Advertisement

Blog

Temperature Measurement, Part 1

This series of blogs was first published on the late, lamented forum, Microcontroller Central. I have modified it very slightly. In my opinion the best place to start whenever you discuss temperature measurement is an Agilent (actually it dates back to HP) application note, which gives a great background to the whole subject. So take time to read it before continuing here because I don’t think I could say it any better.

Finished? Let me say that I disagree with some of the statements in the app note. On page 3 the list of advantages and disadvantages for each approach differentiates between the RTD and the thermistor when both are simply temperature dependent resistors. You can get 2, 3, or 4 wire RTDs and you can even connect a 4 wire as a 2 or 3 wire. Nothing is stopping you from connecting 3 or 4 wires to a thermistor either just as long as your input circuitry caters to it.

The RTD can also suffer from self-heating depending on how you drive it and we will get to that as well. Another minor quibble is that Agilent talks about IC Sensors, when you could just use a diode junction instead. I am not saying that it is better, but it implies that you can only use ICs.

There are one or two other techniques. The first is a pyrometer which measures radiation (normally infra-red) and deduces the temperature from that. Often this is nothing more than a lens that focuses the radiation onto a thermocouple and so it has a familiar interface. A thermopile is made up of several thermocouples normally connected in series to increase the voltage output. It is commonly used in natural gas burners to detect the flame as part of the safety arrangement so that no gas flows if there is no flame.

My impression of the pyrometer technique is that it is rather inaccurate based on the fact that the radiation is affected by the color, size, shape and distance from the object. I got this impression from use of instruments like the Fluke 65. I had thought that this approach was quite esoteric. I came across the TMP006 from TI which appears to integrate a thermopile into a PCB mounting package. The user’s guide provides some more information. Our own Victor Lorenzo pointed out that Melixis makes a similar device MLX90614 as well as GE. Seems there may be a bigger market than I anticipated.

A variation on the temperature dependent resistor is the bolometer which is a blackened metal plate connected in a Wheatstone bridge configuration. It absorbs heat which unbalances the bridge in relation to the radiation. Apparently it can be quite accurate.

Omega is a well known supplier of sensors for the process industry and produces some great catalogues with plenty of information. The temperature sensor you end up choosing is often determined by the application. For instance, if you are measuring the temperature of a furnace, you will probably be limited to a thermocouple because of the temperature range. The cost evaluation should be based on a system approach and not just on the sensor itself.

As a case in point, the thermocouple normally will need a second temperature sensor for its cold junction compensation. Also if you have a narrow enough temperature range you may not have to linearize your readings. Transferring the heat from the source to the temperature sensor can also be challenging. How do you affix it? Do you need to improve the thermal conductivity with thermal compound and how long will it take to reach thermal equilibrium? All are non-trivial factors that must be considered even if not readily apparent.

Table 1: I borrowed this table from Cypress since I prefer it to the Agilent table. I also added a bit.

Temperature is one of those physical properties that held up as proof that the real world is analog and is also a measurement that is made very often. Temperature normally changes very slowly and so there are no high speed demands on the electronics. The first stage of any temperature measurement design is to decide at which point to convert to digital, if at all. Dedicated hardware for thermocouple and RTDs with an analog output abounds and has been around for many years. Before you dismiss the approach as passé, consider this: Sometimes the sensor can be in rather an inhospitable environment and it would really be nice to have the digital electronics in a friendlier place.

Of course you can convert the signal to digital and send it by some protocol, but there are single chips that will condition the temperature to an analog signal like a 4-20mA analog current loop. Many, if not most, industrial plants use this technique extensively and is my employer’s bread and butter. See Figure 1.

Figure 1: This is a 3-wire RTD to 4-20mA converter using only one chip. The temperature range is determined by resistor selection (the two through-hole ones at the centre bottom and the one to the left of the IC).

Figure 1: This is a 3-wire RTD to 4-20mA converter using only one chip. The temperature range is determined by resistor selection (the two through-hole ones at the centre bottom and the one to the left of the IC).

This is an extensive topic and I am going to drag it out over several parts, so this seems like a good place to stop. Next blog I will deal with analog conversion techniques before moving on to ADCs and micros along with digital sensors. Watch this space for more! In the meantime, any comments?

29 comments on “Temperature Measurement, Part 1

  1. Davidled
    March 6, 2014

    I believe that each industry uses a different type temperature sensor. For example, in the Auto industry, in the most case, thermocouple is used to measure temperature around battery and engine compartment. I wonder which application needs the high resolution of temperature measurement.

  2. RedDerek
    March 7, 2014

    I am curious as to the accuracy one wants to measure temperature. I understand if one is performing experiments and temperature is crucial, a 0.001 degree change may be important. But for general use, industrial and commercial, would not an accuracy of 0.2 degrees F, or 0.1 degrees C be sufficient? And even then, I feel this may be too accuracte and doubling, if not tripling these numbers would still be acceptable.

  3. antedeluvian
    March 7, 2014

    Derek

    I am curious as to the accuracy one wants to measure temperature

    In my opinion you are 100% correct. I actually address this in part 4 of the series. In my experience the perison who creates the specification picks a number he thinks is reasonable without considering how you even judge what the actual temperature is in order to validate the specification.

    In part 3 of thi series I touch on some work I did with a calorimeter which measures the temperature rise caused by combustion. In this case accuracy is important, but it was a relative measurement and I don't think it was more accurate than a degree or two.

  4. kendallcp
    March 8, 2014

    Great to see a detailed series of articles on temperature measurement.  One factor I try to get across when I'm training people on that is that accuracy of the sensor is difficult to unpick from effectiveness in getting the sensor to sense where it's needed.  For me there's no doubt that the most cost-effective way of buying accurate temperature measurement is the humble NTC thermistor.  These are mass produced at super low cost with highly repeatable and known characteristics, and require no accurate local voltage reference to get you the right answer.  Sure, these characteristics are not intrinsically linear.  But the non-linearity is very well understood and trivial to correct in the digital back-end these days.

    However, you've got to get the sensor where it's needed.  It often turns out that the things that need the most accurate sensing aren't a good fit for PCB-mounted components.  RTDs and thermocouples with welded cable connections and insulated and/or isolated housings tend to be required.  This kind of construction costs more money – it's the manufacturing that makes up the cost, not the exotic-ness of the materials.

    Looking forward to the next instalment!

  5. antedeluvian
    March 8, 2014

    kendallcp

    But the non-linearity is very well understood and trivial to correct in the digital back-end these days.

    In my 3rd part (if I remeber correctly) I will discuss the PSoC3/5 which has built in IP for linearising the thermistor (as well as thermocouple and RTD).

     

    Thanks for the kind words!

  6. Davidled
    March 8, 2014

    Nonlinearity is the characteristic of sensor. In the most cases, at saturation point, this characteristic would be generated.  Do we really the concern about the nonlinearity range in the normal operation? When system gets the saturation point, the Limp Home mode could be setup.  Cost and location of mount in the PCB board would be considered for sensor selection. Personally, I would like to see the size of temperature sensor as small as possible like MEM based temperature which may have a lot of application in industry area.

  7. Netcrawl
    March 9, 2014

    @Daej you're right high precision temperature measurement can be only achieved through the use of well-designed and suitably calibrated sensors however the accuracy of this tools will be meaningless unless the senosrs are used correctly. For accurate measurement calibration is “must things” and very important, sensors and instruments should be calibrated together as a system

  8. Netcrawl
    March 9, 2014

    @anteluvian I think accuracy is quite confusing, in most electronic devices accuracy means to how closely a certain device reflects actuality and in order to precisely define this ability for temperature sensors you need to prove the traceability to an absolute.  

  9. Netcrawl
    March 9, 2014

    @kendaclip I agree with you RTD, RTD is probably one of the most accurate tempearture sensors, it has a wide range of temperature measurement with some great advantages over other temperature sensors, they're the most accurate and stable of the different temperature measurement devices and they are not linear but RTDs are quite expensive and require current source. 

  10. Netcrawl
    March 9, 2014

    @kendaclip you're right about thermistor, its probably the most cost-effective way of buiyng accurate tempearture measurement, They have fast response and much greater output than RTD, although RTDs still the best choice when it comes to extreme accuracy but Thermistors is not far behind its getting closer. Thermistors provide high accuracy about 0.1 to 1.5 C but only work in limited tempearture range -100 to 300 C.  Thermistors is probably the best choice for small changes in temperature because of its high sensitivity. 

  11. Netcrawl
    March 9, 2014

    @Daej I agree each industry uses different type of temperature sensor and each sensor had its won advantages and disadvantages, choosing the right sensor is very important getting accuracy of the tempearturte measurement.

    Thermocouple is probably the most commonly used sensor in the market. These are low-cost, self-powered and can be run long distances. They can also last or survive in many different environment, even exposure to harsh working environment, perfect for factory . 

  12. samicksha
    March 10, 2014

    When a current flows through a thermistor, it will generate heat which will raise the temperature of the thermistor above that of its environment. If the thermistor is being used to measure the temperature of the environment, this electrical heating may introduce a significant error if a correction is not made.

  13. etnapowers
    March 10, 2014

    Aubrey: you're right, the technique in subject is not accurate, in certain cases the relative error can be figured out as 10%. The strenght of this technique is the wide range of temperatures that can be measured, and the easy usage for the operator.

  14. etnapowers
    March 10, 2014

    The pyrometer technique presents one more weakness: it depends of the resolution of the human eye. The operator has to compare the radiation coming from a heated wire with the radiation coming from the heat source, hence this technique is inaccurate as the human senses.

  15. etnapowers
    March 10, 2014

    ” Transferring the heat from the source to the temperature sensor can also be challenging. How do you affix it? Do you need to improve the thermal conductivity with thermal compound and how long will it take to reach thermal equilibrium? All are non-trivial factors that must be considered even if not readily apparent.”

     

    Aubrey, that's absolutely correct, I think that all the factors that play a role in the temperature measurement process have to be taken into consideration. The correct modeling and designing of the application board can impact the overall accuracy of the measurement as much as the precision of the temperature sensor.

  16. etnapowers
    March 10, 2014
    “Before you dismiss the approach as passé, consider this: Sometimes the sensor can be in rather an inhospitable environment and it would really be nice to have the digital electronics in a friendlier place. “
    Correct. I would add that the requested accuracy of temperature measurement is another factor to be taken into consideration. Let's consider for example that the accuracy of thermistors can be less than 0,0001 °C.The thermistor is utilized for high precision temperature measurement purposes.
  17. Victor Lorenzo
    March 10, 2014

    @kendallcp,

    >> (…)the humble NTC thermistor.  These are mass produced at super low cost with highly repeatable and known characteristics, and require no accurate local voltage reference to get you the right answer .

    I agree with you, thermistors exhibit a very attractive overall performance-to-cost ratio.

    How do you obtain an accurate measurement without using a precision voltaje or current source at a reasonable cost?

  18. Victor Lorenzo
    March 10, 2014

    @DaeJ,

    >> Do we really the concern about the nonlinearity range in the normal operation?

    We normally need to calculate how far these nonlinearities will take us from the 'real' value.

    We must also consider and add the intrinsic sensor error, the repeatability error, the nonlinearity error and all other AFE errors involved in the measurement system to make sure the final accuracy will conform to application requirements.

    In some extreme situations we could need to even take into accout the type of arithmetic libraries we're using.

  19. Victor Lorenzo
    March 10, 2014

    @Netcrawl,

    >> Thermocouple is probably the most commonly used sensor in the market. These are low-cost, self-powered and can be run long distances .

    I agree with you, thermocouples are present in many many applications, specially K-Type in industrial applications, and provide best in class accuracy (comparable to Pt-RTDs).

    I wouldn't say thermocouples are low-cost, at least not compared with thermistors, some IR and most IC temperature sensors. Up to some extent we could say they are “self-powered” but thermocouple AFEs require special cabling (alloy depends on thermocouple type), local temperature measurement for cold-junction compensation and precision amplifiers.

  20. Victor Lorenzo
    March 10, 2014

    @Samicksha,

    >> When a current flows through a thermistor, it will generate heat which will raise the temperature of the thermistor above that of its environment. If the thermistor is being used to measure the temperature of the environment, this electrical heating may introduce a significant error if a correction is not made .

    You're absolutely right on that. Taking into account self heating is crucial, not only in thermistor applications, RTDs also suffer from this effect.

    Correcting self heating is complex and involves taking into account the sensor's thermal resistance (thermal time-constant), the reference current, and how the locally produced heat is conducted to the target object (solid, fluid, gas).

    Some discontinuous sensor excitation techniques can be used to reduce self-heating effects.

  21. Victor Lorenzo
    March 10, 2014

    @Etnapowers,

    >> The correct modeling and designing of the application board can impact the overall accuracy of the measurement as much as the precision of the temperature sensor .

    I agree with you on that.

    A few weeks ago I started writing a blog entry describing a temperature measurement application that exhibited some peculiarities. It is part of a project I worked on and I still have the hardware/software development platform I used on the project (and the client's permission to use it for experimentation).

    If it doesn't interfere with Aubrey's article series and I can find some time to complete and post it, it could serve like a distracting engineering excercise for some of us.

     

  22. kendallcp
    March 10, 2014

    >> How do you obtain an accurate measurement without using a precision voltaje or current source at a reasonable cost?

    Ratiometric measurement against a precision resistor.  A 0.1% accurate resistor is much cheaper than a 0.1% accurate voltage reference,

  23. kendallcp
    March 10, 2014

    >> In some extreme situations we could need to even take into accout the type of arithmetic libraries we're using.

    For sure.  I investigated using a 24 bit fixed point processor to speed up some thermocouple linearization calculations (some systems have a *lot* of thermocouples) and found something like a 0.04degC maximum error on the K-type 0~500degC linearization, using a reasonably optimized Horner's method for expanding the big, ugly, overdetermined polynomial that's commonly employed.  I never got around to writing a Filter Wizard piece about the work (using a filtering engine to do more non-filtery stuff).

  24. samicksha
    March 10, 2014

    Correcting self heating is complex and involves taking into account the sensor's thermal resistance (thermal time-constant), the reference current, and how the locally produced heat is conducted to the target object (solid, fluid, gas).


    I agree you Victor,  what is the best method or process you suggest for correction of self heating, and will it vary with target object.

  25. chirshadblog
    March 11, 2014

    @samicksha: Yes the proposed solution seems to be good but the cost will be a bit high than the current isn't it ?   

  26. etnapowers
    March 11, 2014

    @Victor, thank you for the blog you described. I think it should be very instructive for engineers a blog on this, i'm looking forward to read it, provided that there is no interference with Aubrey's article series, for sure.

  27. Victor Lorenzo
    March 11, 2014

    @Samicksha,

    I think we can get more reliable results taking mitigation approaches than trying to correct it. Reducing the sensor excitation current and its duty cycle to minimum possible values can provide good results, it reduces the total dissipated power in the sensor and thus reduces heating. The higher the sensor thermal time constant and/or mounting thermal resistance, the longer it will take the locally produced heat to dissipate in the surrounding environment.

    In some cases we can determine the dissipation constant and use it to correct the measured value. But best results are obtained when self heating is kept low.

  28. antedeluvian
    March 11, 2014

    Victor

    If it doesn't interfere with Aubrey's article series and I can find some time to complete and post it,

    Even if it does overlap or contradict my blogs, I have no objections at all. However, it is subject to our Editor-In-Chief's decision, so I would talk to him before expending any effort.

     

  29. samicksha
    March 12, 2014

    Thank You @ Victor, again power dissipated in a thermistor is typically maintained at a very low level to ensure insignificant temperature measurement error due to self heating.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.