Temperature Sensor - choice / reliabilty

[picandmix]

For measuing temerature around the 25c mark, off board at about 2 mtrs and to a resolution of 0.1c ,  wondered what device the professional designers  would choose for long term reliabily, a NTC or digital sensor like the DS18B20 ?

[linux-works]

I'm trying out the ds18b20 now.  its ok but its slow and does not seem very accurate.

what I do like is thermocouples.  they are a little expensive but they are not hard to interface to if you use a break-out board (I use adafruit breakouts to test things).

the other thing I dont like is that you can't easily shorten your own tcouple wire.  you can cut it from the bead end, but then if you want to make more from that long wire, you have to re-bead another segment and that's not easy.  I've heard you can use high current (like from a car battery) to spark the metals into forming a bead but I have not done that, only read about it.

tcouples are very high resolution since its voltage based and not a direct digital reading.  I have a fluke tcouple probe and when I watch its millivolt amplified output, I can see actual resolution down to .01 deg F and if I breathe even near the probe, it starts to climb, stop and then decrease again.  that's the kind of sensitivity that I like to see from sensors

the linear ones (lm34,35) are also not too bad.

lm75 works ok and is native i2c; I used an lm75 to manage the PID loop in my espresso machine, for a long time and it worked ok even at very high temps.


tcouples are my favorite, though.

[MichaelCarr]

The DHT11 Sensor is good.

It can be controlled by and Arduino, and the data can be sent to either an external display or the Arduino's serial monitor...

Its really good, and can be set up on a protoboard, breadboard, etc

My Reccomendation Rating? 8/10

[linux-works]

sorry, but that's a crap sensor.  I have one, I was going to use it but gave up right away when I found it to be pretty low resolution and it kept giving me the same number when small changes in temp happened; so I abandoned trying to use the 11.  I have not tried the 22 version (white instead of blue) and it may be better but its still a low quality sensor, from all I can tell.  its easy to use with arduino libs, but with my unit it was not something I found to be precise enough for anything other than a demo or learning exercise.

have you used it in a real application?

[Julian Jameson]

I have a bag of DS1822+-ND. Probably 20-25 of them in there, all new/unused.

I can't really comment on their reliability as I've not used them for anything, but it does look like they match your requirements. If you're interested, make me an offer. 

[bktemp]

I have used DS1820 many years ago, but they had some problems with moisture: After 1-2 years most of them were out of specs.
Today I prefer NTCs: They are cheap and can be connected to the adc input without any additional amplifiers. 10bit gives 0.1°C resolution around 25°C. The accuracy of modern NTCs is good enough for most cases.
Thermocouples and PT sensors provide better accuracy but are also much more expensive and need a more complex amplifier circuit.

[FreddyVictor]

how about Measurement Specialities range which are available @ Farnell
MS5611 has accuracy of +/-0.8deg C with resolution of <0.01 deg C
cheaper is MS5637 accuracy of +/-1 deg C and resolution of <0.01 deg C

[kripton2035]

I always found strange these sensors with a resolution of 0.016mbar, and an accuracy of ... 2.5mbar !

[linux-works]

to try out the tcouple stuff, its pretty easy.  at least get some first-hand experience with it.  its good, but just expensive.

you could start with:

https://www.adafruit.com/products/269
https://www.adafruit.com/products/270

and see if you like it.  you can cut that wire easily and use it easily, but like I said before, to create a new weld point is not just a matter of soldering.  you have to literally weld or fuse the 2 wires together to form a ball.  then, you can cut a small piece of that off, use that and do the same until you run out of that tcouple magic wire

and for a one-off, its definitely good enough.  I ran an espresso machine using the adafruit tcouple breakout board, that same wire stuff and some arduino code.  the temps were accurately reported and consistent, as the boiler went from room temp to literally boiling point of water (and higher, for steam setting).

so, other than the hassle of the magic wire, the need for having to fuse new contact/measure points if you want to make several sensors from a spool of wire, and the breakout board/amp is not cheap - its still the best way to get repeatable resolution.

if you do go to production, then I would hope there is a way to buy short lengths of the magic tcouple wire with the ball end already formed for you.  you can certainly buy and have boards made with the magic amp chip.  there's not much to it and it just plain works.  its just expensive and a hassle, but that may or may not be an issue for your project.

[Paul Moir]

Regarding thermocouples,

but like I said before, to create a new weld point is not just a matter of soldering

This is not strictly true.  It can be done with silver solder and a propane or butane torch.  You want to use an old pair of needle nose pliers or something like that for a heatsink for the sake of the insulation.  With a little care and some grinding you can make a pretty small bead too.

Lately though I just blast them with the TIG torch since it's quicker.  Crimp and the simple twist works too but are bulky and less reliable.

[Mr.B]

I have used DS1820 many years ago, but they had some problems with moisture: After 1-2 years most of them were out of specs...

We have been using DS18B20 for many years on our WiFi temperature transponders.
If the sensor is properly sealed inside a stainless steel probe we have had no issues.
Accuracy and repeatability are quite satisfactory.
It is a legal requirement for us to check calibration of the probes every 6 months. We have not experienced any drift in over 5 years.
The units pictured below can run for 12 months on one battery charge, sampling every 10 minutes, transmitting data every 30 minutes.
Typically they live in a freezer environment at minus 25 degrees Celsius.

[linux-works]

the dallas temp sensors I have used were the plastic transistor case style.

maybe the ones in metal cases (probes) will have better accuracy.  the plastic ones I tried were very clunky; they'd hold the same value for too long and it was like they were stuck and took more change to get them moving again.  it could even be some self-heating issues due to polling them too often (?).

I've not been able to self-heat a thermocouple even if I poll it in a tight loop.

[Mr.B]

Our ones are simply a TO-92 case stuffed down a stainless tube with a bit of thermal paste.
Response time is slowed due to all the thermal mass of the steel etc.
This doesn't affect us as we are only sampling every 10 minutes.

[bktemp]

I have used DS1820 many years ago, but they had some problems with moisture: After 1-2 years most of them were out of specs...

We have been using DS18B20 for many years on our WiFi temperature transponders.
If the sensor is properly sealed inside a stainless steel probe we have had no issues.

I did the same after I had discovered the moisture sensitivity.
If you need the mechanical protection anyway it is no problem, but adding a metal tube only to keep the moisture out is no practical solution, at least for me.

[FreddyVictor]

I always found strange these sensors with a resolution of 0.016mbar, and an accuracy of ... 2.5mbar !

agreed - all it means is that once they are calibrated, they can give accurate results 

[Mr.B]

If you need the mechanical protection anyway it is no problem, but adding a metal tube only to keep the moisture out is no practical solution, at least for me.

Agreed...
But we have stores staff jamming these into frozen product, so they need to be Gorilla proof.

[kripton2035]

I always found strange these sensors with a resolution of 0.016mbar, and an accuracy of ... 2.5mbar !

agreed - all it means is that once they are calibrated, they can give accurate results 

not even sure of that !

[dom0]

Depends on long-term drift and other dependencies. The old analog pressure sensors had such a high TC that they are almost useless without temperature compensation.

[LukeW]

There's a huge spectrum of different temperature sensors, depending on cost, accuracy, temperature range, and environmental ruggedness.

The Microchip MCP970x are an interesting one... linear, good for room-temperature applications, and cheap.

[eV1Te]

As some people hinted, there is a big difference between accuracy and resolution. Any analog temp. sensor will have infinitely good resolution if connected to a infinitely good ADC, but drift over time affects the accuracy of the reading.

My experience is that analog temperature ICs are very good in accuracy and resolution even without calibration, but they only work below ca 125 Celsius and have a slow thermal response.

Thermocouples are good for fast thermal response since they can be made as thin as a human hair, they are also good at high temperature (above 2000 Celsius depending on type). But they drift over time at elevated temperature due to diffusion of other elements into the metal. They also give very weak voltage signal, best case 40 uV/K so to get 0.01 K resolution you need 400 nV resolution on your ADC.

Pt based sensors are then a better choice for resolution and accuracy. But they are more expensive.

For low temperature (crogenic) measurment a normal diode can be used with great accuracy and resolution, since the current is highly dependant on the temperature above absolute zero.

[SeanB]

For a fast response how about the pyrometer based I2C devices used in body temperature measurement. IIRC the devices selected for human body temperature work best in the 10-35C sensing range, and are 0.1c resolution and around 0.25C accuracy in this range. Plus at 2m they do not need wires, just aim it at the surface. I bough a complete unit this week for $20, so the sensor should be cheap enough.

[volthaus]

For plain old hobby electronics the DHT22 gets plenty close enough for me.

[Siwastaja]

I'm trying out the ds18b20 now.  its ok but its slow and does not seem very accurate.

You are probably doing something fundamentally wrong - I have always found DS18B20 to be very accurate. I use it in precision process control with +/- 0.05C spec and it has always met this, at least when the range is not huge, but between 20 deg C and 45 deg C. Not sure how it performs on a wider range.

what I do like is thermocouples.  they are a little expensive but they are not hard to interface to if you use a break-out board (I use adafruit breakouts to test things).

Excuse me? A thermocouple is at least 5-10x less accurate than a DS18B20, and the accuracy of a thermocouple doesn't matter, because you'll need ANOTHER sensor to compare your reading with, because a thermocouple is not a temperature sensor at all; it's a temperature difference sensor. You may have used a module which does this with a NTC or a cheap silicon based device without knowing it, but in this case, you could have just used that NTC to do the thing to begin with and have one one source of inaccuracy less. (The OP specified operation at around room temperature.)

The reason to use a thermocouple is to measure extremely hot temperatures semiconductors cannot survive in, or, in some cases, detect quickly changing temperatures, but it's a bit hard to imagine so many real-world applications for this quick response feature.

tcouples are very high resolution since its voltage based and not a direct digital reading.  I have a fluke tcouple probe and when I watch its millivolt amplified output, I can see actual resolution down to .01 deg F and if I breathe even near the probe, it starts to climb, stop and then decrease again.  that's the kind of sensitivity that I like to see from sensors

Response time and resolution is completely different from accuracy, and it's true that many semiconductor sensors are slow. Depends on case whether this is good or bad. Thermocouple indeed shows quick temperature changes quite well, but the actual readings do drift so you don't have the accuracy of 0.1C specified by the OP (edit: OK, "resolution" was specified, and nothing was said about accuracy, but if they need 0.1C resolution, it's well possible that the typical +/-2C accuracy of a thermocouple is not acceptable).

DS18B20 doesn't either meet 0.1C accuracy by the spec, but in reality, it performs better than the datasheet says, at least under limited conditions; it's just not guaranteed. Typical K type thermocouple, OTOH, will be guaranteed NOT to have even near 0.1C accuracy in any real world application.

Any analog sensor can output any resolution. Accuracy is far more important than resolution, and I'm quite sure that the OP wanted to say "accuracy" instead of resolution.

[nfmax]

For room temperature application, NTC thermistors have the best resolution, moderate accuracy, and good reliability. They are available in a wide variety of packages including stainless steel probes. Platinum resistance probes have the best accuracy, good resolution, and very good reliability, but cost significantly more (avoid the 'film' types which aren't worth the extra cost over NTC thermistors). Again, stainless steel probes are readily available. Semiconductor sensors vary, but the best have good resolution & accuracy, and good reliability within the limitations of their packaging technology. Thermocouples at room temperatures have few features to recommend them apart from small size and consequently rapid response (though miniature NTC thermistors are almost on par)

Without knowing more about your application, it is difficult to be more specific. What are you measuring the temperature of? Is it toxic, corrosive, oxidising, a biohazard, or a food product?

One rule of thumb: traceably measuring temperature to better than ±0.5°C of ITS-90 is HARD

[Mr.B]

...DS18B20 doesn't either meet 0.1C accuracy by the spec, but in reality, it performs better than the datasheet says, at least under limited conditions; it's just not guaranteed. ...

Agree with everything you have said, particularly "it performs better than the datasheet says".

There is a Maxim application note somewhere, just cannot find it right now, that describes the math to improve accuracy significantly.

Some people here have said the DS18B20 is too slow...
It is a little slow thermally due to the plastic package, however as far as reading it is concerned, it is pretty fast.
Sure, if you run it in 'parasite power' mode it will take in excess of 700mS to get a result, but run it 'powered properly' and you will see a dramatic improvement.
Just read the datasheet.
DS18B20 is perfect for the -25 to +50 area I have used it in, and I have used it in quite a number of different designs.

Off to find that app note...

Edit: Found it: http://www.maximintegrated.com/en/app-notes/index.mvp/id/208