BRENNEN

Industries / Industrial equipment and services / Industrial machinery / Fasteners

Inspection and sorting

Inspection and sorting checks that fasteners are what their markings and papers say, first by laboratory tests on a sample from each lot and now often by running every part past a machine that throws out the bad ones.

6

Milestones achieved, from 1988 to 2003.

1

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1988

    Congress finds bogus bolts

    After two years of work, a House subcommittee led by John Dingell wrote a 152-page report that found substandard and counterfeit nuts, bolts and screws, many with false maker's marks, in Army vehicles, NASA shuttle equipment, Navy nuclear submarines and nuclear power plants. It counted 61 general aviation accidents in three years that involved failed fasteners, and in December 1987 an ironworker had died at General Motors' Saturn plant in Tennessee while tightening a substandard bolt.

    deseret.com

    Credit

    Subcommittee on Oversight and Investigations, House Committee on Energy and Commerce

  2. 1990

    A law to test lots

    The Fastener Quality Act, signed on 16 November 1990 as Public Law 101-592, required high-strength fasteners to be tested by accredited laboratories, with records kept. Its rules never took effect, which a 1999 Senate report put down "mainly" to "the lack of a sufficient number of accredited laboratories".

    govinfo.govgovinfo.gov

    Credit

    US Congress; the National Institute of Standards and Technology (NIST), which ran the laboratory accreditation

  3. 1990

    Wrong bolts, same look

    On 10 June 1990 the left windscreen of a British Airways BAC One-Eleven blew out over Didcot, England, as the aircraft climbed through 17,300 ft; the captain was pulled halfway out and held by cabin crew until the co-pilot landed at Southampton. The windscreen had been fitted 27 hours earlier: 84 of its 90 bolts were about 0.026 in (0.66 mm) thinner than specified, and the other 6 were 0.1 in (2.5 mm) too short. The job was not listed as one needing a second inspector, so nobody else checked it.

    LessonBolts that differ by a fraction of a millimetre look the same, so safety-critical fitting needs an independent check.

    gov.ukassets.publishing.service.gov.uk

    Credit

    Air Accidents Investigation Branch (AAIB), report 1/1992

  4. 1999

    Congress narrows the law

    With the 1990 act still unworkable after nine years, the Fastener Quality Act Amendments Act (Public Law 106-34, 8 June 1999) cut it back. By the Senate report's account of the bill, it exempted fasteners made under certified quality systems such as ISO 9000, QS-9000 and AS9000, and fasteners approved by the Federal Aviation Administration; the General Accounting Office (GAO) gives its exemption for spare and replacement parts as packages of 75 or fewer.

    LessonThe law demanded an outside laboratory test for every lot when there were not enough laboratories, so the burden moved to factories' own audited quality systems.

    govinfo.govgovinfo.gov

    Credit

    US Congress

  5. 2001

    Customs tests imported bolts

    The US Customs Service laboratory in Chicago tests graded bolts pulled from import shipments. In a 1998 study, 3 of 77 shipments failed tensile or hardness tests, and Customs estimated that about 5% of 3,097 shipments of that bolt type were substandard; in March and April 2001, none of 66 randomly chosen shipments failed. The GAO reported both studies and noted sampling problems in the first.

    govinfo.gov

    Credit

    US Customs Service; GAO

  6. 2003

    Machines sort every part

    Plants moved from testing a sample of each lot to running every part past cameras, lasers and eddy-current coils, which sense a part's hardness from how the steel responds to a magnetic field. Quality Magazine reported in 2003 that MNP, a fastener supplier in Utica, Michigan, compared sorting machines side by side and then placed its largest single order for sorters, for General Inspection's USS-1000 laser sorter. GEFRA of Germany dates its OPTISORT sorters to 1995, and by 2015 fitted them with up to 14 cameras and a hardness check built into the line.

    qualitymag.comfastenerandfixing.com

    Credit

    GEFRA; General Inspection; MNP

Watch for

  1. Open models check screw heads

    Researchers at the University of Piemonte Orientale and engineers at Dimac, an Italian maker of sorting machines, tested ResNet-50, a freely available image model, on 2,095 screw-head images taken on Dimac machines, 621 of them defective. It classed them as good or defective with 95.5% accuracy (±1.0) in ten-fold cross-validation, a standard test that holds back part of the data each round. The images stay private under non-disclosure agreements; the code is public.

    Achieved whensomeone outside the study gets the same result on images others can inspect, or a plant counts the bad parts such a model lets through in production.

    iris.uniupo.it

    Claimed by

    Santomauro and colleagues, accepted by The International Journal of Advanced Manufacturing Technology in August 2026

Open

  1. 01

    Read the head marks

    A phone app with an open model that photographs a bolt head, reads the grade mark (radial lines for SAE grades, numbers such as 8.8 or 10.9 for metric property classes) and the maker's mark, and looks the maker up in the US Patent and Trademark Office's list of registered head insignia.

    The brief
    Credit · unclaimed
  2. 02

    A test lot for sorters

    A standard test lot that any plant can run through its sorting machine: 10,000 good bolts plus a known number of seeded bad ones of documented types (heads notched to imitate cracks, parts too long or too short, missing threads, a wrong grade mixed in). The plant counts how many seeded parts land in the reject bin, and the escapes are published by machine and by defect type.

    The brief
    Credit · unclaimed
  3. 03

    Cracks confirmed by cutting

    An open set of images and eddy-current readings of cold-formed bolt heads, each labelled cracked or sound only after a magnetic particle check or after it is cut open and polished, with a baseline model trained on it.

    The brief
    Credit · unclaimed

Open 01 · the brief

Read the head marks

Credit · unclaimed

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

What to build

A phone app with an open model that photographs a bolt head, reads the grade mark (radial lines for SAE grades, numbers such as 8.8 or 10.9 for metric property classes) and the maker's mark, and looks the maker up in the US Patent and Trademark Office's list of registered head insignia.

Why it is still open

Portland Bolt, a US bolt maker, says the Patent and Trademark Office "has a long list of registered manufacturer head insignia", yet its staff often answer readers' photos with "we are unable to find that head marking in our literature". Roboflow sells grade-mark reading for factory lines, trained on each customer's own fasteners, with no public data or accuracy figure. A search found no public photo set labelled with grade and maker marks and no app that reads them.

portlandbolt.comroboflow.com

Built with what exists

A phone with a clip-on macro lens; an open detection model such as Ultralytics YOLO; a vision-language model to read worn marks, with each reading checked by a person; the Patent and Trademark Office insignia list; Python.

  1. First step

    Buy 100 bolts of mixed grades from three hardware stores and two online sellers, photograph each head under three lights, and match each mark to the list by hand.

  2. Done looks like

    3,000 photos of at least 150 different head marks released under an open license, and a model that names grade and maker correctly on at least 95% of 500 photos of bolts it has never seen, with every miss listed.

Open 02 · the brief

A test lot for sorters

Credit · unclaimed

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

What to build

A standard test lot that any plant can run through its sorting machine: 10,000 good bolts plus a known number of seeded bad ones of documented types (heads notched to imitate cracks, parts too long or too short, missing threads, a wrong grade mixed in). The plant counts how many seeded parts land in the reject bin, and the escapes are published by machine and by defect type.

Why it is still open

Sorter makers publish no escape rates; Dimac's January 2025 "zero-defect commitment" for its ZED machines gives no defect figures. The one published side-by-side test found, by MNP in 2003, judged sorters on repeatability, ease of use, sorting rate and cost, not on bad parts let through. A search found no public escape-rate test.

qualitymag.comfastenerandfixing.com

Built with what exists

Commercial M8×30 property class 8.8 bolts; a machine shop's wire EDM (electrical discharge machining, which cuts metal with sparks) or a thin abrasive saw to cut controlled notches; calipers and a hardness tester to document each seeded part; volunteer plants or sorting-service companies to run their machines.

  1. First step

    Make 40 seeded parts of four types, measure and photograph each, and ask two sorting-service companies to run them.

  2. Done looks like

    The lot run three times on each of at least five sorters from at least three makers, with escapes published by machine and defect type and a second lab confirming what each seeded part is.

Open 03 · the brief

Cracks confirmed by cutting

Credit · unclaimed

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

What to build

An open set of images and eddy-current readings of cold-formed bolt heads, each labelled cracked or sound only after a magnetic particle check or after it is cut open and polished, with a baseline model trained on it.

Why it is still open

Public screw image sets label parts good or bad by eye; Airlangga University's M2 screw set (5,536 images, CC BY 4.0, June 2026) lists no defect types, and the 2026 Piemonte Orientale–Dimac study kept its images private. A search found no public set labelled for confirmed cracks in fastener heads.

data.mendeley.comiris.uniupo.it

Built with what exists

A turntable with a glass disk and an industrial camera, or time on a plant's sorter; an eddy-current crack probe; a magnetic particle kit; a university lab's saw and polisher; ResNet-50 and EfficientNet-B0, trained with the Piemonte Orientale study's public code.

  1. First step

    Ask a fastener plant for one bin of parts its sorter rejected for cracks and one bin it passed, and check 50 from each by magnetic particle.

  2. Done looks like

    1,000 heads, at least 150 with confirmed cracks, released under an open license with a confirmation photo for each crack, and published detection and false-alarm rates for a baseline model on a held-out fifth of the set.