T.C. measurements on precision resistors

[branadic]

Thanks dieter1 for the script, though I guess the t.c. alone is not enough, but you also need the consider absolute value of the individual resistor to optimize the network for low t.c., as they all have different absolute resistance and not perfect 5k. Thus I've used the resistor values instead of using their t.c. only for my manual "optimization".

-branadic-

[Kleinstein]

The absolute values tend to be not that different. So for TC combinations one could use equal values.

Especially when using sets of 3 and 4 resistors, even picking a fixed combination of resistors should give a reasonable good match.
One also has to take into account that the TC may change after soldering to a board.
In an application like the LTZ1000 or a 7 to 10 V amplifier it is also not so much about the TC, but more about long time drift.

[branadic]

Kleinstein, I don't agree with you and the diagram above tell's another story. The slight difference in absolute value and t.c. leads to some residual t.c. once connected as a divider. However, noone knows how things change after soldering them to a board, I agree with you. And as for long term I would expect at least better behaviour compared to a divider made of single resistors. At least TDP networks have proven to work in W7000 and F7001 over long term.

-branadic-

[dietert1]

It's a simple extension to combine TCs using measured resistor values, and the program won't run much longer. Let's see whether it makes a difference..

Regards, Dieter

PS: Meanwhile i checked and the approximation/simplification i proposed was a good one. I ran several examples of my Nomca data and it always found the best combination. The only noticeable change when taking into account measured resistor values is the optimum residual TC. It may be 10 ** -4 ppm/K off.
In order to avoid changes to the parts after characterization i used SMD solder adapters with gold plated pins in gold plated sockets (construction similar to HP 3456A thin film arrays). It's not only about soldering but also about cleaning, which should both be done before characterization.

[branadic]

I moved on and calculated the change in voltage at pin 6 of LTZ1000 over the temperature range of 10 ... 40°C due to the resistor network only, assuming a nominal zener voltage of 7.15V with the resistor arrangement previously found to perform good. It looks like a change of <3µV can be expected. Since this value is further attenuated by LTZ itself (0.01% --> 1ppm) we can call that zero. I think I need to setup a reference to prove it works in reality. But first I will measure another few samles to see, if one can find similar results within a batch of networks or if they have to be individually measured and connected.

-branadic-

[RandallMcRee]

However, noone knows how things change after soldering them to a board, I agree with you.
-branadic-

What I do is to solder my precision resistors to some sort of carrier or just add wire-wrap leads. Only then measure.

[Kleinstein]

At least for the LTz1000 set point divider there is no need for a TC matching much better than some 5 ppm/K. The reference itself has some TC that for the non A version is adjusted with an extra resistor. If individually fitted this adjustment can also include a small TC from the set point.
So chances are good one can get away with a fixed PCB and use a measurement only as a check to see the final outcome. This could be after soldering the resistor array as one of the first parts.

[branadic]

I'm running the TEC controller with the incubator as a P controller only, everything else lead to large oscillations. My parameters are PID 6.7700, 0, 0
Problem is, that the volume of the incubator is quite large and this is why autotune can't find proper parameters. So I did a manual search for reasonable parameters.

-branadic-

[Eheran]

Maybe I can help with the PID. Basics:
Lower I parameter (K_I) = integration of error has a larger effect
Lower D parameter (K_D) = response to changes is larger

Since you work with larg time scales you need large K_I and K_D to make everything slow and smooth. However, since you may also want to have a fast initial heat-up phase instead of having to wait lots of hours it could be best to use 2 sets of parameters, maybe whatever you use to controll has such a option. Otherwise you can (maybe) change the parameters while the system is running or just turn it off and change them after it reached the target temperature.

What were the parameters with oscillations? With the highest possible K_P with stable oscillation you can calculate good starting points for the parameters. Ziegler–Nichols method


Why im actually here: I want to measure temperature with high precision and resolution (0.1 mK). Regarding the signal processing: Is the LTC2983 good or the best? I would buy the whole evaluation kit and use that.
Any recommendations regarding the sensor? I would prefer Pt1000, but maybe thats nonsense?
What about tolerance? B 1/10 has a max. error of +-1.55°C at 250°C. Doesnt really blow me away. Even AA is +- 0.525°C at 250°C.
Are there easy temperature-references to check things? The triple point of water is a option as well as the meltingpoint of highly pure gallium. What about higher temperatures? Melting point of pure tin/lead/...? How accurate is that? Can I reliably check for absolute precision like that?
There is also the question about linearization parameters [the f in T = f(R)], they only need to be a little bit off to introduce a lot of error. Are they known to be good enough?
Now keep in mind that I want to stay well below 500 € with the thermometer. The 200 € for said evaluation kit is already a lot. I dont need to get the very last bit of precision, but I would need a way to verify it. 1°C off at 250°C is no problem, I just need to know about it.
I have a reference thermometer that was NIST-traceable calibrated until 2018 and had ~1mK error according to the cerfiticate. A quick test with a pure water ice-bath (at 1020mbar in a dewar) resulted in a reading of -0.0043°C averaged over a duration of 4h (0.7mK standard deviation). But I was unable to find a high resolution melting temperature of water ice (Ih). All I found were some models etc. and how they didnt really get it right. They didnt mentioned the melting point with high resolution since they were off >>1°C anyway.

[Kleinstein]

The melting point of pure metals like tin, lead, zink are an option as addition fixed points. The material need be quite pure and the system to measure it may need some care (even if the meting point is OK this does no mean the thermometer is at exactly that temperature. Direct contact may not be possible. Especially molten tin is quite reactive to many metals and the metal would be contaminated. I have used this - but in my case it was easy as the instrument to calibrate was a DSC, so made to measure something like melting points.
The pure metal samples for calibration may be relatively expensive, especially if one needs more material than the few mg for the DSC.
There should be tables to show the melting points as part of ITS90.

For higher temperature PT1000 may not be that good anymore, as isolation gets increasingly difficult. So PT100 or even lower resistance (if really high) would be better with higher temperature.

The Eval kit looks reasonable priced and the resistance accuracy may be relatively good. It depends on the reference resistor used.
I point the kit is likely missing is full protection. So it needs some care using it.

For the PID regulator with a slow process like a larger furnace, I would prefer the Nichols-Ziegler variant starting from the step response or a similar formula starting from the pulse response. Looking for just sustained oscillation can take quite a while. One has to take case how the parameters are entered - some regulators use the inverse of the parameter used in the math.

[guenthert]

  What temperature range are you interested in?   Compared to other fix points the triple point of CO2 seems easier to realize (at least no toxic materials are involved), even if just briefly: https://people.chem.ucsb.edu/feldwinn/darby/DemoLibrary/DemoPDFs/Demo046.pdf
(safety goggles required though 

[Kleinstein]

 The triple point of CO2 may be easier to realize, but it is quite low and one may need good purity CO2 to get a well defined temperature.
The meting points of indium, tin and zink are fix points of the ITS90 scale.
The triple point of CO2 is not in that list, which can indicate that there may be some complication to get it accurate. Chances are the pressure could rise quite high for a fully closed system (no safety valve).

[texaspyro]

Gallium is one of the best triple point materials...

[Eheran]

For higher temperature PT1000 may not be that good anymore, as isolation gets increasingly difficult.

Leakage of current inside the Pt1000 element? How?

What do you mean with "missing full protection" and "some care using" the eval kit? What kind of protection?

The resistor I plan to use is a Vishay Z201 2.5kOhm, tolerance is +-0.005% (0.125 Ohm, 0.013°C absolut error with Pt1000) and 0.05ppm/°C (20°C deltaT results in 2.5mOhm or 0.0003°C with Pt1000). The limiting factor would be the Pt1000 element itself and/or the ADC and its analog stuff.

Temperature range is limited to 400°C with a 2.5kOhm reference resistor since at that point the voltage across the Pt1000 will be equal to the reference resistor and since that is directly the ADC reference --> V_in = V_ref = maximum. At least I hope thats how it works... I only need the positive range of the ADC in the differential measurement. Other than that I plan to use it inside of -40 ... 300 °C, maybe once or twice with liquid nitrogen down to -196°C. But mostly in the range of -20...100°C.

even if just briefly

Hahaha, sadly it wouldnt help if I cant keep it in that state for at least an hour to get a good reference

[SilverSolder]

MAX1978 looks like a nice IC,  a little on the expensive side perhaps...

Can it be amplified externally if you need more than 3 amps, I wonder...

[bgugi]

Uh-oh... be careful, you may wake the Kelvin-Nuts (they tend to hibernate for a month while doing freezing curves on TPW cells).

Ice points: The widely-accepted value for the uncertainty of a "properly-prepared" ice point is around 2-3 mK. https://www.govinfo.gov/content/pkg/GOVPUB-C13-5ce1be74727a73a2d8c89172d042b3a4/pdf/GOVPUB-C13-5ce1be74727a73a2d8c89172d042b3a4.pdf is a decent study on the proper construction and reproducibility of ice points.

the guide to the realization of ITS-90 https://www.bipm.org/en/committees/cc/cct/guide-its90.html offers an absolute flood of information on the realization of temperature primary standards, though there's also a decent amount of information available on essentially any pure substance, and the interpolation equations can be applied to any fixed point (note, some older data may be in IPTS-68 and need conversion to ITS-90). When considering ultra-accurate measurements, it's important to remember that ITS-90 is a practical temperature scale, the associated errors in the fixed points and interpolation functions can reach several mK, not even mentioning the recent redefinition of the Kelvin.

Here's a decent article on a practical home-realization of a TPW cell: https://www.scientificamerican.com/article/tackling-the-triple-point/ no discussion of uncertainty is provided, but you could probably apply some of the same techniques described in the BIPM realization guide to provide estimates of uncertainty. I recall seeing an article somebody did on gallium, but overall I don't think there's any data on the uncertainty of any fixed points with home-practical materials and construction (i don't even want to think about what 7N gallium would cost)

If the reference thermometer you mentioned you have is an RTD, you'll be glad to hear that you can proceed with quite a bit of confidence if your TPW or ice point calibration turn out well - any error in an RTD will affect the TPW reading, you'd need an unreasonably rare combination of factors to affect the overall curve without seeing it in the TPW reading of the thermometer.

Precision/Reproducibility can be quite good for temperature measurement, but absolute accuracy is an incredible feat at the millikelvin scale (I'm not really sure what you mean by "absolute precision"). Expect to spend 5 figures to achieve uncertainties around 25-50 mK over a decently wide range (-200 to 400 C), expand that temperature range or tighten your expectations, and you get into the "if you have to ask..." price range. Sub mK accuracies are the work of NMI's.

[dietert1]

Can you explain a little how your comment relates to "T.C. measurements on precision resistors"? As far as i understand, for a high resolution TC measurement one does not need precision/calibration of temperature nor precision/calibration of resistance measurement. What you need is resolution, i.e. low noise and some short term stability during the measurement. Short term would be a day or some days, until the TC measurement is finished.

We may wonder how long is short term, for example when using a glas encapsuled NTC. How much will it drift within a year? I would try to answer that question by observing different kinds of artefacts to see which ones are more stable than others. I mean if a metrological calibration is very expensive and out of reach.

Regards, Dieter

[Kleinstein]

MAX1978 looks like a nice IC,  a little on the expensive side perhaps...

Can it be amplified externally if you need more than 3 amps, I wonder...

One could add an extra power stage (e.g. half bridge driver + 2 MOSFETS each). However I would expect that there are other similar controller chips that are directly made to work with external power stages. The max1978 is mainly made for constant temperature operation (stabilization of a temperature) and this often only needs low power. At low temperature difference one gets better efficiency if the TEC is only used with reduced power level. So if only some  15 K temperature difference is used it is good to use only some 30% of the nominal current of the TEC.

For high quality ramps (especially if faster) it would probably be better to have a digital control, so that the PID parameters could be better tuned and the nonlinear properties of the TEC taken into account. However this is more like a separate project.

[bgugi]

Can you explain a little how your comment relates to "T.C. measurements on precision resistors"?

Sorry, my reply was related to eheran's stated goals... for stability measurements, triple-point is king (a well-maintained freeze can last a month or more), but it should probably be combined with an equivalent gallium when testing thermistors (fluke quotes their gallium cells will melt up to a week... the curve they advertise shows about 100 hours dead-flat within about 0.1 mK. These are, of course, with comically-pure cells.

In the home lab, you can test melting and freezing curves, reproducibility, probe reentrent effects (to test for "settling" or hysteresis of the probe, and draw your own conclusions.

For stability, reproducibility, sensitivity, and precision in the home lab over the range of 0-100 C, thermistors are very hard to beat: https://www.bipm.org/utils/common/pdf/ITS-90/Guide-SecTh-Thermistor-Thermometry.pdf stabilities are often several mK per year, with "pre-aging and selection" apparently offering drift on the order of tenths of a mK over the same time period. They are reliant on calibration from a PRT to achieve accuracy - they need far too many points to provide good linearity via fixed point calibration.

[julian1]

I currently measure on NOMCA1603 networks. I ran into some issue before, as I use a zero force socket. Obviously the socket was not able to reliable break through the oxide on the leads, which resulted in strange t.c. curves. So with a bit of flux and an almost clean solder tip I had to break the oxide, now I get more reasonable results. Here is a first view on a first samle. The t.c. seems to be almost linear.

Edit: Calculated some first values

R1: 4.333ppm/K
R2: 5.452ppm/K
R3: 4.983ppm/K
R4: 4.667ppm/K
R5: 6.072ppm/K
R6: 5.409ppm/K
R7: 5.986ppm/K
R8: 4.823ppm/K

-branadic-

These results are absolute TCR not ratio TCR right?

That's almost 5x better than the datasheet spec of +-25ppm/C absolute TCR.

https://www.vishay.com/docs/60117/nomca.pdf
 

[branadic]

Hi everyone,

I once bought one of those Ohmite HS520A10K hermetically sealed wirewound 10k, 1.5 W resistors with ±3 ppm/K as a possible solution for a 10kΩ resistor standard, but found the t.c. to be way larger than what was given in the datasheet. So the resistor went into the drawer, until now.

By accident I came across this wonderful article Temperature coefficient compensation of standard resistance (Beta coefficient compensation) by lymex on bbs.38hot.net and thought I give it a play to compensate the t.c. of the formentioned resistor by using a proper sized NTC with serial resistance, some copper wire plus adjustment resistor.

While playing with the Excel provided by lymex I found, that there must be some mistake inside the forumla to calculate the actual Rm at a certain temperature with R@23°C, alpha and beta given, so I simply copied the values of my fit into the relevant columns. Playing with the numbers I found values - Rp=4.85MΩ (4.7MΩ + 150kΩ) plus a 150kΩ NTC with B=3892, a copper resistance of Rc=17.91Ω and some Rs=2.191Ω - to get the curve flat and straight. So up to this point this is just a theoretical excerise.
I wonder if anyone ever tried that in reallity and what results are to be expected (t.c., LTD, ...). However, I already ordered the missing components to try that myself, but thought that could be a nice topic to discuss about, as most of the solutions for resistor standards I saw by now are based on the properties provided by a single resistor (such as VHA518-7) or a resistor arrangement (series + parallel connection of resistors + some trimming for the final value using pots).

Attached is an image of my t.c. measurement of this specific resistor and the Excel file I used to compensate its t.c.

-branadic-

[chickenHeadKnob]

Hi everyone,

By accident I came across this wonderful article Temperature coefficient compensation of standard resistance (Beta coefficient compensation) by lymex on bbs.38hot.net and thought I give it a play to compensate the t.c. of the formentioned resistor by using a proper sized NTC with serial resistance, some copper wire plus adjustment resistor.

-branadic-

Zylmex graced us with his English translation of the article here on the forum:

https://www.eevblog.com/forum/metrology/spread-sheet-aided-design-of-compensation-for-7v-to-10v-step-up-resistor-set/msg896342/#msg896342
and related:
https://www.eevblog.com/forum/metrology/spread-sheet-aided-calculation-for-standard-resistor-measurement/msg902402/#msg902402

At least I think it is a similar article

EDIT: had trouble with translate and see now the bbs.hot article has a different method

[branadic]

Just a small teaser, t.c. results are coming soon as I need to prepare the setup first and get some snake scripts done.

The base resistor is a HS520A 10k hermetically sealed 1.5 W wirewound resistor, that is supposed to have 3 ppm/K, but as posted earlier it was way larger, negative and also showed a decent amount of beta.
Beta was compensated with a parallel circuit of 5 meg resistor in series with a glass encapsulated NTC. I also integrated an GA10kA3 NTC as a temperature sensor, before I started wrapping 0.35 mm diameter copper wire bifilar around it for alpha compensation. It took quite some effort to get the correct length of the copper wire. Two 2.7 ohm resistors paralleled are trimming the resistor it to its final value.
Meanwhile everything is mounted into a case, a Rose 04.08 12 05 with pretty nice quality and look.

https://www.rose-systemtechnik.com/produkte/industriegehaeuse/standardgehaeuse/aluminiumgehaeuse/aluform/

A two hour measurement with my Solatron 7081 (calibrated by Dr. Frank not too long ago) in TrueOhms mode is attached.

-branadic-

[miro123]

I'm beginner in methodology. I followed the thread for alpha and beta TC compensation.

My question why we need to do this in modern days?

I think that adding of external components only increase to uncertainty of the reference. e.g Does the epoxy NTC turns the circuit to humidity sensor with unpredictable behavior.
I don't see the problem if on resistor changes the value but the resistor model is good defined.
another issues is product lifetime - I mean  "Moisture Induced Failure by NTC" - https://www.te.com/content/dam/te-com/documents/sensors/global/te-app-note-ntc-sensor-performance.pdf
Do I miss something? - it looks for me that it solves the non-existing problem but introduces many side effects.

[branadic]

Does the epoxy NTC turns the circuit to humidity sensor with unpredictable behavior.

Beta was compensated with a parallel circuit of 5 meg resistor in series with a glass encapsulated NTC

The GA10kA3 NTC acts as a temperature sensor.

-branadic-