How to calibrate a Pt100

A Pt100 is a platinum resistance thermometer with a nominal resistance of 100 Ω at 0 °C. Industrial Pt100 sensors are made to tolerance classes defined in IEC 60751, but a tolerance class is not a calibration: calibration establishes the actual relationship between the sensor's resistance (or the indicated temperature of the instrument it is connected to) and temperature, with an uncertainty.

Calibration by comparison

• Choose calibration points across the range of use, including the extremes.
• Generate each temperature in a suitable medium — a stirred liquid bath, a dry-block calibrator or a chamber — and allow full stabilisation.
• Immerse the Pt100 and the reference thermometer to sufficient depth, close together.
• Record repeated readings of both; the reference must be traceably calibrated.
• Compute the correction (reference minus indicated) at each point and its uncertainty.

Where the uncertainty comes from

Typical contributions are the reference thermometer's calibration uncertainty and drift, the medium's stability and uniformity, immersion effects, self-heating, the readout resolution and, for the unit under test, repeatability and hysteresis. Each is expressed as a standard uncertainty, combined in quadrature, and expanded with a coverage factor (usually k = 2) following the GUM.

Reporting

The certificate should give the measured values, the corrections or deviations, the uncertainty and its coverage factor, the reference standards used with their traceability, and the environmental conditions. If a conformity statement to a tolerance class is made, the decision rule must be stated.

ValiTrac's AI assistant can retrieve evidence on this procedure from its approved sources and, where a numeric combination is needed, calls the deterministic uncertainty engine rather than estimating the arithmetic.

Choosing the points

Points are chosen from use, not from convenience. Cover the working range including both extremes, because extrapolating beyond a calibrated point is not supported by anything on the certificate. Add a point near any limit the instrument has to police: a logger guarding an 8 °C ceiling should be calibrated near 8 °C, not at 0 °C and 50 °C with the interesting region interpolated.

Three to five points is typical for industrial work. A single point establishes an offset and says nothing about the slope; two points cannot reveal curvature. If the readout applies ITS-90 or Callendar–Van Dusen coefficients, enough points are needed to fit them — which is a different and larger requirement than a few correction values.

Lead resistance, and why three wires are not four

A Pt100 changes by about 0.385 Ω per °C near 0 °C, so lead resistance matters enormously. A two-wire connection adds the whole loop resistance to the reading: 1 Ω of cable is 2.6 °C of error, and it is always in the same direction. Three-wire compensation subtracts one current leg from the other and works only to the extent that the two legs are identical; a 0.1 Ω imbalance from a repaired junction box or differing lead lengths leaves about 0.26 °C behind. Four-wire separates current and sense paths entirely and removes the term. For calibration, use four wires, and if the instrument will be used on three in service, calibrate it that way as well and record which.

Resolution dominates, which is the usual finding for a 0.1 °C logger and the reason a more expensive reference thermometer would barely move this number.

Mistakes that cost a calibration

• Insufficient immersion. The largest avoidable error in the whole procedure, and invisible unless you test for it by immersing 20–30 mm further and watching the reading.
• Reading before stabilisation. Take readings only when both the reference and the unit under test have stopped trending, and record several rather than one.
• Ignoring self-heating. Check it by halving the measurement current; the difference is three quarters of the full effect.
• Entering certificate coefficients into the readout and then treating the readout as contributing nothing. It has its own accuracy, resolution and drift.
• Confusing a tolerance class with a calibration. Class A is a manufacturing limit on a new sensor; it says nothing about the one in your hand after two years in a cold room.