Author Topic: Linearization thermocouple necessary or not?  (Read 4267 times)

0 Members and 1 Guest are viewing this topic.

Offline embedded.world@aol.comTopic starter

  • Contributor
  • Posts: 19
Linearization thermocouple necessary or not?
« on: June 13, 2014, 12:23:42 pm »
One way to measure thermocouple temperature is to:

This short & less processor intensive


1. Measure hot junction voltage
2. Convert it into equivalent temp from lookup table
3. Measure cold junction temp
4. Add the two

Other is hell of processor intensive include double math from NIST:


1. Measure hot junction voltage
2. Measure cold junction temp
3. Calculate the cold junction equivalent thermocouple voltage using the NIST temperature-to-voltage coefficients
4. Add the cold junction equivalent thermocouple voltage calculated in step 3 to the thermocouple voltage calculated in step 1.
5. Use the result of step 4 and the NIST voltage-to-temperature coefficients (the “inverse” coefficients) to calculate the cold-junction-compensated, linearized temperature value.


Why people prefer method 2. Is there better accuracy?
Or method 1 is wrong correc method is second?
I have on many small processor people usually apply method 1?
 

Offline retrolefty

  • Super Contributor
  • ***
  • Posts: 1648
  • Country: us
  • measurement changes behavior
Re: Linearization thermocouple necessary or not?
« Reply #1 on: June 13, 2014, 03:00:42 pm »
I think the answer is application specific. What total range of measurement will the sensor be measuring? A TC can cover a large range and if accuracy is important over such a large range, then the more complex processing may be necessary. Another issue is the accuracy of the specific TC sensor itself. At our refinery we only expected +/- 5F accuracy over larger ranges. We gradually changed to using PT100 RTD sensor for application up to maybe 750F as they were inherently more accurate and their installed costs for new installation Vs TC was not a factor (conduit/labor costs prevailed).

 

Offline IanB

  • Super Contributor
  • ***
  • Posts: 11790
  • Country: us
Re: Linearization thermocouple necessary or not?
« Reply #2 on: June 13, 2014, 03:24:34 pm »
Why people prefer method 2. Is there better accuracy?
Or method 1 is wrong correc method is second?
I have on many small processor people usually apply method 1?

Method 2 is physically correct and therefore is in principle more accurate. Method 1 is not strictly correct and is therefore less accurate.

However, the achievable accuracy in practice depends on many variables, and one always looks to the largest source of error in the system. If you have a 2 degree source of error in one place, there is no point fussing over a 0.05 degree error somewhere else.

- How accurately can you measure the reference junction temperature?
- How accurately can you measure the small thermocouple voltages?
- What is the tolerance and accuracy of the thermocouple itself?
- How repeatable are your measurements?
- How much noise is present in the measurement system?
- How much will the properties of the measurement system drift over time?
- How much accuracy do you actually need for your application?

In short, method 1 may be perfectly fine for your needs.

Note also that over small enough temperature ranges (e.g. 100 degrees) you may not even need  the lookup table. Over small ranges the relationship between thermocouple voltage and temperature is linear to well within other sources of measurement error.

« Last Edit: June 13, 2014, 03:27:40 pm by IanB »
 

Offline Tabs

  • Regular Contributor
  • *
  • Posts: 106
  • Country: gb
Re: Linearization thermocouple necessary or not?
« Reply #3 on: June 14, 2014, 11:37:07 pm »
Ok, you would think that since I work in this field (temperature controllers) I would know the answer to this.
In truth, I only design the hardware and I'm not sure on how the softies do the math.

That being said, I believe there are a couple of problems with method 1.

Thermocouples produce a voltage difference which is a non-linear function of temperature (hence the need to linearise)
The lookup tables produced by NIST use a reference junction at 0 deg C. In real world applications, the 'cold junction' tries to measure the temperature at two junctions (T/C+ line and copper PCB trace & T/C- line and copper PCB trace) and this temperature is usually the ambient air temperature in which the cold junction sensor is placed.
Taking the difference between the voltages at the hot and cold junction results in a voltage that is lower than the just the hot junction voltage alone.
Using the difference voltage puts you at a lower point in the linearisation curve than where you would be if you took the hot junction voltage by itself.

If you approximate the local region around each of these points with a straight line you may find
1) that the two points are on two different lines, each with their own gradient m (as in y=mx+c).
2) You could get lucky and the two points are on the same line.

In case 1, you will get a too high temperature for the hot junction (since its on different linearised curve) with an error that will still exist when you subtract the cold junction temperature (which will probably be on a linearised line near ambient (dont know your application so can say for certain).

In case 2, the principle of superposition applies (forgetting about 'C') and method 1 will work. If you know your going to stay close to ambient (close to the cold junction temp) then your method may work if you can accept the accuracy in your application (as others have said, the answer is specific to your application)

Also,

in your description of method 1,
I can see how you would convert the cold junction sensor (usually a diode connected transistor) into a voltage and then to temperature. This would require your to characterise the transistor Vbe over temperature.
I dont see how you could do this for a thermocouple since there are no conversion tables for an absolute voltage measurement from the thermocouple+ line directly to temperature. An RTD allows such a conversion but thats a different device.
 

Offline IanB

  • Super Contributor
  • ***
  • Posts: 11790
  • Country: us
Re: Linearization thermocouple necessary or not?
« Reply #4 on: June 15, 2014, 12:13:41 am »
To give an idea of the need for linearization of thermocouple voltages, here is a chart of the common type K thermocouple between 0 and 100 degrees Celsius. As you can see, it is rather close to linear. So if you were measuring in this range (for example in a HVAC application), you would not need to be concerned with the rigorous voltage to temperature calculations. The maximum error due to non-linearity is about half a degree, and this is most likely less than the other sources of error in your system.

« Last Edit: June 15, 2014, 12:16:09 am by IanB »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf