TEC Failure

[fourfathom]

I have one of those "reptile incubator" heater/cooler units -- the one that Dave was using for a thermal test chamber ().  I modified it to to better control the TEC and the fans, using an external controller and temp sensor, disconnecting the TEC and fans from the built-in controller. Normally I have a proportional heat / cool cycle, with a reasonable dead-time between heat and cool modes.  But for this test I had my controller just run in bang-bang mode, switching instantly from heating to cooling and back.  It was working well, and then I left it running for about one month in this mode, trying to hold 20 deg C.  The TEC is now dead, drawing only a few tens of mA with 12V applied in either polarity.

So, did I kill the TEC through abuse, or should it have been able to handle this type of cycling?  Would an "off" time between full-heat and full-cool made a difference?



Why did I make my own controller?  The stock unit has about a one-degree C ripple as it cycles to maintain temperature, and it turns the fans on and off as it cycles.  I wanted a more stable temperature and constant internal airflow.  I also wanted remote software control and logging.

[Gyro]

Yes, thermal fatigue is a thing with TECs. They have lots of low temperature solder joints in them for one thing. Such rapid switching of heating/cooling modes would almost certainly accelerate the effect - I'm actually pretty impressed at 1 month of such rapid cycling.

You could maybe try heating up the TEC evenly with very little pressure applied. The trouble is, what temperature to use to reflow / relax the joints without killing the elements, and how much pressure. The elements will all be slightly different lengths and you need the solder to bridge the gaps rather than being squidged out - a bit like BGA mounting. I've no idea what type of flux might be compatible and whether it would help reflow the joints.


P.S. you might find it more successful to use resistors for heating and the TEC just for cooling.

[fourfathom]

Thanks!  I'm probably going to buy another incubator rather than attempt to find a replacement for the TEC element.  I like your idea of using resistors for the heating, and this would actually make the drive circuit simpler: rather than needing a bridge to switch polarity I can just have single-ended switches.  And, if I size the resistor properly I can better-match the heating and cooling rates (the TEC is a much better heater than it is a cooler, so tuning the control loop is more of a hassle).  I hope I can find my notes showing how much current the TEC drew, but I do have temperature-change rate curves so I can always work from those.  Anyway, it's a good opportunity to clean up my controller design.

[Gyro]

Yes, the TEC makes a massively better heater. Last time I tried it that way, it was really difficult to get any sort of sensible heating / cooling balance with a simple controller. You can compensate to some extent by using a diode bridged series resistor, but it's never going to work well. Straight resistor heating is far easier (and it's much gentler on the TEC, fatigue wise, of course).

[magic]

Couldn't it work with some sort of linear control instead of hard switching at full blast?

[Gyro]

Oh yes, certainly. In fact I think I remember reading somewhere that TECs work better with linear current rather than PWM... at the expense of controller dissipation of course!

[fourfathom]

I'm doing duty-cycle control now. With a 10 Hz clock rate. Does anyone know if a faster cycle rate would be better? I'd rather not go full- linear, or use LC filtering

[nfmax]

Just don't use on/off controls with TECs. Not only is it bad for them, but as the I²R heating is proportional to current squared, while the Peltier effect is proportional to the current, they operate less efficiently. Of course, you can always use PWM at a few tens of kHz, with an LC filter to eliminate the switching cycles and give you DC drive.

[T3sl4co1l]

Yeah, proportional control -- PWM into an LC filter is preferred.  If you need bidirectional operation, use an H-bridge directly*, or a controlled current (PWM into LC) into an H-bridge (that's only switched occasionally, to toggle polarity).  I did one of these some years ago, for a dewpoint control, for a major paint-mixer machine supplier (of all things!).

*In which case, treat each leg as its own PWM output, LC filter each.  Why?  If you leave one side pegged to a switching node, you get the full switching noise on all the wiring.  NFG!

Or, switch the one side only occasionally, set PWM on the other side to D or 1-D depending on polarity, and use just the one LC -- this still lets through full switching noise when changing polarity, but that's probably not very often, so is acceptable.  Note the LC filter also doesn't track instantly (it's referenced to GND), so there's a upset when changing direction.  This would suck for most applications (e.g. mains inverter, class D audio amp; maybe motor driver), but doesn't really do anything to the TEC, it's just getting blasted with whatever charge was in the cap.

This probably doesn't make much sense without a diagram, but eh that's okay, if you want to know more just ask.

Tim

[fourfathom]

This probably doesn't make much sense without a diagram

No, it makes perfect sense.  Thanks. 

If I go with the resistor heating element I can use single-ended PWM / LC drive for the TEC (cooling only).  I just have to find a suitable inductor or core -- all the inductor-stuff I generally use is for RF, but I might have some cores somewhere that will be good in the KHz region.

I can use raw PWM for the resistor, I can just switch at a low rate (but faster than the thermal time-constant of the resistor), and do some easy RF filtering if I decide I need it.

Thinking out loud, since I already have a bridge and isolated driver circuit, and it's good well-into the KHz region, I might as well use the TEC for both heating and cooling, nailing one side of the bridge to + or 0 as needed (as you suggested).

[langwadt]

Yeah, proportional control -- PWM into an LC filter is preferred.  If you need bidirectional operation, use an H-bridge directly*, or a controlled current (PWM into LC) into an H-bridge (that's only switched occasionally, to toggle polarity).  I did one of these some years ago, for a dewpoint control, for a major paint-mixer machine supplier (of all things!).

*In which case, treat each leg as its own PWM output, LC filter each.  Why?  If you leave one side pegged to a switching node, you get the full switching noise on all the wiring.  NFG!

Or, switch the one side only occasionally, set PWM on the other side to D or 1-D depending on polarity, and use just the one LC -- this still lets through full switching noise when changing polarity, but that's probably not very often, so is acceptable.  Note the LC filter also doesn't track instantly (it's referenced to GND), so there's a upset when changing direction.  This would suck for most applications (e.g. mains inverter, class D audio amp; maybe motor driver), but doesn't really do anything to the TEC, it's just getting blasted with whatever charge was in the cap.

This probably doesn't make much sense without a diagram, but eh that's okay, if you want to know more just ask.

Tim

the H-bridge with each side PWM'd is basically two sync buck converters, the difference in voltage on the two is across the TEC

e.g. MAX1968 

with only one side PWM, basically a sync buck that either sinks from VCC through the TEC or source to GND through the TEC

[thm_w]

Thanks!  I'm probably going to buy another incubator rather than attempt to find a replacement for the TEC element. 

Unless its hard to remove seems kinda overkill, new TECs are a few dollars.

https://www.aliexpress.com/wholesale?catId=0&SearchText=tec

[fourfathom]

Unless its hard to remove seems kinda overkill, new TECs are a few dollars.

I had assumed it would be hard to find -- not so! I've already removed and measured it (40mm X 40mm) and found the same part# on Amazon.  Arriving on Monday.

The mech engineering on that incubator is pretty sad.  For example, the main heatsink fan is held in place with big screws that sort of wedge between the heatsink fins.  Oh well, what should I expect for so cheap???

[coppercone2]

stick a thermostat on it as a safety measure if it shorts out a diode and starts heating

TEC refrigerators are built like that too, screws in fins.

[fourfathom]

I might end up with a linear output stage on my TEC supply.  It's going to be hard to do PWM faster than 1 KHz with what I'm using, and the inductor required to filter that is pretty big (ballpark 100 mH, 5A).  A regular pass-transistor will dissipate 15W max (6V, 2.5A), and I can build this with my junkbox parts.

[coppercone2]

All I can say is that its already environmentally unfriendly in terms of efficiency and a LDO is not doing it any good.. but this is 15W for science

That inductor should be around the size of half a coke can or so

[Hydron]

I've got the bits from a "black body" (for thermal imager calibration) sitting in a box here, and it's "low-temp" output uses TECs to provide the ability to go above/below ambient with tightly controlled temperature, which sounds similar to what you want to do.

It has a big array (like 8 or something) of large expensive power opamps to drive the TECs linearly, along with heatsinks, a big loud fan etc. Seems that was judged the easiest way to run them reliably, and I guess the cost wasn't a concern.

I was hoping to repackage it in a sane sized enclosure (rather than the multi-enclosure 15kg beast it used to be, though that included a high-temp heat-only side as well), but I think I'd need to ditch the linear drive module, plus probably the power supply (a huge toroidal + caps) running the mess, and replace with a PWM+filter arrangement. Maybe possible to get an amplifier that can be repurposed off ebay/aliexpress?

[langwadt]

I might end up with a linear output stage on my TEC supply.  It's going to be hard to do PWM faster than 1 KHz with what I'm using, and the inductor required to filter that is pretty big (ballpark 100 mH, 5A).  A regular pass-transistor will dissipate 15W max (6V, 2.5A), and I can build this with my junkbox parts.

look in the junk box for a buck converter, since TECs should be driven with constant current an LED driver maybe

[Siwastaja]

TECs should be driven with constant current

They are so much resistive that constant voltage is just fine, too.

LC filtered PWM aka open-loop buck is fine, too, and one of the easiest drive strategies that still give good efficiency. Transistor peak current limit is sensible thing to add, though, to avoid going overcurrent (possibly even saturating the inductor) during startup transient and setpoint changes, because even if the load itself is resistive, the capacitor of the LC filter is AC short circuit.

As others have mentioned, raw PWM at any frequency is not really good due to poor efficiency and mechanical stress.

Poor efficiency is the result of the TEC efficiency being a function of input current. At "full power", efficiency is worse; resistive losses go per I², while cooling power goes per I¹. If you PWM it, you always get this poor "full power" efficiency, even at partial power.

[fourfathom]

I did some searching on driving a TEC with PWM, and while I found mentions of PWM causing TEC inefficiency, and "concerns" about PWM causing thermal-cycling damage, I also found a paper where varying PWM rates (50% duty-cycle, from 0.1 Hz to 10KHz) were used and the TEC characteristics were monitored during the test.  Regardless of the PWM rate they found no significant TEC resistance variation, and no actual damage after 2328 hours of testing.  Here's a link to the paper: https://tetech.com/wp-content/uploads/2013/10/ICT99MJN.pdf

But I took a look at the PWM capabilities of the SAMD controller I am using and saw that it's pretty easy to create a 100 KHz - 1 MHz PWM output.  This means I can use a reasonably-sized 1mH inductor (and 10uF cap) to smooth the PWM output, so I don't have to use a linear variable-voltage supply if I want to supply the TEC with a DC voltage.  I don't even need to use voltage feedback, since I can get close enough by just using my temperature sensor and software to control the pass-transistor duty-cycle.

[Siwastaja]

I have one system where I use raw PWM for TECs as well, and it's still operational. But it does not see large number of operating hours.

Just choose your bias, and you'll find examples of TECs not failing due to PWM, or TECs failing due to PWM.

Obviously, you should not run them with raw PWM because the risk is significant even if not 100% failure rate. Trying to push your luck with something that is well known to cause issues is wasted time.