Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Impact of 2 spheres

Please login with a confirmed email address before reporting spam

I am attempting to model the impact of two spheres in the structural mechanics module. I understand that COMSOL has trouble with contact pairs when performing a transient analysis, and this is the trouble I am having. I attached my file for your consideration. Please keep in mind that I am a novice user.

Any help would be greatly appreciated!


10 Replies Last Post Aug 9, 2013, 5:15 p.m. EDT
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 14, 2012, 12:01 p.m. EDT
Hi

I would not say it's COMSOL that has any major issue with such a model, it's rather that the set-up is delicate, It should work by selecting correct mesh densities and extent of the contact regions, and enough BC to enforce an unique solution, which is not trivial for "rigid" body systems. Furtermore, in transient you have all the resonance modes and pressure waves that are excited, one way to get a better convergence is to remove the "inertial terms" from the solid physics, but then the results are not fully representative. Or to add some damping to minimise unwanted modes.

Unfortunately I do not have 3.5 up running, and translated to 4.3 your model needs some tweaking as entity ID seem to be different

--
Good luck
Ivar
Hi I would not say it's COMSOL that has any major issue with such a model, it's rather that the set-up is delicate, It should work by selecting correct mesh densities and extent of the contact regions, and enough BC to enforce an unique solution, which is not trivial for "rigid" body systems. Furtermore, in transient you have all the resonance modes and pressure waves that are excited, one way to get a better convergence is to remove the "inertial terms" from the solid physics, but then the results are not fully representative. Or to add some damping to minimise unwanted modes. Unfortunately I do not have 3.5 up running, and translated to 4.3 your model needs some tweaking as entity ID seem to be different -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 18, 2012, 7:48 p.m. EDT
Thanks for the fast reply! I would have replied sooner, but I was trying to figure it out on my own.

My model was failing to find an initial value. I discovered that the problem is contained in the quantity, "pn" which is the contact normal penalty factor. Specifically the variable "auglagiter." If I replace auglagiter with a constant, the model will solve. (It will solve until initial contact, then it fails, but I will get to that...) If I replace the entire contact normal penalty factor with a constant value it will solve, again up until initial contact. It gives me the following error: "Nonlinear solver did not converge. Last time step not converged." So I have made progress, but have encountered another problem. I also noticed that if I decrease the penalty factor, say to 0.001, the boundaries will penetrate one another and the model will solve.

My questions are the following:
What is auglagiter, and why is it preventing the model to find an initial value?
What is the contact normal penalty factor?
How do I adjust the penalty factor in order make the model solve without the boundaries penetrating?

Your help is much appreciated!

Ryan
Thanks for the fast reply! I would have replied sooner, but I was trying to figure it out on my own. My model was failing to find an initial value. I discovered that the problem is contained in the quantity, "pn" which is the contact normal penalty factor. Specifically the variable "auglagiter." If I replace auglagiter with a constant, the model will solve. (It will solve until initial contact, then it fails, but I will get to that...) If I replace the entire contact normal penalty factor with a constant value it will solve, again up until initial contact. It gives me the following error: "Nonlinear solver did not converge. Last time step not converged." So I have made progress, but have encountered another problem. I also noticed that if I decrease the penalty factor, say to 0.001, the boundaries will penetrate one another and the model will solve. My questions are the following: What is auglagiter, and why is it preventing the model to find an initial value? What is the contact normal penalty factor? How do I adjust the penalty factor in order make the model solve without the boundaries penetrating? Your help is much appreciated! Ryan

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 19, 2012, 1:46 a.m. EDT
Hi

reread the doc, and check the FORUM, a few years ago there were many discussions about this, including variants of settings to improve contact simulations, I do believe there is a couple of article on the KB (Knowledge Base main COMSOL site)

--
Good luck
Ivar
Hi reread the doc, and check the FORUM, a few years ago there were many discussions about this, including variants of settings to improve contact simulations, I do believe there is a couple of article on the KB (Knowledge Base main COMSOL site) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 28, 2012, 1:47 p.m. EDT
I was able to solve this problem. I went into the solver manager and selected Solve For u. Everything was selected by default. There are a number of variables Tn_cpx_smps, where x corresponds to each contact pair. With these selected the model would not run. With only u selected, the model runs and solves. Can someone explain what is happening and why this is the case?

Thanks!

Ryan
I was able to solve this problem. I went into the solver manager and selected Solve For u. Everything was selected by default. There are a number of variables Tn_cpx_smps, where x corresponds to each contact pair. With these selected the model would not run. With only u selected, the model runs and solves. Can someone explain what is happening and why this is the case? Thanks! Ryan

Nagi Elabbasi Facebook Reality Labs

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Jul 28, 2012, 11:18 p.m. EDT
When you do not solve for the contact variables COMSOL solves the contact problem using a pure penalty formulation (instead of Augmented Lagrange). That seems to be easier to converge for your model.

Nagi Elabbasi
Veryst Engineering
When you do not solve for the contact variables COMSOL solves the contact problem using a pure penalty formulation (instead of Augmented Lagrange). That seems to be easier to converge for your model. Nagi Elabbasi Veryst Engineering

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 11, 2012, 10:24 p.m. EDT
If I was interested in solving for a contact variable, (contact pressure) how could I achieve convergence?

Thanks!

Ryan Musson
If I was interested in solving for a contact variable, (contact pressure) how could I achieve convergence? Thanks! Ryan Musson

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 7, 2013, 3:27 p.m. EDT
Hi!

I am running version 4.3b, trying to solve a transient model with contact, but I'm getting penetration. Would anyone know how to implement pure penalty formulation? As I saw on Nagi's post, this might solve the penetration problem.

Thanks!
Hi! I am running version 4.3b, trying to solve a transient model with contact, but I'm getting penetration. Would anyone know how to implement pure penalty formulation? As I saw on Nagi's post, this might solve the penetration problem. Thanks!

Nagi Elabbasi Facebook Reality Labs

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 7, 2013, 3:55 p.m. EDT
Hi Sylvana,

Check out the “Contact Analysis of a Snap Hook Using a Penalty Formulation” COMSOL Gallery model (#12577). The document supporting the model shows how to implement the penalty contact method (it’s quite straightforward actually). I believe the document also gives the pros and cons of the penalty method compared to the Augmented Lagrangian method.

Nagi Elabbasi
Veryst Engineering
Hi Sylvana, Check out the “Contact Analysis of a Snap Hook Using a Penalty Formulation” COMSOL Gallery model (#12577). The document supporting the model shows how to implement the penalty contact method (it’s quite straightforward actually). I believe the document also gives the pros and cons of the penalty method compared to the Augmented Lagrangian method. Nagi Elabbasi Veryst Engineering

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 8, 2013, 4:23 p.m. EDT
Hi Nagi,

Thank you very much for your reply. I really appreciate it. I took a look at the model you suggested and implemented it on a stationary (parametric) study, with linear elastic materials. However, I would like to use the viscoplasticity material model, which I haven't been able to use with a parametric study, only in a time-dependent study. Would you know how to implement the penalty formulation on a transient model? Or, would you have any suggestions on implementing a parametric study using a viscoplastic material? I get an error regarding the effective creep strain variable (solid.ece):
"There are xx void equations for the variable mod1.solid.ece..."

Thanks again!
Sylvana
Hi Nagi, Thank you very much for your reply. I really appreciate it. I took a look at the model you suggested and implemented it on a stationary (parametric) study, with linear elastic materials. However, I would like to use the viscoplasticity material model, which I haven't been able to use with a parametric study, only in a time-dependent study. Would you know how to implement the penalty formulation on a transient model? Or, would you have any suggestions on implementing a parametric study using a viscoplastic material? I get an error regarding the effective creep strain variable (solid.ece): "There are xx void equations for the variable mod1.solid.ece..." Thanks again! Sylvana

Nagi Elabbasi Facebook Reality Labs

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago Aug 9, 2013, 5:15 p.m. EDT
Hi Sylvana,

Penalty contact as shown in the Model example is the same for both stationary and transient. In transient problems you frequently need a lower penalty factor than in stationary problems to avoid impact oscillations. You cannot use a viscoelastic or viscoplastic material in COMSOL in a stationary analysis (even if you are doing a parametric sweep) since these materials need a time derivative.

Nagi Elabbasi
Veryst Engineering
Hi Sylvana, Penalty contact as shown in the Model example is the same for both stationary and transient. In transient problems you frequently need a lower penalty factor than in stationary problems to avoid impact oscillations. You cannot use a viscoelastic or viscoplastic material in COMSOL in a stationary analysis (even if you are doing a parametric sweep) since these materials need a time derivative. Nagi Elabbasi Veryst Engineering

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.