A reading that stays plausible while it goes wrong.
A process sensor rarely fails outright. It moves, slowly, and the control system follows it, and the first evidence is a yield figure that nobody can trace back to a cause.
What fails
In continuous production the thermometer is inside the control loop, so a sensor that drifts does not produce an alarm. It produces a setpoint that is quietly wrong, and a plant that holds it perfectly. Yield moves a fraction of a per cent, product sits closer to a specification limit than anyone intended, and the cause is indistinguishable from a dozen other things that also drift slowly.
Redundancy helps, but it has limits. Two sensors of the same type in the same environment can drift in similar ways, so agreement between them does not necessarily mean either is still correct.
Why recalibration is not the answer
Removing a sensor from a running process may require a shutdown, a spare sensor, or a thermowell arrangement that introduces additional lag and uncertainty. The recalibration interval therefore becomes a compromise between measurement confidence and operational disruption, rather than a purely metrological decision.
What a noise measurement changes
The sensor itself no longer has to remain stable. We determine temperature from the Nyquist relation, using resistance measured from the same block of data as the noise. If the probe becomes contaminated, oxidises or its resistance changes, the temperature measurement remains valid.
Two engineering results decide whether that is usable on a plant instead of a bench. The first is speed. A random signal takes time to measure, and for a long while that was the objection to noise thermometry in process work. The performance we published is 0.1 per cent in 5.3 seconds, which our paper describes as an adequate uncertainty within a response time acceptable for industrial use.
The second is interference. A plant floor is one of the worst electromagnetic environments there is, and our signal is random and measured in microvolts, which is precisely what makes it vulnerable. Electromagnetic interference had defeated every previous attempt at a practical noise thermometer. Ours went to an accredited laboratory and passed the more demanding industrial category, undisturbed at the highest levels the laboratory could generate.
- Published measurement performance
- 0.1 % in 5.3 s
- Radiated RF immunity, EN 61000-4-3
- 10 V/m, 80–1000 MHz · 3 V/m, 1.0–2.7 GHz
- Unaffected by the applied fields
- The sense resistance is typically 5 kΩ, where prior-art Johnson noise thermometers use around 100 Ω
- 5 kΩ
- The thermometer works over the band 10 kHz to 1.2 MHz
- 10 kHz – 1.2 MHz
Current status
Our electromagnetic work has passed independent accredited immunity testing. That matters because electromagnetic interference has historically been one of the main obstacles to taking noise thermometry out of the laboratory. We have also published the measurement speed, and the ThermoSI consortium identifies high-value heat treatment, alongside nuclear applications, as a target for the current work.
What we have not yet done is the packaging. The instrument is a working second-generation prototype, not a catalogue product with process connections, approvals and a spares list. The final form will depend on the application, which is why we would rather discuss a specific measurement problem than describe a general solution.
If this is your measurement
The thermometer is a working second-generation prototype. Where it goes next will be shaped by real applications. If calibration drift is limiting a measurement you are responsible for, we would like to hear from you.