BRENNEN

Industries / Industrial equipment and services / Industrial services / Calibration services

Electrical

Electrical calibration checks multimeters, calibrators, power supplies and other instruments that measure voltage, current and resistance, by comparing them with reference standards whose values trace back to a national laboratory.

5

Milestones achieved, from 1972 to 2021.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1972

    Volt tied to quantum effect

    Laboratories had kept the volt in standard cells, small electrochemical cells whose voltage drifts over time, depends strongly on temperature and sometimes jumps without warning. From 1 July 1972 the US National Bureau of Standards (NBS, now NIST) based the US volt on the Josephson effect: a junction between two superconductors, cooled close to absolute zero and fed microwaves, produces a voltage fixed only by the microwave frequency and constants of nature. The US ohm stayed tied to a group of wire-wound resistors, as it had been since 1 January 1948. The International Committee for Weights and Measures (CIPM) records that its electrical committee set an agreed value for the Josephson constant, the ratio of frequency to voltage, in 1972.

    nvlpubs.nist.govbipm.org

    Credit

    NBS

  2. 1984

    Arrays reach one volt

    One Josephson junction gives no more than a few tens of microvolts, so a practical standard chains thousands of nearly identical junctions on one chip. By NIST's account the first 1 V standard was produced in 1984, 10 V chips worked by the late 1980s, and agreement between laboratories' dc voltage measurements improved by four orders of magnitude, to within a few parts in 10^10. NIST built programmable versions that can be set to any voltage, 1 V in 1997 and 10 V in 2006, shipped the first 10 V programmable system to NASA's Kennedy Space Center, began producing 10 V programmable standards for others in 2011, and by 2013 had seen similar systems installed in India, Brazil and Taiwan.

    LessonNIST found that many users of its programmable standards did not check that they were working with quantum accuracy, and a result can repeat well and carry a small stated uncertainty yet still be wrong.

    phys.orgnist.gov

    Credit

    NBS and NIST, including Clark Hamilton and Samuel Benz for the programmable designs

  3. 1988

    A calibrator checks itself

    A multifunction calibrator is a bench instrument that puts out precise voltages, currents and resistances for calibrating meters. Fluke's 5700A, introduced in 1988, added what Fluke calls artifact calibration: the instrument carries its own reference standards, a ratio device and a null detector, and adjusts all its ranges against just three outside standards, a 10 V reference and 1 Ω and 10 kΩ resistors, in about an hour instead of several hours, by Fluke's account. Fluke's own study of about 260 instruments returned to its service center found a better than 99% chance of meeting the one-year dc voltage specification at every level except 1 V, where it was 98.4%.

    LessonFluke still recommended a full check against outside standards every two years, because instruments vary more with age and only an outside check finds the few that drift.

    xdevs.comfluke.com

    Credit

    Fluke

  4. 1990

    Volt and ohm reset

    On 1 January 1990, by international agreement, national laboratories began using fixed agreed values for two constants: 483 597.9 GHz/V for the Josephson constant (called K_J-90) and 25 812.807 Ω for the von Klitzing constant (R_K-90). The second comes from the quantum Hall effect, reported in 1980 by Klaus von Klitzing, Gerhard Dorda and Michael Pepper, in which a semiconductor device, cooled and placed in a strong magnetic field, shows a resistance set by constants of nature. The change raised the US volt by 9.264 parts per million (ppm) and the US ohm by 1.69 ppm, and NIST warned that "literally thousands" of standards, meters, calibrators, test systems and avionics would need adjusting; the volt step alone was 19% of a 50 ppm tolerance. The 2018 decision that revised the SI, the international system of units, abrogated K_J-90.

    LessonA smaller US volt change, 8.4 ppm on 1 January 1969, "caused considerable difficulty and many individuals dealt with it incorrectly", so for 1990 NIST published step-by-step guidance six months ahead.

    nvlpubs.nist.govbipm.orgapi.crossref.org

    Credit

    CIPM; NIST's guidelines by Norman Belecki, Ronald Dziuba, Bruce Field and Barry Taylor

  5. 2021

    A robot turns the dials

    Many handheld digital multimeters have no data port, so calibrating one means turning its rotary switch, pressing buttons, moving the test leads and reading the display, step by step. Trescal, a calibration company, built with Thinkbot Solutions and Linestar Automation a six-joint robot arm that does this, first in its Dallas lab; each meter rides a conveyor in a 3D-printed carrier, and a spreadsheet-style file records where that model's dial, jacks and screen are. The British Measurement and Testing Association (BMTA) reported in February 2021 that the robot had passed an assessment by UKAS, the UK's accreditation body, for accredited work at Trescal's Stevenage lab, handles up to 14 meters a session at 18 to 35 minutes each, about a technician's time, and after two years in Dallas calibrated "just as precisely as highly-trained technicians".

    bmta.co.uk

    Credit

    Trescal, Thinkbot Solutions and Linestar Automation

Watch for

  1. Graphene resistance standards spread

    A quantum Hall resistance standard built on graphene, a sheet of carbon one atom thick, has less demanding operating conditions than the gallium arsenide devices that earlier guidelines were written for; NIST ran one in a table-top cryocooler, with no liquid helium, in 2018. A guide published in Metrologia by a task group of the CIPM's Consultative Committee for Electricity and Magnetism (CCEM) says graphene allows more compact and cost-effective cooling and magnet systems, giving national institutes and calibration laboratories wider access to primary quantum resistance standards, and it supports their formal acceptance as primary standards.

    Achieved whengraphene devices are formally accepted as primary resistance standards, or an accredited calibration laboratory outside the national institutes reports resistance calibrations traced to its own graphene quantum standard.

    inrim.itnist.gov

    Claimed by

    CCEM Task Group on Graphene QHR Standards, as reported by Italy's national metrology institute INRiM, June 2026

Open

  1. 01

    Teach the robot new meters

    A bench robot cell that calibrates handheld multimeters it has never seen, learning each new model from a few recorded demonstrations instead of a hand-written configuration file and a custom-made carrier. The robot sets the rotary switch and buttons, moves the leads, reads the display and records the results, at low-voltage test points only (for example up to 30 V, plus resistance and small currents), so nobody works near hazardous voltage.

    The brief
    Credit · unclaimed
  2. 02

    Procedures from manuals, tested

    An AI agent that reads the performance-verification section of a multimeter's manual, the maker's table of test points and allowed limits, and writes an executable calibration procedure in Python; then a test that compares each generated procedure point by point with the maker's table and runs a sample of them on real instruments at low voltage.

    The brief
    Credit · unclaimed

Open 01 · the brief

Teach the robot new meters

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A bench robot cell that calibrates handheld multimeters it has never seen, learning each new model from a few recorded demonstrations instead of a hand-written configuration file and a custom-made carrier. The robot sets the rotary switch and buttons, moves the leads, reads the display and records the results, at low-voltage test points only (for example up to 30 V, plus resistance and small currents), so nobody works near hazardous voltage.

Why it is still open

The one robot found in routine accredited use, Trescal's, needs a separate configuration file for each new meter type and a carrier made from a 3D scan of that meter, and Trescal's UK technical manager said earlier attempts had "either failed or was only successful on a limited range of products". A 2020 station from Gdynia Maritime University reads meters' seven-segment displays with a camera, but each meter type needs its display areas marked out by hand, and the operator still connects the leads and changes the meter's ranges. No published test was found of a robot that handles unseen meter models from demonstrations, and GitHub searches for multimeter robots found none.

bmta.co.ukpmc.ncbi.nlm.nih.gov

Built with what exists

A collaborative arm with force sensing, such as a Universal Robots UR5e, with a gripper and a wrist camera; Hugging Face's LeRobot to record demonstrations and train (see the Note "No download"); an open robot model that learns from a few demonstrations, such as NVIDIA's GR00T 2 (see the Note "Five demonstrations per task"); a multifunction calibrator such as a Fluke 5522A, held to a 30 V limit and driven from Python with the open PyVISA library; each meter's manual for its test points and limits.

  1. First step

    Collect ten different second-hand handheld meters, fix one list of low-voltage test points, and record five demonstrations per meter by remote-controlling the arm.

  2. Done looks like

    Twenty meter models held out of training, each calibrated by the robot at the same 15 low-voltage test points as by a technician, with a published tally of correct switch settings, readings that match the technician's within one display count, minutes per meter, and every failure with its cause.

Open 02 · the brief

Procedures from manuals, tested

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An AI agent that reads the performance-verification section of a multimeter's manual, the maker's table of test points and allowed limits, and writes an executable calibration procedure in Python; then a test that compares each generated procedure point by point with the maker's table and runs a sample of them on real instruments at low voltage.

Why it is still open

Published work stops short of this. In 2023 the chief software architect of GageList, a calibration software firm, had ChatGPT draft a caliper calibration procedure and edited it by hand, without running it or checking it against a maker's procedure. A 2025 benchmark, TMIQ, found that models picked the exact SCPI command, the text commands that remote-control test instruments, 56% to 73% of the time for one oscilloscope. A May 2026 paper on an agent for radio-frequency instruments, RFIA, kept the model to planning and left the instrument commands to fixed, tested code. Hand-written open procedures do exist: one public GitHub project calibrates seven Keysight bench meters from a Fluke 5522A calibrator in Python. None of this work generated calibration procedures and scored them against makers' tables.

qualitymag.comarxiv.orgarxiv.orggithub.com

Built with what exists

An AI model that reads long documents, such as Claude Fable 5.2 (see the Note "Three working devices"); Python with PyVISA; bench multimeters with a remote interface, such as a Keysight 34461A; a multifunction calibrator held to a 30 V limit; makers' service manuals that are free to download.

  1. First step

    Pick five bench multimeters whose manuals are free, type their verification tables into a spreadsheet as the answer key, and score a first batch of generated procedures against it.

  2. Done looks like

    Procedures generated for 30 multimeter models; for each, the share of test points and limits that match the manual's table exactly, checked by a second person; five of them run on real instruments, with readings compared against a technician's run of the maker's procedure; all procedures, prompts and scores published.