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.

Small time step & slow solving process

Please login with a confirmed email address before reporting spam

Hi
I'm building a 2D, time dependent model relating to Solid Mechanics/Heat Transfer in Solids/Coefficient Form PDE. They are coupled by the thermal expansion and swelling of the solid. The creep deformation of the solid is also included. I've generated a mesh of ~4000 elements and of 450 000 DOF. The default solver is used and is running on a computer of 64 cores and 96G RAM.
However, the time step taken by the solver is about 1/1 000 of the total time range and the time cost is huge on the solving. I've tried ways of :
1, BDF method while setting the steps taken by the solver as STRICT, and the initial step is given manually.
2, Segregated instead of Fully coupled
3, refining the mesh to about 8000 element and 850 000 DOF
And still running very slow.

Is there anyone have any idea about accelerating the solving process?

Thank you!
XiaoKang

6 Replies Last Post 30 mai 2017, 05:07 UTC−4
COMSOL Moderator

Hello Donovan Wang

Your Discussion has gone 30 days without a reply. If you still need help with COMSOL and have an on-subscription license, please visit our Support Center for help.

If you do not hold an on-subscription license, you may find an answer in another Discussion or in the Knowledge Base.


Please login with a confirmed email address before reporting spam

Posted: 7 years ago 29 mai 2017, 03:12 UTC−4
Hi,

it will be better to "uncouple" the problem, in the sense to solve firstly for the heat transfer and after that to put the temperature field as a load for the structural analysis using "variables not solved for" in the solver sequence. There is mostly no need of bidirectional coupling, as the deformation is usually small and geometrical nonlinearities are neglected, so that the structure does not influence the heat transfer. If it doesn't help I could give you also some advices for the solver settings and the mesh.

Best regards,

Antoni Artinov
Hi, it will be better to "uncouple" the problem, in the sense to solve firstly for the heat transfer and after that to put the temperature field as a load for the structural analysis using "variables not solved for" in the solver sequence. There is mostly no need of bidirectional coupling, as the deformation is usually small and geometrical nonlinearities are neglected, so that the structure does not influence the heat transfer. If it doesn't help I could give you also some advices for the solver settings and the mesh. Best regards, Antoni Artinov

Please login with a confirmed email address before reporting spam

Posted: 7 years ago 30 mai 2017, 00:04 UTC−4
Updated: 7 years ago 30 mai 2017, 00:09 UTC−4
Xiaokang ,

In addition to doing what Antoni suggested, loose the absolute and relative errors if you have tightened them to a very small value.

My results, for example, wouldn't change if I increased the error from 1e-6 to 1e-4 while a significant improvement in the simulation time was achieved by doing so.

Amin,
Xiaokang , In addition to doing what Antoni suggested, loose the absolute and relative errors if you have tightened them to a very small value. My results, for example, wouldn't change if I increased the error from 1e-6 to 1e-4 while a significant improvement in the simulation time was achieved by doing so. Amin,

Please login with a confirmed email address before reporting spam

Posted: 7 years ago 30 mai 2017, 02:31 UTC−4
Dear Artinov:
Thank you for your suggestion. I'll try to use unidirectional coupling in future works.
The problem seems to be relating to the Coefficient Form PDE. And I've used a Weak Form PDE instead which works well.

Sincerely
Xiaokang
Dear Artinov: Thank you for your suggestion. I'll try to use unidirectional coupling in future works. The problem seems to be relating to the Coefficient Form PDE. And I've used a Weak Form PDE instead which works well. Sincerely Xiaokang

Please login with a confirmed email address before reporting spam

Posted: 7 years ago 30 mai 2017, 02:37 UTC−4
Dear Amin:

Thanks for the advice. I found it really useful to change the relative tolerance when it comes to convergence problems. I've changed the relative tolerance to the default setting (1e-3) and it works well.

Sincerely
Xiaokang
Dear Amin: Thanks for the advice. I found it really useful to change the relative tolerance when it comes to convergence problems. I've changed the relative tolerance to the default setting (1e-3) and it works well. Sincerely Xiaokang

Please login with a confirmed email address before reporting spam

Posted: 7 years ago 30 mai 2017, 04:19 UTC−4
Don't forget to make sure how much the results change when you decrease the tolerence, lets say from 1e-3 to 5e-4. If a significant change was observed, then you need to make the error smaller until a minor change is observed.
Amin,
Don't forget to make sure how much the results change when you decrease the tolerence, lets say from 1e-3 to 5e-4. If a significant change was observed, then you need to make the error smaller until a minor change is observed. Amin,

Please login with a confirmed email address before reporting spam

Posted: 7 years ago 30 mai 2017, 05:07 UTC−4
Hi, Amin

Thanks for the notice. I'll pay attention to that.

Xiaokang
Hi, Amin Thanks for the notice. I'll pay attention to that. Xiaokang

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.