Continue to Site

Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

  • Congratulations KootK on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!

What is the best electronics cooling CFD package? 1

Status
Not open for further replies.

analoguealan

Computer
Aug 26, 2006
8
Hi Folks,

I am an electronics hardware engineer working on set top box development.

I started out using a FVA, finite volume analysis package, Flomerics /Flotherm. Basically we had not heard of any other application and Flomerics do a good job in marketing.

I found I spent 90% of my time in mesh optimisation.
Flotherm only has cartesian mesh capability so could not handle the conical shaped dimples we use to conduct heat from PCB hot spots to the chassis base. These dimples can exceed the max case temp of 50C that we try to keep to.
Therefore an important area to model accurately.

Eventually moved to Fluents/Icepak on the promise that their choice of three meshes,
hexa cartesian , hexa unstructured and tetra mesh
would handle the dimple geometry.

+ Icepak has much more mesh control capability, enabling non-conformal meshing of assemblies, individual control of an objects mesh, etc.

In practice I find Icepak is much more fussy to use and I still spend 90% of my time optimising the mesh taking me weeks of modelling!
My models after careful non-conformal meshing come out at around a million cells and need a fairly large virtual domain ( 750 mm x 650 x 200 mm) to capture all the air flow around the box correctly. This takes about an hour to solve on a power PC. Every time you add something the mesh has to be totally reviewed.
Thoroughly frustrated I have looked around and found a FEA finite element analysis approach from CFDesign. This takes much longer to solve. About 12 hours for my models.

But the meshing is a highly conformal tetra mesh that is very tolerant of poor cell aspect ratio. This covers all the object surfaces handling the most complex of physical geometry.
This means the thermal model looks just like the physical model and not some "leggo" representation.
The meshing operation is fully automatic and done in about twenty minutes. So if it took 48 hours to run I would still win hands down.

We have not moved over just yet, its in the pipeline after
CFDesign have demonstrated they can apply a a fully automated conformal mesh to our most curvy stylish and very complex ABS plastic enclosure.
With plastic tooling costs of around 80 to 100 grand I want a tool that can handle all the details.
This package is not stand alone like Flotherm/Icepak (with their own internal 2D/3D draweing boards) but is embedded in Pro/E MCAD package (which we use) so I now have to learn Pro/E and become a hybrid engineer and work in an integrated modelling environment!

My question then,
Has anyone else gone down the same path and come to the same conclusions regarding CFD for electronics cooling.


analogue alan





 
Replies continue below

Recommended for you

I am a mech. engr. in aerospace communications. I have used the Flotherm package recently and I am not impressed. They claim their result is within 5% accuracy when compared to experimental data. I asked a rep. from flomerics how they can claim this. For one, the cartesian mesh cannot handle complex curved geom. surfaces, and the rep. tried to tell me that was not a problem at all. Tetra meshing with the correct range of angles in the cell corners is ALWAYS superior to rect. meshing, unless all your model geometry is Legos, as you said. Also the time step in Flotherm marches forward, which in one of my CFD college courses, the instructor said can be unstable. The k-epsilon turbulence model is one of the most inaccurate methods possible, yet easy to code, and solves quickly and that is why it is employed in Flotherm. However, it depends upon an accurate value for epsilon, and often it is derived experimentally. It is also only valid for certain ranges in the power spectrum, so if your model flow does not fall into that range, you will not get accurate results.

My experience is that Fluent, which I used extensively in school, is a much more useful program with accurate results, as long as you know what you are doing. It is very easy to use the wrong turbulence model for your application, or poor meshing, and receive junk results. Fluent can accept most solid model files such as pro/e files, and once you get used to the program, you can solve a case and run it on your background, while you do other things.
 
Status
Not open for further replies.

Part and Inventory Search

Sponsor