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 The information provided may be out of date.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.

Natural Convection in simple 2D geometries not converging (despite several adaptations/simplifications)

Please login with a confirmed email address before reporting spam

I am trying to use comsol 4.4 to solve natural convection problems in 3 very simple geometries (cavities). I have considered two flat geometries and a cylindrical geometry. One causes no problems to solve whereas the other two originate strange results (or no convergence at all).

After setting up the simulation for an initial 2D flat geometry with a Height/Width ratio of 0.5 (please see model number 1) and horizontal active walls (i.e. boundaries where the cold and hot temperatures are imposed), the solution was easy to obtain (after imposing a pressure constrain in one point of the domain).

In a second moment, I rotated the geometry 90 degrees so that the active walls would be vertical (please see model number 2). Despite not changing any other parameter, no convergence is achieved with this 2nd geometry, as the solving procedure stops at the 2nd iteration (with problems apparently in the segregated group dealing with flow variables, i.e. velocity and pressure). Every time a solution is attempted, the following error is reported:

“Undefined value found.
- Detail: Undefined value found in the equation residual vector.
There are 461 degrees of freedom giving NaN/Inf in the vector for the variable comp1.p.
at coordinates: (0,0.0236111), (0,0.0222222), (0,0.0208333), (0,0.0194444), (0.000310276,0.0236165), ...
There are 461 degrees of freedom giving NaN/Inf in the vector for the variable comp1.u.
at coordinates: (0,0.0236111), (0,0.0222222), (0,0.0208333), (0,0.0194444), (0.000310276,0.0236165), ...
There are 461 degrees of freedom giving NaN/Inf in the vector for the variable comp1.v.
at coordinates: (0,0.0236111), (0,0.0222222), (0,0.0208333), (0,0.0194444), (0.000310276,0.0236165), ...”

I did try to change mesh density, damping factors in solvers, initial velocity estimates, but obtained always the above type of error. I also tried to ramp up the volume force acting on the fluid elements and obtained a solution when this force was 10% of the actual value. For bigger values (of volume force) the error estimate of the flow-related segregated group reduces initially and then starts to oscillate over iteration (number) in a periodic way. I thought this could indicate that the flow may be unstable but, for the parameters of the current simulation, I do not expect the flow to be unstable… what could be causing this problem? How to address errors like those mentioned above?

In a third moment I tried to solve the same problem but with a 2D cylindrical geometry, basically one cylinder inside a bigger cylinder, with the inner cylinder having the highest temperature (please see model number 3). In this case, the solution is obtained way too fast (just a couple of seconds), and the obtained results show no natural convection and null velocity in between the two cylinders. The fact that the solution is obtained too fast seems to indicate that the flow problem is not being solved. And the fact that no natural convection appears in the temperature/velocity plots even when I increase the temperature difference up to 40ºC is, again, consistent with the flow problem not being solved (since it is very unlikely that the used configuration will not lead to natural convection between the two cylinders). I also tried to change the portion of the cylinders’ surface where I imposed fixed temperature but the results never showed natural convection flow/temperature patterns, which is again highly questionable. Are these results a consequence of wrong definitions in the simulation? How to capture the likely natural convection occurring in this geometry?

Thank you very much for your input.


5 Replies Last Post 25 gen 2014, 08:01 GMT-5
Phillip COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 23 gen 2014, 03:16 GMT-5
Hi,

you forgot to upload your model(s). ;-)


--
Best Regards
Phillip
Hi, you forgot to upload your model(s). ;-) -- Best Regards Phillip

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 23 gen 2014, 05:51 GMT-5

Hi,

you forgot to upload your model(s). ;-)


--
Best Regards
Phillip


Dear Phillip,

Thank you for the tip! :-)
The models are now available.

Regards,

Tiago Sotto Mayor
[QUOTE] Hi, you forgot to upload your model(s). ;-) -- Best Regards Phillip [/QUOTE] Dear Phillip, Thank you for the tip! :-) The models are now available. Regards, Tiago Sotto Mayor

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 23 gen 2014, 15:22 GMT-5
Hi,

Here are new version of models 2 and 3.

Model 2 was working ok by using default solver (Fully Coupled). I also set initial pressure value to 0.

With model 3 I add domain to Volume Force. Due to this there were no flow at all. Default solver was used with "Automatic highly nonlinear" in nonlinear method.

I hope these will help you.

Best regards

Tero Hietanen
Hi, Here are new version of models 2 and 3. Model 2 was working ok by using default solver (Fully Coupled). I also set initial pressure value to 0. With model 3 I add domain to Volume Force. Due to this there were no flow at all. Default solver was used with "Automatic highly nonlinear" in nonlinear method. I hope these will help you. Best regards Tero Hietanen


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 24 gen 2014, 04:55 GMT-5

Hi,

Here are new version of models 2 and 3.

Model 2 was working ok by using default solver (Fully Coupled). I also set initial pressure value to 0.

With model 3 I add domain to Volume Force. Due to this there were no flow at all. Default solver was used with "Automatic highly nonlinear" in nonlinear method.

I hope these will help you.

Best regards

Tero Hietanen


Dear Tero,

thank you very much for your input. I had chosen segregated solver approach because, for the first model (horizontal active walls), it was basically mandatory to solve first the flow variables and only then the thermal part of the problem (to ensure one would actually get the occurrence of NC, instead of a unrealistic stratified thermal/flow field). What I don't understand if why the same approach could not work for the second model (vertical active walls) since (I think) the problem should be more stable (or easy to solve).

Furthermore, for the 3rd model (cylindrical geometry), you had to use "Automatic highly non-linear (Newton) to solve the problem, because the settings used for the 2nd model would not work (on the 3rd model). So, every new change in the problem seems to require new settings of the solvers (after your post, I tried to use the settings of the 3rd model on the previous two models without success), which makes it very difficult to know which settings to apply to a given problem/geometry. For instance, I took the cylindrical model and then increased the dimensions of the circles to 0.45 and 0.5 m and, for this new geometry, none of the previous settings worked. So, my question is... if you know of some literature or documents that may help in understanding the type of approach (or settings) we should use to address the solving of NC problems with COMSOL.

Thank you very much,

Tiago Sotto Mayor
[QUOTE] Hi, Here are new version of models 2 and 3. Model 2 was working ok by using default solver (Fully Coupled). I also set initial pressure value to 0. With model 3 I add domain to Volume Force. Due to this there were no flow at all. Default solver was used with "Automatic highly nonlinear" in nonlinear method. I hope these will help you. Best regards Tero Hietanen [/QUOTE] Dear Tero, thank you very much for your input. I had chosen segregated solver approach because, for the first model (horizontal active walls), it was basically mandatory to solve first the flow variables and only then the thermal part of the problem (to ensure one would actually get the occurrence of NC, instead of a unrealistic stratified thermal/flow field). What I don't understand if why the same approach could not work for the second model (vertical active walls) since (I think) the problem should be more stable (or easy to solve). Furthermore, for the 3rd model (cylindrical geometry), you had to use "Automatic highly non-linear (Newton) to solve the problem, because the settings used for the 2nd model would not work (on the 3rd model). So, every new change in the problem seems to require new settings of the solvers (after your post, I tried to use the settings of the 3rd model on the previous two models without success), which makes it very difficult to know which settings to apply to a given problem/geometry. For instance, I took the cylindrical model and then increased the dimensions of the circles to 0.45 and 0.5 m and, for this new geometry, none of the previous settings worked. So, my question is... if you know of some literature or documents that may help in understanding the type of approach (or settings) we should use to address the solving of NC problems with COMSOL. Thank you very much, Tiago Sotto Mayor

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 gen 2014, 08:01 GMT-5
Hi,

I have not found any good document which tells what setting should be used in various models. Best way (my opinion) is to study models from model gallery and do lot of "trial and error" testing. During time you will learn what settings you should use.

Best regards

Tero
Hi, I have not found any good document which tells what setting should be used in various models. Best way (my opinion) is to study models from model gallery and do lot of "trial and error" testing. During time you will learn what settings you should use. Best regards Tero

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.