Hysteresis in temperature sensors

Hysteresis is the dependence of a sensor's indication on its previous thermal history. Platinum resistance thermometers exhibit it mainly from strain in the element and its encapsulation; thermistors and thermocouples have their own mechanisms.

Detecting and treating it

• Calibrate on both rising and falling sequences and compare readings at common points.
• Treat half the difference as a standard uncertainty contribution with a rectangular assumption if it cannot be corrected.
• Consider whether the service use is one-directional, in which case the calibration direction should match it.

Sensors that have been thermally shocked, dropped or used beyond their range often show increased hysteresis; a change from the previous calibration is a warning sign.

Measuring it, and what the number means

Hysteresis only shows up if the calibration looks for it. A single ascending sequence cannot reveal it, which is why most certificates contain no information about it at all. The detection method is to run the points up and then back down through the same values and compare.

The pattern is characteristic: zero at the turning points, largest in the middle of the loop. The usual treatment is to take half the maximum difference as the half-width of a rectangular distribution, so 0.018/2 = 0.009 °C gives u = 0.009/√3 = 0.0052 °C. That is a real component that a one-directional calibration silently omits.

Sensor types behave differently

• Platinum resistance thermometers: hysteresis comes mainly from mechanical strain in the element and its encapsulation, and grows with thermal cycling and with range. A partially supported wire element shows more than a fully supported film.
• Thermistors: generally small hysteresis, but the encapsulation matters and a bead that has been over-temperature may show a permanent shift rather than a reversible loop.
• Thermocouples: the dominant effect is not classic hysteresis but inhomogeneity — the wire changes composition where it has been hot, so the reading depends on where the temperature gradient sits along the wire, not just on where it came from.
• Humidity sensors: the largest hysteresis of anything in this field, commonly 1–2 % RH and worse after prolonged exposure above 80 % RH.

Living with it

Three practical responses. Restrict the sensor to one direction of approach where the application allows it, and calibrate the same way — a monitoring sensor in a stable cold room approaches its working point once and stays there. Correct for it where the loop is consistent and the direction of approach is known, which is rarely the case in the field. Or include it as an uncertainty component, which is the usual answer.

An increase in hysteresis since the previous calibration is a warning sign rather than a nuisance. Sensors that have been thermally shocked, dropped, or used beyond their rated range show it first in the loop, before the ice point moves.

One practical consequence for mapping: a set of loggers taken from a warm store into a cold room all approach their working point from the same direction, so any hysteresis they share biases the whole study the same way rather than adding scatter. Equilibrating them in the medium before the study clock starts is the cheap control.