Accuracy of a Thermocouple measure system

[Cyclone1]

I am not very knowledgeable in this field. And This is my first time trying thermocouple measurement with a DAQ, I have some question about the main source of error in the measurement system, and technique that can be use to improve the DAQ accuracies. (Let’s assume the voltage measurement isn’t a issue)

Let’s say my DAQ specify thermocouple accuracy of 1C, not including probe error. Is it possible to improve the accuracy of the DAQ by self calibrating the CJC with a good thermistor? Let’s say with a 0.1C thermistor, does it make my DAQ thermocouple accuracy to be 0.1C across all range and different type of TC? (Not including the TC error of course)

Let’s also say that the DAQ internal reference is 0.1C off the thermistor, can I say my DAQ is have atleast ~0.2C accuracies?

I also learned there is a thermocouple simulator equipment like one on the 5520A. The user manual specifically required the use of correct thermocouple in order to properly calibrate the DUT. Since it basically output TC voltage based  on its-90 and there is no thermal gradient between the TC, Why does it still need correct TC wires? In this kind of setup, what would be my test uncertainties from the perspective of my DAQ? Do I still need to include the TC error? Or only the calibrator uncertainties?

[Conrad Hoffman]

I don't know the specific answer to your questions but it comes down to accuracy, sensitivity and trust.

A thermistor has a large resistance change per degree so getting high resolution measurements is pretty easy. 0.001 degree C is easily possible. As for accuracy, you can use circuits that improve linearity but you still need to match the curve and verify it, something not as easy. How many points can you verify? They're usually stable but drift over time is possible. IMO, "trust" isn't as good as a thermocouple.

A thermocouple is based on fundamental material properties and is highly predictable and each type is extensively researched with good data available. Sensitivity isn't as good so it wouldn't be my first choice for ultra-high resolution measurements, but it's highly trustworthy if you eliminate error sources. Every junction is a potential (no pun intended) source of error, thus special thermocouple wires and connectors.

Don't forget the nickel or platinum RTD!

[Cyclone1]

Thank you very much of your reply!

My application does not need to be sensitive, but it does need to meet certain tolerance which it currently doesn't. I was verifying with a Fluke 5520A thermocouple simulator and most of my tests are failing
Then I realized the footnote does mentioned "Does not include thermocouple error" at the Source/Measure section. But I do not understand why there is a need to include thermocouple error if all it does is simulating thermocouple voltage.

[David Hess]

Every junction is a potential (no pun intended) source of error, thus special thermocouple wires and connectors.

Thermocouple voltages are not created at the junctions; the thermocouple voltages are generated along the length of the wire depending on both temperature and composition.

There are several sources of error:

1. The composition and purity of the thermocouple wire.
2. The accuracy of the cold junction measurement.
3. Because the signal levels are so low, the thermocouple potentials between the cold junction and amplifier.  For instance copper-Kovar is 40uV/C so careful thermal layout and balancing is required.
4. Accuracy of the nonlinear correction.
5. And of course sensitivity is low but this is a solved problem even at thermocouple levels of sensitivity.

Resistive sensors like RTDs and thermisters only have the accuracy of the sensor itself and nonlinear correction to worry about.  Neither Kelvin sensing which removes series resistance errors or excitation chopping which removes thermocouple errors and errors from noise are available for thermocouples.

[bgugi]

Fluke has an excellent webinar on thermocouple calibration and sources of error. I don't recall which of these two I watched: https://us.flukecal.com/literature/on-demand-webinars/how-calculate-uncertainty-thermocouple-calibration-system https://us.flukecal.com/literature/on-demand-webinars/how-set-thermocouple-calibration-system, but it covered a lot of the technical detail on how thermocouples work, as well as sources of error in calibration.

The most important thing to remember about thermocouples: they generate voltages along thermal gradients, not at the junctions. When you look at a thermocouple lookup table, you'll notice that, regardless of type, at 0*C the voltage is 0.000... this is because the tables are referenced to 0*C. A typical DAQ will take the mV reading from the thermocouple, then digitally ADD the TC voltage at the CJC temperature, then lookup the sum on the TC table. a thermocouple calibrator does the same process in reverse, only outputting the net voltage to simulate a thermocouple at that temperature.

If your thermocouple calibrator and DAQ were at the exact same temperature throughout, you could use wire of any type to connect them (including cu-cu). It's very difficult to actually achieve this, however, as the thermocouple sockets on either unit is inside or in thermal contact with the each unit, and these microenvironments will typically be somewhat different. Using the proper thermocouple wire will generate the net voltage between the two connectors to make a correct reading. Using cu-cu, you have introduced an error equal to this difference to any readings.

If your DAQ only specifies a single error term for thermocouple measurements, it likely includes the CJC error, the Voltage Sensitivity error, and Voltage offset error. If you don't also see the voltage errors, it may be difficult to characterize which of these is the largest term.

When making measurements near ambient, your largest source of error is typically the CJC. This is partially due to the error in calibrating the reference transducer, but can also come in the form of temperature gradients between the two leads and the built-in temperature transducer. I'd say that most systems utilize thermistors near but not perfectly embedded in the connectors, so it may be very difficult to characterize errors that might be introduced when the daq is in different environments - eg, a daq calibrated on a bench in a calibration laboratory might produce very predictable and accurate results, but have wild errors when stuffed into a warm equipment cabinet, fridge, or environment that experiences temperature swings (like outdoors). I would be very hesitant to expect better uncertainty from the CJC of a daq than would be expected from the manufacturers specifications (look out: "typical" ratings are very common in midrange DAQs)

Best practice for high-accuracy thermocouple calibration often includes the following (in addition to many other considerations, EURAMET provides a good guide https://www.euramet.org/Media/docs/Publications/calguides/EURAMET_cg-8__v_2.1_Calibration_of_Thermocouples.pdf):
1. Use an external cold junction. By connecting the leads of the thermocouple to well-matched, high-quality wire, you create a cold junction that can be placed in a temperature-controlled environment, such as an ice point, or a lag bath monitored by reference probe. (for single-type meters, make sure the unit doesn't use internal TC wire to make the cold junction measurement further inside the unit)
2. manually "chop" the readings - by manually reversing the leads where they connect to the meter and letting them return to thermal equilibrium, you can observe any voltage offsets inside of the meter. There isn't a good way to automate this, as switching will introduce new errors or fail to observe circuit paths where errors are introduced.

All of this ignores errors that are innate to the probe itself. Pure metals are uncommon in thermocouples, and i'm not aware of any types that use two pure metals - alloying precision will affect the accuracy of a thermocouple. In addition, there are other effects that can affect the accuracy of a thermocouple - one of the worst being inhomogeneity of the metals - errors that may be very difficult to quantify, and can be completely missed during calibration - the error may be introduced along different areas along the wire, this can be in the form of contamination (oxidation), alloy depletion, or annealing of the wire. One of the more well-known offenders is type K thermocouples: https://www.osti.gov/servlets/purl/4213761. contamination at the connectors (like oxidation on the copper leg of the type T) can create additional voltage offsets that can be hard to chase down or eliminate if thermal gradients exist in this connection.

Thermocouples excel in certain applications, but are essentially never the most accurate option available. They can be used to make a lot of specialty measurements that would not be possible or practical with other probes, such as surface temperature measurements, incredibly small probes (Physitemp makes probes that can be inserted with a hypodermic needle or used to take rectal temperatures on neonatal mice), or incredibly precise and accurate differential temperatures - like in thermal flux meters. They also excel in harsh environments - sometimes because they are more robust, sometimes because they are more expendable (thermocouple wire can be purchased by the spool for less than a dollar per foot.) In addition, they can be calibrated as a lot, a spool can have representative samples pulled and calibrated, and the whole spool can typically be considered "calibrated," as there will be very little variance within a lot of wire, unlike RTDs or thermistors which will often need individual characterization.