The PT100 is the temperature sensor of the process industries below about 500 °C: a coil or film of platinum whose resistance is 100 Ω at 0 °C and rises by a well-known, slightly curved law. It is accurate, stable and interchangeable, and most of the wrong readings that get blamed on it are the wiring, the installation or the input configuration. This guide is the sensor from the metal up: the curve, the three ways of connecting it and what each costs in degrees, where to put it, how hot the measuring current makes it, what a tolerance class does and does not promise, and how to find the fault when the reading is wrong.
Platinum and why 100 Ω
Every metal's resistance rises with temperature, but platinum does it reproducibly: it does not oxidise, it can be drawn pure, and its resistance–temperature law is the same from one batch to the next closely enough that a sensor from one maker can replace one from another without recalibration. IEC 60751 fixes the law by one number, the mean temperature coefficient between 0 and 100 °C, α = 0.00385 Ω/Ω/°C: a sensor of 100 Ω at 0 °C is 138.51 Ω at 100 °C. The 100 Ω is a convention that happened to suit the bridge circuits of the 1950s; a PT1000, with 1 000 Ω at 0 °C and the same curve, is the same sensor with ten times the resistance and ten times the signal.
Two constructions exist. A wire-wound element is a fine platinum coil in a ceramic or glass body; it covers the full −200 to 850 °C range, it is the more stable at high temperature, and the top tolerance classes are easier to reach with it. A thin-film element is a platinum layer on a ceramic chip, laser-trimmed to 100 Ω; it is smaller, faster, cheaper and more resistant to vibration, and it is what most industrial probes now contain, usually limited to about 500 °C by its construction. Either sits inside a stainless steel sheath of 3–8 mm, with the wires coming out into a connection head or a moulded cable.
The curve: Callendar–Van Dusen, not a straight line
The resistance of a PT100 against temperature is a gently bending curve, described above 0 °C by a quadratic and below it by a quartic that Hugh Callendar (1887) and Melville Van Dusen (1925) fitted to platinum:
R(T) = R0 (1 + A T + B T² + C (T − 100) T³) for T < 0 °C
A = 3.9083 × 10⁻³ B = −5.775 × 10⁻⁷ C = −4.183 × 10⁻¹²
The B term is small but it matters. The slope of the curve, which is the sensitivity in ohms per degree, is 0.391 Ω/°C at 0 °C, 0.379 at 100 °C, 0.368 at 200 °C and 0.298 at 800 °C: it falls steadily, so a fixed resistance error is a growing temperature error as the sensor gets hotter. Treat the curve as a straight line of 0.385 Ω/°C from 100 Ω and the 175.86 Ω a sensor shows at 200 °C reads as 197.0 °C, 3 °C low; at 400 °C the linear reading is 18 °C low. Every RTD input and transmitter linearises the curve properly, and so does the PT100 calculator, but a PLC program that scales resistance to temperature with two points does not. The C term below zero contributes −0.0037 Ω at −40 °C, nothing to worry about in a cold store, but −1.0 Ω at −200 °C, which is 2.3 °C in a liquid-nitrogen line.
Two, three and four wires
The input measures resistance, and it cannot tell the sensor's resistance from the cable's. Fifteen metres of 0.5 mm² copper is about 0.54 Ω per wire; a PT100 changes by 0.38 Ω per degree; so a plain two-wire connection over that cable reads 1.08 Ω, or 2.8 °C, high, and the error changes with the weather because copper's resistance rises 0.39 % per degree of cable temperature. Two-wire is acceptable only for very short runs, for a PT1000 (the same cable is a tenth of the error), or when a transmitter sits in the sensor head and the two wires carry current instead of resistance.
The three-wire connection is the industrial standard and it solves most of the problem. Two wires go to one end of the element and one to the other. The input measures the loop through the sensor (one lead, the element, the other lead) and also the resistance of the two leads that share an end, which are joined at the sensor and so contain no element resistance. Subtracting one from the other removes one lead's worth from the loop and, if the leads are equal, leaves the element alone. What remains is the mismatch between the leads: a corroded terminal on one wire, or two wires of different length, and with a 5 % mismatch on 0.5 Ω leads the residual is 0.025 Ω, which is 0.07 °C. That is why the third wire matters and why a 3-wire sensor must be wired to a 3-wire input with all three wires, not two.
The four-wire connection is the laboratory method, a Kelvin measurement. One pair of wires carries the measuring current through the element; the other pair senses the voltage across it. The sense pair carries almost no current, so its resistance produces no voltage drop, and the input reads the element alone regardless of the cable, the terminals or the temperature. It needs a 4-wire input, which most PLC RTD cards and good transmitters have, and one more core in the cable. The choice is a trade between copper and electronics: a 4-core cable costs a little more than a 3-core, a head-mounted transmitter costs more than either and removes the cable from the measurement entirely.
Where the sensor goes: immersion, thermowells and response time
The sensor reports the temperature of its own element, and the installation decides how close that is to the process. Heat flows along the sheath from the hot tip to the cold connection head, so a probe that is not immersed deeply enough reads low: the rule of thumb is an immersion of at least ten sheath diameters plus the length of the sensing element, which for a 6 mm probe with a 30 mm element is about 90 mm in liquid, and more in gas, where the heat transfer to the sheath is poorer. In a small pipe that means mounting the probe on an elbow, pointing into the flow, or in a tee.
In a pressurised or corrosive line the probe sits in a thermowell: a closed tube welded or screwed into the pipe, so the sensor can be replaced without breaking containment. A well adds mass and a small air gap, both of which slow the response. A bare 6 mm probe in flowing water responds in a few seconds; the same probe in a thermowell typically takes 30–90 s to reach 63 % of a change, and in still air several minutes. A spring-loaded probe that presses its tip against the bottom of the well, a well with a tapered stem, and a little heat-transfer paste all help. For a control loop the response time adds to the dead time and the time constant the controller sees, which is why a temperature loop on a heavy well is tuned slower than the process itself would justify; the PID tuning guide covers what that does to the settings.
Self-heating and excitation current
To measure a resistance the input must pass a current through it, and the current heats the element. At 1 mA through a PT100 at 100 °C the power is I²R = 0.001² × 138.5 = 0.14 mW. How much that warms the sensor depends on how well it is cooled: a thin-film element in flowing water sheds the heat and rises by a few thousandths of a degree; the same element in still air can rise by several tenths of a degree, and a small unsheathed element by more. Industrial inputs and transmitters use 0.1–1 mA for this reason, and some pulse the current to reduce it further. The practical rules are to keep the excitation at or below 1 mA, never to trust a calibration done in still air, and to remember that a PT1000 dissipates ten times the power at the same current, which is one reason PT1000 inputs use smaller currents.
Tolerance classes and what accuracy you really get
IEC 60751:2008 defines four tolerance classes for the sensor, each a fixed part plus a part that grows with temperature: class AA ±(0.10 + 0.0017 |T|) °C, class A ±(0.15 + 0.002 |T|), class B ±(0.30 + 0.005 |T|) and class C ±(0.60 + 0.01 |T|). At 100 °C that is ±0.27, ±0.35, ±0.80 and ±1.60 °C. Class B is the industrial default; class A is worth its small premium wherever the number goes into a calculation or a record; class AA is for reference and calibration work. Classes AA and A are only specified over narrower temperature ranges than B and C, narrower still for thin-film elements, so a "class A" sensor at 500 °C may be class B by the standard's own definition. The tolerance is also a promise of interchangeability: any class A sensor can replace any other without touching the transmitter, which is what makes the PT100 practical to maintain.
The class is the sensor alone, new, in a calibration bath. The installed accuracy is the class plus the wiring error, plus the transmitter or input card (about ±0.1–0.2 °C for a good head-mounted transmitter, often ±0.5 °C or more of span for a PLC RTD channel), plus the installation errors of immersion, well and self-heating. A class A sensor on a 2-wire connection through 15 m of cable into a PLC card can easily be 3 °C wrong, none of it the sensor's fault. When the number matters, check the whole loop, not the sensor: an ice-point check (crushed ice and water, 0.0 °C, the sensor should read 100.00 Ω within its class) and one hot point in a dry-block calibrator, read at the PLC screen, tell you what the loop actually delivers.
Head-mounted transmitters and 4–20 mA
A head-mounted transmitter is a small puck that fits inside the sensor's connection head, measures the element over a few centimetres of lead (3- or 4-wire), linearises the curve, and sends the temperature out as a 4–20 mA current on two wires. Current does not care about the resistance of the cable, so the cable length and the cable temperature drop out of the measurement, the PLC needs an ordinary analogue input rather than an RTD card, and the range (say 0–150 °C over 4–20 mA) and the sensor type are configured at the sensor, often over HART. The transmitter adds its own small error and it needs a loop supply, but for anything more than a few metres from the panel it is the cleanest solution and it is what Ahmedonics specifies by default in the systems it designs. The current loop guide explains the loop and the scaling calculator converts the milliamps back to degrees.
The example in numbers
A class A PT100 in a pasteuriser at 100 °C, wired through 15 m of 0.5 mm² cable at 0.5 Ω per wire to a PLC RTD input.
- Resistance: R(100) = 100 × (1 + 0.39083 − 0.005775) = 138.51 Ω. Sensitivity: 100 × (0.0039083 − 0.0001155) = 0.3793 Ω/°C.
- Class A tolerance at 100 °C: ±(0.15 + 0.2) = ±0.35 °C, or ±0.133 Ω.
- 2-wire: both leads add, 1.000 Ω, which is 1.000 ÷ 0.3793 = +2.64 °C. The PLC shows 102.6 °C, 7.5 times the sensor's tolerance.
- 3-wire with 5 % mismatch: the input subtracts one lead and 0.025 Ω remains, +0.07 °C; the PLC shows 100.1 °C.
- 4-wire: no lead error; the PLC shows 100.0 °C, and the ±0.35 °C of the sensor plus the input card's accuracy is what is left.
- The same cable on a PT1000 in 2-wire: 1 385.05 Ω, 3.7928 Ω/°C, and the 1.000 Ω of lead is +0.26 °C.
In reverse, a resistance of 119.40 Ω measured with the leads accounted for is 50.0 °C, and a cold-store sensor at −40 °C reads 84.27 Ω. The PT100 calculator reproduces all of these and lets you change the cable, the wiring and the class.
Fault-finding a PT100 loop
Disconnect the sensor at the input and measure it with a multimeter across the element (on a 3-wire sensor the two wires of the same colour, red by the IEC 60751 colour code, are joined at one end of the element and should read near zero ohms to each other; the white wire is the other end).
- About 109 Ω at a room temperature of 23 °C, 100.0 Ω in crushed ice and water, about 138.5 Ω in boiling water at sea level: the sensor is fine.
- Open circuit: a broken element or lead, usually from vibration or a cable pulled at the head. Short circuit or a few ohms: crushed cable or a wire touching the sheath. Both give a sensor-fault flag on a modern input and a wild reading on an old one.
- A few ohms low and unsteady: moisture in the connection head putting a leakage path in parallel with the element; it reads low and changes with the weather. Dry the head, replace the gasket and gland.
- Reads 2–3 °C high, steady: the classic 3-wire sensor on an input configured as 2-wire, or a 3-wire sensor wired with only two of its wires. The error is the two leads, as in the example, and it grows when the cable warms up.
- Reads wildly wrong or flags a fault with a good sensor: the wires are crossed (the two same-colour wires must go to the pair terminals), the input is configured for a PT1000 or for a thermocouple, or the sensor is a 0.003916 type on a 0.00385 input, which reads about 1.7 °C high at 100 °C.
- Jumps when the cable is moved: a corroded or loose terminal, which on a 3-wire connection also produces mismatch. Clean and re-terminate.
Most PT100 "failures" in Pakistani plants that we are asked about turn out to be one of the last three: a configuration, a terminal or water in the head. The sensor itself, if it was a decent one and not run above its rating, is usually the last thing to suspect.
References
- IEC 60751:2008, Industrial platinum resistance thermometers and platinum temperature sensors — the resistance–temperature relationship, tolerance classes, wire colours and the reference table
- ASTM E1137/E1137M, Standard Specification for Industrial Platinum Resistance Thermometers
- Callendar, H. L., "On the practical measurement of temperature", Philosophical Transactions of the Royal Society A, vol. 178, 1887; Van Dusen, M. S., "Platinum-resistance thermometry at low temperatures", Journal of the American Chemical Society, vol. 47, 1925
- Manufacturer documentation for the RTD input card or head-mounted transmitter in use — accuracy, excitation current, wiring configuration and the sensor curves it supports