How to model the inside of a server for CFD and thermal studies?

Question

At my workplace I have been assigned to create an accurate 3D model of our Dell PowerEdge R820 servers so we can insert our custom cards and run a CFD and thermal analysis on them. The end goal is to have an exact understanding off all the airflow and heat transfer within the servers and server racks to help in designing future systems and fix overheating issues we are having with some of our current ones.

My problem is that the insides of these servers are quite complex and it is going to take an era for me to reverse engineer all the parts with a pair of verniers and find out how much heat is produced and where.So I am wondering if anyone else has come across something like this before and how they created an accurate CFD of their system.

Would 3D scanning be an option? I am concerned that 3D scanning will be too accurate, in the sense that what should be a straight edge might be turned into 200 different vertices etc. And I am not sure of how much cleanup it would require. I plan to do the CFD in Solidworks which requires solid modelling - it may have issues with the surfacing of the scans. Would like to hear from anyone with experience regarding this.

On the other hand, does anyone have experience with getting 3D CAD models from companies like Dell? I work at a large company and we buy a lot of their products for our projects but I wouldn't have a clue what department to call to see what needs to be organised to get CAD files. Would welcome any input here.

Lastly, for the thermal modelling, my plan of attack is to use some IR cameras to see where the heat is being produced. But I don't know how to work out HOW much heat is being produced at these spots so I can model the heat production from these processors in the CFD. What do people normally do for this?

Answer 1

Your options are a bit limited. Either you get the model from Dell, or you model it yourself.

Getting the model is a long shot, but if you are a really big customer that might happen. However this is not necceserily a godsend, since:

• The model may not be accurate enough for your need. They may have ripped some data out to protect their IP
• It may be too accurate in places that are of no interest for you. And mostly the data is just in your way.
• The model may be unsuitable for your task. So its geometry but your cfd needs the material info and heat generation info.

That said modeling a suitably absteacted version that simulates faster than imported models is not necceserily a huge job. Its just very scary, as it is just raw work.

Would scanning help? In my experience not really, though it might speed up data gathering, as you now have a digital model to check against. So you do not need to go and measure that one dimension you realized you need. Thats a good thing if you can not have the object on the table next to you. However, scanning is another can of worms as a work process.

Concentrate on abstracting things as much as you can. Do you really need every detail?

Also note thermal camera does not directly neasure what you want it measures how much things radiate at a specific band which varies by material and its surface quality. worse some materials like the silicon in your chips are transparent at typical infrared camera ranges. You still need to know how much the materials transfer energy to air. That said the thermal camera is a good validator target for your model.

This is a lot of work no matter how you look at it.

Answer 2

For your IR camera, you need an image of a known object at a known temperature in the same environment as a temperature reference , then when you have the IR image you can cross-reference the colours to approximate the temperatures. Check out some building thermal images for the idea.