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.
One way multiphysics coupling within a transient simulation
Posted Jan 14, 2013, 4:16 p.m. EST Heat Transfer & Phase Change, Studies & Solvers Version 5.0 17 Replies
Please login with a confirmed email address before reporting spam
Hello Fellow Users,
I am solving two physics (fluid flow & species transport) in a transient simulation. The fluid flow is not affected by the concentration distribution. As of now, i have implemented the above using 'fully coupled' feature.
Is there any way to implement one way coupling within a transient simulation i.e. at every time step i want to solve for fluid flow first and then use that estimated velocity field to calculate the concentration distribution. It is not clear to me whether segregated step would do the above mentioned? I tried segregated node separating these two physics (still using direct solver) for both physics but the total simulation time goes up (more than double) compared to the fully coupled case.
my understanding is : if segregated step implements one way coupling then the simulation time should be less than the fully coupled case? any thoughts/comments on this.
Any guidelines or similar example for this kind of coupling in dynamic simulation is appreciated.
Thanks in advance for your suggestions.
I am solving two physics (fluid flow & species transport) in a transient simulation. The fluid flow is not affected by the concentration distribution. As of now, i have implemented the above using 'fully coupled' feature.
Is there any way to implement one way coupling within a transient simulation i.e. at every time step i want to solve for fluid flow first and then use that estimated velocity field to calculate the concentration distribution. It is not clear to me whether segregated step would do the above mentioned? I tried segregated node separating these two physics (still using direct solver) for both physics but the total simulation time goes up (more than double) compared to the fully coupled case.
my understanding is : if segregated step implements one way coupling then the simulation time should be less than the fully coupled case? any thoughts/comments on this.
Any guidelines or similar example for this kind of coupling in dynamic simulation is appreciated.
Thanks in advance for your suggestions.
17 Replies Last Post Jan 25, 2017, 10:21 a.m. EST
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
to fully decouple, you need two study steps, one for one physics and another subsequently, using the first results as initial conditions for the second study. if you use a segregated solver with different steps you order how you start solving, but COMSOL will loop around until it converges so they are coupled. A direct solver usually uses more RAM but often solves quicker, while the iterative solver uses often less RAM for a longer solver time. Sometimes one or the other fails. Often for very non-linear cases the segregated iterative solver is better, but there is no 1 rule.
I usually try the default settings, in case the model is small and strongly coupled I sometimes forces immediatly a direct solver. Only in case of failure, or slow convergence, triggers my attempts to tweak the solvers
--
Good luck
Ivar
to fully decouple, you need two study steps, one for one physics and another subsequently, using the first results as initial conditions for the second study. if you use a segregated solver with different steps you order how you start solving, but COMSOL will loop around until it converges so they are coupled. A direct solver usually uses more RAM but often solves quicker, while the iterative solver uses often less RAM for a longer solver time. Sometimes one or the other fails. Often for very non-linear cases the segregated iterative solver is better, but there is no 1 rule.
I usually try the default settings, in case the model is small and strongly coupled I sometimes forces immediatly a direct solver. Only in case of failure, or slow convergence, triggers my attempts to tweak the solvers
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks Ivar.
The two study steps solution with the first solution being initial condition to the second physics is good for steady state simulation (for both physics) or when first physics is steady and second could be transient.
But is there any way to implement such coupling where both physics need to be solved dynamically i.e. say fluid flow is first and mass transport is second. Both are changing with time but fluid flow is not affected by concentration distribution. How should i implement such on way coupling in a transient solver setup.
Thanks again for your input.
The two study steps solution with the first solution being initial condition to the second physics is good for steady state simulation (for both physics) or when first physics is steady and second could be transient.
But is there any way to implement such coupling where both physics need to be solved dynamically i.e. say fluid flow is first and mass transport is second. Both are changing with time but fluid flow is not affected by concentration distribution. How should i implement such on way coupling in a transient solver setup.
Thanks again for your input.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks again Ivar. I will look for that post/ example and write back if i could not find it.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Dear all,
I would like to ask some help about a simulation that I am trying to implement. I think I have similar problems with the one presented in the thread, but not completely sure.
My simulation is the following:
My system is made of two different materials, one is a good conductor, the other one is a bad one. The good conductor has the shape of a filament sourrounded by the other material.
When I apply a bias current is flowing and the filament heats up faster than the bad conductor because almost all current is flowing through the filament. I make a time dependent simulation where the bias is time-dependent.
This is basically a simple multiphysics current/heat simulation.
The point is that the filament is degradated by the boundary temperature. In the model this means that the internal boundary between the good and bad conductors changes in time depending on the boundary temperature with a velocity which is a function of the temperature.
In other words, I have a coupled time dependent simulation (heat/electrical) with variable internal regions for the two materials which depend on the temperature.
I wonder if this part can be added to the simulation and if yes do you have any suggestion how the boundary dynamics can be implemented within COMSOL 4.2?
Best regards
Alessio
I would like to ask some help about a simulation that I am trying to implement. I think I have similar problems with the one presented in the thread, but not completely sure.
My simulation is the following:
My system is made of two different materials, one is a good conductor, the other one is a bad one. The good conductor has the shape of a filament sourrounded by the other material.
When I apply a bias current is flowing and the filament heats up faster than the bad conductor because almost all current is flowing through the filament. I make a time dependent simulation where the bias is time-dependent.
This is basically a simple multiphysics current/heat simulation.
The point is that the filament is degradated by the boundary temperature. In the model this means that the internal boundary between the good and bad conductors changes in time depending on the boundary temperature with a velocity which is a function of the temperature.
In other words, I have a coupled time dependent simulation (heat/electrical) with variable internal regions for the two materials which depend on the temperature.
I wonder if this part can be added to the simulation and if yes do you have any suggestion how the boundary dynamics can be implemented within COMSOL 4.2?
Best regards
Alessio
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
I undersand that as solid + HT with Joule heating (or TS with EC), it's "just" to sort out which combination of physics or which multiphysics to choose to avoid adding several Temperature variables, or any other double ups.
If you have a large electric conduction behaiour of your two materials, they have probably also quite adifferent thermal response (heat diffusivity) then take care with the mesh density and the sime step versus temperature gradients
You will probably need a fine mesh and the material interface (I sometimes uses "boundary mesh" here as these often line up perpendicular to the heat gradients
the turn on transients might be tough for the solver, sometimes its better to start witha stationary case to settle any transients and define all gradients and then ramp up the current with a time series study, starting from the overall steady state case
--
Good luck
Ivar
I undersand that as solid + HT with Joule heating (or TS with EC), it's "just" to sort out which combination of physics or which multiphysics to choose to avoid adding several Temperature variables, or any other double ups.
If you have a large electric conduction behaiour of your two materials, they have probably also quite adifferent thermal response (heat diffusivity) then take care with the mesh density and the sime step versus temperature gradients
You will probably need a fine mesh and the material interface (I sometimes uses "boundary mesh" here as these often line up perpendicular to the heat gradients
the turn on transients might be tough for the solver, sometimes its better to start witha stationary case to settle any transients and define all gradients and then ramp up the current with a time series study, starting from the overall steady state case
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Many thanks for the fast response Ivar and thanks for the advices.
I have still a question, I don't know if you have already answered to me :)
Is it then possible to add the "variable internal boundary" part within the model? I have seen that there is a physical model called "Moving interface" that in theory should be exactly what I would like to do with mine thermal/electrical simulation. The only point would be to be able to connect the moving of the internal boundary to the solution of the thermal solution.
Best regards
Alessio
I have still a question, I don't know if you have already answered to me :)
Is it then possible to add the "variable internal boundary" part within the model? I have seen that there is a physical model called "Moving interface" that in theory should be exactly what I would like to do with mine thermal/electrical simulation. The only point would be to be able to connect the moving of the internal boundary to the solution of the thermal solution.
Best regards
Alessio
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
you have two "moving" "physic"s: ALE and DG deformed geometry, they are slightly different and act at different levels of the "frames" chain.
But I'm not sure I catch everything from your explanations, as TS (thermal stress) alows mateial (linear elastic or non linear) to deform from the thermal load, and then other "solid" material will be "pushed" to deform by the expanding internal material, even if not (yet) subject to the same thermal load. And this works within "SOLID" without DG nor ALE
--
Good luck
Ivar
you have two "moving" "physic"s: ALE and DG deformed geometry, they are slightly different and act at different levels of the "frames" chain.
But I'm not sure I catch everything from your explanations, as TS (thermal stress) alows mateial (linear elastic or non linear) to deform from the thermal load, and then other "solid" material will be "pushed" to deform by the expanding internal material, even if not (yet) subject to the same thermal load. And this works within "SOLID" without DG nor ALE
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Ivar,
thanks again. I will try to be clearer in this email.
Imagine a filament sourrounded by a second material. Current is flowing and heating the filament as well as the other material. However, most of the current flows inside the filament, then it should heats up more than the other material.
In particular I have a temperature field at the boundary between filament and sourrounding material.
What I need is to change the geometry of the filament (it shrinks its diameter with a velocity which depend on the temperature, something like dr/dt = f(T(r), t), with r= radius of the filament at time t) and T(r) the temperature at the boundary.
In theory after a certain time the filament should break as the filament in a light bulb, however I know that this is a topological change of the geometry and could be difficult, I just need to know the velocity of filament shrinking at different sweep velocity of the bias in time and as function of maximum bias reached (imagine to apply a triangular bias weep to the device: the sweep reaches a maximum bias in a certain time).
The volume left empty by the filament shrinking is filled by the sourrounding material...in other words the boundary between the two materials should change in time with the function dr/dt = f(T(r),t). I attach a figure with caption took from a paper that make a similar simulation. In the figure the green line represents the variation of the filament diameter in time (at different biases) as a function of the thermal heating. The more the filament heats up, the faster the diameter drops. the figure comes from (Phase Transitions Vol. 84, No. 7, July 2011, 570–602).
I wonder if something similar could be implemented within COMSOL 4.2. I have a discrete understanding how to couple the thermal and electrical parts, but for the inserting of the dynamic change I really don't know.
Best
Alessio
thanks again. I will try to be clearer in this email.
Imagine a filament sourrounded by a second material. Current is flowing and heating the filament as well as the other material. However, most of the current flows inside the filament, then it should heats up more than the other material.
In particular I have a temperature field at the boundary between filament and sourrounding material.
What I need is to change the geometry of the filament (it shrinks its diameter with a velocity which depend on the temperature, something like dr/dt = f(T(r), t), with r= radius of the filament at time t) and T(r) the temperature at the boundary.
In theory after a certain time the filament should break as the filament in a light bulb, however I know that this is a topological change of the geometry and could be difficult, I just need to know the velocity of filament shrinking at different sweep velocity of the bias in time and as function of maximum bias reached (imagine to apply a triangular bias weep to the device: the sweep reaches a maximum bias in a certain time).
The volume left empty by the filament shrinking is filled by the sourrounding material...in other words the boundary between the two materials should change in time with the function dr/dt = f(T(r),t). I attach a figure with caption took from a paper that make a similar simulation. In the figure the green line represents the variation of the filament diameter in time (at different biases) as a function of the thermal heating. The more the filament heats up, the faster the diameter drops. the figure comes from (Phase Transitions Vol. 84, No. 7, July 2011, 570–602).
I wonder if something similar could be implemented within COMSOL 4.2. I have a discrete understanding how to couple the thermal and electrical parts, but for the inserting of the dynamic change I really don't know.
Best
Alessio
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
if you can express the thermal shrinking as a negative thermal expansion, normally the TS would take that, depending now if the linear material behaviour is correct for your case, there are other models in the non linear material section.
Indeed as you have large deformations, at least the geometric non-linearities should e turned on, but perhaps you are better off with a fluid approach and ALE, but then you need to have your material properties accordingly, and I'm not sure the volumetric thermal expansion/compression DoF for fluids is implemented in COMSOL (yet ?)
--
Good luck
Ivar
if you can express the thermal shrinking as a negative thermal expansion, normally the TS would take that, depending now if the linear material behaviour is correct for your case, there are other models in the non linear material section.
Indeed as you have large deformations, at least the geometric non-linearities should e turned on, but perhaps you are better off with a fluid approach and ALE, but then you need to have your material properties accordingly, and I'm not sure the volumetric thermal expansion/compression DoF for fluids is implemented in COMSOL (yet ?)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Ivar,
thank you very much for the many advices. Actually I menaged to get what I wanted using partially this thread and partially another one (I don't remember the exact title) where you explained how to simulate the bending of a capacitor induced by the charge!
I used ALE (on top) coupled to EC and HT simulations, putting a temperature dependent formula on the boundary between the two materials. And it works!
there is still a lot of work to do to generalize in 3D, get all the right parameters, but I think I am on the right track.
Thank you very much for your help :)
Best
Alessio
thank you very much for the many advices. Actually I menaged to get what I wanted using partially this thread and partially another one (I don't remember the exact title) where you explained how to simulate the bending of a capacitor induced by the charge!
I used ALE (on top) coupled to EC and HT simulations, putting a temperature dependent formula on the boundary between the two materials. And it works!
there is still a lot of work to do to generalize in 3D, get all the right parameters, but I think I am on the right track.
Thank you very much for your help :)
Best
Alessio
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi there,
a short question. I am still working on simulations of thermal effects on the filament.
I use three modules: ALE, EC, HT.
The temperature calculated using the EC module (Joule heating) is used to move the boundary between two materials. On the boundary is defined a normal velocity (v_n) which depends on the temperature.
It is a time dependent simulation: I apply a bias sweep (a triangle function, 1 second in total).
I have noticed that the simulation is very delicate (it can easily not converge, especially the ALE part). I wonder if there are some tricks for time-dependent/ALE simulations to improve the convergence probability (changes in the solver, etc). In particular I wonder if it is possible to make a first step "stationary" without ALE and then start the time dependent simulation, because in many cases the time dependency didn't ocnverge because problems in finding "initial values". I thought that maybe providing to the code a better first guess, things can improve.
Best
Alessio
a short question. I am still working on simulations of thermal effects on the filament.
I use three modules: ALE, EC, HT.
The temperature calculated using the EC module (Joule heating) is used to move the boundary between two materials. On the boundary is defined a normal velocity (v_n) which depends on the temperature.
It is a time dependent simulation: I apply a bias sweep (a triangle function, 1 second in total).
I have noticed that the simulation is very delicate (it can easily not converge, especially the ALE part). I wonder if there are some tricks for time-dependent/ALE simulations to improve the convergence probability (changes in the solver, etc). In particular I wonder if it is possible to make a first step "stationary" without ALE and then start the time dependent simulation, because in many cases the time dependency didn't ocnverge because problems in finding "initial values". I thought that maybe providing to the code a better first guess, things can improve.
Best
Alessio
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
Hello Ivar,
I am in need of a way to simulate in COMSOL a coupling boundary condition - the pressure generated out of the piezoelectric material at each time step needs to be applied as the wall boundary condition for the fluid....I have posted my queries in COMSOL forum, but couldnt get any response....Please if you could help me,as I am stuck with this part.....
Regards,
Rahul Kishor
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
Does anybody know where I could find this post? I've been searching for hours and couldn't find anything.
Many thanks in advance!
Florian
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
Does anybody know where I could find this post? I've been searching for hours and couldn't find anything.
Many thanks in advance!
Florian
Hi Florian and others
At least in Comsol 4.4 you can solve this by doing a study with two time dependent steps. The first step solves the first "independent" part of your problem, and the second solves the second "dependent" part of your problem, and in the second step you click "values of variables not solved for", choose "method" -> "solution", "study" -> that same study, "time (s)" -> "all". In this way you get a one-way coupling between the two time dependent problems, however it solves the first step for all times, and then the second step for all times, thus you need to save the solution for the first part at all steps taken by the solver!
If anyone knows a way to set up an iterative solver with one-way coupling between two time-dependent physics, i.e. the solver takes one step in the first "independent" part, then one step in the second "dependent" part, repeatedly, allowing for you not to save the solution at all time steps, please let us know. I am not sure whether this was what Ivar referred to in his comment, I was not able to find the discussion he talks about.
Good comsolling
Peter
Please login with a confirmed email address before reporting spam
Posted:
10 years ago
Hi
two transient solver cases where one depends on the next, by using "t" time as a pointer into the solution from study 1 should be possible, some months ago it was discussed here on the Forum, I believe i dropped an example too, but there were some limitations in the versions, not sure any more if it was limited to latest one (4.3a)
--
Good luck
Ivar
Does anybody know where I could find this post? I've been searching for hours and couldn't find anything.
Many thanks in advance!
Florian
Hi Florian and others
At least in Comsol 4.4 you can solve this by doing a study with two time dependent steps. The first step solves the first "independent" part of your problem, and the second solves the second "dependent" part of your problem, and in the second step you click "values of variables not solved for", choose "method" -> "solution", "study" -> that same study, "time (s)" -> "all". In this way you get a one-way coupling between the two time dependent problems, however it solves the first step for all times, and then the second step for all times, thus you need to save the solution for the first part at all steps taken by the solver!
If anyone knows a way to set up an iterative solver with one-way coupling between two time-dependent physics, i.e. the solver takes one step in the first "independent" part, then one step in the second "dependent" part, repeatedly, allowing for you not to save the solution at all time steps, please let us know. I am not sure whether this was what Ivar referred to in his comment, I was not able to find the discussion he talks about.
Good comsolling
Peter
Dear Peter,
thanks for your answer.
I try to simulate instationary flow and reactive transport (water purification tank).
In my case the one-way coupling (method" -> "solution", "study" -> that same study, "time (s)" -> "all".) seemed to work, but the transport developed heavy oscillations. If I solve simultaneously for flow and transport, the transport seems to be ok.
Nevertheless a decoupling would be preferable . I would be able to experiment with the transport, while using an existing flow field.
I let you know, if I find another workaround.
Greetings,
Florian
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Hi comsol fellows,
I'm using comsol to solve a thermal-hydric study. My case study is to calculate the temperature (T) and pressure (Pl) distribution in the domain 1D which varies from 0 to 250h. I create two studies, study 1 solved for heat transfer in solid to calculate T(x,t) ; and study 2 solved for coefficient form PDE to calculate Pl(x,t). In order to use result of T(x,t) from study 1 as input for study 2, I follow what you guys mentioned above by selecting "Values of variables not solved for” – Method: Solution –Study: Study 1, Time Dependent –Solution: Solver 1–Time: All / Automatic / Last.
When I chose "Time: Automatic or Last", it works and provides the same result Pl(x,t) i.e calculation of study 2 based on the last result of study 1, T(x,250h).
A problem is when I chose "Time: All", it has encountered an error : May have reached a singularity, Failed to evaluate expression :
d((-mod1.Da2(mod1.T/unit K cf,mod1.Pl/unit Pa cf)*mod1.Plt*test(mod1.Pl)-mod1.C2(mod1.T/unit K cf,mod1.Pl/unit Pa cf)*mod1.Plx*test(mod1.Plx))*(dvol),{test1})
I just want to precise that the thermal result was correct, but I think it tried to derive the coefficient Da2 (T,Pl) and C2 (T,Pl) against T and Pl, instead of deriving them just against Pl as T is known from study 1.
Can anyone please tell me how to solve this problem? Where can I make an explicit calculation? Moreover, in Result part, when I plot the graph of Pl, is it normal that it shows different curves once I choose Study 2 and Study 1 in Data set.
Thank you in advance for your help,
Ponleu
I'm using comsol to solve a thermal-hydric study. My case study is to calculate the temperature (T) and pressure (Pl) distribution in the domain 1D which varies from 0 to 250h. I create two studies, study 1 solved for heat transfer in solid to calculate T(x,t) ; and study 2 solved for coefficient form PDE to calculate Pl(x,t). In order to use result of T(x,t) from study 1 as input for study 2, I follow what you guys mentioned above by selecting "Values of variables not solved for” – Method: Solution –Study: Study 1, Time Dependent –Solution: Solver 1–Time: All / Automatic / Last.
When I chose "Time: Automatic or Last", it works and provides the same result Pl(x,t) i.e calculation of study 2 based on the last result of study 1, T(x,250h).
A problem is when I chose "Time: All", it has encountered an error : May have reached a singularity, Failed to evaluate expression :
d((-mod1.Da2(mod1.T/unit K cf,mod1.Pl/unit Pa cf)*mod1.Plt*test(mod1.Pl)-mod1.C2(mod1.T/unit K cf,mod1.Pl/unit Pa cf)*mod1.Plx*test(mod1.Plx))*(dvol),{test1})
I just want to precise that the thermal result was correct, but I think it tried to derive the coefficient Da2 (T,Pl) and C2 (T,Pl) against T and Pl, instead of deriving them just against Pl as T is known from study 1.
Can anyone please tell me how to solve this problem? Where can I make an explicit calculation? Moreover, in Result part, when I plot the graph of Pl, is it normal that it shows different curves once I choose Study 2 and Study 1 in Data set.
Thank you in advance for your help,
Ponleu