Thermocouple accuracies

[Bomar2]

Can anyone point me to information about thermocouple system accuracy calculations? In particular, there are many sources of error such as cold junction compensation, TC errors, data acquisition errors, (are there errors associated with the TC wire itself?).

I'd like to also find information about how to compensate for these errors. For example, we are calibrating thermocouples and generating a curve to "curve fit" our actual reading to. Should I be calibrating the TC with the final wire attached?

I'm trying to get as accurate as possible.

 

[itsmoked]

Bomar2;

Yes all those errors are certainly present. It is the name of the T/C game.

T/Cs vary all over out of the box. They are probably off by at least a degree maybe 3 due to the fact that they are made from a melt of elements that can only be repeated up to a certain extent.

With T/Cs you want a system that is easy to calibrate because you will need it.

A standard method is to use a T/C calibration source. This is a voltage generator that puts out the expected voltage for your T/C at a certain temperature. The factory of product X can run along and calibrate all 20 channels to their T/C calibration source. Once the device is installed in the field with the actual T/Cs they will be off but not by more than a few degrees. This is usually fine for most applications.

If you want right-on calibration you generally need to place the 'actual' installed T/C into a controlled thermal bath and then ultimately use the systems calibration features to apply the required corrections or offset. This is the only way to get absolutely the best accuracy. This method can be problematic if you want to calibrate at some elevated temp over about 300F.

You have not stated what you are trying to control the temperature of. In reality most processes don't care about actual temperature that much. What they care about is repeatability and consistency for their process. Be careful you don't make an expensive decision if it actually isn't needed.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[kontiki99]

From memory, TCs are only considered to be accurate to a couple of degrees of error max. I believe that’s a repeatability issue over time.

Their response is not linear either. The thermoelectric response of each metal is linear, but in a junction I guess one metal dominates the effects of the other at different temp ranges.

Of course as with any electric sensor circuit there is an error budget or an uncertainty analysis to be made.

Very often the calibration is performed so the system is most accurate in the critical temperature.
You might check books on metrology. I know there are some books on measurement uncertainty analysis out there (maybe ISO).

 

[Bomar2]

Thanks for the inputs.

You have backed up what I have been reading and researching about T/Cs.

I am not trying to control the temperature of a process. Instead, I'm trying to measure the temperature of a dewpoint. (Dewpoint Thermocouple) We are trying to develop a high temperature 200 - 400 degF dewpoint thermocouple. The application needs to have an accurate temperature (what is really the temperature) and of course repeatability.

Here's what I'm thinking I need: (I'd like to know if I need anything else or if I'm not doing something correctly)

Thermocouple & Wire (Special Limits): We currently took a set of T/Cs and performed a calibration sweep on them and took data points. We did this for 8 runs and generated data curves. From day to day there were diferences at the different temperature settings but we averaged the 8 data points at each temperature and developed a curve. I was going to use this curve in a DAS as a correction factor. The more I've learned, led me to believe I should also be using the actual T/C wire that I will be using for the test in this calibration process to also get the curve. Does this sound correct?

Cold Junction Compensation: We usually just rely on National Instruments SCXI boards to do the CJC with its built in thermistor doing the compensation. However, the more I learn it seems that to be more accurate, I need to use an "Ice bath" at the junction. So I was considering the use of Isotech's (

http://www.isotechna.com)ICE

Point Reference Model 938 to eliminate (as much as possible) the error associated with the CJC.

Data Acquistion System (DAS): I was going to read the mV output of the Model 938 with a highly accurate HP Voltmeter.

Does any of this sound right?

 

[unclesyd]

I don't know the specifics but we use hundreds of thermocouples or RTD's for temperature measurement and control in the 150°C-300°C range. All probes have two couples or RTD's. We have an instrument group that essentially does nothing but calibrate and verify every probe.
Omega did have a very good manual on thermocouples and temperature measurements.

http://www.omega.com/techref/

 

[itsmoked]

hmmmm. I just have to ask. Why in the world are you using T/Cs for this? They will probably give you an ulcer.

RTD's are vastly more accurate, more repeatable, more linear, and carry no such intermediate junction headaches. Why are you not using RTDs?

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[IRstuff]

Platinum resistance thermometers (PRTs) can be used for calibration transfer and measurement:

http://ts.nist.gov/ts/htdocs/230/233/calibrations/thermodynamic/pubs/106-1-01.pdf

TTFN

 

[danw2]

I agree with ItSmoked, why are you using thermocouples when the need for accuracy is primary? Your temperature range fits RTDs very nicely. The inherent RTD accuracy is so much greater than even limit of error TCs.

RTDs don't require ice point compensation, which can be a source of drift. The cal shop I deal with does an ice bath in a thermos jug for their critical apps - just to eliminate the cold junction source of error.

RTDs do have a self heating characteristic, that might me more easily characterized for a commercial application than the hobgoblin mess one gets with TCs.

FYI, this month's ISA InTech magazine June 2006) has an article on troubleshooting thermcoouples.

Dan

 

[JLSeagull]

The need for accuracy generally drives one toware the 100 Ohm Platinum Resistance Temperature Detectors (RTD) instead of thermocouples. Both work. Most temperature transmitter transmitters these days accept the common RTD and T/C element types.

 

[Bomar2]

I've gotten a lot of feedback to use RTDs. However, with my application I can not use them. The instrument needs to be at the tip of the probe.

 

[IRstuff]

What dimensions?

TTFN

 

[itsmoked]

[red]"The instrument needs to be at the tip of the probe."[/red]

May I ask why this would preclude an RTD?

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[IRstuff]

http://www.sisweb.com/ms/sis/prt.htm

shows a ceramic coated PRT that's 0.04" OD and 0.045" length which is quite small, even when compared to a thermocouple.

TTFN

 

[jgv]

Here you can find what you want:

http://www.omega.com/