Thermocouple measurement with a DAQ

A thermocouple is two wires of different alloys joined at the tip. It is cheap, tough, covers a wide temperature range and needs no excitation, and nearly every test bench has a few. A type K produces about 40 µV per °C near room temperature, so a DAQ input with a few millivolts of offset can be wrong by tens of degrees while the readings still look reasonable. The thermocouple calculator converts between millivolts and degrees with the NIST ITS-90 reference functions, including cold junction compensation.

How a thermocouple produces a voltage

The Seebeck effect produces a voltage in any conductor with a temperature gradient along it. Two different alloys produce different voltages across the same gradient, and the thermocouple’s output is the difference between them. The voltage is generated along the wires wherever the temperature changes.

• The output reflects the temperature difference between the tip and the point where the thermocouple alloys end and copper begins. That point is the cold junction, or reference junction, and on a DAQ it is usually the screw terminal.
• Every length of wire that crosses a temperature gradient contributes. Extension wire must be the matching thermocouple alloy, and a copper splice or a generic connector sitting in a warm zone adds its own error.

Thermocouple types, ranges and signal levels

NIST publishes ITS-90 reference functions for eight letter-designated types. The ranges below are the spans those functions cover. The usable range for a particular probe is usually narrower and depends on wire gauge, sheath and atmosphere. Sensitivities are calculated from the NIST functions at 25 °C.

A type K junction at 100 °C, referenced to 0 °C, produces 4.096 mV. Types R and S produce under 6 µV/°C near room temperature, so resolving 1 °C with them takes a measurement chain that resolves a few microvolts.

Wire tolerance

Under ASTM E230, as listed by wire makers such as Pelican Wire, standard limits of error for types K and N are ±2.2 °C or ±0.75% of the reading, whichever is greater, from 0 to 1260 °C. Special limits wire tightens that to ±1.1 °C or ±0.4%. Type J has the same figures from 0 to 760 °C. Type T is ±1.0 °C or ±0.75% (standard) and ±0.5 °C or ±0.4% (special), from 0 to 370 °C. A standard-grade type K probe at 400 °C can be 3 °C off before any electronics are involved.

Cold junction compensation

Cold junction compensation (CJC) measures the reference junction temperature and adds back the voltage a thermocouple would produce between 0 °C and that temperature. The hardware measures it with a temperature sensor (a thermistor, RTD or IC sensor) mounted near the input terminals.

Converting with CJC

Thermocouple output is nonlinear, so the compensation has to happen in voltage:

1. Measure the thermocouple voltage, V_meas.
2. Measure the terminal temperature T_cj and convert it to the voltage the same thermocouple type would produce at that temperature, referenced to 0 °C: V_cj = E(T_cj).
3. Add them and convert back: T = E_inv(V_meas + V_cj).

A type K probe reads 3.096 mV with the terminals at 25 °C. The NIST function gives E(25 °C) = 1.000 mV, so the compensated voltage is 4.096 mV, which converts to 100.0 °C. The thermocouple calculator runs these steps on your own readings.

Two shortcuts give the wrong answer. Converting 3.096 mV straight to temperature (75.9 °C) and adding 25 °C gives 100.9 °C. Dividing by a fixed 40.5 µV/°C and adding 25 °C gives 101.4 °C here, and the error grows with temperature. For a true 500 °C, the fixed-coefficient method reads 510 °C.

CJC error on general-purpose hardware

An error in the reference junction temperature passes almost one for one into the result. In the example above, a CJC sensor reading 26 °C instead of 25 °C moves the answer from 100.0 °C to 101.0 °C. On general-purpose hardware, the CJC sensor is often the largest term in the error budget:

These figures assume the sensor and the terminals are at the same temperature, as MCC states for the USB-TC. A DAQ next to a warm power supply, in sunlight or with a fan blowing across one end of the terminal strip has a gradient across its terminals that no datasheet covers. Keep the terminals out of drafts, give the device its warm-up time (MCC specifies 30 minutes for the USB-TC), and check a probe in an ice bath or against a reference thermometer before a long run.

LabJack’s thermocouple note points out that when several thermocouples share one terminal block, the CJC error is common to all of them, so the differences between channels are more accurate than the absolute values.

Input range and resolution on a general-purpose DAQ

Dedicated thermocouple inputs use small ranges, ±80 mV on both the NI-9211 and the MCC USB-TC, each with a 24-bit delta-sigma converter. A multifunction DAQ can read a thermocouple if it has a small enough range and low enough offset.

The USB-TC figure covers the whole system. For a multifunction device you build that budget yourself from input accuracy converted to degrees, CJC error and wire tolerance. ADC resolution vs accuracy shows how to read the accuracy tables.

Noise and 50/60 Hz rejection

At 40 µV/°C, 1 mV of mains pickup on the leads is a 25 °C swing. Long thermocouple runs next to motor wiring or heater cables pick up that much easily.

Integrating converters average the input over a window, and a window that spans a whole number of mains cycles cancels mains pickup. A 100 ms window covers 5 cycles at 50 Hz and 6 at 60 Hz, so it rejects both. Dedicated thermocouple hardware is slow for this reason. The USB-TC samples each channel at 2 S/s and specifies at least 90 dB of normal mode rejection at 60 Hz. The NI-9211 takes 70 ms per channel and specifies at least 85 dB of rejection at 50 and 60 Hz.

On a fast multifunction DAQ you can get similar rejection in software by sampling quickly and averaging over a whole number of mains cycles, for example 100 samples at 1 kS/s. Don’t log at a rate that divides evenly into the mains frequency without averaging. As shown in sample rate and aliasing, 60 Hz hum sampled at 10 S/s aliases to DC and shows up as a slow drift. Twisted thermocouple wire, a differential input and a shield grounded at one end handle most of the remaining pickup.

Grounded, ungrounded and exposed-junction probes

Sheathed probes come with the junction either welded to the sheath (grounded) or insulated from it (ungrounded). A grounded junction responds faster, but once the sheath touches grounded metal, such as a pipe, a heater or a chassis, the thermocouple is connected to a ground that can sit at a different potential from the DAQ’s ground. On a single-ended input that difference lands in the reading. On a differential input it becomes common-mode voltage, which is rejected only while it stays inside the input’s common-mode range. Ungrounded probes avoid the problem and respond more slowly. Exposed junctions respond fastest and have no mechanical protection.

An ungrounded thermocouple on a differential input is a floating source, so it needs a path to ground or its common-mode voltage wanders until the input saturates. LabJack’s example uses about 100 kΩ from the negative input to GND. Single-ended vs differential inputs covers bias resistors and ground loops.

Open-thermocouple detection

A broken thermocouple leaves a high-impedance input that can float to almost any value, and on a slow test that value can look like a believable temperature. Dedicated hardware checks for this, and on the USB-TC and USB-TEMP open-thermocouple detection is enabled automatically when a channel is configured for a thermocouple, with a maximum detection time of 3 seconds. On a general-purpose input you can add a high-value pull-up resistor on the positive lead, which drives an open channel to full scale where an alarm limit will catch it. Size it so that its current through the thermocouple loop resistance causes negligible error.

Choosing hardware

For temperature-only work a dedicated module is simpler. The MCC USB-TC has eight differential thermocouple inputs. The MCC USB-TEMP uses the same eight inputs for thermocouples, RTDs, thermistors and semiconductor sensors. The NI-9211 puts four thermocouple channels in a C Series module for a cDAQ chassis. All three handle CJC on board.

A multifunction DAQ with a ±0.1 V or ±0.2 V range, such as the LabJack T7 or the NI USB-6211, makes sense when thermocouples share a test with faster signals. Plan for an external CJC sensor at the terminals, do the conversion in software and average for mains rejection. DAQAtlas, which is in development, includes a simulated USB-TC, so a temperature dashboard can be built before the hardware arrives.

If you need better than about half a degree near room temperature, consider an RTD instead. A Pt100 changes by about 0.385 Ω per °C, needs an excitation current and a 3- or 4-wire connection, and has no cold junction to compensate. The RTD calculator converts resistance to temperature.