Calibration of PRTs

[Anders Petersson]

An account of my attempt at in-house calibration of PRTs (platinum resistance thermometers).
This is the next logical step after my previously post about my reference resistor, where I concluded that my 6.5 digit DMM6500 has 4-wire resistance noise of +- 2 counts, corresponding to +- 0.5 mK for a PRT. The end goal is to characterize my homemade climate chamber and to calibrate sensors for atmospheric research.

The equipment is a DMM6500 with a 2000-SCAN card. This allows me to measure 5 PRTs in 4-wire resistance mode. My sensors are:
#1 PT100 Pentronic 10 cm probe, class B/3, probably wire-wound
#2 PT100 Labfacility 2420300 probe, class B, probably wire-wound
#3 PT1000 element IST P1K0.161.4W.B.010, thin-film
#4 PT100 thin-film Heraeus 32208550, Class A, Digikey 1759-1021-ND
#5 PT100 thin-film Heraeus 32208550, Class A, Digikey 1759-1021-ND

The sensor elements of #3-5 are merely 1-2mm in size, chosen for their fast response time in air. We cast their hair-thin legs and 4-wire solder points in epoxy with 3mm diameter for mechanical protection and electrical isolation against the water. The casting works quite well after some practice (except the cable stiffens from wicking up epoxy) but I don't know how the long-term measurement stability will be affected. Also, these are thin-film PRTs, instead of wire-wound which are said to be more stable over time.

Let's do an ice bath calibration!

This paper goes into great detail on precision ice bath measurements.
(NIST "Reproducibility of the temperature of the ice point in routine measurements" https://www.govinfo.gov/content/pkg/GOVPUB-C13-5ce1be74727a73a2d8c89172d042b3a4/pdf/GOVPUB-C13-5ce1be74727a73a2d8c89172d042b3a4.pdf )
It shows the overwhelming importance of using ice from pure water to achieve 0.002 C accuracy of the melting point.
However, the calibration company Pentronic told me they use ice from (good quality) tap water, consistently getting +- 0.003 C. In my case, I used store-bought Norwegian ice and tap water.

As dewar, I used a glass vase ca 25 cm high and 10-15 cm diameter, wrapped in a bath towel. I used a consumer ice crusher to break the ice into pieces < 1 cm in size. A piece of styrofoam over the top helped to keep the sensors roughly in the middle of the vase. After inserting the sensors, I filled with ice, then with water pre-cooled in the fridge.





Playing around with the setup for an hour, the readings were mostly stable within 1-2 mK, with three caveats:
1. Some drifting occurred, especially in #3 which didn't seem to settle
2. Some sudden jumps between plateaus when stirring and siphoning off meltwater, notably #2 which lowered its plateau by 29 mK
3. Some individual readings are a fair bit (~ 20-30 mK) lower than the average, hinting that the average might be elevated
See attachments of multimeter plots for details.

For those interested, a cheat sheet for conversion between ohm and Kelvin for PT100:
  1 K = 0.385 ohm
  1 ohm ~= 2.597 K

My results are:
#1 Class B/3 (tolerance +- 0.1 K) Stable at 100.0116 ohm = 0.030 K above nominal
#2 Class B (tolerance +- 0.3 K) Stable at 100.0440 ohm = 0.114 K above nominal
#3 Class B (tolerance +- 0.3 K) Measured 999.23 ohm at lowest = 0.20 K below nominal. Too much drift to record full resolution.
#4 Class A (tolerance +- 0.15 K) Dips down to 99.9760 ohm = 0.062 K below nominal
#5 Class A (tolerance +- 0.15 K) Dips down to 99.9640 ohm = 0.093 K below nominal, but otherwise stable at 0.069 K below nominal.

All sensors are within tolerance. From the drift and dips in readings, my guess is that all sensors were exposed to the range 0 - 0.03 C at their lowest reading. This is a magnitude worse than the 0.003 C accuracy mentioned by the calibration company and I hope to improve on it. I made an earlier test with sloppier technique, getting less consistent results.
The process of stirring the ice and adjusting the sensors while looking for changing readings feels somewhat arbitrary (and time-consuming). Ice bath calibration is easy and cheap in theory, but my conclusion is that good results take some preparation and experience.
 
It will be interesting to repeat the test, attempt other temperature points and eventually send some sensors for 0.015 K calibration.
I'm learning as I go, comments are welcome.

[Anders Petersson]

Can someone explain to me what I'm doing wrong when trying to inline four of the pictures? 

[DrG]

Can someone explain to me what I'm doing wrong when trying to inline four of the pictures? 

Thanks for an interesting hands on report.

As for the inline pictures problem, it is not so much anything that you are doing wrong but rather some "modifications" to picture posting to eliminate a nasty situation where pics got replaced automagically. There are a number of threads about this.

If you look here https://www.eevblog.com/forum/chat/attention-eevblog-admins-attachments-mixup/msg3179834/#msg3179834 you can see one way of doing inline pictures. That is my opinion on how to handle it at present. Basically, do you attachments first and then modify your post to add links, to those attachments, in your post. Again, maybe there is a better way, but I don't know it.

[Conrad Hoffman]

I've always wanted to try the triple point cell from the old Scientific American article. Should be online if you search.

[Anders Petersson]

I've always wanted to try the triple point cell from the old Scientific American article. Should be online if you search.

This one I suppose: https://www.scientificamerican.com/article/tackling-the-triple-point/
Sounds like an interesting process. But for me, definitely more work than it's worth. A professional calibration costs a few hundred USD.

[Dr. Frank]

Thank you for this nice report on T-sensor measurements.

I had to solve the same problem @ university, when I had to compare / calibrate different uncalibrated sensors from LAKE SHORE CRYOTRONICS  (e.g. PT103, CGR-1-1000 - carbon resistor, DT450 - diode, AuFe/NiCr - Kondo effect based thermo element) versus a fully calibrated (max. +/-6mK) diode sensor DT470K. I coupled the DUT and the reference thermometer on a metal block and measured the output signals (resistance or voltage) over the whole temperature range needed.
Lake Shore offers uncalibrated sensors, which only follow their standard curves (typ. +/- 0.5K), 2/3 point calibration (typ. +/-0.25K) and full/best calibration (table, down to +/-10mK)

That's the first bug of your measurements, which I would not call a 'calibration'.
You only do a 1 point comparison at 0°C, a sort of Null reading, but a calibration would also require to determine the Slope value(s) by measuring at one elevated temperature at least.
This 1 point comparison with supposed 3mK uncertainty is not meaningful if you want to really use your sensors, i.e. at other temperatures.

The different tolerance classes (e.g. +/-0.1K @ 0°C) you mention - please check how this is really defined, over the complete T-range.

PT100 sensors urgently require 4W Ohm mode but also Offset Compensation, which your DMM6500 offers.
Please check if you engaged this method.

You should test self heating effects when you really want to achieve 3mK uncertainty. 100 Ohm @ 1mA are 100µW, which might create enough temperature rise => use low power mode, if available.

An Ice Bath for 0°C poses two challenges for such low uncertainties:
At first, you need to continuously stir the bath, as the self generated heat has to be transported away from the sensor.
Second, these 0°C apply for the 'mixture' or coexisting phase of solid / liquid water. Therefore, w/o stir, the temperature is too low if your sensor is in direct contact with the solid water only, and too high, if it's in contact with the fluid only. Standard cells are designed to overcome this problem.

This might be the root cause of these jumps, you already explained partly by yourself.

Your epoxy glued TF sensors look quite nice, but I wonder whether they would be suitable (i.e. stable enough) for a 15mK calibration. The substrate is partly glued into the epoxy, and the sensor element is exposed to the environment.  It is not evident, whether moisture may enter the PT film, or into the epoxy, but that may create errors.
Thermal conductivity into the epoxy and similar effects might deteriorate accuracy.

Fully (glass) encapsulated PT100, like the ones from Lake Shore might only be qualified to be calibrated to such low uncertainties.

So do further analysis, before you sink money. See below links about PT sensors, and the different calibration grades.

In the end, that might really evolve into a nice temperature metrology experiment, please go ahead!

Frank

https://www.lakeshore.com/products/categories/overview/temperature-products/cryogenic-temperature-sensors/platinum
https://www.lakeshore.com/docs/default-source/product-downloads/literature/lstc_appendixd_l.pdf?sfvrsn=6b9ac223_4
 

[dietert1]

What i found important with precision temperature is to have a certain cable length near the sensor at the same test temperature, maybe 20 or 30 cm. Heat conduction through the wiring is one of the major error sources. Dr. Franks remarks are a bit ambiguous. Hermetic enclosure helps on the long term (years). It may save you future calibration work because hermetic sensors are more long term stable. The fluctuations observed by you probably have other reasons, like moving cables while stirring the mixture. If you had another precision DVM you could try to understand those fluctuations by observing two sensors at the same time. Then you could probably "recognize" which observations are real and which ones are noise.

Regards, Dieter

PS: After reading more carefully: In this case the scanner replaces another DVM. If the DVM + Scanner is good for 10 ppm, then with the 4E-3 Pt coefficient basic precision should be about 2.5 mK. This work becomes science, if it repeats. If you take a pair of your sensors and try to reproduce their calibration difference near 0°C once or twice, will you get consistent results?

[Anders Petersson]

Thank you for this nice report on T-sensor measurements.

Thanks for the insightful thoughts, Frank!

I agree, I was too generous in calling this a "calibration". At the edge of a +- 50 C range, the class A tolerance widens by 0.1 K and class B and B/3 both widen by 0.25 K. I need to determine the slopes too.

Self-heating is indeed a real concern for the small sensors; Heraeus states 0.4 K/mW. Naively, 100 Ohm at 1mA gives 100µW as you mention, for 0.04 K heating. However, the test is in water and the bias current is pulsed, throwing the simple calculation out the window. I will do a test for self-heating.
I think DMM Offset Compensation was set to "auto" which is not explained in the manual. Thanks for the tip.

Moisture ingress could be a real problem; Heraeus datasheet states "Environmental conditions: unhoused for dry environments only"
https://www.mouser.se/datasheet/2/619/M222_HST-USA-1131469.pdf
These fast-response sensors might be best calibrated dry, against a reference sensor attached to or inserted into a metal block. Then the epoxy doesn't need to attach to the sensor for water-proofing.
Lakeshore PT-111 should manage being wet but costs $179... 50x that of Heraeus 32208550.  https://shop.lakeshore.com/index.php/temperature-products/temperature-sensors/platinum/uncalibrated.html

[Anders Petersson]

What i found important with precision temperature is to have a certain cable length near the sensor at the same test temperature, maybe 20 or 30 cm. Heat conduction through the wiring is one of the major error sources.

Great observation, I will change this.

PS: After reading more carefully: In this case the scanner replaces another DVM. If the DVM + Scanner is good for 10 ppm, then with the 4E-3 Pt coefficient basic precision should be about 2.5 mK. This work becomes science, if it repeats. If you take a pair of your sensors and try to reproduce their calibration difference near 0°C once or twice, will you get consistent results?

Correct, the scanner lets me compare disturbances between channels and I will try this. The noise-free DVM resolution is 0.5 mK and although DMM6500 1-year stability is only speced as 100 ppm (25 mK), my reference resistor gives some additional confidence. I can't know for sure if the sensors avoid drift over time but I'll definitely repeat the experiment, with your and Franks tips.

[jbb]

I think I saw a simplified triple point calibration video somewhere on YouTube. Sorry I can’t search for it; super busy at home.

The epoxy bead could be an issue in terms of mechanical stress and moisture diffusion. I guess you have to weigh up the cost of fancy probes vs how irritating it might be if your results are off.

Maybe a blob of neutral cure silicone RTV would be better? It’s a little flexible so hopefully imparts less stress. If the sensor is left wet continuously (as much as practical) then may be you would reach constant humidity too.

[cellularmitosis]

I tried my hand at this as well.

I used a PT102A1 (+/-0.06%) thin-film PT1000, affixed it in 4-wire mode to a popsicle stick using hot glue, with the submerged portion of the wiring being 30AWG and the non-submerged wiring being two pairs of 22AWG from an ethernet cable, terminated with a pair of Pomona 4892-0 dual-banana plugs.

Made up an ice bath, plugged it into the 34401A.  Assuming my meter is perfectly in-cal, and that my filtered tap water is perfectly pure, the RTD has about a -0.1C offset.  The spec for this part is +/-0.15C, so that sounds reasonable.

I've used several of the I2C temp + humidity breakout boards (Si7021, BME280, etc), and some of them disagree with each other by nearly a degree, so hopefully I can settle their little dispute now 

[cellularmitosis]

I just realized that digikey no longer carries this part, which surprised me, because I ordered these like, just a couple of years ago, but then never got around to doing this little calibration experiment.

Then I realized... yeah... that was a little more than just a few years ago 

[HighVoltage]

In order to have an absolute temperature reference, I bought a Fluke PRT that came with a detailed calibration certificate.

My other PT100 are tested in an oil bath for comparison.

[Pip]

Thanks for sharing, Anders and previous posters. Here are my thoughts and suggestions for improvement.

If you have access to an insulation resistance tester with a 100 V range, or a 100 V DC source and a DC ammeter capable of measuring low current, conduct an insulation resistance test of each PRT at ambient temperature. Apply the voltage between the sheath or outside of the PRT, and all four wires together. The PRTs  must have an insulation resistance of at least 100 MΩ to be satisfactory. This test could also be performed at say 50 V if 100 V is not available. I recommend that you do this before and after your ice point check to confirm that moisture is not affecting your sensors.

Use distilled or de-ionised water in your ice point. Having said this, the error contributed by the use of ordinary tap water could still be quite small. It will depend on the minerals in the water. Reverse osmosis or rain water may also be suitable.

The ice should be shaved. If you use crushed ice, the pieces should be no larger than say 1 mm (ideally) to 3 mm in diameter. This will allow close connection or good thermal contact of the PRT with the melting ice. There should be no air gaps between the ice and the PRT, only ice and water. This is easier to achieve with small ice pieces.

The ice should be translucent, not white.

When placing your PRTs that are at ambient temperature into the ice, initially place them into the ice off to the side until they have reached near zero then remove them and insert them to a position near the centre of the bath. This minimises melt back of the ice from the PRT. This will be fine for the metal-sheathed PRTs. I'm not sure if your 'home-made' PRTs are able to be pushed into the ice.

Heat conduction can be a concern but it looks like you may have sufficient immersion depth for those short PRTs. You will know if it's sufficient by withdrawing the PRT 20 mm without significant increase in ice point resistance.

Make sure that the tip of the PRT is always surrounded by melting ice. Over time, the ice melts and water pools below the ice, if using the slush method and water is not being siphoned off. If the PRT tip is in the warmer water, there will be an error introduced into your readings.

Metal-sheathed PRTs can be 'annealed' at a high temperature to stabilise their resistance. This is not easy to achieve without the use of a high temperature source such as a dry-block calibrator.

You can easily check the resistance of your PRTs at (close to) 100 °C in a large pot of boiling distilled or de-ionised water. This may be sufficiently accurate to allow for the determination of their alpha values and an indication of conformance with their rated accuracy classes.

I recommend the use of an excitation current of 1 mA. Although you could use lower or higher currents, noise could be an issue at lower currents, and self-heating could be an issue at higher currents.

If any of the PRTs are old/used, check continuity of all wires as it is not uncommon for PRTs to have an intermittent break near where the wire enters the metal sheath. Use a DMM with an audible continuity indication to detect an open circuit while carefully flexing the cable.

The Measurement Standards Laboratory of New Zealand has a very good publication on making ice points, available for download here https://www.measurement.govt.nz/resources/#collapse-control-1-5. You just have to enter your email address.

BIPM also has a very good document that covers the two methods for making an ice point: https://www.bipm.org/en/search?p_p_id=search_portlet&p_p_lifecycle=2&p_p_state=normal&p_p_mode=view&p_p_resource_id=%2Fdownload%2Fpublication&p_p_cacheability=cacheLevelPage&_search_portlet_dlFileId=34734582&p_p_lifecycle=1&_search_portlet_javax.portlet.action=search&_search_portlet_formDate=1659679906652&_search_portlet_query=ice+point&_search_portlet_source=BIPM

A water triple point is not necessary for checking most industrial PRTs. The ice point is an excellent secondary reference temperature if made correctly.

I hope this helps.

Regards,

[Pip]

Correction: I recommend the use of an excitation current of 1 mA for pt100 sensors and 0.1 mA for pt1000 sensors.

[mzzj]

You can easily check the resistance of your PRTs at (close to) 100 °C in a large pot of boiling distilled or de-ionised water. This may be sufficiently accurate to allow for the determination of their alpha values and an indication of conformance with their rated accuracy classes.

Good points, you have been obviously involved in the dark arts of temperature metrology. 
Just wanted to add that if you use boiling point of water as a reference  you should compensate for the atmospheric pressure. Local pressure variations depending on weather can cause roughly +-1cel error on boiling point temperature and if you live at high altitude the effect is even bigger. (City of Colorado Springs in US has water boiling point around 94..95Cel)

Another tidbit that might be usefull or bite your butt is if you use air pressure data from nearby airport: Airports don't inform real atmospheric pressure at that location but one calculated to sea reference level. (same applies to weather services?)
Denver International  airport( alt. 1650m) showing 1000 mBar(hPA) pressure means that the actual absolute pressure is only 820 mBar

[David Hess]

Another tidbit that might be usefull or bite your butt is if you use air pressure data from nearby airport: Airports don't inform real atmospheric pressure at that location but one calculated to sea reference level. (same applies to weather services?)

That is right, as I had to learn again when calibrating my barometer.  Weather stations work with barometric pressure "corrected" to sea level so that comparisons can be made between stations at different altitudes.  Finding the absolute pressure requires a correction for altitude.

If you have a GPS receiver with barometric pressure sensing, then it should be able to report relative and absolute pressure.

For less demanding temperature calibration, I have been thinking of picking up a pair of inexpensive precision 0.1C thermisters and using them in a bridge configuration.

[DavidKo]

As a part of the practical workshop on university we had some experiment with gas thermometer. As far as I remember:
- ice was done from distilled water (stirring was necessary)
- for boiling water was also used distilled water
- for checking the actual atmospheric pressure there was an mercury barometer with vernier scale and thermometer 
 
Formula for water boiling point that we have used is in attachment (b is atmospheric pressure in Torr).