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.
New user, New software, New errors (mechanical problem)
Posted 24 juin 2016, 08:37 UTC−4 Modeling Tools & Definitions, Parameters, Variables, & Functions, Studies & Solvers, Structural Mechanics Version 5.2 2 Replies
Please login with a confirmed email address before reporting spam
Hello,
I have started using comsol two weeks ago for my research and I'm a little bit disappointed. I had a first contact with this software during my studies and i remember a taste of "it never converge !".
So i usually use ANSYS but i thought COMSOL will be better for a young researcher because of the more "mathematical" environment and all the manual things you have to do in this software (control the computation, optimize the mesh at strategical position, define the boundary contact, only use coherent physical parameters...) that ANSYS could do for you, and so you don't really know what you are really doing.
To learn i do a solid mechanics 2D model to define contact between aluminium and steel, I do my own model based on the comsol template "cylinder roller contact" but i just converge the computation one time for a couple of parameters. It doesn't converge if i change values of parameters like the pressure, the dimension or even the mesh ... i can't verify if my model is just coherent with the physic. So i come back to the template and it's the same, the model converges only for a couple of values, if you change something it doesn't works. They also had to define a good initial contact pressure to make the model stable and they finally verified the answer values with analytic because it's a really simple case.
So why this model is so fragile ? if you don't use symmetry, change geometry, change dimension, increase/decrease the load, change the mesh ... it doesn't converge. THERE IS NO INTEREST. The error looks basic: Failed to find a solution.
Maximum number of segregated iterations reached.
In Segregated Step:
There was an error message from the linear solver.
The relative residual (0.002) is greater than the relative tolerance.
Returned solution is not converged.
- Feature: Stationary Solver 1 (sol1/s1)
So why this model is so fragile ?
My ideas are:
- contact is really hard to compute,
- comsol has not enough robust solver for this,
- i miss something,
Could you help me on this please ? Be indulgent it's my first model.
I have started using comsol two weeks ago for my research and I'm a little bit disappointed. I had a first contact with this software during my studies and i remember a taste of "it never converge !".
So i usually use ANSYS but i thought COMSOL will be better for a young researcher because of the more "mathematical" environment and all the manual things you have to do in this software (control the computation, optimize the mesh at strategical position, define the boundary contact, only use coherent physical parameters...) that ANSYS could do for you, and so you don't really know what you are really doing.
To learn i do a solid mechanics 2D model to define contact between aluminium and steel, I do my own model based on the comsol template "cylinder roller contact" but i just converge the computation one time for a couple of parameters. It doesn't converge if i change values of parameters like the pressure, the dimension or even the mesh ... i can't verify if my model is just coherent with the physic. So i come back to the template and it's the same, the model converges only for a couple of values, if you change something it doesn't works. They also had to define a good initial contact pressure to make the model stable and they finally verified the answer values with analytic because it's a really simple case.
So why this model is so fragile ? if you don't use symmetry, change geometry, change dimension, increase/decrease the load, change the mesh ... it doesn't converge. THERE IS NO INTEREST. The error looks basic: Failed to find a solution.
Maximum number of segregated iterations reached.
In Segregated Step:
There was an error message from the linear solver.
The relative residual (0.002) is greater than the relative tolerance.
Returned solution is not converged.
- Feature: Stationary Solver 1 (sol1/s1)
So why this model is so fragile ?
My ideas are:
- contact is really hard to compute,
- comsol has not enough robust solver for this,
- i miss something,
Could you help me on this please ? Be indulgent it's my first model.
2 Replies Last Post 27 juin 2016, 06:56 UTC−4
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Hi
Indeed when you are used to ANSYS you come from the engineering world, here with COMSOL you must understand the physics too, it's a bit trickier but at least you know the underlying hypothesis used, something from my experience is far from true with older software.
But this means also we end users must do a sever mind (re)set. I'm an "old" product developer 35+ years of professional experience, and I still prefer COMSOL by far from the older tools I used before, so sorry I do not follow you there.
Now if you have a classical engineering job to be done rapidly, and you are not searching new exotic features, then often the older tools are quicker, within one or at most a few physics (this my personal opinion and I'm not linked to COMSOL, I'm a paying user as you).
Now to my suggestions for your issue:
1) contact models, the true physics that COMSOL tends to solve, are very complex and fragile (i.e. numerical instabilities due to the limited representation of real numbers in binary). Most older software adds "a little bit of clever contact softening" and get around like that, the thing is that depending on the size and scale of your model, this "little" bit of softening has severe impacts, you as end users have no idea of what is truly going on! Such adapted model looks great, but the numbers are far from representative! Once you try to validate them.
2) there are many things to respect for contact problems like correct "source" and "destination" definitions and respective curvature and mesh densities, the latest COMSOL versions starts with physics controlled meshing far better now.
3) offset values can change the convergence and the scaling of the pressure value is essential
4) discretisation settings of the physics: mostly linear is suggested (may have changed in v5.2a)
5)...
6) if the standard Augmented Lagrangian is not working, have you tried the penalty method ?
Read carefully the doc about contact, it's tricky to do it correctly, but fully feasible.
If you go to the conferences or the training courses, you will see in the slides presented for contact problems quite a long list of criteria to check and adapt. In fact for me true contact problems are among the most complex to set up, as there are very many combinations that might appear to have an automatic set up. But mostly I manage to get my contact problems running, so it should work for you too.
--
Good luck
Ivar
Indeed when you are used to ANSYS you come from the engineering world, here with COMSOL you must understand the physics too, it's a bit trickier but at least you know the underlying hypothesis used, something from my experience is far from true with older software.
But this means also we end users must do a sever mind (re)set. I'm an "old" product developer 35+ years of professional experience, and I still prefer COMSOL by far from the older tools I used before, so sorry I do not follow you there.
Now if you have a classical engineering job to be done rapidly, and you are not searching new exotic features, then often the older tools are quicker, within one or at most a few physics (this my personal opinion and I'm not linked to COMSOL, I'm a paying user as you).
Now to my suggestions for your issue:
1) contact models, the true physics that COMSOL tends to solve, are very complex and fragile (i.e. numerical instabilities due to the limited representation of real numbers in binary). Most older software adds "a little bit of clever contact softening" and get around like that, the thing is that depending on the size and scale of your model, this "little" bit of softening has severe impacts, you as end users have no idea of what is truly going on! Such adapted model looks great, but the numbers are far from representative! Once you try to validate them.
2) there are many things to respect for contact problems like correct "source" and "destination" definitions and respective curvature and mesh densities, the latest COMSOL versions starts with physics controlled meshing far better now.
3) offset values can change the convergence and the scaling of the pressure value is essential
4) discretisation settings of the physics: mostly linear is suggested (may have changed in v5.2a)
5)...
6) if the standard Augmented Lagrangian is not working, have you tried the penalty method ?
Read carefully the doc about contact, it's tricky to do it correctly, but fully feasible.
If you go to the conferences or the training courses, you will see in the slides presented for contact problems quite a long list of criteria to check and adapt. In fact for me true contact problems are among the most complex to set up, as there are very many combinations that might appear to have an automatic set up. But mostly I manage to get my contact problems running, so it should work for you too.
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Thanks a lot for your answer, it’s really helpful. I will try the penalty method and check if my model respect the needs you showed me.
I said before that in ANSYS i don’t really know what i do and it’s one of the reason i come to COMSOL, but what you say about contact treatment in this “engineer” software is really frightening. One of my previous work was to show the constraint in a wheel section with all the contact between caoutchouc, steel rim, fiber reinforcement … These contact was automatically defined by the software so i found that really easy but maybe it wasn’t physically correct.
If i succeed my contact model in COMSOL i will confront the results with the same model but done in ANSYS.
Thanks,
Nicolas
I said before that in ANSYS i don’t really know what i do and it’s one of the reason i come to COMSOL, but what you say about contact treatment in this “engineer” software is really frightening. One of my previous work was to show the constraint in a wheel section with all the contact between caoutchouc, steel rim, fiber reinforcement … These contact was automatically defined by the software so i found that really easy but maybe it wasn’t physically correct.
If i succeed my contact model in COMSOL i will confront the results with the same model but done in ANSYS.
Thanks,
Nicolas