Henrik Sönnerlind
COMSOL Employee
Please login with a confirmed email address before reporting spam
Posted:
7 months ago
Jul 11, 2024, 12:12 p.m. EDT
Updated:
7 months ago
Jul 12, 2024, 2:29 a.m. EDT
You cannot have nonlinearities in a modal study, so you must use a standard Time Dependent study.
This kind of analysis, where you are forced to run periodic problems in time domain is computationally very expensive. For each frequency, you must time-step many periods until you reach a stabilized solution that is independent of the initial conditions.
You can get some improvement by trying approximate the initial conditions using a linear frequency domain study.
As an inspiration, take a look at this example:
https://www.comsol.com/model/download/1254561/bracket_general_periodic.pdf
https://www.comsol.com/model/download/1254561/bracket_general_periodic.mph
-------------------
Henrik Sönnerlind
COMSOL
You cannot have nonlinearities in a modal study, so you must use a standard Time Dependent study.
This kind of analysis, where you are forced to run periodic problems in time domain is computationally very expensive. For each frequency, you must time-step many periods until you reach a stabilized solution that is independent of the initial conditions.
You can get some improvement by trying approximate the initial conditions using a linear frequency domain study.
As an inspiration, take a look at this example:
Please login with a confirmed email address before reporting spam
Posted:
7 months ago
Jul 15, 2024, 12:11 p.m. EDT
Thank you very much Henrik,
Im not clear about the initial conditions using a linear frequency domain study. Do you mean I need to find the band gaps from freuncy domain and just consider them?
Im also using parameteric sweep to repeat the simulation for couple of frequncies.
Also, Im running simulation for 2D problem using Time dependent and simplify the model as simple as possible for just one unit cell, but after one day it was progressoed for 7%. To capture the contact I refined my mesh as much as possible and reduced time step. I used columb friction and applied a sinosouidal displacement.
Otherwise it cannot capture the contact. I can see that reciprocal of step size convergance graph remained constant and didnt changed. Hence, I guess that simulation is correct but only requires imporvement of solver configuration.
Thank you!
Thank you very much Henrik,
Im not clear about the initial conditions using a linear frequency domain study. Do you mean I need to find the band gaps from freuncy domain and just consider them?
Im also using parameteric sweep to repeat the simulation for couple of frequncies.
Also, Im running simulation for 2D problem using Time dependent and simplify the model as simple as possible for just one unit cell, but after one day it was progressoed for 7%. To capture the contact I refined my mesh as much as possible and reduced time step. I used columb friction and applied a sinosouidal displacement.
Otherwise it cannot capture the contact. I can see that reciprocal of step size convergance graph remained constant and didnt changed. Hence, I guess that simulation is correct but only requires imporvement of solver configuration.
Thank you!
Henrik Sönnerlind
COMSOL Employee
Please login with a confirmed email address before reporting spam
Posted:
7 months ago
Jul 17, 2024, 2:30 a.m. EDT
I just meant that for each frequency in your sweep, you need a new set of initial conditions for the time-dependent analysis.
I would add two study steps inside the parametric sweep over frequencies: One frequency-domain step with a linear version of the model (Friction node suppressed), and one time-dependent step using the results from the first study step as initial conditions.
The success of a such an approach hinges on the assumption that the friction gives a small correction to the response. If not, it is hardly worth the effort.
Note that 'good' initial conditions typically have almost zero displacement, but maximum velocities. If you start from a complete zero solution, then it takes more time to find the stable (nonharmonic) cycle in the time-domain analysis.
A final general remark on nonlinear problems with harmonic excitation: It is not always they have a periodic solution at all. Also problems with friction can exhibit chaotic behavior.
-------------------
Henrik Sönnerlind
COMSOL
I just meant that for each frequency in your sweep, you need a new set of initial conditions for the time-dependent analysis.
I would add two study steps inside the parametric sweep over frequencies: One frequency-domain step with a linear version of the model (*Friction* node suppressed), and one time-dependent step using the results from the first study step as initial conditions.
The success of a such an approach hinges on the assumption that the friction gives a small correction to the response. If not, it is hardly worth the effort.
Note that 'good' initial conditions typically have almost zero displacement, but maximum velocities. If you start from a complete zero solution, then it takes more time to find the stable (nonharmonic) cycle in the time-domain analysis.
A final general remark on nonlinear problems with harmonic excitation: It is not always they have a periodic solution at all. Also problems with friction can exhibit chaotic behavior.
Please login with a confirmed email address before reporting spam
Posted:
7 months ago
Jul 22, 2024, 11:06 p.m. EDT
Updated:
7 months ago
Jul 25, 2024, 3:32 p.m. EDT
Thsnk you very much Henrik,
I think the effect of friction should be significant.
I recelty tried ACOUSITC>> SOLID >> ELASTICWAVE study and the speed of convergance us pretty fast. However, I see discountinuty in my displacement responses from the input and output probe. Is this normal? Can I use this module instead?
Thank you!
Thsnk you very much Henrik,
I think the effect of friction should be significant.
I recelty tried ACOUSITC>> SOLID >> ELASTICWAVE study and the speed of convergance us pretty fast. However, I see discountinuty in my displacement responses from the input and output probe. Is this normal? Can I use this module instead?
Thank you!