Actions
Forum Thread Details
  • Replies 25 replies
  • Subscribers 340 subscribers
  • Views 6465 views
  • Users 0 members are here
Related

Any ideas for low-cost Thermocouple Interfacing methods?

shabaz

Note: See here for a project that resulted from this discussion: BLE EasyTempProbe 

Hi,

Since thermocouple measurement ICs are getting expensive/hard to find, I wished to use a single channel ADC, for thermocouple measurements (actually, I want to use a dual channel ADC for two thermocouples, but it's likely the same problem just doubled!).

The trouble is, the cold end of the thermocouple needs measuring too, and I was thinking of using a thermistor for that because that's easier to obtain (and cheaper) than an IC sensor. In summary, I wished to multiplex a thermistor and a thermocouple.

I've come up with the diagram below so far and wish to use it in an environment where the cold junction might be in the range of -40 to +50 deg C, and the thermocouple might be in the range of -40 to +400 deg C (maybe a Type J thermocouple). I think it will have an error of a few deg C. The ADC is 16-bit, and I likely won't be using the whole range of it. The ADC has a PGA of up to 8X, so it will be set to 8X when measuring the thermocouple.

The main benefit of the proposal below is that it is cheap since it just needs a couple of transistors and a few resistors. I could think of more complex circuits for a more accurate measurement, but it would be nice to see if this is good enough, or if it could be tweaked to be good enough unless anyone has other suggestions.

If it works, this would be a cheap way (under $5, ADC included) to have two thermocouples each with their own compensation, with the drawbacks that accuracy might be a few degrees at best, and no isolation either, unfortunately. 

Any ideas would be gratefully appreciated, since I'm sure I may be missing some great techniques, missing the wood for the trees, etc! Has anyone come across any low-cost methods to do such a thing? Any mistakes I'm making?

image

  • Could you use a microcontroller with built-in temperature sensor to avoid the thermistor for the cold junction? TI might even have an MCU with 16 bit A/D and internal temp sensor.

  • Thinking out loud:

    Both transistors off, no path on input to 0V so ADC pins 2 and 3 pulled to 3.3V, probably outsiede ADC common mode range.

    Both transistors on, both sides of thermocouple pulled to 0V, but you can measure thermistor.

     How about this:

    image

    (Do you like the new Kicad version 7 ?)

    R2 = R3

    R1 = TH1 @ 25C (or what suits)

    If your ADC is true differential and can' measure the two inputs independently then you need R4 and the sums are harder

    If you can measure each input then you don't need R4 (reduce to zero)

    The switch can be a P channel MOSFET or a PNP bipolar.

    The ADC inputs are never at the rails.

    Watch out for the common mode range of the built in amplifier (and its noise).

    It won't be that good, the big series resistance when measuring the thermocouple will make things noisy but the capacitors will help.

    The measurement of the voltage drop across R4 involves several resistors.

    @Doug, the cold junction sensor has to be thermally coupled to the junction so one in a processor is never that good - although it does get rid of a lot of other problems. In the end it might be  a close call for which works best, although micro ADCs tend not have PGAs.

    MK

  • in reply to dougw

    Hi Doug,

    As a slight variation, instead of an internal temp sensor, I was thinking of using (say) an internal 8-bit ADC in the microcontroller for measuring just the thermistor, which could be fine over a limited temperature range, but over -40 to +50 degC it's not going to be granular enough, no matter if I try slightly linearizing the thermistor with resistors. A higher-res ADC in the microcontroller could be the answer to that, but I'm hoping that using the external ADC could be as good or better, and eliminates needing particular microcontrollers.

  • in reply to michaelkellett

    Hi Michael,

    That's a neat trick to use the thermistor circuit to also set the common mode voltage.

    The particular ADC I'm going to use is MCP3426 (16-bit, low-cost and 5k in stock at Farnell, so maybe it's not that popular!). It's a dual-channel ADC so I wanted to use both channels for two thermocouples.

    It is only fully differential. It's common-mode range extends to the supply rail, I thought that was unusual, but looking at another PGA-input ADC, it too has common-mode extending to the supply rail. The 3.3V connection could be set to something else, but that involves more circuit, so if 3.3V works it would be nice to keep it, but I really have no idea (I have not tried it). It is really neat to use the thermistor circuit to set the common-mode voltage, but it does then require the additional resistance as you say, and also then means the thermistor is always powered (although the 3.3V could easily be switched to eliminate that). 

    I've created a spreadsheet to see expected values for a generic thermistor when in the circuit and expected thermocouple values, but I have not tried to calculate error yet. The simulation looks like this (the 250k resistance will increase when measuring the thermistor, because I can reduce the gain, but when measuring the thermocouple, it will be that ballpark). I wish the circuit was more generic, but unfortunately probably the performance will hinge on (say) using 1k thermistor and so on, rather than having more freedom to alter values.

    If it's not very accurate (I don't have a spec, but was hoping for a few degrees) I may just give up, and use two of the ADC chips : ( They are not expensive, it would be nice though to reduce to the single part. 

    image

  • in reply to shabaz

    Thermistor linearization is a bit tricky, but well documented and can get within a few degrees without too much problem, especially if you have a high resolution A/D. Your solution should be a good low-cost method.

    The TI MSP430F20x3 family (with 16 bit A/D and internal temp sensor) is listed by TI as under $1 in 1K qty. So it may be competitive to the cost of an A/D chip plus MCU plus thermistor. Although it does not have an amp.

  • The TC resistance is just 1 or 2 ohms per foot so do you really need Q2? The TC voltage can be subtracted from the thermistor reading.

  • in reply to michaelkellett

    Yes - the MCU and TC would need to be thermally connected with an isothermal block.

  • in reply to shabaz

    There will be  a problem with the thermistor being in series with the input, introducing noise and attenuation caused by the ADC input differential resistance. As you've said keeping the thermsistor a low value will help. A big capacitor across the ADC inputs will reduce the noise a bit. You kind of know the thermistor resistance so could compensate for its effect but you would need to know ADC input resistance was constant.

    MK

  • in reply to dougw

    Hi Doug,

    Everything interacts a bit, at a first approximation it's just the thermocouple voltage, but I'm down in the single-digit millivolts and it's harder to model with that precision without Spice, which the microcontroller won't have. I need to spend half a day or so properly modeling this, and start getting a feel for how much overall error there could be.It might be that the Q2 is not needed, if there isn't much difference in error with any model I can achieve in Excel.

  • in reply to shabaz

    Hi Shabaz,

    I am curious about “It might be that the Q2 is not needed, if there isn't much difference in error with any model I can achieve in Excel.” I am interested in hearing more detail and how how that exercise goes when you are finished.