Discussion Forum

Posted: 1 decade ago 18 janv. 2010, 18:08 UTC−5

I am struggling with a similar problem in 3D radiative heta transfer myself right now. Do you know if there is a way to use symmetry BC with radiation heat transfer (as in mirror symmetry) to reduce the problem size? If you had a 3D square brick shaped cavity with a sphere in the center, is there a way to get away with modeling only a quadrant of the geometry taking advantage of symmetries? I am not sure if COMSOL can somehow account for the radiative exchange between physically modeled and omitted parts of the geometry or one must include the entire geometry for the radiation calculation to be correct. Going back to your question here are some ideas that you may consider trying: 1- good initial guess is important for convergence of nonlinear problems, so if you can, make an educated guess at the answer and use this as initial conditions for starting the iterations 2- if you have enough RAM use a direct solver like UMFPACK 3- If nothing works, you may try running the problem as transient and let the solution "evolve" into the right steady state solution. This may take longer to solve but should have better convergence characteristics. Please let me know how your troubleshooting turns out Thanks, Ozgur [QUOTE] Hello my name is Luis. We are working on a thermal model of a cubesat (nano satellite). Well, imagine a hollow cube but his faces have thickness. Now imagine a little solid cube intro the other and in the middle. Well this is the 3D model. We are using the “Heat Transfer Module 3D” each face have been irradiated with a different heat radiation, and have different emissivity. We have different condition at the same time for different faces, “surface to surface” and “surface to ambient”. Well the objective is, take the temperature of the little solid cube. But we have a problem, the solution is not converging. Mesh: The “mesh” is “Extremely coarse” because the RAM is not enough. Solver Parameters: The “analysis” is “stationary”. Linearity -> Nonlinear “Damped Newton” has a tick “Highly nonlinear problem” has a tick Solver Manager: “Initial value expression” has a dot “Use setting from initial value frame” has a dot So the question is, How is possible that solution is not converging? And, What will we do to fix it? Thanking you in advance for any tip about these issues you can give us. Best Regards, Luis [/QUOTE]

Posted: 1 decade ago 18 janv. 2010, 18:58 UTC−5

Hi Ozgur, thanks for your answer. Well, let me say you that, it is my first work with comsol so I don’t know the correct answer for your problem. But I have been thinking it. Did you try making only the part of the model that you want? I mean, the symmetric part, maybe is foolishness. Well, now let me tell you how I did to this stuff works. In “htug.pdf” in pag 135 (from the pdf) or in pag 126 (from the document). ”First, with enough experience you can set up the mesh yourself in the Mesh Parameters dialog box” So, I kept the “Extremely coarse” but I refine the mesh of the internal box, (“Free Mesh Parameters” -> “Subdomain” -> “Maximum element size” = 0.005.) I guess that was luck. The solver UMFPACK works too, thanks!! The option 3) I couldn’t make it works. It is not converged But now I have a lot of other problems. 1) In the COMSOL's “Heat Transfer User Guide” it says that the emissivity and absorptivity are equal. But they don’t make difference with the radiation spectrum. So how can I make this deferens? 2) How can I spin the cubesat in COMSOL? I mean, I want simulate the spin of the cubesat and the interaction with the radiation. The radiation is different in the time and in the space coordinal. I think that this have a relationship with the fact that I want see the temperature behave of one parte of the model and not the “steady-state” Thanking you in advance for any tip about these issues you can give us. Best Regards, Luis

Posted: 1 decade ago 19 janv. 2010, 13:53 UTC−5

Hi Luis, e=a and independence of e from wavelength are common assumptions made in the formulation of all but the most sophisticated radiation models. I don't know if you can modify this and make e wavelength-dependent in COMSOL. I would ask COMSOL support. As far as spining the cubesat, your options are 1- spin the irradiation source i.e. calculate how irradiation will look like with time and impose it as a time varying boundary condition 2- it might be possible to use ALE formulation and prescribe a rigid body rotation to make the cubesat spin hope this helps Ozgur [QUOTE] But now I have a lot of other problems. 1) In the COMSOL's “Heat Transfer User Guide” it says that the emissivity and absorptivity are equal. But they don’t make difference with the radiation spectrum. So how can I make this deferens? 2) How can I spin the cubesat in COMSOL? I mean, I want simulate the spin of the cubesat and the interaction with the radiation. The radiation is different in the time and in the space coordinal. I think that this have a relationship with the fact that I want see the temperature behave of one parte of the model and not the “steady-state” Thanking you in advance for any tip about these issues you can give us. Best Regards, Luis [/QUOTE]

Posted: 1 decade ago 19 janv. 2010, 16:01 UTC−5

Hi Ozgur Thanks The point 1), well i'm working on it. The point 2), I'm studing it. About the e=a I’ll try ask it to the Comsol´s support. Let me explain the situation. The cubesat have two heat situations. 1- The day (x seg) and 2- the eclipse (y seg). In The day the cubesat has some faces with a determinate heat inward, and in the eclipse it has other faces with other radiation. With transient analysis I make the simulation with this time (0, (x+y)/100, x+y) and works. Here I use in the boundary condition’s “q” something like that (G1*(ty)). But this is one period. I want to repeat the simulation in the time, eg: (0, (x+y)/100, 10(x+y)) -> 10 periods And in the “q” of the boundary conditions look like [G1((t

Posted: 1 decade ago 19 janv. 2010, 16:57 UTC−5

Hi there Well I know from experience that when you put to many step functions (with abrupt steps) the solver has problems on the jumps, I often smoothen them with a heaviside function (search for heaviside on the doc files), this often helps. I agree too that it becomes tedious to write out 10 periods, but you can also build functions and use these, this I have seeen that it will be better systematised in the "to come" V4 Good luck Ivar

Posted: 1 decade ago 20 janv. 2010, 11:36 UTC−5

Thanks so much Mr. KJELBERG Luis