Industries / Industrial equipment and services / Industrial machinery / Bearings
Testing
Testing is how makers and laboratories check bearings before and after sale: running batches until they fail to measure fatigue life, spinning each finished bearing to grade its vibration, and naming the damage on bearings that failed.
7
Milestones achieved, from 1901 to 2013.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1901
Ball bearings put to test
Richard Stribeck measured how much load hardened steel balls could carry and how much friction ball bearings had. He began the tests at the technical university in Dresden in 1894, continued them after leaving Dresden in 1898, and published "Kugellager für beliebige Belastungen" (ball bearings for any loads) in the journal of the German engineers' association, VDI, in 1901. His entry in the Neue Deutsche Biographie says the work gave engineers formulas they could use and confirmed and extended Heinrich Hertz's 1881 theory of how elastic bodies press on each other. In 2003 Bo Jacobson of Lund University wrote that Stribeck's method for sharing the load among the balls is still used, and that his measured friction matches the coefficient of 0.0015 that SKF's 1989 catalogue gave for deep-groove ball bearings.
Credit
Richard Stribeck
1956
Four makers pool their tests
A bearing's life falls steeply as its load rises, and makers disagreed whether life falls with the cube of the load or with its fourth power. Several of the larger makers pooled their fatigue test records through the American Standards Association, which asked the National Bureau of Standards to analyse them. J. Lieblein and M. Zelen fitted 213 test groups, 4,948 deep-groove ball bearings from four companies, and found the data consistent with the cube. They printed every group's summary in their paper. In 2003 a NASA team compared computer-simulated tests with 51 sets of 5,321 real bearings and found that lives calculated with the fourth power for ball bearings agreed with the data.
Credit
J. Lieblein and M. Zelen, National Bureau of Standards; the American Standards Association's ball and roller bearing committee
1959
Five balls stand in
Running whole bearings until they fail from fatigue takes months. W. J. Anderson of NASA's Lewis laboratory conceived a small tester built around five balls in late 1958, and its first data came in March 1959. By 1981 eight of these rigs, run 24 hours a day, had logged about 500,000 test hours on how contact angle, steel, hardness, heat treatment, lubricants, temperature and vacuum affect fatigue life, and NASA reported that their results correlated with those of full-size bearings.
Credit
W. J. Anderson, E. V. Zaretsky and R. J. Parker, NASA Lewis Research Center
2003
Shortcut tests checked
Sudden death testing splits the bearings into groups, runs each group until its first bearing fails and stops the rest, to save rig time; Leonard G. Johnson described it in 1964 as one of three ways to shorten fatigue tests. In a NASA study, Brian Vlcek, Robert Hendricks and Erwin Zaretsky ran 30,960 simulated deep-groove ball bearings through 33 sudden death test plans. The spread in the estimated life depended on how many bearings failed, not on how many were tested. Sudden death testing saved no more time than running bearings to their expected median life and stopping after a set number of failures, which cut test time and calendar time by up to 40%.
LessonCertainty in a life test comes from the number of failures, so a shortcut that yields fewer failures buys less certainty.
Credit
Brian L. Vlcek, Robert C. Hendricks and Erwin V. Zaretsky; Leonard G. Johnson
2004
Vibration measured alike
How quietly a bearing runs depends on the form and finish of its rolling surfaces, whose flaws are measured in nanometres. Makers therefore spin each finished bearing on a quiet spindle and measure how fast its outer ring moves. S. J. Lacey of Schaeffler UK describes the common test: radial velocity in three bands, 50 to 300, 300 to 1,800 and 1,800 to 10,000 Hz, with a limit in each. ISO 15242-1, first published in May 2004 and revised in 2015, set out the measuring methods and how to calibrate the measuring systems. Testers such as Sugawara Laboratories' Anderon meter turn the inner ring at 1,800 rpm and report each band in anderons, a unit of which 0.1 is about 0.77 micrometres per second.
Credit
ISO technical committee 4, rolling bearings
2004
Damage given shared names
ISO 15243, first published in February 2004, sorts the ways bearings fail in service into modes and gives each one's appearance and likely causes, with photographs, so that people examining a failed bearing use the same names. The 2017 edition limits itself to features that can be seen without cutting the bearing, and says further investigation is needed when the cause cannot be judged from appearance.
Credit
ISO technical committee 4, rolling bearings
2013
Rigs sized for wind bearings
Wind turbine bearings can be metres across and carry the shifting loads of a rotor, too large for ordinary test stands. Schaeffler built Astraios, a 350-ton rig about 16 m long that loads bearings up to about 3.5 m in diameter through eight load cylinders, built over two years for about €7 million; the trade magazine Windpower Engineering, describing it in June 2013, called it the largest high-performance rig of its kind. In 2017 SKF commissioned two rigs at Schweinfurt, Germany, built by Renk Test System, one of which applies a bending moment of 40 MN·m (meganewton-metres), with about €3.5 million of Bavarian and German federal funding.
Credit
Schaeffler; SKF; Renk Test System
Watch for
Twenty years in eight months
The China Bearing Industry Association reported that R&D Test Systems had installed a 560-tonne load cell in what it calls the world's most powerful main bearing test rig, at the "Linde Offshore Renewable Energy Center" (LORC). The rig is to apply bending loads above 100 MN·m and axial thrust above 10 MN to bearings up to 3.5 m across and 15 tonnes, and the report says accelerated tests on it can reproduce 20 years of offshore loads in 6 to 8 months.
Achieved whenthe rig has run a main bearing through such a program and the damage found is published beside the damage on bearings of the same design from turbines in service.
Claimed by
R&D Test Systems, as reported by the China Bearing Industry Association, August 2025
Open
01
Every life test, tabulated
An open, machine-readable table of published bearing fatigue tests. One row per test group: bearing type and size, load, speed, lubricant, steel, how many were tested and how many failed, the measured L10 (the life that 90% of the group outlived), L50 and Weibull slope (a measure of how widely lives scatter), the life calculated for it, and the source page.
The briefCredit · unclaimed02
Vibration grade against life
A public test of whether a bearing's end-of-line vibration grade predicts its fatigue life. Measure at least 120 new small ball bearings in the three standard velocity bands, then run every one of them in an accelerated life rig to its first spall (a flake of metal lifting from a raceway) or a fixed cut-off time.
The briefCredit · unclaimed
Open 01 · the brief
Every life test, tabulated
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, machine-readable table of published bearing fatigue tests. One row per test group: bearing type and size, load, speed, lubricant, steel, how many were tested and how many failed, the measured L10 (the life that 90% of the group outlived), L50 and Weibull slope (a measure of how widely lives scatter), the life calculated for it, and the source page.
Why it is still open
The data exist on paper. The National Bureau of Standards printed summaries of 213 test groups, 4,948 bearings, in 1956, as tables in what is now a scanned paper, and NASA compared simulations with 51 sets of 5,321 bearings in 2003 without publishing them as a file. In November 2026 searches of Zenodo and Hugging Face for bearing fatigue life data returned nothing, and a web search found no compiled public table.
Built with what exists
The NASA Technical Reports Server and its API for the reports; pdftotext and Tesseract OCR for scanned tables; WebPlotDigitizer for Weibull plots; an AI model to draft rows from long reports, with every value checked by a person (the Note "Seven more flagged" covers an AI reviewer of long documents); Python with the reliability package for Weibull fits.
First step
Type the four appendix tables of the 1956 Bureau paper into a CSV file and reproduce its fitted exponent.
Done looks like
At least 300 test groups from at least 20 published sources, each row linked to its page; a second person checks 10% of rows and the disagreement rate is published; a chart of measured against calculated L10 by bearing type and decade.
Open 02 · the brief
Vibration grade against life
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of whether a bearing's end-of-line vibration grade predicts its fatigue life. Measure at least 120 new small ball bearings in the three standard velocity bands, then run every one of them in an accelerated life rig to its first spall (a flake of metal lifting from a raceway) or a fixed cut-off time.
Why it is still open
Makers grade every bearing by vibration, and S. J. Lacey of Schaeffler UK writes that waviness of the rolling surfaces "may affect bearing life", but no public data set ties a bearing's vibration grade to its life. Paderborn University showed the life test is within a university's reach: its rig ran 108 bearings of type 6203 and produced damage in 18 of them. The nearest prior work found, a 2009 patent of Chugoku Electric Power, relates vibration to remaining life for bearings in service with failing lubricant, not to grades measured at the factory.
Built with what exists
A quiet spindle turning at 1,800 rpm with a velocity sensor on the outer ring; a four-bearing life rig with a spring-loaded radial load and thin oil, after Paderborn's design, behind a guard; a USB data acquisition card; Python with SciPy and the reliability package; bearings of one size from at least three makers.
First step
Build the measuring spindle and measure 50 new 6203 bearings twice each to see how repeatable the readings are.
Done looks like
At least 120 bearings, each measured twice before testing to show repeatability, then run to spall or cut-off; the rank correlation between each band's reading and life, with its confidence interval; all readings and lives published.