Industries / Mining and materials / Steel and metals
Quality and testing
Quality and testing checks that metal has the chemistry, inner structure and strength it is sold as.
4
Milestones achieved, from 1863 to 1941.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1863
Steel under the microscope
Henry Clifton Sorby of Sheffield ground, polished and etched steel in acid and examined it under a microscope by reflected light, starting in summer 1863, and showed etched specimens at the British Association meeting in Newcastle that autumn. His full account was not published until 1887. The study of metal structure this way is called metallography.
Credit
Henry Clifton Sorby
1900
Ball hardness test
Johan August Brinell, chief engineer at the Fagersta ironworks in Sweden, first showed his test at the 1900 Paris Exposition. A 10 mm steel ball is pressed into the metal under a heavy load, and hardness is worked out from the load and the size of the dent. It is still one of the most widely used hardness tests.
Credit
Johan August Brinell
1940
Ultrasonic flaw detection
Floyd Firestone filed a patent on 27 May 1940, granted on 21 April 1942, for sending high-frequency sound into a part and timing the echoes. A hidden crack sends an echo back sooner than the far end of the part.
Credit
Floyd A. Firestone
1941
Liberty ship fractures
Some of the 2,710 Liberty ships built in the United States between 1941 and 1945 suffered serious brittle fractures, and a few broke in half. At Cambridge, Constance Tipper and Professor Baker later wrote about the ductile-to-brittle transition, the temperature below which steel that would bend snaps instead, and about how welds make brittle fracture more likely. Design codes now set Charpy impact test requirements, which measure the energy to break a notched bar, where low temperatures are expected.
Lessontoughness must be tested at the temperature the steel will meet in service.
Credit
Constance Tipper
Watch for
AI reads microstructures
NISCO, working with Huawei, says an AI model now checks metallographic samples with over 95% accuracy and has cut time per sample from 10 minutes to 3 to 5 minutes. No test method, sample count or measurer is given.
Achieved whena blind comparison against qualified metallographers on the same samples is published.
Claimed by
NISCO and Huawei, October 2026
Open
01
Answer key for mill certificates
A public set of redacted mill certificates, the document a mill ships with each batch listing its chemistry and test results, each keyed by hand into an answer key: heat number, grade, standard, chemistry, yield and tensile strength, elongation and impact energy. Add a scoring script so reading and checking tools can be compared.
The briefCredit · unclaimed02
Surface defects from several mills
An openly licensed set of steel strip surface images from at least three mills, with defect classes named against a written catalogue and a labelled test set held back for scoring.
The briefCredit · unclaimed
Open 01 · the brief
Answer key for mill certificates
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of redacted mill certificates, the document a mill ships with each batch listing its chemistry and test results, each keyed by hand into an answer key: heat number, grade, standard, chemistry, yield and tensile strength, elongation and impact energy. Add a scoring script so reading and checking tools can be compared.
Why it is still open
Tools that read certificates are sold, but their accuracy is self-reported. One vendor's case study gives "96% extraction accuracy on chemical composition fields," with no named dataset or measurer.
Built with what exists
Redaction software; two independent people keying each certificate; a Python scoring script; current AI document readers as the first baselines (the Note "Seven more flagged" covers an AI reviewer of long documents).
First step
Ask fabricators and university laboratories for certificates they hold and may share after redaction, and draft the answer-key fields.
Done looks like
At least 300 certificates from at least 30 mills in at least three languages, double-keyed with disagreements resolved, and at least three tools scored field by field.
Open 02 · the brief
Surface defects from several mills
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An openly licensed set of steel strip surface images from at least three mills, with defect classes named against a written catalogue and a labelled test set held back for scoring.
Why it is still open
The largest public set, from Severstal, has 18,074 images in four numbered classes that follow the company's internal categories; what each class is has not been published, and the test labels are not public.
Built with what exists
Inspection cameras mills already run; open labelling tools; Hugging Face for hosting.
First step
Write the defect catalogue with example photos and approach the quality departments of five mills.
Done looks like
At least 10,000 images from three or more mills, at least six named classes, a permissive licence, held-out test labels, and a baseline that trains on two mills and is scored on the third.