Industries / Industrial equipment and services / Testing, inspection and certification
Product testing labs
Independent laboratories test products against published standards for makers, buyers and regulators, and let the products that pass carry their mark.
6
Milestones achieved, from 1896 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1896
A bureau to test lamps
Thomas Edison started the Lamp Testing Bureau in the United States in 1896 to test electric lamps, by Intertek's account. It was later renamed Electrical Testing Laboratories, and its ETL mark is still issued: the British group Inchcape bought the lab for its testing arm, which became Intertek Testing Services after a buyout by Charterhouse Development Capital.
Credit
Thomas Edison; Intertek
1978
Lab records become law
In the 1970s, FDA inspections of laboratories that run safety studies on animals found studies that had not been done as reported. One was Industrial Bio-Test Laboratories (IBT), a contract lab that tested drugs, pesticides and chemicals for their makers; three men who had run its toxicology work were convicted of fraud and false statements over a 1976 report to Monsanto on TCC, an ingredient of deodorant soaps, and a report to Syntex on the arthritis drug Naprosyn whose blood and urine data were found to be fabricated, and a federal appeals court upheld the convictions. FDA's Good Laboratory Practice (GLP) rules, 21 CFR Part 58, issued on 22 December 1978 and in force from June 1979, set how such studies must be planned, recorded and reported, and FDA began regular inspections and audits of the raw data.
Lessona report is only as good as the raw records behind it, and someone outside the lab has to be able to check them.
Credit
US Food and Drug Administration; federal prosecutors in Chicago
1981
Tested once, accepted everywhere
On 12 May 1981 the Council of the OECD, an intergovernmental organization of mostly industrialized countries, decided that data from chemical safety tests run under the OECD Test Guidelines and its Principles of GLP must be accepted by every country that adheres to the decision, so a maker no longer has to repeat the same test for each market. The OECD estimated in 2019 that this saves governments and industry more than €309 million a year.
Credit
OECD Council; the OECD expert group that wrote the GLP principles in 1979 and 1980
1988
More labs may approve
US workplace safety rules said electrical equipment had to be approved by two named private firms, Underwriters Laboratories (UL) and Factory Mutual Research Corporation. MET Electrical Testing Company, a Baltimore lab, sued; under a 1983 settlement the Occupational Safety and Health Administration (OSHA) agreed to drop the names, and in December 1987 a federal judge ordered it to finish the rule within 120 days. The rule of 12 April 1988 deleted the two names from 23 provisions and accepted any "nationally recognized testing laboratory" (NRTL) that OSHA recognizes after checking its test equipment, staff, written procedures and follow-up inspections at factories.
Lessona rule that names private testers shuts out their competitors, and here it took a lawsuit to open it.
Credit
MET Electrical Testing Company; OSHA
1992
Textiles tested part by part
OEKO-TEX has issued textile certificates since 1992 through member testing institutes and calls its STANDARD 100 label its original safety standard. Every thread, button and accessory of a labelled article is tested against a list of more than 1,000 harmful substances, with the strictest limits for articles for children up to three years old. Parts that already hold the label need not be tested again in the finished product, so the cost of testing is spread along the supply chain.
Credit
OEKO-TEX and its member institutes
2020
Retests overrule lab reports
On 3 April 2020, to get enough respirators for US hospitals, the Food and Drug Administration (FDA) authorized Chinese-made respirators that met one of three conditions; one was a test report from an independent accredited laboratory. From April, NIOSH, the federal institute that approves US respirators, tested respirators bought on the market and found that many admitted on test reports filtered less than the expected 95% of particles. On 7 May 2020 FDA reissued its authorization without the test-report route, and more than 70 products came off its list, by the count of the American Industrial Hygiene Association (AIHA).
Lessona report describes the sample a lab was sent; testing units bought on the open market showed the gap.
Credit
NIOSH; FDA
Watch for
Safety marks for AI products
UL Solutions says it has issued the world's first certifications under UL 3115, its Outline of Investigation for the Safety of AI-Based Products, to the Qcells energy management system for data centers and the Omniconn Platform 4.0 smart building system, after evaluations that it says confirmed their robustness, reliability, transparency and human oversight. An outline of investigation is a set of requirements written by the certifier itself, not a standard agreed through a consensus process.
Achieved whenthe requirements are published as a consensus standard, or a regulator, buyer or second certifier repeats the evaluation of a certified system and accepts the result.
Claimed by
UL Solutions, March 2026
Open
01
A robot test finger
A robot cell that runs the jointed test finger check. The jointed test finger, test probe B of IEC 61032, is a metal finger 12 mm across with two joints that bend to 90°; a tester pushes it into every gap in a product's case with a force of 10 N and watches a contact indicator to see whether it reaches a part that would be live or dangerous. The cell holds the probe on a force sensor, finds each opening with a camera, logs force and probe position at every push and films it. It works only on products that are never connected to mains: the parts that would be live are wired to a low-voltage contact circuit instead.
The briefCredit · unclaimed02
Flames timed from video
An open set of videos of glow-wire tests, in which a wire heated to as much as 960 °C is pressed into a plastic part, and the times when a flame appears and goes out decide whether the part passes. Each video gets frame-by-frame flame times agreed by two engineers, and a vision model is trained to time the flames. The set also compares the engineers' frame times with the times the operator recorded on the tester, to show how much hand timing varies.
The briefCredit · unclaimed
Open 01 · the brief
A robot test finger
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A robot cell that runs the jointed test finger check. The jointed test finger, test probe B of IEC 61032, is a metal finger 12 mm across with two joints that bend to 90°; a tester pushes it into every gap in a product's case with a force of 10 N and watches a contact indicator to see whether it reaches a part that would be live or dangerous. The cell holds the probe on a force sensor, finds each opening with a camera, logs force and probe position at every push and films it. It works only on products that are never connected to mains: the parts that would be live are wired to a low-voltage contact circuit instead.
Why it is still open
The check is done by hand. A probe maker's instructions tell the tester to hold the insulated handle, guide the tip into vents, panel seams and control slots, apply a steady 10 N and check the connected detector. Web, GitHub and arXiv searches in November 2026 found no robot or cobot set up to run it.
Built with what exists
A collaborative robot arm with a wrist force-torque sensor, such as a Universal Robots UR5e; a commercial IEC 61032 probe B; a depth camera to find openings; a low-voltage continuity circuit and a data logger; teaching by demonstration (see the Note "Five demonstrations per task").
First step
Build a plastic test box with ten openings of known size, some leading to a wired contact plate, and teach the arm to probe all ten.
Done looks like
20 household products and enclosures, every opening probed by the cell and separately by two test engineers; the share of openings where the cell and both engineers agree on touch or no touch; force traces showing every push held within the probe maker's 10 N ± 0.5 N; all of it filmed and published.
Open 02 · the brief
Flames timed from video
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of videos of glow-wire tests, in which a wire heated to as much as 960 °C is pressed into a plastic part, and the times when a flame appears and goes out decide whether the part passes. Each video gets frame-by-frame flame times agreed by two engineers, and a vision model is trained to time the flames. The set also compares the engineers' frame times with the times the operator recorded on the tester, to show how much hand timing varies.
Why it is still open
On a typical tester the operator stops the clock by hand: one maker lists its ignition timer as "automatic recording with manual pause". GitHub and Hugging Face searches found no dataset or model for timing glow-wire flames; video flame detection exists for wildfires and buildings, not for this test.
Built with what exists
Videos filmed by a partner laboratory that already runs the test, so the builder never operates the hot wire; a phone or machine-vision camera at 60 frames a second; open-source detection models such as Ultralytics YOLO, fine-tuned on frames the engineers label; OpenCV and Python for timing and scoring.
First step
Ask two accredited laboratories that run IEC 60695-2 glow-wire tests whether they will film ten routine tests each with a fixed camera.
Done looks like
500 test videos from at least three laboratories with agreed frame times; the model's flame start and end times within 0.5 s of the engineers' in a published share of videos; a table of how far the recorded hand times differ from the frame times.