Temperature Sensors
Thermocouples, RTDs, thermistors and infrared sensors: types, accuracy, wiring, cold junction and lead-resistance corrections.
Choosing a temperature sensor
Temperature is the most measured quantity in industry, and the sensors range from a few pence for a thermistor to expensive platinum standards. The three workhorses are the thermocouple (wide range, rugged, cheap, modest accuracy), the RTD (accurate and stable, up to about 600 °C) and the thermistor (very sensitive over a narrow range). Infrared sensors measure without contact. The electronics for reading them are covered in Sensors and Signal Conditioning; this page is about choosing, wiring and installing them in a plant.
| Sensor | Range (typical) | Accuracy (typical) | Strengths | Limits |
|---|---|---|---|---|
| Thermocouple | −200 to 1 800 °C | ±1 to 2.5 °C or 0.4 to 0.75 % | Wide range, rugged, fast, cheap, small | Low signal, needs cold-junction compensation, drift, needs special cable |
| RTD (Pt100, Pt1000) | −200 to 600 °C | ±0.15 to 0.5 °C | Accurate, stable, near-linear, interchangeable | More costly, slower, self-heating, lead resistance |
| Thermistor (NTC) | −50 to 150 °C | ±0.1 to 1 °C over a narrow range | Very sensitive, small, cheap | Non-linear, narrow range, self-heating |
| Semiconductor (IC) | −55 to 150 °C | ±0.5 to 2 °C | Linear voltage or digital output, cheap | Limited range, a chip that needs a supply |
| Infrared (non-contact) | −50 to 3 000 °C | ±1 % of reading | No contact, fast, moving or hot targets | Emissivity, optics, ambient reflections |
| Bimetal, filled system, thermostat switches | −40 to 500 °C | ±1 to 5 % | No power, a set point switch | Slow, no analog signal |
Thermocouples
When two different metals are joined at two points at different temperatures, a small voltage appears (the Seebeck effect). A thermocouple measures the difference between the hot junction (at the process) and the reference (cold) junction at the instrument. The output is only tens of microvolts per degree and depends on the metals used; each pair is a letter type with a standard table (IEC 60584).
| Type | Metals | Range (°C) | Notes |
|---|---|---|---|
| K | Chromel / alumel | −200 to 1 260 | The general-purpose standard; about 41 µV/°C; alumel is magnetic |
| J | Iron / constantan | −40 to 750 | Cheap, higher output; the iron rusts in damp or oxidizing places |
| T | Copper / constantan | −200 to 350 | Good for low temperatures and moisture; stable |
| E | Chromel / constantan | −200 to 900 | The highest output (about 68 µV/°C) |
| N | Nicrosil / nisil | −200 to 1 300 | More stable than K at high temperature |
| R, S | Platinum-rhodium / platinum | 0 to 1 600 | Very stable and accurate at high temperature, expensive, low output |
| B | Platinum-rhodium (30/6) | 600 to 1 700 | Furnaces; almost no output below 100 °C |
Tolerance classes (IEC 60584-1): for type K, class 1 is ±1.5 °C or ±0.4 percent of the temperature (whichever is greater); class 2 is ±2.5 °C or ±0.75 percent. The sensor’s accuracy at the tip depends on the whole circuit, including the compensation and the wire.
Cold-junction compensation
The voltage a thermocouple produces depends on the difference between the hot and the cold junction. The cold junction is where the thermocouple wire meets the copper terminals of the instrument, so the instrument must measure the temperature there and add the voltage that the cold junction would produce: cold-junction compensation (CJC). Nearly every thermocouple input does this with a built-in sensor at the terminals (keep them out of drafts and away from heat).
Worked example: a type K measurement
- A type K thermocouple gives 12.0 mV across its terminals while the terminal block is at 25 °C. At 25 °C the cold junction itself would give 1.000 mV (table value).
- Adding it, the equivalent voltage relative to 0 °C is 12.0 + 1.000 = 13.0 mV, which from the table (12.209 mV at 300 °C and 16.397 mV at 400 °C) is about 319 °C.
- Without the compensation the instrument would read 295 °C: an error of 24 °C, the whole cold-junction offset. A 1 °C error at the terminals gives a 1 °C error in the reading, whatever the process temperature.
- The tiny signal also means the input must resolve about 41 µV per degree: a 16-bit converter over ±78 mV (about 2.4 µV per count) resolves roughly 0.06 °C.
Thermocouple wiring and junctions
- Use the right wire all the way. The thermocouple wire (or a matching extension or compensating cable) must run to the instrument; ordinary copper wire adds an extra junction and a wrong reading.
- Polarity matters. Colors differ by standard: for type K, IEC 60584 uses a green sheath with a green positive and white negative, ANSI uses a yellow sheath with a yellow positive and red negative (red is always negative in ANSI). Check the code.
- Junction types: an exposed junction is fastest but unprotected; a grounded junction touches the sheath (fast, but can make a ground loop); an ungrounded junction is isolated from the sheath (slower, immune to ground loops and noise).
- Open thermocouple: most instruments detect a broken wire with a tiny bias current and drive the reading up-scale or down-scale, a useful fail-safe.
- Noise: millivolt signals are easily disturbed. Use shielded twisted pairs, keep them away from power and motor cables, ground the shield at one end only (see Installing and Troubleshooting Sensors), and use a transmitter close to the sensor for long runs.
- Do not splice carelessly: a mismatched connector or a screw terminal made of the wrong metal is an extra thermocouple. Use the proper connectors.
RTDs
A resistance temperature detector uses the fact that the resistance of a metal (platinum) rises almost linearly with temperature. The standard Pt100 has 100 Ω at 0 °C and a coefficient α = 0.00385 per °C (about 0.385 Ω per degree): the Callendar–Van Dusen equation for temperatures above 0 °C is R = R0 (1 + A T + B T²), with A = 3.9083 × 10-3 and B = −5.775 × 10-7. A Pt1000 has ten times the resistance and so ten times the signal and a smaller lead-resistance error.
Worked example: Pt100 readings
- At 200 °C: R = 100 × (1 + 3.9083×10-3 × 200 − 5.775×10-7 × 200²) = 175.86 Ω. At 300 °C it is 212.05 Ω.
- Non-linearity: a straight line with 0.385 Ω/°C would give 215.50 Ω at 300 °C: 3.45 Ω too high, equal to a 9.0 °C error. A transmitter or input card linearizes the curve for you.
- Accuracy classes (IEC 60751): class A = ±(0.15 + 0.002 |T|) °C, class B = ±(0.30 + 0.005 |T|) °C. At 200 °C that is ±0.55 °C and ±1.30 °C.
- Self-heating: the measuring current heats the element: 1 mA through 138.5 Ω (at 100 °C) dissipates 0.14 mW, which may cause a tiny error in still air, but 5 mA would be 25 times worse (the heat goes as the square of the current). Use 1 mA or less, or a pulsed current.
Two-, three- and four-wire RTDs
Worked example: lead resistance
- A Pt100 at 200 °C has 175.86 Ω. With two wires of 0.5 Ω each (a long cable), the instrument sees 175.86 + 2 × 0.5 = 176.86 Ω, which it reads as 202.7 °C: an error of 2.7 °C (1 Ω is 2.6 °C).
- With three wires the instrument measures the lead resistance in a second loop and subtracts it; this works if the leads are equal (same wire, same length, same temperature) and is the industrial standard.
- With four wires two wires carry the current and two sense the voltage directly across the sensor, so the lead resistance does not matter at all: used for the best accuracy and in laboratories.
- Use thicker wire or a Pt1000 to cut the effect of the leads in a two-wire installation, and always write down which type the input expects.
Thermistors
A negative-temperature-coefficient (NTC) thermistor is a semiconductor whose resistance falls steeply with temperature, by about 4 percent per °C near room temperature, so a small circuit gives a large signal. The β-parameter equation R = R0 × exp(β (1/T − 1/T0)), with T in kelvin, describes it well over a range of a few tens of degrees; more precise work uses the Steinhart–Hart equation.
Worked example: a 10 kΩ, β = 3950 thermistor
At 25 °C it has 10 kΩ. At 50 °C: 1/T − 1/T0 = 1/323.15 − 1/298.15 = −2.595 × 10-4 K-1, so R = 10 kΩ × exp(3950 × (−2.595 × 10-4)) = 3.59 kΩ. At 0 °C it is 33.6 kΩ. The change between 25 and 50 °C is 6.4 kΩ for 25 °C, a sensitivity of about 260 Ω/°C against 0.385 Ω/°C for a Pt100, which is why a thermistor needs so little circuitry.
The divider and the conversion are in Sensors and Signal Conditioning, and the standard-value resistors in the Resistor Network Calculator. Thermistors are also used as inrush limiters and over-temperature switches (PTC).
Infrared sensors
An infrared sensor (pyrometer) measures the radiated heat of a surface and so needs no contact: it suits moving, hot, hazardous or delicate targets. The radiation depends on the target’s emissivity ε (1 for a perfect black body; shiny metals can be 0.1 to 0.3 and plastics, paint and organic matter 0.9 to 0.97), so the emissivity must be set correctly. The sensor also needs a clear view of the target: the distance-to-spot ratio (for example 20:1) says the spot size, and the target must fill it.
Worked example: the effect of emissivity
A bare steel surface at 400 °C (673 K) has an emissivity of about 0.3, but the pyrometer is set to 0.95. The radiated power goes as T4, so the instrument computes a temperature lower by the fourth root of the ratio: 673 × (0.3 ÷ 0.95)1/4 = 505 K = 231 °C: an error of -169 °C. Paint a patch with matt black paint or use a two-color pyrometer for shiny metal.
Mounting, thermowells and response
- Immersion depth: a probe in a pipe must be immersed deep enough that the heat conducted along the stem does not cool the tip: a depth of at least 10 times the stem diameter (more for a large pipe, and at the middle of the flow).
- Thermowells protect the sensor and allow replacement under pressure, but add a thermal lag: a typical time constant moves from 1 to 3 s (bare probe in water) to 10 to 30 s with a thermowell. Use a fast sensor, a thermal paste or a spring-loaded tip for good contact.
- Surface sensors need good thermal contact and insulation on the outside so that the reading is of the surface, not of the air.
- Orient the probe against or across the flow, and keep it away from heaters, drafts and dead zones.
- Response time: for a first-order response the sensor reaches 63 percent in one time constant and 95 percent in three (see Sensor Basics and Selection); a control loop must account for it.
- Vibration and pressure: choose a sensor rated for them; thermowells in fast flows can fail from resonance (ASME PTC 19.3 TW).
Transmitters and outputs
A temperature transmitter sits near the sensor (head-mounted or on a DIN rail), conditions and linearizes it, compensates the cold junction, and sends a 4 to 20 mA signal (or a fieldbus value, HART, IO-Link) over any distance without the noise or lead-resistance problems of a raw sensor. It is also the best place to configure the range.
Worked example: scaling a 4-20 mA temperature signal
A transmitter is ranged 0 to 200 °C for 4 to 20 mA. A current of 12.0 mA is (12.0 − 4) ÷ 16 = 50 percent of the span, so the temperature is 100 °C. A reading below 3.6 mA or above 21 mA (NAMUR NE43) means a sensor or wiring failure: use it in the program to raise an alarm. The Analog Scaling Tool does the arithmetic.
Troubleshooting
| Symptom | Likely cause | Check |
|---|---|---|
| Reading is wrong by a constant offset | Cold-junction error, wrong thermocouple type, an extra junction | The terminal temperature, the sensor type setting, the cable and connectors |
| Reads very high or low, or jumps to full scale | An open wire or sensor (up-scale or down-scale burn-out) | Resistance of the sensor with the wires disconnected; wire continuity |
| Reading goes the wrong way | Reversed polarity (thermocouple) | Color code and terminal polarity |
| Noisy reading | Interference on the millivolt wires, a ground loop with a grounded junction | Shield, route, ungrounded junction, a transmitter at the head |
| RTD reads high | Lead resistance in a 2-wire connection, a wrong wire configuration | Wire count, cable length, the instrument’s setting |
| Slow response | Thermowell, poor contact, a thick sheath | A faster sensor, thermal paste, the tip touching the well bottom |
| Drift over months | Thermocouple aging, contamination, moisture | Replace and compare with a reference; use a sealed, protected sensor |
| Infrared reading depends on the surface | The emissivity setting, reflections, a dirty window | Set ε; shield from other heat sources; clean the lens |
Related
- Sensors and Field Devices: the overview of this guide.
- Sensors and Signal Conditioning: the circuits for thermistors, RTDs and thermocouples.
- Sensor Outputs and Wiring and Installing and Troubleshooting Sensors: 4-20 mA wiring, shields and grounding.
- 4-20 mA / 0-10 V Analog Scaling Tool.
- Material Properties: thermal conductivity and expansion.
FAQ
Thermocouple or RTD?
Use an RTD below about 500 °C when accuracy and stability matter, since it is 5 to 10 times better. Use a thermocouple for high temperatures, fast response, vibration, small spaces and low cost.
Why do thermocouples need special wire?
The signal comes from the junction between two particular alloys. Joining them to copper elsewhere creates new junctions that add their own voltages, so the alloys (or a matched extension cable) must run all the way to the instrument.
What is the difference between Pt100 and Pt1000?
The same platinum element, with 100 Ω or 1 000 Ω at 0 °C. A Pt1000 gives ten times the signal for the same current, so lead resistance matters ten times less, and suits two-wire and low-power use.
How do I check a temperature sensor?
Compare it with a reference in an ice bath (0 °C) or a dry-block calibrator, and read the resistance or voltage with the wires disconnected from the instrument to separate sensor faults from instrument faults.
References
- IEC 60584-1 and -3 (thermocouples: tables and tolerances, extension cables); IEC 60751 (industrial platinum resistance thermometers).
- NIST, ITS-90 Thermocouple Database and Temperature Measurement Handbook (Omega Engineering).
- Nicholas, J. V. and White, D. R., Traceable Temperatures, Wiley.
Last reviewed: October 2026. Tables and tolerances are typical and rounded; use the standards and the manufacturer’s data for real work.