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 The information provided may be out of date.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.
whats wrong with comsol / my understandig of a differential equation? - thermal conductance
Posted Oct 8, 2012, 10:30 a.m. EDT Heat Transfer & Phase Change Version 4.2a 6 Replies
Please login with a confirmed email address before reporting spam
a simple system:
2d
a very small region (we call it region 1), which is defined to be a heat-source, and a neighboring large region (we call it region 2), that shares a boundary with the source. everything is isolated except for one boundary which is a heat-sink. The initial temperature is constant all over the system.
If this system is simulated over time comsol produces smth that is, in my opinion, rather unphysical. ->
Intuitively, i'd say that initially the temperature supposed to be the initial-temperature and depending on the parameters the heat diffuses through the respective materials.
What comsol gives is for some reason already at the beginning a finite temperature that is HIGHER than the initial one, hence for some reason there is a (unphysical) Temperature-JUMP right at the start.
I'm interested in the heat up mechanism so I would like to get rid of this jump since it's unexplainable to me and my colleges!
It's not a matter of the scale. For instance already after e-9 seconds this jump is shown.
And it has also nothing to do with the 'thin thermal resistive layer' that can be defined between the two regions, which i've also checked.
So what the heck is comsol actually doing?
I simply want to get T=T_initial for t=0 as it is defined !!?
2d
a very small region (we call it region 1), which is defined to be a heat-source, and a neighboring large region (we call it region 2), that shares a boundary with the source. everything is isolated except for one boundary which is a heat-sink. The initial temperature is constant all over the system.
If this system is simulated over time comsol produces smth that is, in my opinion, rather unphysical. ->
Intuitively, i'd say that initially the temperature supposed to be the initial-temperature and depending on the parameters the heat diffuses through the respective materials.
What comsol gives is for some reason already at the beginning a finite temperature that is HIGHER than the initial one, hence for some reason there is a (unphysical) Temperature-JUMP right at the start.
I'm interested in the heat up mechanism so I would like to get rid of this jump since it's unexplainable to me and my colleges!
It's not a matter of the scale. For instance already after e-9 seconds this jump is shown.
And it has also nothing to do with the 'thin thermal resistive layer' that can be defined between the two regions, which i've also checked.
So what the heck is comsol actually doing?
I simply want to get T=T_initial for t=0 as it is defined !!?
6 Replies Last Post Oct 10, 2012, 9:17 a.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
this is often coming from a heat diffusivity and mesh density incompatibility with the small time steps around t=0. The physics tells us to take care as these three items are linked. Have you tried to make a real fine mesh in your "small region" AND for some extent along the common boundary on the larger domain and checked that the effect does not dissapear. It might also be anumerical effect, always probe your initial conditions gradients againstyour local mesh density.
Note, I often add "boundary mesh" on BOTH sides of a common boundary between two materials with large differences in alpha (heat diffusivity) for time solving HT models
--
Good luck
Ivar
this is often coming from a heat diffusivity and mesh density incompatibility with the small time steps around t=0. The physics tells us to take care as these three items are linked. Have you tried to make a real fine mesh in your "small region" AND for some extent along the common boundary on the larger domain and checked that the effect does not dissapear. It might also be anumerical effect, always probe your initial conditions gradients againstyour local mesh density.
Note, I often add "boundary mesh" on BOTH sides of a common boundary between two materials with large differences in alpha (heat diffusivity) for time solving HT models
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
thank you for your quick reply...
i made the meshgrid finer and introduced boundary layers in the meshgrid-tree for all boundaries.
for some reason nothing significantly changes.
maybe somebody wants to take a look on the file itself, attached as 'test.mph'
a 1d-piont graph has to be added after the computation to see the temporal evolution.
thanks in advance, looking forward to resolve this weirdness and curious about every hint?!
i made the meshgrid finer and introduced boundary layers in the meshgrid-tree for all boundaries.
for some reason nothing significantly changes.
maybe somebody wants to take a look on the file itself, attached as 'test.mph'
a 1d-piont graph has to be added after the computation to see the temporal evolution.
thanks in advance, looking forward to resolve this weirdness and curious about every hint?!
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi, Florian,
I'm not sure about the "initial temperature jump" you mention. I understand you refer to temperature at point 6, near the tiny upper piece.
Perhaps this is too simple, but a finer time resolution seems to solve the problem. Particularly, I have tried times: 10^{range(-15,0.5,-5)}, and then I see no jump about t = 0 (temperature raises continuously from 293.15 K). Those times are very very short, but the power density you are entering at top piece is very very high (920 W at 6e-10m^2 = 1.53e12 W/m^2).
Bye,
Jesus.
I'm not sure about the "initial temperature jump" you mention. I understand you refer to temperature at point 6, near the tiny upper piece.
Perhaps this is too simple, but a finer time resolution seems to solve the problem. Particularly, I have tried times: 10^{range(-15,0.5,-5)}, and then I see no jump about t = 0 (temperature raises continuously from 293.15 K). Those times are very very short, but the power density you are entering at top piece is very very high (920 W at 6e-10m^2 = 1.53e12 W/m^2).
Bye,
Jesus.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
thanks a lot! I get the same, but I need to solve the system up to one second!
So, I cranked up the accuracy of the meshgrid and set the 'relative tolerance' in the time-dependent study to a tiny number (from 1e-20 down to 1e-40) but the jump starts to appear again already for 10^{range(-15,0.5,-3)}.
The way I understand the option 'relative tolerance' is that it sets in some way the 'temporal-meshgrid', the timestep that is used under the hood by comsol. Apparently I'm wrong, otherwise it should help to decrease it.
So what can I do to increase the accuracy of the solver in the time-domain?
thanks again, I sense that I'm almost there!
So, I cranked up the accuracy of the meshgrid and set the 'relative tolerance' in the time-dependent study to a tiny number (from 1e-20 down to 1e-40) but the jump starts to appear again already for 10^{range(-15,0.5,-3)}.
The way I understand the option 'relative tolerance' is that it sets in some way the 'temporal-meshgrid', the timestep that is used under the hood by comsol. Apparently I'm wrong, otherwise it should help to decrease it.
So what can I do to increase the accuracy of the solver in the time-domain?
thanks again, I sense that I'm almost there!
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hello, Florian,
It seems that a fine-tuning of the "Time-Dependent Solver" (inside Solver Configurations/Solver1) arranges the problem. I have set:
relative tolerance: 1e-5
In the Time-Dependent Solver: Method: scaled, tolerance: 1e-10, tol. for time-derivatives: 1e-8,
In Time stepping: Initial step: 1e-10, maximum step: 1e-3.
It works.
Bye,
Jesus
It seems that a fine-tuning of the "Time-Dependent Solver" (inside Solver Configurations/Solver1) arranges the problem. I have set:
relative tolerance: 1e-5
In the Time-Dependent Solver: Method: scaled, tolerance: 1e-10, tol. for time-derivatives: 1e-8,
In Time stepping: Initial step: 1e-10, maximum step: 1e-3.
It works.
Bye,
Jesus
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
thank you so much!