Industries / Industrial equipment and services / Industrial services / Non-destructive testing
Magnetic and dye methods
Magnetic particle testing finds cracks at or just under the surface of iron and steel, where fine iron powder gathers over a crack that disturbs a magnetic field; penetrant testing finds cracks open to the surface of almost any solid, where a dye soaked into the crack seeps back out and shows against a white coat or glows under ultraviolet light.
4
Milestones achieved, from 1929 to 2015.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1929
Current runs through the part
Railway workshops once found cracks with the "oil and whiting" test: parts were soaked in thinned oil, wiped, coated with chalk and struck with a hammer, so that oil seeped out of any crack and stained the chalk. In magnetic particle testing, steel is magnetized and covered with fine iron powder, which gathers where a crack bends the field out of the surface; William Hoke had described the principle in a 1922 patent. A. V. de Forest and C. E. Betz of Magnaflux wrote in 1945 that the method in use "actually dates from 1929", when two changes were made: circular magnetization, in which electric current is passed through the part so that the field runs around it and shows cracks lying along it, and powders made for inspection.
Credit
Alfred V. de Forest and Foster B. Doane, the co-founders of Magnaflux, whom the company credits with developing the method in the early 1930s
1941
Penetrants that glow
Robert C. Switzer's US patent 2,259,400, "Flaw Detection", issued on 14 October 1941, first disclosed and claimed fluorescent penetrant inspection: a liquid holding a fluorescent dye soaks into surface cracks, the surface is cleaned, and under ultraviolet "black light" the dye left in a crack glows. A 1960 patent by Joseph L. Switzer and Donald W. Parker, assigned to Switzer Brothers, calls it the basic patent and says the method was adopted by American industry at about the time it issued. Magnaflux, working with the Switzer brothers of Cleveland, introduced it as Zyglo in 1942.
Credit
Robert C. Switzer; Magnaflux
1996
A visible crack is missed
On 6 July 1996 the fan hub in the left engine of Delta Air Lines Flight 1288, an MD-88, broke apart at the start of its takeoff roll at Pensacola, Florida, and debris entering the cabin killed two passengers. The crack had grown from metal damaged when the hub was drilled during manufacture. The National Transportation Safety Board (NTSB) found that at the hub's last fluorescent penetrant inspection at Delta the crack was about 0.46 inch (12 mm) long on the surface, well above the 0.10 inch (2.5 mm) minimum detectable length in a published reliability handbook, and blamed the failure of Delta's inspection process to detect it. It asked for uniform rules on drying parts after cleaning, on covering every surface with developer powder, on tracking development times, and on fixtures that do not hide surfaces. An FAA team formed on 12 September 1996 reviewed six penetrant shops for engine parts and found "poor quality assurance practices at most" of them; in 2001 Iowa State University's Center for Aviation Systems Reliability began FAA-funded tests of each penetrant step with airlines and engine makers.
LessonA penetrant inspection is only as good as the cleaning, drying and developing done before the inspector looks.
Credit
a failure; investigated by the NTSB; follow-up by the FAA
2015
LED lamps get a standard
Fluorescent penetrant and fluorescent magnetic particle inspections need ultraviolet A (UV-A) light to make the dye glow. In 2015 ASTM International published the first edition of E3022, a practice for measuring and certifying light-emitting diode (LED) UV-A lamps for these inspections; it does not apply to mercury vapor and other discharge lamps. Makers must test each lamp design and each lamp they ship, and the standard bans pulse-width modulation and pulse firing, ways of switching LEDs rapidly on and off, because radiometers may not measure such lamps correctly.
Credit
ASTM Committee E07 on nondestructive testing, Subcommittee E07.03
Watch for
AI reads magnetic particle indications
KARL DEUTSCH, a German maker of testing equipment, says its CRACKVIEW AI photographs parts after magnetic particle testing and uses a neural network it trained to mark likely crack indications, so that parts are sorted as OK or not OK "without a human inspector having to view them". It describes running the system beside an experienced tester until it reaches the wanted accuracy and then letting it take over the decision, and says training usually needs several thousand images.
Achieved whena user or an outside body publishes how many real cracks the system missed and how many sound parts it rejected, against human inspectors on the same parts.
Claimed by
KARL DEUTSCH, in a product leaflet dated March 2026 (an earlier leaflet is dated July 2024)
Open
01
Penetrant photos with measured cracks
A public set of photos of fluorescent penetrant indications on metal test pieces with real fatigue cracks, each crack's length measured under a microscope, taken under UV-A light with a phone and with a machine-vision camera. Add a baseline model that marks indications and a script that computes, for the model and for human inspectors looking at the same photos, the share of cracks of each length that are found.
The briefCredit · unclaimed02
UV lamps tested in public
An independent test of UV-A LED lamps sold for penetrant and magnetic particle inspection against the limits ASTM E3022 sets for makers: a peak wavelength of 365 nm, plus or minus 5 nm; current ripple of no more than 5%; no pulsed LEDs; and controlled visible light, which washes out the glow of an indication. Publish the results lamp by lamp.
The briefCredit · unclaimed
Open 01 · the brief
Penetrant photos with measured cracks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of photos of fluorescent penetrant indications on metal test pieces with real fatigue cracks, each crack's length measured under a microscope, taken under UV-A light with a phone and with a machine-vision camera. Add a baseline model that marks indications and a script that computes, for the model and for human inspectors looking at the same photos, the share of cracks of each length that are found.
Why it is still open
Groups have trained models on penetrant images they made themselves: on car steering knuckles at Donghua University (2022), whose data is available only on request, and on aircraft wheel tie bolts at Universiti Tun Hussein Onn Malaysia (2026), which reports no accuracy figures. Inspection Tech Ltd reported a fully automated dye inspection system in 2023, and the FAA-funded program at Iowa State recorded indications digitally on blocks with cracks 0.021 to 0.118 inch (0.5 to 3 mm) long. A developer's MIT-licensed project on GitHub trained a YOLO11 model to mark penetrant indications, with one "defect" class, and reports 0.79 mean average precision on 836 validation images; it gives no crack sizes, and its trained weights sit behind a gated Hugging Face page. Searches in November 2026 found no public penetrant image set with measured crack sizes.
frontiersin.orgperiodical.uthm.edu.mybindt.orgcnde.iastate.edugithub.com
Built with what exists
Low-cycle fatigue crack specimens in nickel alloy, titanium and steel from a specimen supplier or a university fatigue lab; a penetrant line run by a technician certified to ISO 9712 Level 2; a UV-A LED lamp certified to ASTM E3022 and a UV-A radiometer; a phone and a machine-vision camera with a UV-blocking lens filter; Meta's Segment Anything Model 2 to speed outlining; Ultralytics YOLO11 segmentation as the baseline.
First step
Ask a university fatigue lab for ten cracked specimens it has finished testing, and photograph them after one penetrant run at a local NDT training school.
Done looks like
At least 1,000 photos of at least 100 cracks between 0.5 and 5 mm long released under an open license, with the crack length found 90% of the time published for the model and for two Level 2 inspectors who judge the same photos.
Open 02 · the brief
UV lamps tested in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent test of UV-A LED lamps sold for penetrant and magnetic particle inspection against the limits ASTM E3022 sets for makers: a peak wavelength of 365 nm, plus or minus 5 nm; current ripple of no more than 5%; no pulsed LEDs; and controlled visible light, which washes out the glow of an indication. Publish the results lamp by lamp.
Why it is still open
E3022 is written for makers to certify their own lamps, and Magnaflux tells users it is "not intended to be referenced by UV lamp users". A 2019 Quality Magazine article warns that "several lamps are falsely marketed as compliant" but names none and gives no count. Searches in November 2026 found makers' own certificates and no independent test of lamps on the market.
Built with what exists
A compact spectrometer such as a Hamamatsu C12880MA board or an Ocean Insight unit; a calibrated UV-A radiometer such as the Spectroline DSE-100X used in the FAA-funded penetrant studies; a fast photodiode and an oscilloscope to see pulsing and flicker; a dark box; Python; UV-blocking safety glasses.
First step
Buy three lamps sold as E3022-compliant and two general-purpose 365 nm flashlights, and measure their peak wavelength and flicker.
Done looks like
At least 20 lamp models bought at retail, each measured for peak wavelength, irradiance at 15 inches (381 mm), visible light and flicker, with all raw data published and each maker sent its results before publication.