Industries / Industrial equipment and services / Industrial services / Calibration services
Pressure and temperature
This work checks pressure gauges, pressure transmitters and thermometers against reference instruments whose own values trace back, step by step, to a national laboratory's standards.
5
Milestones achieved, from 1849 to 2014.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1849
Bourdon's curved tube
Eugène Bourdon, a French watchmaker and engineer, invented in 1849 a gauge built around a curved, hollow tube sealed at one end: pressure inside makes the tube flex, and a linkage turns a pointer over a dial. It could read far higher pressures than the liquid-filled U-tubes of the time, and Bourdon founded a company in 1850 to make it. Because the tube is a spring, not a standard, each gauge has to be checked against a reference before its readings can be trusted. WIKA, a gauge maker, says millions of Bourdon gauges are still sold every year, and notes that Bernard Schaeffer of Magdeburg patented a diaphragm gauge the same year.
Credit
Eugène Bourdon
1927
One temperature scale
The 7th General Conference on Weights and Measures (CGPM), the treaty body that governs the metric system, adopted the International Temperature Scale of 1927 to replace the differing national scales and to avoid the practical difficulty of measuring temperature with gas thermometers. The scale assigned values to fixed points, temperatures at which pure substances melt, freeze or boil in a repeatable way, from the boiling point of oxygen up to the gold point. Between them it named the instruments to use: a platinum resistance thermometer for the lower range, a platinum-rhodium thermocouple above it, and an optical pyrometer above gold. Calibration laboratories have worked from such a scale ever since.
Credit
the 7th CGPM
1990
A flawed scale replaced
Soon after the 1968 scale (IPTS-68) was adopted, metrologists found it had a lower limit of 13.81 K, disagreed with true thermodynamic temperature, and was poorly reproducible between about 630 °C and 1064 °C, the range where a thermocouple was the defining instrument. The Consultative Committee for Thermometry built a new scale, the International Committee for Weights and Measures (CIPM) adopted it in September 1989, and the ITS-90 took effect on 1 January 1990. It replaced the thermocouple with platinum resistance thermometers up to the freezing point of silver (961.78 °C) and with radiation thermometry above it. NIST's guide told every standards and calibration laboratory to convert as soon as possible; at 630.6 °C, for example, the two scales differ by 0.125 °C.
LessonA reference instrument that cannot be reproduced well limits every calibration below it, so the scale itself had to change.
Credit
the CIPM's Consultative Committee for Thermometry; B. W. Mangum and G. T. Furukawa of NIST for the guide used to put it into practice
2011
Mercury thermometers lose NIST
The US National Institute of Standards and Technology (NIST) stopped calibrating mercury thermometers on 1 March 2011, ending a service it had offered since it opened in 1901. It acted with the US Environmental Protection Agency and standards groups to phase out mercury; by NIST's count about 300 of roughly 700 ASTM standards had been changed to allow mercury-free thermometers. Two years later the Minamata Convention, signed on 10 October 2013, set 2020 as the phase-out date for mercury barometers, manometers and thermometers, but it exempts products used to calibrate instruments or kept as reference standards.
LessonA calibration chain can drop a reference instrument for safety reasons even where the law still allows it, and the replacement thermometers then need their own route back to the standards.
Credit
NIST, with the US EPA; the parties to the Minamata Convention
2014
Pressure measured with light
NIST's Fixed-Length Optical Cavity (FLOC), about the size of a travel mug, measures gas pressure from the change in the gas's refractive index, the amount by which it slows light, inside a laser cavity. NIST reported that it resolves 0.1 mPa, 36 times finer than its 3-meter mercury column, the US primary pressure standard, and works about 100 times faster. Outside NIST, researchers in the European QuantumPascal project, led by Germany's PTB, built optical pressure standards, and in 2023 a transportable refractometer built by Umeå University and RISE in Sweden was carried to national laboratories in Sweden, Germany, Italy and France and compared with their dead-weight pressure balances, piston devices that set pressure with known masses, from 10 to 90 kPa.
Credit
NIST, with Jay Hendricks as principal investigator; C. Forssén, M. Zelan and co-authors for the transportable instrument
Watch for
A chip that recalibrates itself
NIST holds a US patent, number 12,625,007, for a thermometer on a single chip that pairs a fast photonic sensor with a slower optomechanical sensor that measures absolute temperature and corrects the first one while it works. NIST says this gives "permanent, drift-free accuracy" and pitches it for factories, but gives no accuracy figures.
Achieved whena laboratory outside NIST runs a packaged device beside a calibrated platinum resistance thermometer for several months and publishes how far it drifts without being recalibrated.
Claimed by
NIST (inventors Daniel Barker, Nikolai Klimov and Thomas Purdy), June 2026
A robot verifies dial gauges
Sinopec, China's state oil refiner, says it led the development of a device that verifies four dial pressure gauges at once: a robot arm mounts and removes them, a camera centers on and reads each dial, and software raises the pressure, judges each gauge and issues the certificate, in 5 to 6 minutes against about 20 minutes by hand. It claims a reading recognition rate of 99.7%, says the device meets the national standard, and notes that its Tianjin refinery alone has about 20,000 pressure gauges in use.
Achieved whenChina's metrology authorities or an outside laboratory publish the device's readings and pass-or-fail verdicts next to manual verification of the same gauges, or a site reports a year of routine use with counts.
Claimed by
Sinopec, November 2026
Open
01
Read test gauges to tenths
An open set of photographs of dial pressure gauges taken at known setpoints during calibration, with careful human readings as the answer key, and an open reader that turns each photo into a reading good to a tenth of a scale division, the level a calibration needs.
The briefCredit · unclaimed02
A thermometer that checks itself
An open-hardware Johnson noise thermometer, which reads temperature from the tiny random voltage that heat makes in any resistor, built from audio-grade parts. Because that noise depends only on temperature and resistance, the reading does not drift the way an ordinary sensor's does, and the instrument can be checked against a reference resistor held in an ice bath instead of being sent away for calibration.
The briefCredit · unclaimed
Open 01 · the brief
Read test gauges to tenths
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of photographs of dial pressure gauges taken at known setpoints during calibration, with careful human readings as the answer key, and an open reader that turns each photo into a reading good to a tenth of a scale division, the level a calibration needs.
Why it is still open
Published open methods read gauges far more coarsely. A 2024 ETH Zurich system reads gauges "in the wild" to within 2% of the reading, and an August 2026 study found the best fine-tuned vision-language model still erred by 2.4% to 4.4% of the gauge's range, while a tenth of a division on a 100-division dial is 0.1% of range. Closed machines claim more without public proof. Additel's 710 reader says its datasheet, revised in June 2026, reads to "better than 0.1 division", but it cannot read narrow-pointer test gauges and has no independent test. Sinopec's November 2026 verification robot gives only a 99.7% "recognition rate", not an error in scale divisions. A 2022 academic system was tried on one gauge in three trials. No public set of gauge photos with calibration-grade answers was found.
Built with what exists
A pneumatic hand pump and test manifold kept to low pressure, with gauges of 0 to 2 bar range; a digital reference gauge such as a Fluke 700G or Additel 681 with a current calibration certificate; a fixed camera such as the Raspberry Pi HQ Camera; Label Studio for marking pointer tips and scale marks; an open vision-language model such as Qwen2.5-VL-7B fine-tuned with QLoRA, as in the 2026 study, and an Ultralytics YOLO keypoint model as a second approach.
First step
Borrow 10 gauges, photograph each at 10 setpoints on a hand-pump bench, and have three people read every photo with a magnifier to measure how much human readers disagree.
Done looks like
At least 30 gauges of five kinds, including mirror-band test gauges with knife-edge pointers, dual-scale gauges and vacuum gauges, each photographed at 10 setpoints rising and falling; three trained readers per photo; the reader's error published for every photo in scale divisions, with at least 95% of readings within 0.1 division of the readers' average, and the resulting gauge errors matching a calibration done by hand on the same gauges.
Open 02 · the brief
A thermometer that checks itself
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-hardware Johnson noise thermometer, which reads temperature from the tiny random voltage that heat makes in any resistor, built from audio-grade parts. Because that noise depends only on temperature and resistance, the reading does not drift the way an ordinary sensor's does, and the instrument can be checked against a reference resistor held in an ice bath instead of being sent away for calibration.
Why it is still open
The UK's National Physical Laboratory (NPL) says the industrial Johnson noise thermometer it helped Metrosol develop reached its target of better than 1 °C, and NIST's self-correcting chip thermometer exists only as a patent with no published accuracy. Low-cost builds so far are teaching tools or unfinished. A March 2026 preprint describes a cheap two-channel setup that measures the Boltzmann constant to about 1%, which at room temperature is about 3 °C; a University of Ljubljana group described a thermometer built around a studio sound card but had only preliminary results; and the University of Szeged's open sound-card thermometer software is for teaching. No open design with published accuracy against a reference thermometer was found.
Built with what exists
The March 2026 preprint's two-channel circuit as a starting point; a two-channel 24-bit audio interface; two low-noise amplifier channels whose outputs are cross-correlated so that each amplifier's own noise averages out; metal-film resistors as sensor and reference; Python with NumPy and SciPy; an ice bath and a stirred water bath from 0 °C to 90 °C; a calibrated platinum resistance thermometer and readout as the reference.
First step
Build the two-amplifier front end, record a resistor in an ice bath and in a 50 °C bath, and report how much the ratio of their noise varies over 100 runs.
Done looks like
Readings at 10 temperatures from 0 °C to 90 °C within ±0.5 °C of the reference thermometer, repeated 30 days later with no adjustment, with the uncertainty stated at about 95% confidence, parts costing under US$500, and design files and raw data published.