Author Topic: Thermistors  (Read 18766 times)

0 Members and 1 Guest are viewing this topic.

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Thermistors
« on: November 24, 2012, 04:27:29 AM »
I've been doing a bit of research on thermistors...okay...a lot of research.

Here's the bottom line that I wanted other folks to be aware of:

You cannot just go out and buy any old 10k ohm thermistor for a few dollars, stick it into  an OS sensor, and expect to get accurate temperature readings. Not gonna happen.

Want to know why? Read on...

I have seen at least one discussion here about replacing the thermistors in OS sensors,  which led me to wonder if I could improve the accuracy of the OS sensor by replacing the thermistor with a more accurate one. For example, US Sensor company makes a line of thermistors accurate to 0.1 degC and they are available from Digikey in single-unit quantities for less than $10 US. Hah! So maybe here's a neat way to improve OS sensor accuracy...? That's how I got started researching this and I wanted to share what I've learned. Please feel free to correct anything that I may have gotten wrong below.

To make a long story short, I've learned that there are two or three important properties which describe a thermistor's behavior:

1) The resistance at some reference temperature, usually 25 degC.

2) The beta-K value which indicates how much the resistance changes between two reference temperatures (often 0 and 50 degC).

3) There's also a detailed table (or polynomial curve fit) of resistance versus temperature that usually goes along with each beta-K value so that might also be considered another important property. I suspect that it is not enough to know the beta-K value for a thermistor -- the detailed table values might be different for two thermistors with the same beta-K value.

Any replacement thermistor must have the same nominal resistance at 25 degC and must also have the same beta-K value. Get either of these wrong and you get erroneous readings from the OS sensor. It goes farther than this too; the sensor requires an accurate temperature value before it can compute relative humidity readings, so the wrong thermistor can also create humidity errors.

So, what kind of thermistor does an OS sensor need?

This is the cool part.

I used a reverse-analysis technique to determine how the THGN801 converts resistance to temperature. First, I removed the thermistor from my THGN801 sensor. Then I connected many different calibrated resistance values to the thermistor terminals of the THGN801 and recorded the resulting temperature reading. If anyone wants to see my measurement data, post back here and I'll provide it.

The thermistor in the THGN801 appears to have a standard resistance of 10k ohms at 25 degC (I measured 9830 ohms), and the beta-K value I measured is 3186. Specifically, the resistance values that cause the THGN801 to read 0.0C and 50.0C are 25,960 and 4,270 ohms, respectively.

I went looking on the internet for a thermistor that would match these characteristics. The closest match I've found so far is a thermistor made by General Electric with a Material Type C6.1. The resistance values track what the THGN801 expects to within plus or minus 0.2 degC over a temperature range of about -5 degC to +40 degC. Problem is, these particular thermistors seemed to be used in applications other than temperature sensing (inrush current limiting) and are not available with the desired resitance value at 25 degC (10k ohms).

As it stands now, I have not yet found a high-accuracy replacement thermistor that would be a drop-in replacement for the THGN801's thermistor. I have not tested any other OS sensors...but I may do that soon. Given what I've learned, I could install a high-accuracy thermistor from US Sensor, and figure out a set of calibration offsets that would make things accurate to perhaps 0.3 degC...but relative humidity accuracy would suffer and I don't know by how much at this point. Not sure I want to go there just yet.

In summary, those of you who have replaced thermistors in OS sensors are at a significant risk of increased temperature errors. In general, these errors increase as the temperature departs from 25 degC, but unless you have the calibration data for the replacement thermistor, it is not possible to estimate the amount of error.
« Last Edit: November 24, 2012, 04:33:34 AM by aweatherguy »

Offline DanS

  • Chiang Mai weather
  • Forecaster
  • *****
  • Posts: 5434
    • ThaiWx
Re: Thermistors
« Reply #1 on: November 24, 2012, 05:05:44 AM »
Also have to be sure to get the proper 'direction' PTC or NTC thermistor (pos /neg temp coeficient).
« Last Edit: November 24, 2012, 05:07:17 AM by DanS »

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #2 on: November 24, 2012, 08:30:15 PM »
Good point, Dan...installing a PTC thermistor would be like missing the forest for the trees!

I've done a little more work now and have two plots here -- showning the errors that I would get with my THGN801 if I replaced the thermistor with either a US Sensor KS103G2 (0.1C accuracy, $17 at Digikey) or a PR103J2 (0.05C accuracy, $23 at Digikey).

The errors are small enough that perhaps RH readings would not suffer too much above 0C with the KS103G2, but they probably would with the PR103J2 thermistor. Correcting the temperature errors with either setup would require a correction table with entries every 1-2 degC over the range of interest (e.g. -20C to +40C) -- so maybe a total of 60 entries or so. I'm using software that supports multi-point temperature calibration offsets so that would work. Alternatively, the corrections can be done more accurately with a third-order polynomial...but I don't know if there is any software out there right now that has that option.

If I go with the more accurate thermistor, I could just ignore RH readings from that sensor and use a second OS sensor to acquire the humidity data. Between the OS sensor's 0.1 degC resolution and the 0.05C accuracy of the thermistor, I suspect the whole package might be accurate to plus or minus 0.2C or perhaps a bit better.

Note that in order for this to work it requires the calibration step of attaching various accurately known resistors to the OS sensor as part of the process. I should probably also shove the OS sensor in the freezer (with a known resistor outside at room temperature) to be sure the OS sensor does not have significant temperature drift itself. If that pans out I may give this idea a try...

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #3 on: December 01, 2012, 05:01:21 AM »
Okay, I did it. Went and bought one of the $22 super-accurate (0.05C) thermistors and replaced the one in my THGN801. I went through quite a bit of work on the math behind this, but it all distilled out okay in the end. As a result, I think I've now got a THGN801 sensor that's accurate to about 0.3C over the range from -20C to +50C. Maybe even a bit more accurate above 0C.

So far, I've verified that the temperature readings are correct to within 0.1C at several temperatures from -20C to +25C.

This mod can be done by folks with soldering skills (no surface-mount work required), about $30US and a bunch of free time on your hands. If you like tinkering and have too much free time on your hands, this is for you!

I'll be writing up the process and posting it (or a link to it) here in the coming days. The quick synopsis is this:

1) Purchase a handfull of very accurate fixed resistors (less than $10US from Digikey)
2) Connect each of them to the OS sensor and record the resulting temperature readings.
3) Shove the OS sensor in the freezer with one of the fixed resistors connected and note how much the temperature reading changes.
4) Run some math on the results to generate a set of corrections over temperature. I will provide some tools that will run in FreeMat to compute the corrections, so that part will be relatively painless.
5) Install the fancy new thermistor in the OS sensor
6) Enter the corrections from step (4) into your weather software. I'm using WSDL, but I'm pretty sure that at least one other package out there (WD maybe?) can accept multi-point corrections for temperature...can someone verify this?

There are (at least) two thermistor options -- the less acurate option (0.1C accuracy) probably does not hurt humidity readings too much but the more accurate option (0.05C) might. If you do this with a temperature-only sensor, then there's no worry about messing up humidity readings. I'm planning to try this with at least one temperature-only sensor too.



Offline Bushman

  • Forecaster
  • *****
  • Posts: 7549
    • Eagle Bay Weather
Re: Thermistors
« Reply #4 on: December 01, 2012, 07:49:55 AM »
That's cool.  Which model number?  (And I wish you could do this for a Davis unit)
Need low cost IP monitoring?  http://wirelesstag.net/wta.aspx?link=NisJxz6FhUa4V67/cwCRWA or PM me for 50% off Wirelesstags!!

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #5 on: December 01, 2012, 06:43:00 PM »
I am using one of the US Sensor interchangeable thermistors http://www.ussensor.com/products/thermistors/leaded-thermistors/interchangeable-thermistors. Specifically the PR103J2 model. I think you could also use one of the standard precision models, like PS103G2 or PS103J2. The G2 model has a Beta-k value a little closer to the original OS thermistor so the temperature corrections would be a bit less and RH errors at low and very high temps would be less (still don't know what the RH errors would be, though).

Lots of manufacturers make interchangeable thermistors accurate to 0.1C or 0.2C, and many of those would also work as long as it was 10k-ohms at 25C. Beta-k values close to the OS thermistor (3186) are preferable, but the PR103J2 Beta-k is 3890 and that's working okay. There's a slight reduction in total temperature range with the higher Beta-k, and values over 4000 might start to be an issue, even for those who don't live in Phoenix or Yellowknife.

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #6 on: December 01, 2012, 08:33:43 PM »
Not sure if I understood your question about what model? First I thought you were asking about the thermistor. If it was about the OS unit -- I modified a THGN801 but the base console could be anything that works with the THGN801 and has a USB connection for data logging (WMR100 or WMR200 for example). The temperature displayed on the base console will be incorrect with this modification and it requires the logging software to be capable of correcting the temperature errors. I'll also be experimenting with a THGR122NX sensor to see if that could be modified too.

Offline Bushman

  • Forecaster
  • *****
  • Posts: 7549
    • Eagle Bay Weather
Re: Thermistors
« Reply #7 on: December 01, 2012, 08:36:30 PM »
Sorry for being vague.  I mean which thermistor.  You answered  my Q.  Thanks!
Need low cost IP monitoring?  http://wirelesstag.net/wta.aspx?link=NisJxz6FhUa4V67/cwCRWA or PM me for 50% off Wirelesstags!!

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #8 on: December 04, 2012, 06:02:37 AM »
Just an update. I've got two sensors now with the 0.05C thermistors installed (THGN801 and THGR122NX).

I'm running some tests to see how well they track together. Also comparing them with a Sensirion SHT15 sensor at the same time. It's a painfully slow process to get stable readings at the level of 0.1C !

My test setup is an insulated box containing the sensors. The box is then placed into a picnic cooler with a fan and heat source (light bulbs of various wattages). First, I have to futz around with my test setup to create a specific stable temperature (choose light bulb wattage, etc). Then wait for it to ramp to the target temperature, which takes up to an hour or so. Then it has to sit there for 1-2 hours at that temperature before the readings settle down enough that I can make a valid comparison.

So far, I've got good results at four temperatures: 18, 26, 33 and 37C. The two sensors are tracking within 0.2 to 0.3C of each other at each temperature.

The SHT-15's accuracy at these temperatures is +-0.3 to 0.4C depending on the temperature. The readings are tracking the SHT-15 values within that level of accuracy too -- but that's not saying a whole lot since this thermistor setup should be at least as accurate as the SHT-15, perhaps even a bit better. For example, if the SHT15 is +-0.4C and the new thermistor setup is +-0.3C then the readings could be off by as much as 0.7C with both sensors meeting their accuracy specs.

Not sure just yet how I'm going to make measurements at cold temps...

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #9 on: December 09, 2012, 01:46:42 AM »
Wow, this actually seems to be working. I've modified two OS sensors with 0.05C accurate thermistors (US Sensor PR103J2 in OS THGR122NX and THGN801). They are tracking to within 0.2 or 0.3C at several test temperatures from -20C to +38C.

I've also performed the calibration process on a third sensor (THGN123N Ultra-Cold model) and the data is as well behaved as the other two, so I suspect that most OS sensors are capable of handling this mod.

They also agree with a Sensirion SHT15 sensor to within its accuracy specification (assuming the new OS sensors are good to +-0.2 or +-0.3C). The SHT15 accuracy degrades quickly away from 25C, so this mostly says, "yeah, they could be accurate to 0.2 or 0.3C", but it is far from proof. As I don't have anything more accurate to compare against, I cannot claim for certain that I've achieved this level of accuracy...but the odds are good methinks.

There's a document available in the zip file (link below) that describes how to do this modification. The zip archive includes analysis scripts for Matlab or FreeMat and data for the US Sensor G and J series models if you wanted to go with those units. They are available in 0.05, 0.1 and 0.2C accuracy levels. It is a fair bit of work but relatively inexpensive. If you have rudimentary soldering and mechanical skills it should be doable.

In principle, there's no reason this mod would not work on other brands of wireless temperature sensors too.

www.osengr.org/Downloads/OS-Thermistor-Calibration.zip


Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #10 on: January 17, 2013, 08:51:24 PM »
Another thing I've discovered about thermistors -- getting them too hot during soldering can instantly cause a permanent shift in temperature readings. I had to replace some of the 0.05C accurate thermistors because I was not careful heat sinking them during installation. They got too hot and now they are reading 1 degF off or more. From now on I'm going to just crimp them in place to eliminate the risk of over-heating.

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #11 on: September 22, 2013, 10:28:03 AM »
Assuming you attached the new, more accurate thermistor as a probe on the end of a wire, you could temporarily sheath it (to waterproof it) and submerge it in ice water to get a highly accurate zero reference.  That should be easy to do and should also should beat putting it in the freezer and having to guess at the temperature inside the freezer (or having to rely on a measured value which could be inaccurate).

I don't know of a similar phase change material to establish a super accurate 25C datapoint, but since you're good at research, maybe you can find one?

How long does your calibration remain accurate?  i.e. How often do you need to re-calibrate to maintain accuracy? 

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #12 on: September 23, 2013, 01:17:21 AM »
I think I kind of skipped a full explanation of the use of the freezer. Sorry about that.

First, I'm relying on the thermistor manufacturer to have provided me a sensor that meets the +-0.05C spec. From this point of view, there's no need to confirm against a known standard like ice water. I just need to turn the OS electronics into an accurate ohm-meter and leave the temperature standards to the thermistor manufacturer.

The OS sensor electronics appear to have a slight amount of drift, which behaves like a small offset current that varies with temperature. The following behaviors led me to this suspicion:

1. Attach a small fixed resistor (e.g. 10k) and run the OS electronics between room temp and freezer -- no change in reading
2. Attach a larger fixed resistor (e.g. 90k) and there is a noticeable change in reading when the OS electronics are heated/cooled.
3. Double the size of the resistor in (2) (e.g. 180k) and the change in reading is doubled between room temp and freezer.

If this is correct, and because the offset or error current would be very small, it is only noticeable when the thermistor resistance is very large -- i.e. when it is cold. At most this amounts to 0.5C at an equivalent temperatur of maybe -30C or so.

The freezer experiment is a way to measure the drift. The goal is to subject the electronics to as large a change in temperature as possible while a stable, fixed large resistor is connected by wires outside the freezer (and not subject to temperature changes).

In summary, I'm assuming several things here:

1. That this small drift is repeatable.
2. The drift is stable over time.
3. The drift is due to a small electrical "offset" or "error" current.
4. The offset current varies linearly with temperature between -40C and +30C or so

I have not completely proven these assumptions (although I think they are reasonable), so there is some risk there.

Anyway, the idea is to install a very large fixed resistor on the OS electronics to simulate a very cold thermistor. Then read the sensor at room temperature and in the freezer and see how much the reading changes. Then a little math and I've got some additional corrections for the calibration offset table -- and these numbers are very very small until things get quite cold. At temps above 0C they can be completely ignored.

It will have been about a year since I calibrated one of these units in December. I'll go back then and re-check it to see how the calibration has held up.


Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #13 on: September 23, 2013, 10:47:06 AM »
Thanks for clarifying that.  So, one of the concerns is that the fixed resistor might change its resistance too much if it were put in the freezer?  Since your assuming the thermistor meets the 0.05C spec, which is pretty tight, would another (not quite as good, but almost as good) way to do it be to just attach the thermistor instead of the fixed resistor and put the whole kit-and-caboodle into the freezer?  If so, then maybe you double zip-lock bag the whole kit-and-caboodle and submerge it in icewater and wait for it to settle out at exactly 0 degrees C.  Before that, just put the thermistor into the icewater (sheathed), and measure with the electronics at room temperature.  i.e. maybe you gain some accuracy from the precise 0 degree reference, even if you lose some accuracy by using the thermistor rather than the constant temperature fixed resistor.  Just an idea.  I'm only guessing as to whether that might be a good trade-off.  Seems like it might be, but maybe it isn't.  Most people keep their freezers somewhere safely below freezing rather than right at freezing.

By the way, on a related but slightly different topic: if I already have a pretty accurate ohm meter, can I simply measure the ohms of the thermistor and--after maybe collecting some calibration points and doing some calculations--have a pretty accurate reference thermometer?  I'd venture that my Fluke ohm meter is better at being an ohm meter than the THGR810 is.  I'd like to have a darn good reference thermometer, and I suppose this would be a lot cheaper than just buying one.  I found a $70 NIST traceable digital thermometer on Amazon that claims to be accurate to +-0.3C, but your 0.05C thermistor might cost less (remind me the rough market rate for a quantity 1 purchase?) and might even be more accurate if driven by my Fluke volt-ohmmeter (which, to put it in perspective, cost me around $200 when I bought it a couple years ago).

Offline SLOweather

  • Global Moderator
  • Forecaster
  • *****
  • Posts: 3456
    • Weatherelement Moline IL
Re: Thermistors
« Reply #14 on: September 23, 2013, 11:02:37 AM »
If you already have a Fluke DMM, you could consider getting the thermocouple accessory for it. The model will depend on the meter you have.

Look for model 80TK, 80BK-A, etc. I have an 80TK which I find very useful and very fast reacting.

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #15 on: September 23, 2013, 12:32:44 PM »
If you already have a Fluke DMM, you could consider getting the thermocouple accessory for it. The model will depend on the meter you have.

Look for model 80TK, 80BK-A, etc. I have an 80TK which I find very useful and very fast reacting.


Thanks for pointing that out.  However, if I'm understanding correctly, the US Sensor PR103J2 is insanely accurate (+- 0.05C) and only costs about $8.12 (cf Digikey) in single unit quantities.  Not 0.5C, but actually 0.05C.  The Fluke 80TK's accuracy appears to be 0.5% +-2C and costs more than $100.   So, unless I'm overlooking something huge, that would seem to indicate the 80TK is 10x the price but more than 40x less accurate.  Is that right?  I don't make thermometers for a living, so please do call me on it if I've gone off the rails.

It appears to be a simple table lookup (http://www.ussensor.com/ultra-precision-interchangeable-thermistors-pr103j2-table) to convert from ohms into degrees C.  Also, there's a fair number of ohms separating each 0.05C (e.g. 25C is 10,000 ohms, and 24.95C is 10,021.95 ohms).

So, on the face of it, it would seem that all I need to do is attach the thermistor to my DMM, and bang, I've got an accurate thermometer as long as my DMM is "accurate".  Let's call doing that simple thing the "essential notion."  Does the essential notion actually capture the essence of how to convert my DMM into an accurate thermometer?  

If so, just how accurate does my DMM need to be to cash-in on the accuracy afforded by the PR103J2 thermistor?  What spec should I look for on my DMM's spec sheet to know if it's accurate enough or not?  I guess if it can discriminate between the ohms of each increment or decrement of 0.05C, then it should be accurate to at least 0.05C?

I realize I may have to do something similar to what aweatherguy is doing with the THGR810 to achieve maximum accuracy.  I'm not sure what, exactly, as I haven't thought that part of it through yet.   If nothing else, interpolating between a bunch of calibration points would add confidence and wouldn't hurt.  On the other hand, maybe two calibration points would be sufficient if that's all that's needed for the THGR810.  

Anyhow, setting aside calibration for a moment, can the basic method really be that easy?  Am I correctly capturing the essential notion of how to make my DMM into a surprisingly accurate thermometer?  Somehow it sounds too good to be true.  It sure would be awesome if it was true though.
« Last Edit: September 23, 2013, 02:41:53 PM by NeverDie »

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #16 on: September 23, 2013, 03:31:33 PM »
Yes, that's the concern. A typical semi-precision fixed resistor will have a temperature coefficient of 50 or 100 parts-per-million-per-degC. Over the test temperature range of 40C then, it would change by either 0.2 or 0.4%. The Thermistor changes by 0.2% per degC (roughly), so this amount of variation in the "fixed" resistor would be too much. It's really not that much trouble to attach the wires.

Also, keep in mind that since I'm doing this measurement over a very wide temerature range and the resulting corrections are quite small (0.5C or less) that errors in my measurement of the freezer and room temperatures have to be very large to have a significant effect on the outcome. I already have an OS wireless sensor in the freezer and as long as it is within 2-3C I'm good to go.

Certainly you can use your ohm-meter with any thermistor. If you have an interchangeable unit with calibration table that's all you need. If you don't have the calibration table it would be very difficult to do it yourself (even with 2-3 good calibration points) because you don't even know what the actual curve looks like.

If you plot the logarithm of thermistor resistance against the inverse of absolute temperature (e.g. Kelvin or Rankine) you get something that's close to a straight line. The slope of that line (between 0C and 50C usually) is called beta-k. But if you want say 0.1C accuracy, then it is not really a straight line; a high-order polynomial can be fit to that with pretty good accuracy so you don't need the calibration table though. Anyway, without data from the mfr, or your own adjustable accurate temperature reference, just measuring 2 or 3 points on the curve is not good enough to nail down the whole curve.

The US Sensor thermistor in question changes resistance from 10,000.00 ohms to 10,021.95 ohms from 25.00C to 25.05C. If your DMM has a 1-ohm resolution and perhaps 2-5 ohms accuracy at 10,000 ohms (0.02% - 0.05%) it would not significantly degrade the accuracy of the thermistor. These resistances are high enough that you would not need to worry about a 4-wire ohms measurement.

Everything needs to be clean and free of dirt, oil and other residues. At 0C for example where the thermistor is at 32k-ohms, all it would take is spurious parallel resistance of 13 meg-ohms to create an additional 0.05C error. You need to keep that unwanted stuff larger than say 60meg-ohms at 0C to avoid additional errors. And it gets worse the colder you go.

You can learn more about thermistors at many places on the internet. I think US Sensor has some educational materials and so does Omega (www.omega.com). Or just do an internet search...

Also on the subject of mesuring temperature to high levels of accuracy, measuring air temperature to an accuracy of say 0.1C is an interesting proposition. If you had a probe that could accurately and instantly measure the temperature at one small location like the head of a pin, you would get a surprise. The spot temperatures would fluctuate widely over time and position (at least several tenths C, perhaps more) in a normal room with quiet air. There are little pockets of warmer and cooler air just floating around all the time; I tried this with a thermistor and was surprised how noisy the measurements really are. I wound up adding a thermal mass (small piece of brass) to the thermistor just to filter out the short-term temperature fluctuations. I think this is partly why OS sensors have the thermistor inside a small plastic pocket with limited access to outside air -- that also acts like a low-pass filter to reduce the short-term fluctuations.

You may see a reference to a "well-stirred oil bath" on some of the thermistor and high-accuracy thermometer data sheets. They have found that if you place a thermistor in an oil bath with a known accuracy thermometer, the oil bath must be well-stirred to eliminate temperature differences between the two thermometers. Even in a liquid like oil, there can be significant temperature changes over short distances if the oil is not continuously stirred.

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #17 on: September 24, 2013, 08:02:35 AM »
First and most importantly, thank you for your awesome post.   =D>


The US Sensor thermistor in question changes resistance from 10,000.00 ohms to 10,021.95 ohms from 25.00C to 25.05C. If your DMM has a 1-ohm resolution and perhaps 2-5 ohms accuracy at 10,000 ohms (0.02% - 0.05%) it would not significantly degrade the accuracy of the thermistor. These resistances are high enough that you would not need to worry about a 4-wire ohms measurement.


I looked at taking the next step, and it looks like I would need at least 4-1/2 digits (19999 counts) on my DMM to take the measurements in your example.  Is that correct?  Most Flukes, including mine, have only 3-1/2 digits.  The 4-1/2 digit models cost >$400.  Fluke doesn't make handheld DMM's with more than 4-1/2 digits.  

Admittedly, that's just one vendors line-up.  I'm not beholden to Fluke, and there might be good  alternatives.  However, it looks like number of digits would be an important selection criteria.  So far, it's looking like more digits implies more dollars.

It's ironic that it would take such a high-end DMM to barely do the job.  I haven't reasoned beyond the above, so I'm not sure whether even 4-1/2 digits would be enough to cover all the measurements that would be needed for a useful range of temperature measurements.  

Even a dirt-cheap calculator has 8 digits, so I guess there must be basic reasons relating to electrical measurement as to why even fancy DMM's have so few digits.  

Anyhow, given all of that, it's even more impressive that you managed to turn the THGR810 into a better ohm-meter than my Fluke. Most Flukes, actually.   At least for me, that's a counter-intuitive outcome.  Also, at least for me, a counter-intuitive outcome is often (though not always) a sign that there's a disconnect somewhere.  I'm not saying that's necessarily true in this case, but it does have me scratching my head a bit.
« Last Edit: September 24, 2013, 09:02:46 AM by NeverDie »

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #18 on: September 24, 2013, 03:59:03 PM »
So, what would you get with a 3-1/2 digit DMM? Resolution is 10 ohms at 10k ohms, which is about 0.025C resolution. That's half of the thermistor accuracy...not great but still workable perhaps. Just below that (between 9k and 10k) the resolution is the same -- 10 ohms and remains that way down to about 3200 ohms or so where the DMM switches back to 4 digits. Consider a measurement at 50C (3602 ohms on the thermistor). The 10-ohm resolution amounts to about 0.07C temperature change. This, and at about 32k where it switches from 10-ohm to 100-ohm resolution are the worst points and your DMM would not quite provide you with 0.05C resolution.

The next question is accuracy.

The percentage change in resistance with the thermistor varies with temperature. The worst case is at 50C where you get about 0.2% change in resistance per 0.05C change in temperature. To comfortably eliminate your DMM as a source of error you want it to be 10x more accurate -- 0.02%. At -20C things are better -- you get 0.3% resistance change per 0.05C change in temperature.

So, I don't know what the accuracy spec on your DMM is and what the recommended calibration interval is or if you've had it recently calibrated. If it is outside the calibration interval then Fluke does not guarantee the accuracy -- even though it may be in spec...they just won't guarantee it. I've got an old Fluke 77 that's probably 30 years old and it still meets specs whenever I check one of the functions against an in-calibration 6-1/2 digit DMM. It's never been re-calibrated.

I looked at the Fluke website; the 27/28 models do not specify any accuracy on resistance. The model 83V is 0.4% plus one least-significant digit, and the 87V is 0.2% plus one digit. These two numbers correspond to 0.10C and 0.05C error respectively, if the thermistor is at 50C. So, you can see that getting enough accuracy is also a problem.

You can still do this...but the DMM will be a significant contributor to the total accuracy that you get. In the above example, your worst case error at 50C would be 0.05C (thermistor) + 0.07C (1 DMM digit) + 0.10C (83V accuracy) for a total of 0.22C accuracy. If you want get statistical and RSS these numbers, you would compute 0.13C accuracy. Still not too bad, just not as good as you'd get with a more accurate DMM.

So why are 4-1/2 digit DMMs so expensive, and wireless sensors so relatively inexpensive? Because DMMs have to do it all -- DCV, ACV, DC current, AC current, ohms -- and they have to do it over a huge range of values, from fractions of an ohm to 10's of megohms or more. A wireless sensor only has to measure resistance, and only over a relatively small range of values -- say from 3k ohms to 150k ohms or something like that. These sensors can use circuits and components which are limited in scope and make use of design tricks that would not be applicable to a DMM.

For example, the OS sensors use the thermistor in a circuit very similar to the old 555 timer IC to convert the resistance value to a frequency -- then they measure the resulting frequency. That's a very inexpensive circuit that only requires a couple of components that are very stable -- everything else is a don't care. You could never get away with that in a DMM.
« Last Edit: September 24, 2013, 04:03:34 PM by aweatherguy »

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #19 on: September 25, 2013, 11:43:00 AM »
That's awesome. Thanks for explaining that. With that as background, the result no longer seems counter-intuitive. 

As far as temperature measurements go, does it matter at all which device gets upgraded to the more accurate thermistor?  Would they all report the temperature about the same (within the accuracy tolerance) because the temperature circuitry is the same in all of them, or are there differences in the temperature circuitry that would make--after thermistor upgrade--one model more accurate than another as far as the temperature it reports?

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #20 on: September 25, 2013, 03:38:16 PM »
I think your choices on "which one" fall into two camps -- the old version 2.1 sensors (e.g. THGR122NX or the ultra-cold model), and the newer ones (801/810). While the older units are slightly better behaved in terms of the actual correction data (and considerably less expensive), the difference is not large and you don't have a choice anyway if you're using an OS console which requires the new 3.0 version sensors. There's no difference as far as I can tell between the 801 and 810 units.

One recent discovery I made working with modified sensors has to do with high humidity. I had a unit in an outdoor shield and after about 36 hours of continuous fog, it started reading higher temps (by about 0.5C). Turned out that condensation had formed on the thermistor body and the leakage current corrupted the readings. There is very little distance between the two wire leads as they leave the thermistor body. After the fog left, the readings were still a few tenths high and did not return to normal until I cleaned off the thermistor body with some isopropyl alcohol.

I fixed this problem by embeding the lower part of the thermistor body and the wire leads in some electonics-grade epoxy. The thermistor sits within a small plastic pocket in the OS sensor body -- I arranged for the wire leads to exit the pocket through separate small holes about 1/4" apart (drill new holes if necessary). I placed a small blob of epoxy so that once inside the pocket, the wire leads are completely contained with the small blob of epoxy. The lower part of the thermistor body is also inside the epoxy. With this fix, the unit has been through several days of 100% humidity now with no problems.

If you plan to have a modified sensor in high humidity I'd highly recommend this tweak.

P.S. If you try this and run into any problems, you can PM me on this site and I'll try to help. It may take as long as a week or so for me to respond to PMs though, so be patient  :grin:


« Last Edit: September 25, 2013, 03:40:53 PM by aweatherguy »

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #21 on: September 25, 2013, 04:24:54 PM »
Thanks.  If you ever get a chance to post a picture of your epoxy handiwork, then it will probably be self evident what to do.  I'm not sure what qualifies as electronics grade epoxy, though.  I'm planning to put an 810 into the Davis 7714 when it arrives today.  It does get humid here, especially during the summer.  Much of the time this summer it felt almost tropical.

I have five 810's plus a bunch of other 433Mhz wireless temp sensors, some of which also report humidity.  I'm using an RFXCOM to read them because it can handle reading multiple sensors on the same channel.  It can read all 3 versions of Oregon Scientific simultaneously.  I'm aware of someone reading twenty 810's (2 per channel) with RFXCOM for over a year without experiencing any problems.   It can handle different brands, different sensors, and even different protocols (like X10) all simultaneously.  The antenna is easy to upgrade, and you can also create a network of them to further extend coverage.  I have no affiliation with the vendor other than as an end-user customer.   I only mention it in case you ever find you want or need those capabilities and because I noticed you are/were building something similar with an arduino.  
« Last Edit: September 25, 2013, 04:44:55 PM by NeverDie »

Offline aweatherguy

  • Senior Contributor
  • ****
  • Posts: 288
    • Weather Station Data Logger
Re: Thermistors
« Reply #22 on: September 26, 2013, 12:16:44 AM »
You just want to get an epoxy that has high electrical bulk resistance (10^10 ohm-cm or greater). I suspect that most epoxies will work but many do not specify volume or surface resistivity. There's a Potting Epoxy from GC electronics (19-823), and they also sell a silicon sealant (19-155,19-158,19-159) that would work and you could disassemble later too.

If you can't find anything quickly, just get a quick-set epoxy from the hardware store. Arrange two clean bare, small-guage copper wires very close together (maybe 0.050 inch spacing) and let a blob of epoxy harden on them for a day or two -- then check the resistance between the wires with the DMM -- it should be off-scale overrange -- no reading.

Since you're using a non-OS receiver, the THGR122NX or ultra-cold model would be cheaper if you're buying a new unit. It is a bit more difficult to open the thermistor pocket on these; it has a plastic cover that you need to pry up with an Xacto knife or similar tool, and is sealed with some really goopy, sticky stuff -- but it will come off just by pulling a lot.

I've been using a "WxShield" receiver (as shown over there on osengr.org) for several years now -- same characteristics as what you have in terms of sensor support. I'm running 2-3 THGR122NX sensors on each of 3 channels; no problems and no complaints.

I'll try to post a photo of the epoxy job at the next opportunity I have to take the sensor apart.

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #23 on: October 03, 2013, 12:30:29 AM »
I have an arduino on order now.  What is the most accurate temperature sensor that plays well on an arduino?  Would it be the same US Sensor PR103J2, or would it be a digital temperature sensor instead?

Offline NeverDie

  • Senior Member
  • **
  • Posts: 87
Re: Thermistors
« Reply #24 on: October 03, 2013, 08:46:07 AM »
If nothing else, the SHT25 (http://www.sensirion.com/en/products/humidity-temperature/humidity-sensor-sht25/) would be more accurate for temperature and humidity than what the Davis is currently using (the SHT11 if I remember right), and it's output is (I^2)C digital.  It looks to be more accurate than the DHT22, which seems to be the most accurate of what I've seen in starter kit collections of arduino sensors.

Presumably, as per earlier discussion above, using an analog sensor like the PR103J2 would require some circuitry to make the arduino into an accurate ohm meter.  Would that approach yield better temperature accuracy than the SHT25, which wouldn't need that?

I'd like to have at least one sensor that's extremely accurate to serve as a reference for generating corrections for my other (mostly Oregon Scientific) less accurate sensors.
« Last Edit: October 03, 2013, 08:50:59 AM by NeverDie »

 

anything