Making Do with What You Got/ Building a Thermal Stability Fixture

[Vtile]

The periodic temperatur indicate oscillation due to poor PID parameters. The waveform is however a little odd - normally one gets more like a near sine.
[...]

I suspect the reason for asymmetrical waveform is the different gain (as in energy transferred) of the TEC when heating/cooling.

Might be just bang-bang control mode and what we see is thermal mass integrating the function. Too early and not enough information to say anything.

[antintedo]

NIST claims ice melting point can be easily realized with 0.01C uncertainty. One point is not ideal for tracking long term drift but still better than nothing.

[View[+]Finder]

Might be just bang-bang control mode and what we see is thermal mass integrating the function. Too early and not enough information to say anything.

Integrating? Probably, however the TEC controller is not supposed to be "bang-bang" as it is bipolar and designed to maintain the frequency of lasers by stabilizing temperature. One addition to my tracking protocol will be the amperage through the TEC. Voltage is currently tracked.

Any suggestions as to what should be expected from a ThorLabs TEC2000? ThorLabs is a high-end supplier of optical gear to universities and other research facilities. I would like to think they know their job.

[View[+]Finder]

The "ice melting point" used for temperature calibration is in the latency zone where water temperature is stable due to the latent heat. Ice by it self is colder and water with no ice is warmer. There should be a mixture at the bulb of the thermometer. Not just touching ice or in the melt water.


Is there a place in Metrology for temp-nuts? Probably.

[Vtile]

Might be just bang-bang control mode and what we see is thermal mass integrating the function. Too early and not enough information to say anything.

Integrating? Probably, however the TEC controller is not supposed to be "bang-bang" as it is bipolar and designed to maintain the frequency of lasers by stabilizing temperature. One addition to my tracking protocol will be the amperage through the TEC. Voltage is currently tracked.

Any suggestions as to what should be expected from a ThorLabs TEC2000? ThorLabs is a high-end supplier of optical gear to universities and other research facilities. I would like to think they know their job.

...Reality is that it is just fast response thermostat (it seems) and Bang-bang controller can be bipolar. There could have been a bang-bang mode for some odd use case, implementing that costs nothing if done with software which I do suppose is the case in here. The ripple looks like a non-symmetric saw-wave (on-off-controller if you wish) with thermal mass rounding it (acting as integrator). Well research can be done with really odd pieces of equipment (means poor or modest), I'm not stating that it would be the case here with ThorLabs, what I thought that the device is some "as is delivered / lastly used" mode and doesn't do it job as out of tune / wrong mode for the job in hand. 

With quick skim TEC2000 is PID-only (as pointed). Only PID and there is a rather good (at least longish) tuning instruction there.
https://www.thorlabs.com/drawings/241fa595cd73d614-F2A048EE-AD74-E27F-6E71BEA26112560B/TEC2000-Manual.pdf

But first thing you might try out is to make some step-response test (or pulse response by doing up and then down steps) to your chamber with increasing and decreasing set-point (desired set process temperature) as one discrete Heaviside step function (or as fast it goes.), magnitude of 0,5..1,5 degrees maybe in this case. With making a known disturbance to the system you actually can see something about quality of control and behavior of the system.

I have politically incorrect place in mind where all temp-nuts might belong. 

...But I'm not here to argue, these are interesting projects and subjects and I will follow them (..and learn probably some new things or viewpoints along the way).

PS.  shodan@micron, I totally didn't notice two things A) it is for medical B) the NIST traceability, I need to order one or two of these... 

Also it is full traceability sensors: "The TMP117 units are 100% tested on a production setup that is NIST traceable and verified with equipment that is calibrated to ISO/IEC 17025 accredited standards"

[Vtile]

Still very horrible and useless as thermostat. Putting DUT in a towel and inside metal trashcan will get better stability than this... 

Hello Illya,
Are you sure about the can and towel? Or is that a suggestion for a comparison test? We have recycling here and metal trash cans are nowhere to be found, but I could maybe use a cook-pot . . .

Have a look at the attached plot that covers a two hour period last night from roughly 22:40 to 00:40. Ambient temperature was 19C and inside the box was a relatively toasty 28.4C +/- 0.4C. Did I give the impression that this project was completed? That I was satisfied with the results? I don't think so.

There are two very good reasons to participate in a forum like this: to learn something new from folks with experience and to share experience with others so that they might learn. Sometimes sharing one's efforts, mistakes and failures can be more instructive than showing off a finished product as if it were easily done. My mathematical vocabulary developed in a narrow segment of finance, metrology is like learning a new language so if I have given offense, tut mir leid.

How familiar you are with control tuning and control theory (in this case Proportional-Integral-Derivative aka PID)?
Please do set-point step with that setup something like 2 degrees, it is oscillating if you are using the TEC2000 for control. Read the manual from above post, you can try to tune the system from the instructions given there. If I would need to give an educated guess there is too much integral time set to controller, maybe also derivative which tries to stabilize the thing ending up steady oscillation. One way to solve this is if you educate yourself with some articles of Ziegler–Nichols method PID tuning method (which is some times duped as oscillation method, but I can not remember anymore what books did say) or step-response PID tuning method, for fast and harmless system oscillation method might be usable approach.

If I would need to tune this box with specialist methods aka intuition (or expert method aka autotuning button  ) I would go with step method with both integral and derivative terms turned off so the controller is only proportional. Setting the set-point nominal value to ambient temperature and doing steps at both directions. First by just adjusting the P-term so that I would find fast response to set-point step. As there I would enable so slightly the integral term of the TEC2000 so that the offset-error would go away. D-term is a bit two fold and I would leave it out from the first test round. After I would have noted that the system is following the set-point step with damping oscillation, without offset error and rather rapidly I would run first process test (aka leave it logging for a few hours and see how it react to ambient).

[View[+]Finder]

Thank, I'm certain there will be many mistakes as "learning opportunities" for all. I'll try to report what I observe without bias as to the (assumed) provenance of the hardware.
1) I don't know anything about PID setup. Other than an excellent trading video from ABB on YouTube last week, my history with industrial control was "send a guy out with a wrench" trial and error, followed by "if it's working, don't f--k with it."
2) I have the manual for the TEC 2000 from ThorLabs and have read it. The link in the post above gets a 404 error, but it is probably the same manual. No worries.
3) "Bang-bang" is what I see from the data I'm collecting. A sawtooth with 240sec period with about 10sec turnaround at top and bottom. Not a sine curve at all.
4) I have decreased the set temp from 33c to 30C and controlled the ambient to 25C +/- 0.5C
5) The manual setting instructions start with PID all fully CCW--it looks like 'no effect'--so i set 'P' to middle (better), so then full CW and now nice little sine wave, 80sec period.
6) Now the measured temperature at the DUT is 30.4C +/- 0.05C

Now what? The manual suggest tweaking 'D' next . . . .

7) Setting 'D' to max CW made small improvement.
8) Moving 'I' off full CCW made things worse.

Amps are set to max for the TEC installed. Any reason to adjust to lower value?

[Vtile]

Small correction, which probably is because my average english. Bang-bang controllers are nickname for on-off-controllers, iirc one description is hysteresis controller. You are (at least by manual) using PID-controller. Why I did refer to bang-bang controller at beginning is that I falsely (which quick reading of manual titles etc.) thought that it does have such included, now it seems that I was mistaken.

I think the best thing for you is play with the controller a few days and try to google some hands-on PID tuning instructions. No need to go full academic with mathematics, since you do not have a model of your system anyway (and I think the TEC2000 is also black-box, so actual algorithms on controller is not known) and probably it will be changed many times more.

The D-parameter try to resist the change what I&P terms do, it doesn't hurt you to tweak it, but I would imagine that it is still not fully optimal at all since you are saying there is steady (but small  oscillation going on).

The a graphed setpoint step is good way to see how system behaves.

You have a good setup to familiarize yourself for tuning of PID.
Like I said try to find some good (authorative) hands-on articles about PIDs. While this is TEC-controller it is still just a PID controller in the end. After also reading a bit about the control subject return to tec2000 manual, it most probably open differently what everything do mean on it.

You have a good learning setup there, also play a bit the box and sensor itself to see how it might effect to controller outcome. That is with different measurement point and added thermal mass (ie. Cold soda can to the box etc.).

Not so quickly from phone.

Ps. Graphs should have (if possible) setpoint, controller output and desired target (box or dut temperature at this case) with those three you can see how the controller behaves. For long term measurements add athmosperic temp as that is source of disturbamce for you (by not knowing exactly all the whys and wheres).

[Vtile]

And these advices are given for you for reason you get understanding (of PID tuning) only then you can start to get where you actually wan't to go. Because if you do not have basics sorted out you will not get where you wan't to go as metrology thermal box.

[Kleinstein]

A TEC is a nonlinear device, especially near its rated current.  Usually the rated current in the DS is for a perfect heat sink. Real world the maximum useful current is lower (e.g. some 10-30%) and with more current it will not cool more but less. It absolutely makes sense to limit the current to a lower level, more like 50% of the rated current, so the TEC is in the relative linear range and no danoger of overheating. I would use more only if absolutely needed to reach the final temperature.

There are many different methods to tune a PID system.
It absolutely makes sense to record the step resonse for the system, that is apply a defined step in the TEC current (e.g. off to some 20% of rated current - maybe more if the response is weak, or less if the temperature gets too high/low). This could show the characteristic of the system an tell how easy it is to regulate or if one may need a different postion of the sensor.  A system with a long delay gets hard to control, especially with an anolog control loops with limited paramters.

For the step response on can calculate approximate suitable PID setttings via several approximate rules made for slightly different systems.
2 such systems are  Ziegler, Nichols and Chien, Hrones, Reswick.
https://infosys.beckhoff.com/english.php?content=../content/1033/tf4100_tc3_controller_toolbox/9007199500260107.html&id= 
This should be at least a suitebale starting point and OK to check if the parameters are feasable with the given controler.

This needs a little more math, but it can be worth it. The try and erros method is really time consuming with a slow system. A 3D parameter space is quite large when every try takes hours.

[Vtile]

Kleinstein did write it up clearly. I want to give a warning those tables can be extremely confusing, since the controller (I don't know if it is the case with this one OP have) might not have the same input parameter levels what those gives out as values and the markings differentiates highly (as seen on those two tables). Also there is differences on those tables at which way the parameters do go, I have forgotten the 'why', but relates to the mathematical model.

I'm actually surprised that this is TEC-unit is still analog  , but it seems it is rather old (does its job). Makes sense in that regard. Uh, I feel relieved I do not need to use analog day to day, there is enough headache from million different digital implementations to remember, but auto-tune algorithms and process tuning programs are from heaven.

4) I have decreased the set temp from 33c to 30C and controlled the ambient to 25C +/- 0.5C
6) Now the measured temperature at the DUT is 30.4C +/- 0.05C

What I was after was to adjust the set temp "zero level" to be ambient temperature (room temperature) and then make these steps below and above from that zero level. By doing so you can see both cooling (down) and heating (up) and how the response of controller do look like for each direction (when graphed, if you have real time graphing possibilities that makes this so much easier).

5) The manual setting instructions start with PID all fully CCW--it looks like 'no effect'--so i set 'P' to middle (better), so then full CW and now nice little sine wave, 80sec period.

The tuning instructions usually (you can read always) are written from state that controller is turned to as close as possible to P-only controller, removing I and D -terms effect. (this leaves usually permanent offset (error value) between set-point to response, mathematically always, but at practice especially with analog devices not always). Also notice how the manual says to open the device and removing jumper if you need to disable I-term (removing capacitance from operator amplifier circuit most probably).

Kleinsteins method is way to go if you can wrap your head around those tables, I were hoping you would end up to some articles explaining the tuning and these tables.

There is caveats you will find out, ie. when you modify your sensor position the system will change and the tuning might not be as good anymore (or improve).

[Kleinstein]

Those tables for PID tuning are partially heuristic and partially based on a model for the system. Slighlty different models give slightly different values for the tables. The other point is that there are different targets on what to consider a good tuning. This can be just without overshoot or with an "acceptable" level to get fast relatively close.
There is an awful lot of tables and tuning methods - though the principle is the same: do a test (e.g. step response, pulse response, limit of oscillation or a bang bang type crude regulator) to learn about the sytem and than calculate suitable PID parameters based on approximate formulas. It depends on the details of the system which approximate model fits best - so there is no univesal best method. The resulting parameters are still rather similar  most of the time.

Within the range given by different tables is about where one can see with try and error and look at the specific target which one fits best. One may still have to look at the parameters as sets - mixing the gain from one table with the times from another one may not work. Often there is also a good enough - so the first result from the tables may work good enough - especially if the system is easy to control. Thermal systems are often slow, but still usually relatively well behaved. If really bad, it may be more about sensor placement than fine tunig the PID parameters.

Many of the so called auto-tune regulators one can buy do a similar test and calculation method inside the box. This allready is a big plus for the users. I find especially the method starting with a crude band bang regulator interesting though this is more a method to integrate in a regulator, not so much to do by hand.
There are true auto-tune algorithms, that do the system test while doing the regulation, but this is not without problems and mathematical complicated. Controling the speed of learning may be as tricky as setting the PID paramerters  .


The PID parameters can be confusing as there are different formulas used to expess the regulator. So one can have the gain or the invers and similar one can expess the I and D parts with times for the cross over or as factors including the gain. The units give a hint, at least for the I and D part.

With analog regulators one may not have a meaningful scale   .

[View[+]Finder]

Well, thanks to all the advice and references, things are looking better. The interior of the DUT box has been holding 30.4C +/-0.01C and I've managed to cpntro; ambient to 25C +/- 0.5C using a "small-room" air conditioner. Adjusting 'P' was made more difficult due to the 'P' control being 'high' at the CCW setting, not 'low' as the instructions seemed to indicate. Right now all PID are max CW and temperature is stable in the box. It looks like I'm have reached the "don't f--k with it " stage for now.

Screenshots attached.

[Kleinstein]

For testing the control loop, one should intentionally not have a really stable room temperature - more like the opposite of intentionally increasing the external temperature varaition to see how much lower the variations are at the inside.   A stable temperature is  more like something for later to help the meters and in case the regulation does not work well.

The temperature still seems to show some relatively fast oscialltion. Because of the low level I am not sure if this real, or just something like a beat frequency to mains.
So the parameters may be still far off.

Another way to test is to look at steps, like the turn on transients, or a short disturbance, like opensing the case for a short time while running.

[View[+]Finder]

The fixture being measured is designed to hold a small calibration reference, most likely a 10VDC source for monitoring autocal on DMM's. The plan is to leave it powered up and connected, probably to the 3458 for some long time testing.

The same effect as changing ambient can be had by making a change in the set temperature. That was just done and the attached photo shows the measured temperature on the plate side of the TEC rising to the new setting without overshoot and settling in with the same relative swings in temperature as before the change. The periodicity is about 65 seconds, which seems more consistent with latency in the fixture than a beat frequency. There is some new 'space-age' insulation to be installed and that should reduce the fluctuations.

[CatalinaWOW]

Unless something very strange is going on I would not expect your new insulation to reduce the magnitude of these oscillations, though it could possibly lengthen the period.  Your graph seems to show either an oscillation or a dead band.  It would be very telling to show TEC current on the same trace as the temperature.

The real question is does this behavior meet your needs?  For what I would be doing with voltage references it would be more than good enough.

[View[+]Finder]

There are two temperature sensors in the system: one is connected to the TRC controller and reads the temperature of the 'cold plate' in the chamber; the other is close to the DUT (a TI REF102-based 10VDC reference) and reads the temperature affecting its voltage output. Both are thermistors of the 10kOhm type. The TEC controller reports 'actual' temperature as resistance, leaving the calculation of temperature TBD. There is another source of actual temperature--this time reported on a 0 to 10VDC scale--presumably for a more precise laser diode controller.

The short-term fluctuations shown in the plots above are of the 0 to 10VDC signal representing the cold plate temperature. To be honest, given that the thermistor temperature (resistance) is inverted, I am not sure whether the voltage signal is as well. In any case, my focus is on its magnitude and periodicity compared to the DUT temperature. In rough numbers, the cold plate fluctuates about one percent of the full scale on a cycle of sixty seconds and the DUT temperature is stable at 25.5C +/-0.05C.

The TEC controller may not yet have enough data to integrate the fluctuations. As reported above, the current is 800mA and there is a fairly rapid shift of polarity at the inflection points.

The real test of "good enough" will be the stability of the DUT over a lengthy testing period. We'll see . . .

[Kleinstein]

If the TEC power is really changing polarity and reaching some 800 mA this sounds quite a bit like oscillation of the regulator loop. So it would be a good idea to look at the loop tuning again. Even if not used for the tunig, the setp response of the sensor to actuator step (not a step in the set point) could tell if the system is easy to control or more difficult and may thus need a different sensor position. 

The suppression of external temperature variation (some 15 min cycle) is also not that good. Some 0.1 K ambient and 0.015 K inside. Not bad, but also not especially good.  With so little change the suppression factor is a little hard to judge.

[Andreas]

As reported above, the current is 800mA and there is a fairly rapid shift of polarity at the inflection points.

A rapid change of polarity only generates heat.
I would avoid that.

And always think of: the TEC acts also as generator when there is a temperature difference on both sides.
So if you change polarity quickly the voltage generated on the TEC adds to the supply voltage.
This increases the intended current. (maybe above the system limits).

With best regards

Andreas

[View[+]Finder]

Here's the latest:
First, a link to a YouTube video on PID control 
It is by a drone pilot--with a lot more at risk than a voltage reference .5 PPM variation--and by his own admission, an over-simplification. A graphic, math-free explanation of P, I and D contribution to process control.

Second, my power to the TEC at 800mA was the cause of oscillations. I reduced it by halves down to 50mA, checking the effect at each stage, and no more oscillations. Considering that this particular DUT is drawing just over 1mA from the PSU, what is left to improve?

Finally, I got over "making do" with a 4.0C swing in ambient temperature in my lab and found a small "room air conditioner" on Amazon that now keeps the ambient at +/-0.4C. It sends the hot air into the attic via a 5" flex hose.

[dietert1]

Is that an Arroyo 5305 (12V 5A)?
Ambient temperature is 29°C, is this the water temperature or room air temperature?
The datasheet of the quad module has limits of 14.8 V and -50 °C. Did you use both Krytherm modules on top of each other or or only the quad module? Will the "tower" crack when going lower?

Regards Dieter

[Kleinstein]

The 4 stage version is for an even smaller system, more like a IR photodiode.

If a slightly lower temperature is needed, there may still be the option to relace the external radiator with a bucket of ice water. It is low tech, but OK for short time use.

[Kleinstein]

The temperature control looks really good. Getting a nice stable temperature is easy, a nice fixed slope is a little more tricky,  but the sharp corners are the art.

The orange foam looks a bit messy. For the low temperature range is can be OK, but in the higher temperature range the ready made isolation foam is usually better.

[CatalinaWOW]

Very nice work. 

A controlled slope is important for some types of tests.  But square corners are kind of like ornate artwork on baroque buildings.  Pleasing to the designer, and satisfies a currently popular style but really has little or no functional importance.  A minor savings in test time is about all you can claim for it.

[Anders Petersson]

Hi Shodan. I follow with interest since I'm doing a similar project.

You may see, temperature of Arroyo feedback have large shift between DUT temperature, when Tset away from ambient.
Reason is simple: heat leakage trough DUT measurement wires

Could you make the DUT temperature wires loop around inside the box before exiting? In that way, the wires can cool down better, with less ambient temperature conducted to the sensor tip.