Industries / Industrial equipment and services / Industrial machinery / Bearings
Steel and materials
This work chooses, cleans and hardens the steel or ceramic that rings, balls and rollers are made from, so that they survive rolling under heavy load for years.
6
Milestones achieved, from 1905 to 1983.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1905
One steel for a century
A steel with about 1% carbon and 1.5% chromium, sold under names such as 52100, 100Cr6, SUJ-2 and GCr15, is still the main bearing steel. H. K. D. H. Bhadeshia's 2012 review says Richard Stribeck's ball bearing tests of 1901 showed it suited the job, that Fichtel & Sachs of Schweinfurt adopted it for bearings in 1905, and that it makes up most of the roughly six million tonnes of bearing steel made each year; its fatigue life has improved since mainly through cleaner steel. Erwin Zaretsky of NASA writes that a French bearing steel whose analysis was published in 1907 closely matches 52100, and that the American AISI grade was first specified about 1920.
Credit
Fichtel & Sachs
1960s
Bearings grew on the shelf
Hardening steel means heating it until its crystals take a high-temperature form called austenite, then quenching it so that most of the austenite turns into hard martensite; some can be left over, called retained austenite. Zaretsky reports that in the early 1960s a major US aircraft engine company had to throw away unused bearings of M50 steel because their bores had grown larger than the engine shaft they were made for: retained austenite had turned into martensite and bainite while they sat at room temperature. For most critical aerospace bearings, retained austenite has since been held to 2 to 5%.
LessonA bearing can change size without ever running if its steel is still changing inside.
Credit
E. N. Bamberger of General Electric, whose account Zaretsky quotes from a 1963 conversation
1970s
Steel melted twice in vacuum
Hard oxide particles left in steel, called inclusions, act as starting points for fatigue cracks, Zaretsky writes. Vacuum melting came into steelmaking in the 1950s, and in the early 1970s the two vacuum methods were combined: an ingot made by vacuum induction melting is remelted by an electric arc under vacuum (VIM-VAR). For M50 steel, figures in Bhadeshia's review put oxygen at 67 parts per million by weight after air melting and 6 after VIM-VAR. The Society of Tribologists and Lubrication Engineers (STLE) gives VIM-VAR steel a life factor of 6 against 1 for air-melted steel, and Zaretsky calls vacuum processing the single most important variable that has increased bearing life.
LessonElectroslag remelted steel keeps tens of parts per million of oxygen yet can resist fatigue better than vacuum-degassed steel, so the size and spread of the largest inclusions matter more than total oxygen, as Bhadeshia notes.
Credit
rolling-contact rig tests by E. N. Bamberger and J. C. Clark (1982); STLE life factors by W. J. Anderson, E. N. Bamberger, W. E. Poole, R. L. Thom and E. V. Zaretsky (1992)
1972
More alloy, shorter life
Through the 1950s and 1960s the bearing industry assumed that steels with more alloying elements, which stay hard when hot, would also last longer. R. J. Parker and Erwin Zaretsky at NASA's Lewis Research Center ran 720 tests of eight through-hardened steels in the NASA five-ball fatigue tester, all made from single heats and hardened alike, and found that rolling fatigue life fell as total alloy content rose; tests of 120 bearings at 316 °C agreed. Zaretsky writes that M50 became the steel of choice above 149 °C by the mid-1980s, while 52100, the leanest alloy, lasts longest below that.
LessonTension, compression and rotating-beam tests could not predict rolling fatigue life, so only rolling tests could settle the choice of steel.
Credit
R. J. Parker and Erwin V. Zaretsky, NASA Lewis Research Center
1974
Ceramic balls pass fatigue tests
Parker and Zaretsky tested balls of hot-pressed silicon nitride, a ceramic, in the five-ball tester and reported that, running against steel, they gave fatigue lives comparable to or greater than steel against steel, and far longer than any other ceramic or cermet tested. Their computer analysis found no life gain in 120 mm bore bearings at very high speeds from swapping steel balls for ceramic ones. Bearings with ceramic balls and steel rings, called hybrid bearings, later went into the high-pressure oxidizer turbopump of the Space Shuttle main engine, where, a 2004 NASA report says, they solved "a highly persistent and troublesome bearing problem". Bhadeshia notes that silicon nitride is stiffer than steel, so hybrid races carry higher contact stress and need harder steel.
LessonTurbine engines and helicopter gearboxes detect failing bearings by the magnetic debris in their oil, and silicon nitride debris is not magnetic, so NASA's Paula Dempsey and colleagues wrote that a new sensing system had to be developed before ceramic parts could be used safely in aerospace.
Credit
R. J. Parker and Erwin V. Zaretsky, NASA
1983
A tough core for engines
A ring hardened all the way through is brittle: a tight fit on a shaft stretches the inner ring and can crack it right through, Zaretsky writes. Zaretsky credits E. N. Bamberger of General Electric, citing his 1983 paper, with lowering the carbon and raising the nickel in M50 to make M50 NiL, a steel that is case-carburized: carbon is diffused into the surface before hardening, which leaves a hard skin over a tough core. Zaretsky gives its core fracture toughness, a measure of how large a crack a steel can carry before it breaks, as over 60 MPa·m½ against 29 for M50, and says carburizing leaves compressive stresses above 210 MPa where contact stresses peak. Above 149 °C, he writes, M50 NiL is the carburized steel of choice.
Credit
E. N. Bamberger, General Electric
Watch for
A steel for engine bearings
SKF says ARCTIC15, a patented temperature-resistant, corrosion-tolerant steel for aircraft engine bearings, has a higher load and temperature capacity than existing steels and allows smaller bearings. It was developed over nearly a decade with EU Clean Sky funding, and SKF says demonstrator and full-scale bearings combining it with ceramic balls have passed endurance tests, including loss of oil supply; it gives no figures. A ground test with an unnamed engine maker was planned for early 2026.
Achieved whenan engine maker or test body outside SKF publishes results from the ground test, or the steel is approved for a certified engine.
Claimed by
SKF, November 2025
Bearings from hydrogen-made iron
SKF and voestalpine Wire Technology made a prototype spherical roller bearing from steel containing hydrogen direct reduced iron (H-DRI), iron made from ore with hydrogen, which Machine Design calls a low-emission alternative to conventional steelmaking. SKF's chief technology officer, Annika Ölme, said: "We have data that shows that it works just as well as any other steel." No figures were given.
Achieved whenfatigue life and cleanliness data for H-DRI bearing steel are published and checked by a laboratory outside SKF and voestalpine.
Claimed by
SKF, September 2025
Open
01
Inclusion ratings, tested in public
An open set of microscope images of polished bearing steel sections in which every inclusion is outlined and rated by type and size by at least two metallographers working separately, with a scorecard on which any automatic rater, commercial or open, can be tested.
The briefCredit · unclaimed02
Oxygen, inclusions and life, pooled
An open table of published rolling fatigue tests on 52100-type and M50 steels that report the steel's total oxygen, melting route or largest inclusion size, with a model that predicts relative life from them, tested by leaving out one study at a time.
The briefCredit · unclaimed
Open 01 · the brief
Inclusion ratings, tested in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of microscope images of polished bearing steel sections in which every inclusion is outlined and rated by type and size by at least two metallographers working separately, with a scorecard on which any automatic rater, commercial or open, can be tested.
Why it is still open
Zhu and colleagues (Micromachines, 2023) write that inclusions are still mainly rated by hand, which is slow, error-prone and inconsistent between raters. They trained a DeepLabv3+ model on 1,864 images labeled by experts under ASTM's test methods for inclusion content and reported 90.35% correct classification, cutting rating time for a sample from 30 minutes to 15 seconds, but their data availability statement reads "Not Applicable". A 2026 study of cast 52100 rings that used 2,400 surface images has only the paper in its Zenodo record. Searches of GitHub, Hugging Face and Zenodo in November 2026 found one undocumented classification repository and no labeled image set of inclusions in bearing steel.
Built with what exists
Polished sections of bearing steel bar from several heats; a light microscope with a motorized stage; Fiji (ImageJ) for stitching; CVAT or LabelMe for outlining; segmentation models such as DeepLabv3+ or a U-Net trained on one graphics card; images published under a CC BY license.
First step
Write rating rules from the chart method, have two people rate the same 100 fields from one sample, and measure how often they agree.
Done looks like
At least 2,000 microscope fields from at least 20 heats of steel, each rated by two people, with their agreement reported for each inclusion type; at least two automatic raters scored against the people, with precision and recall for each type and the worst-field rating of each sample compared.
Open 02 · the brief
Oxygen, inclusions and life, pooled
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of published rolling fatigue tests on 52100-type and M50 steels that report the steel's total oxygen, melting route or largest inclusion size, with a model that predicts relative life from them, tested by leaving out one study at a time.
Why it is still open
Bhadeshia's 2012 review says fatigue life correlates with total oxygen but with noise, because the length and spread of inclusions matter, and electroslag remelted steel with tens of parts per million of oxygen can outlast vacuum-degassed steel. Bearing life calculations still use one multiplier per melting route, such as STLE's 1.5 for vacuum degassing and 6 for VIM-VAR. Searches of Zenodo and GitHub in November 2026 found no pooled public data set of bearing steel cleanliness and fatigue life.
Built with what exists
Published tables and charts, digitized with WebPlotDigitizer; NASA technical reports, which are free to download; Python with pandas and SciPy for Weibull fits (the statistical model used for bearing lives); an AI document reader to find and tabulate studies, every value checked by hand against the paper (the Note "Seven more flagged" covers an AI reviewer of long documents).
First step
Digitize Zaretsky's 2012 chart of M50 life by melting route and the life data in NASA's five-ball test reports, and enter them as the first rows.
Done looks like
At least 40 test series from at least 8 independent sources in one open file with a link for every row; for each study left out in turn, the model predicts its relative life within a factor of two, with all errors published beside those of the fixed STLE multipliers.