Long-term stability of temperature sensors

[2N3055]

Yep, very good rant Splin!!

Merry Christmas to all!

Regards,

Sinisa

[MisterDiodes]

Just another thought, based on experience:

We use LM35A and have had as good or better results than Sensirion parts, and keeping it all analog keeps the system quiet.  The LM34/35A are nice in that they are ready to work if you can read an analog voltage accurately on your system anyway.

Our situation demands the least amount of digital hash around the measurement area.  We feed the temperature measure voltage into the same low emf scanner that's running other analog voltage tests on the system, and just treat it like any other analog Vmeasure in the test script.  That's actually much easier for us than trying to integrate an I2C or SPI bus into the test apparatus - and then trying to deal with the resulting noise.

We also do a cal test every year and the drift is very minimal, usually < .02C or less, and over 2 or 3 years time it tends to average out to even less - we don't really see a major long term trend up or down.  The datasheets are fairly conservative for the "A" parts - they run around $25~ or so.  Not the cheapest solution but they work very well - the cans are nice since they don't pick up board stress / vibration nearly as bad as the SMT parts.  If you have to, Add a 4-20mA current driver to the part and you're ready to ship the temp data over very long lengths of wire with virtually zero digital noise impact.


 

[Echo88]

Interesting read! Wouldnt have thought that an ice-bath could be simply produced with 2mK uncertainty; i thought the atmospheric variation error would be bigger.

[splin]

Single point calibration of NTCs raises the question of exactly how they drift - if the shape of the transfer curve drifts then it wouldn't be sufficient. Fortunately though it sems that NTC drift is typically thermometric - ie. the temperature change is constant, or an offset across the range. This was from a paper found on the Measurement Specialities website, but having been taken over by TE I can't find  an online copy. I have thus attached a copy I made earlier. (And why I always try to take copies of anything I find interesting).

...

[EDIT] Turns out I can't attach the paper as it's too big  (1.7MByte v 1.2M limit). PM me if you want a copy.

I tried a few online pdf compressors and pdf2go.com did the biz. The quality of the attached is only slightly worse than the original.

I also forgot this rather relevant link to a National Bureau of Standards paper, 'An Investigation of the Stability of Thermistors':

http://nvlpubs.nist.gov/nistpubs/jres/83/jresv83n3p247_A1b.pdf

The conclusion is to avoid disc type NTCs at all costs, and that bead types can be very stable indeed, particularly if keep below 30C. (See figs 15B-1 and 15B-2 on page 260). Some drifted less than 2mK over 2 years! Unfortunately they didn't reveal the manufacturers and part numbers for the 145 devices tested. 

[SteveP]

The article mentioned by mycroft (back on page 1) may well be an excerpt from an MSc. thesis by Anupama Kulkarni at TU Delft which is freely available online. He tested 9 NTCs and gives product numbers. Test setup is described in detail.

I can't seem to create a decent direct link to it, but search for "Low Drift, Wireless Temperature Sensor" and you should find copies from citeseerx and Delft University.

HTH,
--Steve

[CatalinaWOW]

Perhaps another way to imagine the problem is needed to see the point Andreas and others are making.

Start with an assumption of a perfect temperature sensor.  No drift.  No bias errors.  Small.  Near zero time constant. Cheap.

Now instead of trying to make the temperature sensitivity of the voltage reference zero you just use your magic sensor to servo the voltage reference to a fixed temperature.  As you start to think through the errors that might occur in such a system you can start to understand the difficulty of making a precision measurement of "the" temperature. 

If you can control "all"  of the variables it is possible to imagine precision measurements of the relative temperature over time, but there are many to control.  How does the thermal conductivity and heat capacity of the PWB vary with humidity for example.

[2N3055]

The article mentioned by mycroft (back on page 1) may well be an excerpt from an MSc. thesis by Anupama Kulkarni at TU Delft which is freely available online. He tested 9 NTCs and gives product numbers. Test setup is described in detail.

I can't seem to create a decent direct link to it, but search for "Low Drift, Wireless Temperature Sensor" and you should find copies from citeseerx and Delft University.

HTH,
--Steve

This one ? :

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1010.1132&rep=rep1&type=pdf

[SteveP]

Yup. Dunno why I couldn't make that link work. Thanks.

[massivephoton]

Just out of curiosity, how much drift should one expect from an old school mercury/quicksilver thermometer?
I know there are capillary effects, parallax and variations on the wall thickness of the glass element.
But once it is calibrated, what drift level can possibly be achieved?

[zhtoor]

The article mentioned by mycroft (back on page 1) may well be an excerpt from an MSc. thesis by Anupama Kulkarni at TU Delft which is freely available online. He tested 9 NTCs and gives product numbers. Test setup is described in detail.

I can't seem to create a decent direct link to it, but search for "Low Drift, Wireless Temperature Sensor" and you should find copies from citeseerx and Delft University.

HTH,
--Steve

NCP15XH103D03RC is the best of the lot according to this article and pretty cheap also:-

https://www.digikey.com/scripts/DkSearch/dksus.dll?Detail&itemSeq=248253183&uq=636508288350905852

regards and thanks Steve 

-zia

[MK]

Just out of curiosity, how much drift should one expect from an old school mercury/quicksilver thermometer?
I know there are capillary effects, parallax and variations on the wall thickness of the glass element.
But once it is calibrated, what drift level can possibly be achieved?

i believe that NPL in teddington still use a lot of glass/mercury thermometers, so they must be pretty stable.

[branadic]

Inspired by the discussion here I ordered a few NCP15XQ102J03RC from RS Components and connected two of them to the RDC unit of a PCap01 which is able to handle resistance from 250? up to several k? forming a RC circuit and measuring discharge time with several ps resolution. This works pretty good by the looks of it. This gives a simple but high resolution thermometer for temperatures below 0°C up to 60°C without any analog circuit. 
Results will be shown soon, after some more investigation and calibration.

-branadic-

[cellularmitosis]

This maybe of interest Investigation of Long-Term Drift of NTC Temperature Sensors with less than 1 mK Uncertainty
 [url]http://ieeexplore.ieee.org/document/7281460/]http://ieeexplore.ieee.org/document/7281460/]
 [url]http://ieeexplore.ieee.org/document/7281460/[/url]

I came across a copy of this paper recently (attached).  This is the paper which surprisingly found that the SMD part NCP15XH103D03RC actually drifted less than the glass-encapsulated types (but the glass types still drifted typically less than 2mK per year).

Abstract— Long-term drift of temperature sensors is critical to applications requiring high reliability. However, documentation and knowledge regarding long-term stability is limited. Usually manufacturers promulgate drift margins down to 10-20 mK/year, while the performance of the sensors might be much better. For some advanced industrial applications, which demand drift rates down to a few mK/year, this information is inadequate. In this paper, we present our investigation of the long-term drift of a few sets of small footprint, off-the-shelf NTC (negative temperature coefficient) temperature sensors, based on an extremely stable test setup guaranteeing stability of better than 1 mK between two calibration intervals. The results show that the SMD type sensor from Murata manufacturing (NCP15XH103D03RC), intriguingly, is the most stable sensor among the sensors tested with a drift rate of 0.492 mK/year peak-to-peak. Most of the other sensors tested have drift rates of lower than 1 mK/year, making them suitable for temperature sensing applications requiring long-term stabilities in the mK range.

Hmm, but I wonder if this is why that SMD part performed so well:

In order to keep the measurement environment uniform and to avoid temperature fluctuations, all the test sensors are enclosed in hermetically sealed, stainless steel tubes.

Our results in a non-hermetic environment might not be as good.

Also interesting was that the level of excitation voltage also affects drift (0.6VAC performed better than 1.2VAC):

...the SMD type NTC from Murata Manufacturing (NCP15XH103D03RC) performs better than the glass encapsulated bead-type NTC. With an excitation voltage of 0.6 V AC, it has a drift margin as small as 0.492 mK/year pp.

When the same SMD type sensor (NCP15XH103D03RC) was given an excitation voltage of 1.2 VAC, the drift rate was found to be about 1.704 mK/year pp. With higher voltages the effect of self-heating is more pronounced, the as the overall drift increases. Conversely, at low excitation voltages (0.6 VAC), self-heating does not noticeably affect most of the tested sensors.

and apparently DC is also worse than AC:

The glass encapsulated bead-type NTC from Measurement Specialties (46036) exhibits drift rates of 0.546 mK/year pp when supplied with 0.6 VAC. The results are in good agreement with the previous studies presented in the literature that have indicated glass encapsulated bead-type NTCs to be one of the most stable sensor types [11], [13]. However, when supplied with 0.6 VDC, the drift rate increases to 0.9 mk/year pp.

[thermistor-guy]

...
how does 1 include (or mitigate?) uncertainty of the self heating?
i could gather from the datasheet, the voltage bias needs to be very low, could it be some kind of whetstone/bridge setup and not a simple divider?
(just running some numbers in a spreadsheet, with 1 v biasing, the self heating could be around 0.04mW. which is likely a few mK of offset?)

One way is to measure the thermistor resistance with two different excitation currents, e.g. I and I/sqrt(2), then extrapolate to the zero-power resistance. Once you have a zero-power resistance versus temperature characteristic, you can estimate the thermistor's internal temperature rise due to self-heating at each calibration (temperature) point.

This first-order correction works ok if your self-heating is small (e.g. tens of uW as in your case). In principle, you could use more than two excitation currents, to estimate zero-power resistance, but I have not heard of that done in practice.

[HighVoltage]

Really nice info in this thread!
Thank you all.

I have a Fluke 5616 Secondary Reference PRT that was bought new from Fluke with a calibration certificate.
- The calibrated accuracy is ± 0.011 °C at 0 °C
- The short-term repeatability is better than ± 0.010 °C (± 0.004 °C is typical)
- The long term drift is ± 0.007 °C at the triple point of water when exposed up to its maximum temperature (420 °C) for 100 hours

Since I am using this probe carefully only for up to 100 °C, I would not expect any measurable long therm drift.
Or do I assume too much?

What I find interesting as well:
All my other 4 wire PT100 sensors have a TCR of 0.00385 Ω/Ω/°C (As far as I know)
This Fluke 5616 PRT is listed with 0.003925 Ω/Ω/°C
What is the reasoning behind a different TCR for a high end probe?

[Vgkid]

From what I've read, the .003925 (alpha) is the tc of pure platinum.  That is why this value is listed for many (s)prt thermometers.

[HighVoltage]

Thanks Vgkid

I looked a little deeper and found this:

the TCR of 0.00385 Ω/Ω/°C for RTD sensors is the European standard,  DIN/IEC60751 (IEC751)
the TCR of 0.003925 Ω/Ω/°C is the USA standard.
I had no idea that the USA and Japan would still use a different standard for RTDs.

TCR of highest purity platinum is 0.00393 Ω/Ω/°C

[texaspyro]

TCR of highest purity platinum is 0.00393 Ω/Ω/°C

Well,  there are now RTDs made of isotopically pure platinum... just send money... lotsa, lotsa money.  Not sure how swoopty they are.  Send me a few and I'll play with them. 

[Conrad Hoffman]

Just out of curiosity, how much drift should one expect from an old school mercury/quicksilver thermometer?
I know there are capillary effects, parallax and variations on the wall thickness of the glass element.
But once it is calibrated, what drift level can possibly be achieved?

Just as a reference point, the old Julie SC-106 standard cell oven used a dual oven setup, both controlled by mercury thermoregulators- nothing more than mercury thermometers with electrodes stuck through the glass. As the mercury rises and falls in the column, contact is made or broken, activating the heaters. Now, the advantage they have over an ordinary thermometer is the reservoir is large and the entire column is thus only a few degrees. They had to be purchased for a specific temperature. AFAIK, they were quite stable, the standard cells providing an immediate indication of any drift.

As for regular mercury thermometers, I've never seen one go off calibration unless the mercury column got separated, and that's usually fixable.

[czgut]

Glass exposed to temperature changes gets stress build into it. The smaller changes of temperature, the better stability. For 0'C to 30'C changes [From measurements at NBS / NIST (Special Publication 250-23)] achievable repeatability + stability/year was about 0.017 deg 3 sigma std dev.