Industries / Industrial equipment and services / Industrial services / Calibration services
Calibration in the field
Field calibration is checking and adjusting instruments where they are installed, in plants, pipelines and power stations, by bringing reference equipment to them or by watching their signals, instead of shipping them to a lab.
5
Milestones achieved, from 1990 to 2018.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1990
Transmitters that leaked oil
A pressure transmitter is the field instrument that turns a pressure into an electrical signal, and in many designs a sealed fill oil carries the pressure to the sensing element. In March 1990 the NRC's Bulletin 90-01 reported that some transmitters made by Rosemount could lose their fill oil, and US nuclear plants responded by monitoring the affected models more closely or replacing them. A utility working group's report, reproduced in EPRI's 1998 online-monitoring study, says that in a US Department of Energy demonstration at the Crystal River plant in February 1996 a pattern-recognition program, MSET with a sequential probability ratio test, identified a Rosemount oil-loss failure several weeks before the instrument showed any visible symptom.
LessonA sensor can fail slowly between scheduled calibrations, so watching its signal over time catches faults that a periodic check can miss.
Credit
US Nuclear Regulatory Commission; US Department of Energy; EPRI
1994
Calibrators that keep records
A documenting process calibrator is a handheld instrument that supplies or measures the test signal, such as a pressure or a 4–20 mA current (the standard loop signal, in which 4 mA stands for the bottom of an instrument's range and 20 mA for the top), and also stores the readings found before and left after adjustment, so the technician no longer writes them on paper to be typed in later. Fluke introduced its first two, the 701 and 702, at the ISA Show in Philadelphia in May 1994. Beamex, a Finnish maker founded in 1975, released its MC5 multifunction calibrator in 1999 and says it was among the first to include HART, the digital protocol that rides on the 4–20 mA wires and lets a calibrator read and adjust a smart transmitter.
Credit
Fluke Corporation; Beamex
2000
Regulator accepts watching signals
Online monitoring (OLM) checks installed instruments for drift while the plant runs, by comparing each one with redundant instruments or with a model of the process, so that only those found drifting need a hands-on calibration. EPRI's 1998 study of the method combined transmitter calibration data from 18 nuclear plants, and in July 2000 the NRC issued a safety evaluation that authorizes using OLM on safety-related instruments under a plant's technical specifications. Each plant must still ask for its own license amendment and meet 14 requirements that the NRC set.
Credit
EPRI and Edan Engineering (study); US Nuclear Regulatory Commission
2016
Meters that check themselves
Some flowmeters test their own electronics in place against built-in references that were calibrated at the factory, and produce a pass or fail report without being taken out of the pipe. Endress+Hauser calls its version Heartbeat Technology, and Emerson's Micro Motion meters offer Smart Meter Verification. An Endress+Hauser engineer described the method in Processing magazine in 2016 and wrote that its self-monitoring detects more than 94% of potential failures, by the maker's own failure analysis. M.E. Simpson, a US firm that sells field flow testing, wrote in February 2026 that water utilities "often incorrectly assume that electronic calibration of a meter's instrumentation is equivalent to an accurate flow test".
LessonA self-check of the electronics confirms the meter's circuits, not the flow it reports, so it complements a flow test instead of replacing it.
Credit
Endress+Hauser; Emerson
2018
A sensor 21 degrees off
An angle-of-attack sensor is a vane on an airliner's nose that measures the angle between the wing and the oncoming air. One was serviced in 2017 at Xtra Aerospace in Miramar, Florida, and fitted to a Lion Air Boeing 737 MAX the day before the plane crashed into the Java Sea on 29 October 2018, killing all 189 people aboard. Indonesia's National Transportation Safety Committee found that the new sensor read 21 degrees differently from the one on the other side and was "most likely improperly calibrated at Xtra Aerospace", and that investigators could find no paperwork showing it had been calibrated properly. Hours after the report came out in October 2019, the US Federal Aviation Administration revoked Xtra's repair station certificate, saying the shop had not followed the procedures in its own repair station manual. The crash's design causes are covered under Aerospace and defense.
LessonA shop's calibration record does not prove a part is right once fitted, so a check after installation, such as a comparison with its twin on the other side, is the last chance to catch the error.
Credit
Indonesia's National Transportation Safety Committee (investigation); US Federal Aviation Administration
Watch for
Plants reading certificates automatically
Researchers at BAM, Germany's federal institute for materials research and testing, with a co-author from DAkkS, built a demonstrator that monitors a process temperature and uses the sensor's DCC to check conformity automatically, with no person transferring calibration values by hand, and published the code.
Achieved whena plant or lab outside BAM reports using DCC-based checks on installed sensors in routine operation, with the number of sensors covered and the errors caught.
Claimed by
BAM, April 2026
Open
01
Time a paperless calibration
A timed, filmed bench trial in which technicians calibrate the same set of pressure and temperature transmitters two ways: with a reference gauge, a multimeter and a paper sheet typed into a spreadsheet afterwards, and with a documenting calibrator that records results itself. Count the minutes per instrument and every wrong value in the final records.
The briefCredit · unclaimed02
Open data for sensor drift
A public data set from a small bench water loop fitted with three or more redundant pressure, flow and temperature transmitters, logged continuously for months, in which known drifts are introduced on purpose by changing one transmitter's calibration trim in small recorded steps, and natural drift is recorded against a reference. Add a scored comparison of drift-detection methods such as auto-associative kernel regression (AAKR), which predicts what each sensor should read from the others.
The briefCredit · unclaimed
Open 01 · the brief
Time a paperless calibration
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A timed, filmed bench trial in which technicians calibrate the same set of pressure and temperature transmitters two ways: with a reference gauge, a multimeter and a paper sheet typed into a spreadsheet afterwards, and with a documenting calibrator that records results itself. Count the minutes per instrument and every wrong value in the final records.
Why it is still open
Makers publish large savings with no published trial behind them. Fluke says route-based documenting calibrators "can save as much as 50 percent" of the time and cost of manual methods. Beamex's product marketing manager wrote in Quality magazine in February 2022 that the typical manual data-entry error rate is about 1% and that about 40% of records entered by the two-step paper method contain errors, citing no study. A search found no independent measurement.
Built with what exists
Ten used smart pressure transmitters (for example Rosemount 3051) and five temperature transmitters with RTD probes on a bench; a hand pump kept below 7 bar and a reference pressure gauge; a dry-block temperature calibrator; a rented documenting calibrator such as a Fluke 754 or Beamex MC6; a HART modem; a spreadsheet standing in for the maintenance system; cameras to film each run so that errors can be counted afterwards.
First step
Buy five used transmitters, write the test protocol and score sheet, and time one person doing five calibrations each way.
Done looks like
At least 60 calibrations each way by at least three people in randomized order, with the median minutes per instrument and the count of wrong values in the final records for each method, published with the run logs and a protocol fixed in advance.
Open 02 · the brief
Open data for sensor drift
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public data set from a small bench water loop fitted with three or more redundant pressure, flow and temperature transmitters, logged continuously for months, in which known drifts are introduced on purpose by changing one transmitter's calibration trim in small recorded steps, and natural drift is recorded against a reference. Add a scored comparison of drift-detection methods such as auto-associative kernel regression (AAKR), which predicts what each sensor should read from the others.
Why it is still open
The nuclear studies behind online monitoring rest on utilities' own records; EPRI's 1998 study pooled transmitter calibration data from 18 plants. A March 2026 paper on predicting when instruments need calibration had to simulate drift by adapting NASA's C-MAPSS jet-engine data set, and a 2017 drift study built its own 24-sensor wireless test bed. The nearest open bench data, the Skoltech Anomaly Benchmark (SKAB) from a water-circulation test bed, labels process faults such as closed valves, leaks and rotor imbalance, not drift in a sensor's calibration. No open data set of real, labeled drift in process transmitters was found.
Built with what exists
Used HART pressure and temperature transmitters; a small pump, tank and pipe loop running tap water at low pressure; a reference transmitter checked at an accredited lab; HART modems on a Raspberry Pi for logging; Python with scikit-learn and the open aakr package for AAKR and other baselines; SKAB's published scoring code as a model for the benchmark; Zenodo to publish the data.
First step
Build the loop with three pressure transmitters on one header and log a first week of data to see how closely they track each other.
Done looks like
At least six months of one-second data from at least nine channels with at least 20 injected drifts of known size and timing, and for each detection method the time to detect drifts of 0.25%, 0.5% and 1% of span and the false alarms per month, published with the code.