# How to calculate thermal conductivity of potting compound around complex shape

I am currently evaluating the performance of various cooling strategies of a high-dissipation toroidal core inductor. The inductor specs:

• 20mm diameter core
• 9 turns (=very sparse), 16ga
• 5W core losses (max)
• 2W Joule losses (max)
• High voltages may occur between heatsink and the conductors
• Heatsink on one side of the inductor, coplanar with the core at about 1.5mm distance from the outside of the turns

I am currently evaluating cooling strategies for this relatively high power density component. One of these strategies is using thermally conductive potting compound. However, I am a bit lost. Can you help me with:

• Evaluating whether this compound is the best I can get (for evaluation purposes - price is not terribly important)
• Explaining how I would go about (roughly) calculating the temperature drop over the potting compound if I use this compound to space-fill between the inductor and an aluminum heatsink

The shape is obviously quite irregular, so it's not immediately obvious to me how to attack this calculation. Especially as the core is the main source of heat here.

The material you suggest is electrically insulating and has reasonable thermal conductivity. There are thermally better materials available such as 3M TC-2810 which is about 2x as good.

It's possible to predict the heat flow if you have access to appropriate thermal analysis software (often mechanical FEA programs will do this). An example would be MSC NASTRAN. However, cost and learning curve are both steep.

You can easily measure the temperature rise of a built inductor by measuring the resistance delta of the coil, since the temperature coefficient of copper is known. You would turn the power off and make a measurement before the temperature has a chance to change much. That is a standard method used for transformer temperature rise measurement. You can also attempt to embed sensors in the epoxy, however it's difficult to get a measurement during operation if there is a lot of EMI floating around and the wires close to the high voltage may present other hazards.

Edit:

You could fill the potting housing with salt water, with a copper plate at the position the heatsink would be in, and wrap the toroid in foil, immerse to the expected position and measure the resistance (use AC). Take the ratio of that to the (measured, with AC) volume resistivity of the salt water gives units of length. Multiply that by the thermal conductivity and you get W/°C.

• I'm intimately familiar with Patran/Nastran, but I really don't feel like modeling this just for feasibility calculations. I was kind of hoping for a good 'rule of thumb'-type approach to get started. Thanks for the 3M material link, that helps a lot! Commented Jul 27, 2015 at 15:41
• There is a lot of symmetry in the configuration, and the epoxy is a pretty good insulator compared to the metal, so most of the heat flow will be from the heat sink side of the outer part of the torus (including winding) to the surface of the heatsink. Guesstimate spacing from outer part of toroid to heatsink + 50-100% and calculate area and average thickness might get within 25% or so. Commented Jul 27, 2015 at 15:49
• Your edit about the salt water idea as well as the other answer are giving me ideas on how to attack this problem experimentally. I think it's going to be faster to just desolder the inductor, 'pot' it and do some synthetic measurements than it's going to be to do proper FEM. I think I'm going to mark this as answered, this should suffice. Commented Jul 27, 2015 at 16:09

My approach in these situations is to embed thermocouples in the device, then actually running it under operating conditions and log the data.

I would glue one thermocouple on the core (either between or under the windings, another thermocouple on the outside of the windings, plus any other thermocouples that might provide useful information. These last thermocouples are obviously device specific and only you can decide where and how many are required.

We make our own thermocouples from bulk 28 AWG K-type thermocouple wire. You can either crimp or solder the conductors at the measuring point - we have a special solder that adheres to both conductor materials in K-type thermocouples (Chromel & Alumel) which is suitable for temperature measurement up to the melting point of the solder.

We use this thin thermocouple wire to minimize the temperature measurement errors caused by the thermocouple conductors bleeding heat away from the measurement point.

You should also monitor and log the ambient temperature as well as any other thermal barriers or heatsinks that are part of the thermal path.

If you have the time and resources, you can use this technique to evaluate several different epoxy compounds. Simply make as many prototype devices as you need and evaluate each of them.

You can probably calculate and arrive at an estimate of the epoxy performance but nothing beats direct measurement if this is possible.

For what it's worth, we use the epoxy that you link to for our hazardous-location cabinet heaters Trinity Electronics Cabinet Heater with good results. We used exactly the approach that I described above to choose that particular epoxy compound out of the several products that we were evaluating.

• At the moment, I have already done thermal measurements with both a thermal camera and a thermocouple in free air. Unfortunately, in still air the temperature never stabilizes and the experiment needs to be shut off within 2 minutes to avoid uh... destructive interference :P Experiments are definitely an option, but it's quite annoying to test multiple compounds as we only have 2 test setups (so we need to remove the potting gunk every time we want to re-test). Commented Jul 27, 2015 at 15:46
• Ouch! I consider my test prototypes to be disposable and don't worry about trying to reuse them with different epoxy compounds. The prototype cost is usually much less than the labor cost of removing the epoxy from the previous test. Commented Jul 27, 2015 at 16:02
• Unfortunately, these tests are done on devices that we get externally and that can take weeks to replace. It's a shitty situation, which is why we're trying to go the modeling route wherever it's convenient enough. Commented Jul 27, 2015 at 16:04