Convergence Issues in COMSOL Modelling of a Laboratory Burner: Seeking Help

[martingotu]

Hi everybody. This is my first post in this forum. I am working with COMSOL MULTIPHYSICS and I am trying to model a laboratory burner (Meker burner) heating a copper disc placed above the burner. I am modelling the burner as a boundary setting of heat flux entering the domain. I have obtained a solution for low values (1000-1800 W/m^2) but when I increase this value the model doesn´t converge and by my calculations the burner should give in order of 2exp6 W/m^2 (burner diameter 38 mm).
Could someone tell me if this value is correct and bring me some help in this issue.
Thanks anyway.
Martin.

 

[Mech_Engineer]

How is the heat flowing out of the copper disc? Is it subject to convection, or is it conducting into a table or something? Are you doing a transient or steady-state analysis? What non-linear effects are you taking into account that are preventing convergence?

 

[martingotu]

The sorroundings of burner and disc is air and I set all properties as constants. The flow is free convection and density is calculated by the Boussisnesq aproximation as a temperature function. The analysis is steady state because I want to solve the problem for the real value of heat flux of Meker burner before, and then make a transient analysis. I use the non isothermal flow model from general heat transfer module. Do you know the value of heat power of a propane Meker burner?
I hope with this explanations you can say me something.
Thanks.
Martin.

 

[Mech_Engineer]

You could try setting the convective coefficient on the outside of the part to a constant value and see if the model converges. Maybe try estimating the convective coefficient using analytical methods.

That sounds like the most complex portion of the model, other than that you could check the solver being used, perhaps changing it to a different type.

 

[martingotu]

What I need to know is why there is a limit value for heat flux for which the model doesn´t converge, keeping all other settings the same. If the problem is the grid and I have to refine it I will have to run it in other machine. Also, if I have to reach 2exp6 W/m^2, I don´t know what size of grid element will be necessary. Maybe I will need a lot of RAM for solve it. I don´t know because I don´t have any experience in FEM analysis.
Should be possible that I can send you the mph file and you take a look if it will be a very complex problem to solve. I think the geometry is quite simple and I tried to simplify everything, setting constant values for air properties, etc...
Anyway thanks for your interest in help me.
greetings.
Martin.

 

[Mech_Engineer]

You wouldn't be able to send me the model because I don't have Comsol, so I couldn't do anything with it. With respect to the RAM issue, I do remember having problems with a demo version of Comsol where problems of a certain mesh density would not solve due to a problem with the solver not being able to use virtual memory, only system RAM. Not sure if this problem has been fixed, but it's something to keep in mind since it is quite possible you are using up your ram depending on how many elements you are working with. You could always try making your mesh coarser to see if the problem converges.

Technically speaking, there aren't any fundamental reasons you can't raise the flux past a value before the solver doesn't converge, it's really a case-by-case thing based on the problem being solved... You mentioned that you were calculating the density of the air using an equation (I'm assuming a temperature-density relationship). I would suggest removing nonlinearities like this for a first attempt. First, try to solve the simplest problem- heating bottom of the plate, constant h-value on the sides and top. After that, you can begin to add degrees of complexity. It's also possible it's better to model a burner as a constant temperature source rather than a uniform flux source, as a flame will have a combustion temperature associated with it...

If during the solve you're getting semi-infinite convective currents, it may mess up the convergence of the solution depending on the solver's settings for determining convergence. If, after adding a level of complexity the solution refuses to converge, you know the culprit and can adjust things accordingly.

Also, you may also consider trying to solve this problem analytically, to see what numbers you should expecting. An iterative process between a steady-state 1-D conduction model and free convection model using resistive-thermal equations should give you a good idea of the average temperature of the plate, as well as convective currents outside of it. It's possible you're barking up the wrong tree and the copper plate will melt before it reaches steady state if you don't take into account radiation effects... Better guesses at the start of the process could yield better results at the end.

 

[martingotu]

Ok, thank you for your answer. I´ll try to do what you say. Bye.

 

[hedenqvist]

Many thanks for your comments.

 

[akshayvohra]

Hi All,

I am trying to solve a problem where the solutions are expected to be zero for a very long time.

I have refined the mesh with the maximum element size of 0.002 and am using the GMRES method with preconditioning quality of 0.01.

However, inspite of it, the system is unable to process the solution and gives an error for residual computation citing singular matrix as the problem.

Can anyone please suggest some ways to solve the problem? I need it very urgently!

Akshay