Very stable temperature control

[HendriXML]

My room temperature varies way to much to be useful. 

Does it? OK, if you have a room that gets lots of sun, or leave the windows open, but the OP asked for "20min" stability, and over that time i suspect the actually air temperature doesn't change that much.  Find a shady corner of your house, put the device there, and i see no reason to chase complex PID controllers etc, just put on a couple of watts of heating, leave it on till the temperature reaches stability and you're done!

You know how it goes, first fluctuations within a degree are OK. But then engineers want to go better. For most people it is: if it aint broken, then don't try to fix it.
For engineers it is: if it aint broken, it hasn't got enough features.

[Rerouter]

If we haven't reached our budget or a hard limit, why stop now. 

In reality the stacked peltier system joeqsmith has built is approaching what I am hoping for with one of my projects, but he really has it locked down to laboratory reference level of stability.

[HendriXML]

So I did some testing. With the 10uF capacitor it kept oscillating slowly, never reaching a balance. I could have then turned down the reactivity, but I went for just wacking 5 more capacitors in parallel. Nearing the 63 uF I "analyzed". Now its stable after just a few ups and downs. The wattages varies only a few % in less than a second. Not having the proper tools to measure the temperature variations, I need to make a guess.
With no power the temp drops 0.006 deg per sec. If the average power is stable, the 1% wattage change would correspond with 1% of 0.006, thus 0.00006 deg.
It won't have this low variation in reality, mainly because of the used components and reference voltage stability.
However I think it will do well under better testing. Not because anything special in the PID, but having a lot of thermal mass. Water is special. Also I found out the heatsink consists copperplates (if the specs are correctj, so the heat transfer to water might actually be very good.
https://www.engineeringtoolbox.com/specific-heat-capacity-d_391.html

[HendriXML]

I havn't had an eBay account for at least 10 years now but thought I would have a look on eBay for cheap used bench meters.   Now I'm glad I closed my account as I would have been in a world of trouble buying some this old vintage gear.   May have had to join that TEA thread to recover.       There are some decent meters out there, and some very expensive junk.

Buying an uncalibrated (used) one would lead to the continuous question whether the meter is accurate or not, if one doesn't own another accurate device. So I don't even check eBay stuff. Maybe in the future I'll buy something like the Siglent SDM 3055. But 6.5 digits would be nicer..

[joeqsmith]

Buying an uncalibrated (used) one would lead to the continuous question whether the meter is accurate or not, if one doesn't own another accurate device. So I don't even check eBay stuff. Maybe in the future I'll buy something like the Siglent SDM 3055. But 6.5 digits would be nicer..

This is why we have cal houses.    I bought two brand new HP bench meters for home.  Based on my experience with them, if I ever have a need to upgrade (doubt) or replace one (possible after that last lightning strike), I will buy one from Keysight.  Say I spent $1000 USD.  I've have had the 34401A for well over 20 years now, so $50/year.  That's what, a trip to the bar, dinner,  wife's trip to having her nails and hair done ...      I've always looked at my hobbies as an investment in myself.   

My room temperature varies way to much to be useful. 

Does it? OK, if you have a room that gets lots of sun, or leave the windows open, but the OP asked for "20min" stability, and over that time i suspect the actually air temperature doesn't change that much.  Find a shady corner of your house, put the device there, and i see no reason to chase complex PID controllers etc, just put on a couple of watts of heating, leave it on till the temperature reaches stability and you're done!

OP's title is "Very stable temperature control".   I'm just offering my perspective.   

Your "change that much" is subjective.  For you, maybe that means a degree.   Trying to control down to 0.001 C, small air currents and body heat can be a problem.   Again, a basic beach towel can combat a lot of that but its not going to get you to the level I am showing.   Automating the system can be helpful for running experiments.   

[joeqsmith]

If we haven't reached our budget or a hard limit, why stop now. 

In reality the stacked peltier system joeqsmith has built is approaching what I am hoping for with one of my projects, but he really has it locked down to laboratory reference level of stability.

My advice is BIG heatsinks and more Peltiers and lots of copper spreaders.   You can never have enough.     

[max_torque]

and how are you measuring this "0.001" degC ?

[David Hess]

So I did some testing. With the 10uF capacitor it kept oscillating slowly, never reaching a balance. I could have then turned down the reactivity, but I went for just wacking 5 more capacitors in parallel. Nearing the 63 uF I "analyzed". Now its stable after just a few ups and downs.

That indicates to me that a zero needs to be added to the transfer function like a resistance in series with the integration capacitor.

The wattages varies only a few % in less than a second. Not having the proper tools to measure the temperature variations, I need to make a guess.

The output of the error amplifier can be measured to tune the transient response.

[joeqsmith]

and how are you measuring this "0.001" degC ?

From the pictures in the links I provided, you may notice a sensor located pretty much in the center of the box.  There is a fair bit of air flow inside and it should have little dead volume.   The sensor is a platinum RTD that is read through my HP 34401A bench meter.   The PC reads the meter and performs the conversion.   The conversion is supplied by the manufacture.   Keep in mind that my goal is not to have absolute accuracy but rather short term stability, followed by repeatability and then accuracy.   If you have specific questions, feel free to ask. 

***
I forgot to mention that normally when I am running experiments where I need stable temperatures, I will normally have other test equipment near by controlling and collecting data.   All of that equipment also creates air currents and heat which can cause problems with my measurements.  I expect OP may have a similar problem.   

***
A link showing some of the basic differences between sensors:
https://www.omega.co.uk/temperature/z/thermocouple-rtd.html   

[HendriXML]

and how are you measuring this "0.001" degC ?

From the pictures in the links I provided, you may notice a sensor located pretty much in the center of the box.  There is a fair bit of air flow inside and it should have little dead volume.   The sensor is a platinum RTD that is read through my HP 34401A bench meter.   The PC reads the meter and performs the conversion.   The conversion is supplied by the manufacture.   Keep in mind that my goal is not to have absolute accuracy but rather short term stability, followed by repeatability and then accuracy.   If you have specific questions, feel free to ask. 

***
I forgot to mention that normally when I am running experiments where I need stable temperatures, I will normally have other test equipment near by controlling and collecting data.   All of that equipment also creates air currents and heat which can cause problems with my measurements.  I expect OP may have a similar problem.   

***
A link showing some of the basic differences between sensors:
https://www.omega.co.uk/temperature/z/thermocouple-rtd.html   

OP needs only a square cm for his diode, so his task is easy. I've put a 10 cm foam on top, so above the heatsink there's only a tiny room of air. The diode and sensor are glued to the heatsink, so most of the heat will be taken from that. I guess the air might be a tiny bit cooler.
If I where to scale this setup to a larger box, I would probably have a box in a box where the inner box would be surrounded by water. The outerbox would then have styrofoam isolation on the outside. Then have maybe the outerbox made of alu or something with heaters to keep the water at the requested temperature. Maybe with dual heaters, one water boiler 2300W for fast heating, and one much less powerful to keep it at the target temperature. Maybe throw in a waterpump as well, pumping water trough copperpipes mounted on a isolated alu lid. Inside the box the temperature will eventually be the same as the water temperature.
In the inner box, there also should be a fan to move the air around.

Maybe one day I'll have a purpose for this idea.

I assume that the peltier aficionados know the following vids/channel:
https://youtu.be/aSGwaQWA05U
I really like the stuff they're doing.

[joeqsmith]

I was asked about the control system in a PM.  Would rather keep discussions in a public forum so others may benefit.

I posted about how I monitor the temperature in my previous post.   The PC uses this feedback to run a simple PI controller (D set to zero).  Output from the PI drives a power supply in voltage mode which in turn powers the Peltiers.   The voltage is limited to a safe operating area.   Similar to the video posted, the Peltiers are different for the two stages.  Lower power on the cold side.   It's nothing too complex.

If the goal is to run below -40 with any sort of volume and loading, I would still suggest a more conventional system.   Faster ramp times, more efficient, maybe longer life depending how you plan to use your system.  The datasheets for the Peltiers talk about their life and how they are cycled. 

[HendriXML]

The best way to use this setup is to start with boiling water and let it then cool down. The regulation will than eventually kick in. To keep the temperature stabile it needs about 2 W

if you have the temperature at x watts, you can likely estimate the RC time.
overall heat leakage (water temp - ambient temp) / watts = "C/W"
the water "thermal mass" = ((liters)*4200) = J/K
the container also have a thermal mass est about 100s of J/K
the heater and circuit also have a mass est about 50s of J/K
the 1RC time is approx = "C/W value" x "J/K value"
look like the RC is going to be in 000s of seconds

000s second damped op amp? 2RC ? or faster? 0.5RC ?

I had to think about this for a while. I see the usage of the capacitor as adding a way to store state. In this circuit it should store the avg difference voltage that is needed to counterbalance the temp leakage. By oscillating a few times it will approximate  that voltage. This will happen without problems because the heating power is low to the thermal capacity and the cooling power is even lower. Except when the capacitor gets near its max and min voltages too soon while heating and cooling. When that's the case it's "memory function" is not entirely effective anymore.
So I guess the calculation of the value of C should mainly take that into account. It is like the rough analysis I did before, but without the conclusion is should be much less than 68uF. It should be around 68uF, so the  the capacitor is kept in a more effective range of it's charging and discharging curve.
So I think half the capacitor voltage at half of the temp difference would be a good rule of thumb, when doing exact calculations -due to missing parameters- is impossible. There is for example unknown thermal resistance to the thermal mass. (And the most power will be used to heat the mass a bit more.)
If my reasoning seems broken, please let me know.

[HendriXML]

Also I think the circuit can be improved by hooking up C3 after R16. Charging C3 with more voltage than what will ever be used by the following stage is not helpful in getting the balancing voltage in that cap.
It wouldn't surprise me if that would make a 10uF as C3 stabilizing as well.

[HendriXML]

The modification did not mean only a 10uF cap could be used. It wasn't the heating that unbalanced things. It is the slow spread of heat to thermal mass in the cooling down cycle part. That part of the cycle is also the part that was taken in the RC analysis. Don't know whether the analysis was correct, but it seems to work.

[Kleinstein]

Moving the capacitor to the other side of the resistor makes sense. It is a way to implement anti-windup. It helps for the initial approach, but no more in the final approach, when the heater no longer goes into saturation. 
The change should not effect the choice of the capacitor values.

[HendriXML]

I started this post with not much (formal) knowledge of PID control systems.

In the meantime I found some good video's giving more insight into PID control.
https://youtu.be/NVLXCwc8HzM

[mzzj]

Look for INA330 peltier controller if you want stability.. noise level below 0.0001C

Thermistor is the way to go if you want maximum sensitivity but don't care about long-term stability.
Platinum resistance thermometers(Pt100)  or SPRT's have less sensitivity but better long-term stability.

[LaserSteve]

See if you can get a copy of A microcontroller-based driver to stabilize the temperature of an optical stage to within 1 mK in the range , using a Peltier heat pump and a thermistor sensor
A W Sloman, Paul Buggs, James Molloy and Douglas Stewart     Measurement Science and Technology,  Volume 7,  Number 11

Bill Sloman hangs out on sci.electronics.design  (google groups) and is the principal author.  He might just send you a copy.

Steve