Not sure if this is going to be helpful but I am taking a step back here.

The fact that the implicit prediction is matching the expectation does not imply the model is appropriately set-up to solve for the physics of interest. The physics involved with a friction coefficient of 1.0 is tricky and it gets tricky very quickly if compliant materials are involved. I could make the counter-argument: Explicit is showing stick-slip behavior whereas implicit isn't and I would expect to see some stick-slip phenomenon at a high friction coefficient. So, some questions to consider: Have you characterized the friction coefficient between the two materials with rigor? Can you assume the frictional behavior is Coulomb-like or is it intricate? What does the experimental force-displacement measurement look like? Are you confident about the material properties of the two bodies?

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

 

[SinaPeugeot]

Thanks IceBreakerSours for your reply. Please note that this is a simple model setup to check whether the Coulomb friction model in Abaqus provides reasonable results. The exact same behaviour is observed even with a Mu=0.3. We're not attempting to capture any particular experiments with this simple model - the only intention was the ensure that the Coulomb model is working as it should be.

Hope this clarifies.

 

[SinaPeugeot]

Hi FEA way and thanks for your reply. Yes, I've been monitoring the energy balance and have noticed quite high ALLVD. I tried to reduce the Bulk Viscosity - by doing so ALLVD disappeared, however, the results got even worse. What I could see from the model is that, in the explicit analysis, once the vertical force is applied, the blocks start to 'wobble' and move around, and I suspect this adversely affect the frictional resistance at the interface. Reducing the Bulk Viscosity makes this effect more pronounced and causes the blocks to move even more prior to the start of Step 2. Note that I tried to increase the loading duration, however, you'd need a very long analysis to make sure that the blocks don't move around when the vertical force is applied (which is not feasible for our main model).

So, in essence, this might be caused by the fact that in this particular model we have a lot of parts which are difficult to keep in place under a compressive force in an explicit analysis. Nevertheless, I'd imagine that there should be a way to overcome this as this is quite a basic model. Any suggestions?

 

[FEA way]

Instead of reducing bulk viscosity, try significantly reducing or even disabling tangential contact damping:

*CONTACT DAMPING, TANGENT FRACTION=…

 

[SinaPeugeot]

Thanks FEA way. After quite a few runs with different settings, I've started to better understand how the model behaves. As you correctly pointed out, the energy balance helps a lot in determining whether the results are reliable or not. In doing so I've noticed two issues which I'd like to receive your advice on if possible:

1. In addition to the model above, I've prepared other similar models with different number of blocks etc. I've noticed that if I manage to keep ALLVD reasonably low, the friction results are generally accurate. However, in some models, removing bulk viscosity does not control ALLVD. In those cases, I believe the contact damping should also be reduced to control ALLVD. My question is, is ALLVD only related to bulk viscosity and contact damping? Are there any other contributors to ALLVD?

2. Another energy component that the manual briefly mentions that needs to be checked is ALLPW (contact penalty work). However, the problem with ALLPW is that it's not clear (at least to me) what influences it. In other words, in some models where ALLPW is high, I'm unable to find what could cause it and how it could be reduced. Just wondering if you have any advice on this as well?

Thanks again for your helpful comments.

 

[FEA way]

ALLVD stands for viscous dissipation energy. So basically everything associated with damping but not with stabilization. It includes contributions from bulk viscosity, contact damping, material damping, viscous regularization in fracture mechanics analyses and so on. You can also request viscous forces output to see which regions are particularly affected by those effects.

ALLPW is work done by contact penalties. Just like ALLCW, it should be close to zero. Otherwise, there might be some overconstraints, for example.

 

[SinaPeugeot]

Thank you FEA way. I'm still having problems containing ALLPW - It is not clear what causes high ALLPW in some models... Two questions from you:

1. You mentioned viscous forces could be output. I wonder could you somehow show interface damping forces as well? I'm trying to isolate the resistance due to interface damping from the pure frictional resistance.
2. As an example, you noted overconstraining could cause high ALLPW. However, I don't have any overconstraints in my model - they are just blocks sitting on each other. I haven't been able to find any other guidance regarding ALLPW in the manual or online. Just wondering if you might have any suggestions.

Many thanks again for your time.

 

[FEA way]

You can request the CDSTRESS (CDPRESS and CDSHEAR) output variable. It can be compared with regular contact stresses.

High ALLPW might be also related to incorrect contact definition (main-secondary assignment and use of node-based surfaces with surface-to-surface discretization for instance). You could also run Abaqus/Standard analysis to make sure that there are no overconstraints that can have various causes.

 

[SinaPeugeot]

Sorry FEA way, but my understanding is that CDSTRESS cannot be output in Explicit. Am I missing something here?

Thanks.

 

[FEA way]

This would apply to that additional Abaqus/Standard analysis that you can perform to check for overconstraints and also compare the results obtained with those two solvers. For Abaqus/Explicit you can use ELVD output variable. It’s not related to contact but may show some regions with largest contribution to ALLVD.

 

[SinaPeugeot]

Thanks again FEA way for all your help. Very much appreciated.

 

[3DDave]

Are the blocks infinitely rigid?

 

[SinaPeugeot]

Hi Dave, no, they are elastic with a reasonable stiffness.

 

[3DDave]

In that case the blocks nearest the force application break loose first, compressing against the friction of the remaining blocks. This allows dynamic friction to develop, lowering the local friction. This break-away progresses until the last block breaks loose, so there isn't as much force applied and the nearest blocks stop and stick again, allowing all the rest to stop and stick again.

That's what I would expect of a "real" part. Not sure if that is what is happening in the simulation, but the repeated ripples are what I would expect to see.

Normally one sees this in moving a chair where the legs take turns sticking and slipping and it makes a low-pitched sound.

 

[SinaPeugeot]

Thanks Dave and I think you're partly right. I also thought of this behaviour; so, I did two tests:
1. Applied the displacement more slowly
2. Applied the displacement uniformly to all the top blocks at the same time

In both case, the force was still notably lower than the implicit analysis. I still think what FEA way is suggesting regarding the energies is the main reason, however, fully resolving it in more complex models with a lot of blocks is still challenging.

 

[3DDave]

Diagnostics wise, I would try a variety of contact friction values, say .5 , .75, 1, 1.25. and 1.5 to graph the results and see if there is a trend or some instability.

 

[SinaPeugeot]

Hi Dave, this was done previously and all will yield the same results; i.e. if no extra effort is put in to resolve the issue, explicit will by default result in a much lower friction resistance than implicit (in this particular model I mean).