Peltier cooling - weird behaviour

[jeremy]

Hi everyone,

I'm trying to build a peltier chamber which goes down to -20 C. I'm ignoring the control aspect at the moment and just powering them with DC supplies (if I can get them to easily go below -20C, then I will add control). I have two "12710" peltiers from Aliexpress and a Corsair CPU AIO watercooler with a single fan. The peltiers are powered by independent power supplies, and they are mounted on a die cast aluminium enclosure so that they are thermally in parallel. There is an aluminium bar (maybe 5mm thick) covering the hot side of both, and then the water cooling block is attached on top of the aluminium bar. The watercooling block uses the stock thermal paste, and the peltiers both have thermal pads (I think roughly 1mm) on each side. The die cast aluminium box (which is roughly 140x120x25mm) is surrounded by about 40mm thick of closed cell insulation foam, except for the water cooling pump area where I tried to seal around it as best as possible. I also have two thermocouples mounted on the inside of the die cast enclosure taped down with kapton roughly in the location of the coolers.

Presumably, these coolers should be run at 12V 10A. However, if I run them at 12V, they draw about 3.5A or so. I have also found that they actually perform much better at 8V 2.4A, where my setup gets down to nearly -2C in a 25C ambient room. If I increase the voltage to 9V (~2.8A) or 10V (~3.1A), there is a quick burst of cooling (maybe an extra 0.5C or so) and then the performance gets worse and the whole system begins warming up slowly. Reducing the voltage back down improves the situation.

Can anyone help me understand what is happening with the transient behaviour here, and why the current draw is so low? Do I just have dud peltiers? Also, does a cooling limit of -2C for 25C ambient sound about right with 40mm of foam and 2 peltiers running at 20W each (although they should be running at 120W each...)? I was hoping for a little better than that. From what I understand the theoretical limit is about 70C difference between both sides, so I am just over 1/3 there - though I am not sure exactly how hard it is to approach the theoretical limit.

[wraper]

Don't trust Chinese specs, and specs from reputable source too. Maximum power spec is very far from useable. Peltier modules rapidly decrease efficiency with increasing the power applied. So they will start to dissipate so much heat you can hardly cool them enough. Also you may want staging them for achieving lower temperatures (cold side module needs less power applied to work the best). You may see something like 70^oC temperature difference at max power spec, but don't forget you need to keep the opposite side at constant temperature to see such an effect. In practice at some point you will have an opposite effect since the hot side temperature will raise more than temperature difference across the module.

[wraper]

40x40mm modules need around 16V for 10A current.

[metrologist]

My first play at these devices yielded something similar. I had to revisit some heat transfer principles as the main issue was not enough heat transfer out of the system. The specs provided are nom operating and maximums, so you will not get 120W at 12V.

You will want to become familiar with the datasheet and your device performance curves.

https://datasheetspdf.com/pdf/634370/HB/TEC1-12710/1

This site helped me: https://curiousscientist.tech/blog/how-to-read-peltiers-performance-chart

[Conrad Hoffman]

As said above, it's hard to get the theoretical maximum out of a TEC because sufficient cooling is nearly impossible in the real world. Start thinking solid copper mounts with drilled water channels, or two pieces bolted together with a serpentine channel pattern milled into one.

[mzzj]

Kryotherm software from the russian manufacturer of TEC's is best thing since sliced bread https://kryothermtec.com/
You can pick up some TEC module that is close to chinese specifications and expect to get 30% worse results than that.

But even without diving into details and numbers I can tell the same that others have already mentioned: hot side cooling is always the weak link.
If you think you have huge heatsink on hot side you should double it. 

[metrologist]

If this is of interest here is my testing of a similar device. Best I was able to do is make a 20ml ice cube from ambient temps in about an hour.

https://www.eevblog.com/forum/beginners/how-to-read-this-peltier-device-chart/msg4354738/#msg4354738

[TheMG]

The 5mm aluminum you're using to carry the heat from the Peltier modules to the water cooling block is very likely hampering the cooling. Aluminum is not the greatest heat conductor, copper is better, heatpipes or direct contact with the water block even better.

Use thermal paste, not pads.

Peltier modules have a limited delta-T they can achieve, so the cooler you can keep the hot side, the colder the cold side will get.

[jeremy]

Thanks everyone for your suggestions.

The watercooling block is already a copper block with skived microfins, so I don't think that is the limitation. I will try replacing the pads with paste - I was just worried about putting too much pressure on the peltiers and cracking them with thermal expansion, but the watercooling block has springy mounting points that I didn't notice at the start so hopefully that will be enough. I will also see if I can find a piece of copper to use instead of the aluminium as the heatspreader. I also think I need to put a thermocouple in the hot side block so I can at least know what the temperature difference is.

FWIW, I just measured the system again and at 8V 2.4A * 2 it takes approximately 17 minutes to go from 25C to 0C.

[wraper]

Do you think water cooler is able to keep ambient temperature at hot side Peltier module? If temperature rises to something like 45^oC, it's already basically impossible to achieve the desired temperature. And temperature will rise to this level with power dissipation you use. It's like looking on 150W power dissipation figure for MOSFET in TO-220 package and designing your circuit to use that. Yes, it's theoretically possible, but good luck keeping its tab at 25^oC. Look at some PC liquid cooler testing and temperatures they get. It will become obvious it's not likely to achieve -20^oC the way you are doing it. However using thermal pad on a hot side will completely screw the performance.

[Circlotron]

Use thermal paste, not pads.

^ This.

[amyk]

Peltier coolers are extremely inefficient. Unless your requirements specifically prohibit it, I would go with a mini (compression cycle) refrigeration system.

[Kleinstein]

The peltier modules are really demanding on the thermal contacts, especially on the hot side. The thermal pads are just not good enought / too thick. The main point of the paste is to desplace air - onece that is done,  less paste is bettter. The thermal paste is still pretty poor thermal conductivity compared to the ceramics  - it is just much better than air and can be used as the really thin layer.
Once attached there is no need for a high pressure / force - as a thin layer the paste gets more like sticky clay.

The effect described matches what is expected from poor cooling of the hot side.
The modules have a current where cooling is most efficient. For the ideal case with perfect thermal contact this gives the nominal current / maxium current. With added thermal resistance the current at which the best cooling is achieved will shift down to lower and lower current. For a short time a higher current works, but after while the hot side gets warmer and this counteracts the posive effect of more current.
The best current for the setup gives a good hint on how good the thermal contract is.

[mawyatt]

Peltier coolers are extremely inefficient. Unless your requirements specifically prohibit it, I would go with a mini (compression cycle) refrigeration system.

That's the truth!!

Long ago (~1980) when developing a Remote Sensing Chemical Agent detector (XM21 later called M21) we had planned on using a pyramid of stacked Peltier TE devices to cool a HgCdTe detector on a cold-finger to 80K in a closed loop system, ended up needing 6 levels of stacked TE devices and massive amounts of raw power (well beyond what the TE OEM had indicated). We changed the original design concept to a miniature Split-Cycle Stirling Cryogenic cooler which was developed under the program that proved much more efficient.

Best,

[jeremy]

Just a small update: I replaced the thermal pads with arctic silver thermal paste, added a third peltier (the driver I wanted to use outputs 24V, so 3*8V = 24V), changed the aluminium heatspreader to a 2mm copper sheet and changed the radiator to a slightly larger AIO cooler. For 24V @ 1.3A I get to -10C in about 25 minutes from 26C ambient, and the hot side copper heatspreader rises to about 38C. I still don't really understand why the elements are running at close to 1/5th their rated current (other than that they are just mislabeled) but -10C is probably good enough for my purposes so I will leave it at that I think. I can probably just put the radiator in front of the air conditioning outlet to get an extra 10C or so anyway.

Total power usage including the water pump/fans/peltier driver is about 35W. Total cost of the system not including the PC watercooler or power supply (though it's just a 24V DC supply) is much less than AUD$50.

[Kleinstein]

38 C at the hot side is still quite hot. Chances are the 12 K drop to ambinent would be an area for possible improvements. The temperature difference from +38 C to -10 C is already quite high.
Really reaching -20 C would likely need a 2 stage design, as the elements are usually not made for such a large difference.


The rated current is for a best case (ideal heat removal) scenario. It is normal  to get better performance with somewhat lower current, like -10% or -25%.

[wraper]

I still don't really understand why the elements are running at close to 1/5th their rated current (other than that they are just mislabeled) but -10C is probably good enough for my purposes so I will leave it at that I think.

I already said you need about 16V for such current.

[jeremy]

Yep, I am aware that I could probably go lower with a 2 stage setup, but I only have single stage elements available at the moment, so if I stacked them I would need to drive the different stages at different voltage/current and that makes the controller a bit more complicated.

I’m struggling to come up with ideas as to how to close that cooling gap to ambient (without more stages), I guess I would need a lot higher water flow rate, and maybe a larger water block (although the microfins should have a lot of surface area).

The only thing on the cold side is maybe the 4x M3 stainless screws that are holding the sandwich together - they are in contact with the cold plate. But I’m not sure I trust plastic screws here. I might try to add some washers.

[jeremy]

At 16V I get about 3A in the elements.

[wraper]

Then those elements most likely are counterfeit.

[Siwastaja]

Wrong current is simply due to wrong specifications, you did not get what you ordered.

Poor performance is due to insufficient hot side cooling (as always), maybe combined with unrealistic expectations about TEC modules (rated cooling power only applies at zero delta T, and rated delta T applies at zero cooling power. Interpolate linearly in-between), and maybe also because of poor module quality (random junk).

1mm thick thermal pad was the biggest initial mistake as you found out. Even the best pads are something like 20W/(mK), random generic ones more likely 2-4 W/(mK). Compare that to aluminium (around 200) or copper (around 400)! Extremely thin layer is the key. High quality (15-20W/mK) pad might work out if you buy 0.2-0.3mm thickness. You can of course do the math.

[wraper]

Looks like TEC1-12703 sold as 12710.