The milling solution
One of the limiting factors for Peltier systems is the thermal contact between the Peltier’s outer ceramic and the rest of the thermal system, such as an aluminium cooling block as shown. Thermal paste is the go-to material, but in reality this is hit and miss and even under ideal conditions, can reduce the system’s potential by around 10-20% due to relatively high thermal resistance of the paste. In experimental set ups, it’s not unusual for things to slide around or detach while trying to set things up, adding variability and uncertainty to the experiment, and risking a “bad joint” by introducing air. In long term applications the thermal paste often dries out. And in general I just hate working with the stuff, it gets everywhere.
A typical “cell” with the Peltier glued between two water cooling blocks
After experiments, it was found that sanding the aluminium blocks flat (since they are often not particularly flat when purchased) and then sliding the Peltier on to the blocks with Japanese water based glue (nori) worked well. The nori can be spread very thin and doesn’t lock into place like other glues, allowing it to be aligned and adjusted. After fixing, a “cell” (one Peltier with cooling blocks on each side) is clamped and allowed to dry for at least 12 hours.
This sanding/nori technique was the first time to get some truly cool results (pun intended), with cells clocking in at a COPc over 10 at low power, and COPh of 2 - 3 at useful power levels.
This method then unleashed a new wave of applications such as daisy chaining 7 cells to create a 200W heat pump, the 15min beer cooler along with many other advancements in understanding Peltiers, modelling and measurement.
And yet, the method also turned out to be a bit hit and miss, with measurements indicating the total thermal resistance ranged from 1.8 K/W to 2.4 K/W (Kelvin per Watt is a measurement of thermal resistance, lower is better). For serious heat pumping, that extra 0.6K/W is an application killer. It’s suspected that despite the best efforts, air bubbles and other defects still occasionally got in the way of good contact, and since each “cell” has two thermal joints, a 20% failure rate for a single joint would lead to a fairly high chance that at least one side in a cell is dud. In a batch of 10 cells, I was getting around 3 that were good, 5 that were OK and 2 complete duds.
A cell based on a milled out contact area (second side ready for final assembly)
While contemplating various ideas to improve this method, I thought: why not skip the aluminium contact altogether and just allow the water to contact the surface directly? It’s not an entirely new idea, but by my earlier attempts failed to understand that the water needs to have turbulent contact with the ceramic. Water is, surprisingly, a good insulator so its ability to pick up heat quickly relies heavily on movement. The M shape inside the aluminium cooling blocks is there for the express purpose of creating turbulence and increasing chance of each water molecule bumping into the hot (or cold) part; any interface with the ceramic would need something similar. Why not just mill out the aluminium and allow the water to contact the ceramic directly?
And it worked!! The test results were slightly better than the best “nori” cells, but the results were consistently good for every cell made so far.
The main benefit of the method is that it now produced a reliable contact between the water and the Peltier, without having to worry about a nervous gluing process that only has a 80% success rate.
The disadvantage: at the moment hand milling (Hozan K-280 X/Y bed) takes about 40min per side, so about 2 hours to make a single cell.
I tried a CNC machine (Genmitsu 4040-PRO) but the 75W motor is too small for this kind of work, the Hozan 200W is much better suited.
The next step may be 3D printing a shell with the M shape, although it’s not clear if aluminium still helps to move heat to (or from) the water.
Since the cells take so much effort, and good quality Peltiers are getting more expensive, it makes sense to eliminate the other weak point: the wire connection. This is especially important for TEC 12710 cells that have only 0.6Ω resistance in the cell, and need relatively high currents for normal operation, so a dud electrical joint can have a major effect on efficiency.
Cell with separate terminals to relieve stress on the Peltier solder joints
Completed cell, the pine wood also helps to insulate
Test results:
TEC 12708 based cell (rated at 8.5A, 1.5Ω):
Very low level, 0.5A input: 5.15W cooling with just 0.370W input, COPc = 13.9 (yes, this is true!)
Low level test at 1A input: 10.0W of cooling, with 1.48W in (COPc = 6.8)
Mid level test at 3A input: 27.0W cooling with 13.3W in (COPc = 2.0)
Implied thermal resistance: 1.78 K/W (hot to cold)
TEC 12710 based cell (rated at 10A, 0.6Ω):
Crazy low level test, 0.5A: 3.70W cooling, 0.170W in (COPc = 24.1)
Mid level test at 3A input: 19.64W of cooling, with 5.415W in (COPc = 3.63)
High level test at 5.72A input: 33.0W cooling with 20.0W in (COPc = 1.65, COPh = 2.65)
Implied thermal resistance: 1.81 K/W
These tests are with the hot and cold side at the same temperature and the low level tests, while amazing, aren’t often useful in most applications. They do however help for comparative testing and to confirm the cell is working OK. Real world applications need the higher currents and have to deal with temperature differences which leads to much low COPs, but still useful. For example the TEC 12710 operated at 5.72A with 20K difference between hot and cold is still expected to have 20W cooling with 20W input (COPc = 1.0, COPh = 2.0). This kind of COP is on par with real world air conditioning units that also need to deal with large temperature differences, such as in winter.
Although 20W cooling may seem small, for a modern fridge with good insulation this is often enough. For higher power applications the cells can be daisy chained and it is expected that 400W heating with 200W input is possible with 10 cells, even with temperature differences of 20K, making it suitable for gentle, constant warming of a mid sized room in winter.