[_Tiny_]

Hi,

After many years of avoiding it, I need to invest some time in CFD. I always have the same problem to solve. The heating and cooling of electronics. I started initially playing with ElmerFEM, and it looked like we might get some sensible results, but as I increased the complexity of the problem, it failed to converge on a solution (I was using coupled Heat/Navier stokes). It also only runs single threaded (out of the box). Before I bite the bullet and invest a load of hours getting to grips with OpenFOAM, I'd like to gross error check that it will calculate the answer I need.

The problem is typically some parts on a PCB in a box, each part dissipating some fixed amount of power power. We want to know the die temperature within the part as we try to design the mechnical solution for spreading / removing heat. In elmerFEM, we model the individal parts using a multi body model of the known internal geometry of the part and the associated thermal characteristics.

We want to include the turbulent convection and conduction of heat, and possibly include some forced air flow.

I think the chtMultiRegionFoam solver can do this, but documentation seems a bit sparse. Critically I am unsure if it will model (multiple) heat sources in solid regions. I'd like to avoid the need to modify source files if possible.

I'd really appreciate any wise words on whether this is possible before embarking on a probably long learning curve.

thanks!

[groblerjf]

Hi Tiny,

I am also interested in the same functionality as you are. The systems we are analyzing are already pretty complex with a lot of interfaces (too many in my opinion) between the PCB and where the heat is dissipated.

I am still learning about the capabilities of OpenFOAM andI am trying to justify whether I should spend the time to learn OpenFOAM for the purpose of analyzing electronics cooling, or bite the bullet and recommend we buy Ansys which can do what I need, knowing that I do not constantly require simulation solutions.

[akidess]

You will spend a lot of time getting OpenFOAM results, so you should consider getting professional support. Once you've got it going, it will be free forever, which is nice. If you want a simpler option, I'd recommend you have a look at 6sigmaET and Flotherm/FloEFD, which are fairly simple to learn (immersed boundary really reduces meshing effort, and expert settings are kept minimal) and geared towards people looking at electronics cooling.

[groblerjf]

Originally Posted by akidess

You will spend a lot of time getting OpenFOAM results

Hi akidess,

Why do you say that? I understand that if I want to go super detailed that it would take time, but I am mostly interested in getting the heat from the PCB to the outside, and doing that efficiently. I have no real desired at the moment to obtain a detailed PCB thermal model.

If I have to spend a week or two learning the software and getting it to work, then that is still cost effective when the other option is a commercial software. Or is it more like a month or two??

[akidess]

It really depends on how much knowledge you already have (CFD, general meshing, and Linux) and if you need to learn everything from scratch or if you have someone to teach you. It also really depends on the complexity of your model. For OpenFOAM you will need a clean model, which CAD is usually not. You will need to manually specify the interfaces between contacting materials, which can be a lot of work.

[groblerjf]

Thanks, that makes sense. I come from the structural sides of things, so I am not well versed in CFD. Something like 6sigmaET or Ansys IcePak makes mores sense then. I will probably learn OpenFOAM on the side then.

Thanks for the help!!

[akidess]

ICEpack also uses body-fitted meshes which means more work compared to 6sigma or Flotherm. Check out their webinars and see which works best for you though.

[groblerjf]

If you dont mind another question. I have also been looking at NX CAE and Femap from Siemens. Specifically Femap, since it integrates directly with our CAD package (Solid Edge). The other option also is FloEFD from Mentor, that also integrates with our CAD as I understand. In some blogs people has also recommended Comsol.

[_Tiny_]

I still can't determine whether OpenFoam will actually solve the problem.

We have been using ElmerFEM with limited success. We have set the problem up with the relevant boundaries, materials, heat sources etc and solve for a coupled navier stokes and heat equation system, using the Boussinesque approximation for convection. Initially, it looks promising, then the whole thing goes unstable and my box with a chip in it end up hotter than the sun.

I know that this system is not guaranteed to be stable. We've mucked about with various settings to try and make it more stable, including using adaptive time-stepping, various mesh settings and types and using relaxation.

My conclsuon (rightly or wrongly) is that Elmer just isnt that refined / stable / good at solving big linear algebra problems. An it takes for ever on a single thread. This is a shame, becuase setting up the case is easy and it takes the input data in a form which is easy for me to work with. I think the developers focus is elsewhere.

So we though we'd try OpenFoam, using ChtMultiRegionFoam, in the hope of better and faster linear algebra and more refined flow stuff. The biggest problem, its seems, is to actually specify a heat source in watts, without fiddling around with the source.

So I remain unsure as to whether openfoam can solve the problem out of the box.

I did cast around for a commercial solution, but the solidworks bolt on was £15K and IcePak was £30k. Simply out of reach for me. So we need an opensource solution still, but where?

[akidess]

I cannot help you with Elmer, but I'm 100% confident that given the correct settings you should be able to obtain a stable solution. You might want to describe your problems on the appropriate forum if you haven't so already. Depending on your case it may be impossible to expect to solve the case with a single core in reasonable time, but everyone has access to multicore processors these days. Elmer runs parallel out of the box, as does OpenFOAM.

To specify a volume heat source in OpenFOAM, you need a scalarSemiImplicitSource (you define the regions via cellSets). At surfaces, you specify temperature gradients, that is the heat source in [W/m2] divided by the thermal conductivity. So it's possible, but I'm not sure if your experience will be any better than Elmer.