Advanced Nondestructive Evaluation Techniques for Rail: What Is Field-Ready

Advanced nondestructive evaluation techniques for rail, such as phased array UT, EMAT, air-coupled UT, ACFM, guided waves and machine vision, add to the conventional multi-probe ultrasonic search but do not replace it. Under 49 CFR 213.237, Class 4 and 5 track still needs a valid internal search at least every 370 days or 30 MGT.

This page sorts the newer methods by how ready they really are: field-proven, limited use, pilot or research. For each one it says what the method finds, what evidence exists, and how it could fit a compliant testing program. The standard methods program (multi-probe UT, induction, eddy current basics, visual inspection) is covered in our guide to rail flaw detection methods, so it isn't repeated here.

What does 49 CFR 213.237 require, whatever the technology?

Section 213.237 is technology-neutral: it requires internal rail inspections, a valid search, a qualified operator and equipment that can see between joint bars, and it names no method. Any advanced technique is judged against those conditions, not against a list of approved devices. The word "ultrasonic" does not appear in Part 213; ultrasonic testing is industry practice, not a federal mandate.

The rule's core definition is in 213.237(j):

"Valid search means a continuous inspection for internal rail defects where the equipment performs as intended and equipment responses are interpreted by a qualified operator as defined in § 213.238." (49 CFR 213.237(j))

Five more conditions apply to any equipment used for the search (49 CFR 213.237 on eCFR):

Two related rules matter for new technology. 49 CFR 213.240 allows continuous (non-stop) testing in place of stop-and-verify, provided suspect locations are field-verified by a 213.238-qualified person within 84 hours (36 hours for suspected A, A2 or B defects), and the data is reviewed by a qualified interpreter. And 49 CFR 213.17 lets an owner petition FRA for a waiver "from any or all requirements prescribed in this part", filed under Part 211. Continuous testing itself was trialed under 213.17 waivers from 2009 before the 2020 rule wrote it into Part 213.

Key takeaway: a new technique earns a place in a 213.237 program by meeting the valid-search conditions, not by being newer.

How mature are advanced nondestructive evaluation techniques for rail?

Most advanced nondestructive evaluation techniques for rail are still at the pilot or research stage in North America; only eddy current RCF scanning and machine vision have broad field use, and neither is an internal inspection. The table uses four levels:

• Field-proven: in routine revenue-service use by several owners, with published field data
• Limited use: in service on some networks or for specific tasks
• Pilot: prototypes tested on a test track or in revenue service, not sold as routine
• Research: laboratory, rig or modeling work

The maturity column is a judgment from the published sources cited below. Vendors may describe products at a higher level; ask for test data before accepting a claim.

What can phased array UT add for rail welds and rail?

Phased array UT (PAUT) steers and focuses the beam electronically, which helps on worn rail, under shells and in welds, but in rail it remains a verification and research tool rather than routine test-car output. The physics difference from single-element UT is explained on our UT vs PAUT page.

Vehicle PAUT. TTCI built a PAUT rail inspection prototype that fits in place of a conventional roller search unit, and tested it at TTC in Pueblo, Colorado. Its Technology Digest TD-16-050 (Witte and Poudel, November 2016) reports detection of "transverse defects under shells, defects in welds, and web defects at speeds up to 20 mph", and states that the prototype "performed equivalent or better than a conventional UT inspection truck on every type of rail defect tested." One example was a gauge-corner TD of about 5.9% head area beneath a shell 5.8 in. long. The same digest is candid about the limit: PAUT "is still a head contacting inspection technique and so suffers the limitation of being unable to inspect the rail base flanges."

Steering also addresses a known weakness of fixed-angle probes. TTCI's TD-16-030 notes that on worn rail fixed probes can inspect "the wrong volume of material", and TD-16-050 says PAUT can "correct for rail profile wear". Those are test-track results; the digest itself said service-environment comparison was future work.

Manual PAUT on thermite welds. TTCI's TD21-027 (Poudel, September 2021) studied thermite welds after installation and found that existing practice is "limited to visual inspection". Full base coverage needed a "full skip" off the rail, and a 5 MHz, 32-element array performed best. That is a research technique, not field practice. For a method comparison on rail welds see PAUT vs TOFD for rail weld inspection.

Sizing caution. The weld PAUT standard ASTM E2700-26 warns: "Extreme caution should be used when attempting to size indications using phased arrays". A PAUT image looks precise; the length and height it shows still carry beam-width and calibration error.

Key takeaway: PAUT can see past shells and correct for wear on a test track, but it still cannot reach the base flanges from the head.

Is EMAT a practical non-contact UT for rail?

Electromagnetic acoustic transducers (EMATs) generate ultrasound inside the steel with a magnetic field and a coil, so they need no couplant, but their low efficiency has kept rail EMAT at the research and pilot stage for four decades. That gap between promise and practice is the main thing a buyer should know.

FRA studied rail EMAT as early as 1980. The final report Development of a Prototype EMAT System for Inspection of Rails (FRA/ORD-80/45, Rockwell International, September 1980) found that "virtually all of the critical rail flaws can be detected in the laboratory by EMAT's" with sensitivity comparable to conventional piezoelectric transducers (TRID record). The key words are "in the laboratory".

The reason it didn't become routine is signal strength. ASTM's EMAT guide, E1774-17(2022), describes EMAT as having "very low efficiency", with insertion losses of 40 dB or more, and notes that corroded surfaces can make it unsuitable. A rail head carrying RCF, rust and contamination is exactly that kind of surface. The transducer must also stay at a very small, constant gap at speed.

The most recent integrated attempt found was the EU RIFLEX project (TWI, 2014–2016), which used flexible EMAT coils on the rail head with an actuator holding the gap. It validated detection on reference rails with artificial defects on a rotating rig at up to 80 km/h (about 50 mph), and described the result as a prototype "ready for further development" (CORDIS project report). Its higher speed and coverage figures are project projections, not field results.

Key takeaway: EMAT removes the couplant but not the coupling problem; treat any rail EMAT offer as a pilot.

How far have laser and air-coupled UT come?

Air-coupled and laser UT for rail have reached instrumented field trials on test track, with one passive system tested at up to 80 mph, but none is yet a sold, routine rail flaw detection system. These methods send or receive ultrasound through air, so the probe never touches the rail.

The longest-running program is at the University of California, San Diego, funded by FRA. FRA's report On-line High-Speed Rail Defect Detection: Part II (DOT/FRA/ORD-12/02, March 2012) describes a prototype with non-contact transducers, guided waves and real-time statistical pattern recognition, field-tested seven times on FRA's hy-rail vehicle (ROSA P record). Its abstract reports blind-test reliability "exceeding industry average"; the report gives the test conditions, and the figure applies to those tests only.

Later papers moved to fully air-coupled designs:

• Mariani, Nguyen, Zhu and Lanza di Scalea, "Field Test Performance of Noncontact Ultrasonic Rail Inspection System", Journal of Transportation Engineering, Part A 143(5), 2017 (doi 10.1061/JTEPBS.0000026): two tests at TTC's Rail Defect Test Facility at 1 to 15 mph (abstract, as indexed).
• Datta and Lanza di Scalea, Experimental Mechanics 63:1043–1054, 2023 (doi 10.1007/s11340-023-00972-9): a passive system that records guided waves generated by the wheel-rail contact itself. The authors report TTC tests showing "possible test speeds as high as 80 mph" and evaluate performance with ROC curves across speed, sensor position and signal quality.
• Kenderian, Cerniglia, Djordjevic and Garcia reported a laser-air hybrid technique for rail in Insight 47 (2005) (doi 10.1784/insi.47.6.336.66454).

The passive 80 mph result matters because it hints at monitoring from revenue trains. It doesn't yet answer the 213.237 questions: can it see inside joint bars, how does it report untested rail, and who is the qualified operator?

Key takeaway: non-contact UT is the most promising pilot on this list, and still a pilot.

Can ACFM and eddy current measure RCF depth?

Eddy current can estimate the depth of surface rolling contact fatigue (RCF) cracks and is used on grinding scanners and inspection trolleys; ACFM gives a stronger depth signal but remains a research and trial method in rail. Neither looks inside the head, so neither is a 213.237 internal inspection. The defect itself is explained in our guide to rolling contact fatigue in rail.

FRA's RCF review, Rolling Contact Fatigue: A Comprehensive Review (DOT/FRA/ORD-11/24, Magel, November 2011, ROSA P), sets out why depth matters. Surface fatigue "prevents the successful transmission of ultrasonic waves to the base of the rail and back", producing a no-test that then needs hand testing. Conventional UT analysis also ignores "the first few millimeters" under the running surface. The review's conclusion is practical: "current best practice is to combine nondestructive evaluation with preventative rail grinding".

The two electromagnetic methods differ in how they report depth:

TD-14-007 says it plainly: the tested eddy current vehicle "relies on the user to enter an assumed crack angle, as the actual angle cannot be determined through nondestructive testing." Most RCF is shallow. TTCI's TD-18-027 (2018) sectioned rails removed after low-confidence UT and found 90% of 880 cracks and spalls were 0.040 in. deep or less. The real value of an RCF depth map is deciding how much metal to grind before the UT retest, as the worked example below shows. Grinding itself earns no 213.237 credit; see rail grinding and ultrasonic testing.

Key takeaway: use eddy current to size the grind, and UT to make the search.

Where does guided-wave monitoring fit?

Guided-wave (long-range) ultrasonics is in limited service use for detecting complete rail breaks between fixed stations, while detecting partial defects at range is still research. Guided wave and long-range UT are the same method under two names.

The best-documented installation is South Africa's heavy-haul Sishen–Saldanha line, where a permanently installed system "detects complete rail breaks at long-range using guided wave ultrasound" (Ramatlo, Loveday, Long and Wilke, MATEC Web of Conferences 347, 2021, doi 10.1051/matecconf/202134700034). The authors note that real damage signals in service are very hard to obtain, so they evaluate partial-defect detection by simulation.

Range figures need their mode of operation stated. In pulse-echo, Loveday, Long and Ramatlo detected and located an artificial defect at 370 m on operational heavy-haul track (Structural Health Monitoring 19(6), 2020, doi 10.1177/1475921719893887). Kilometer-scale figures come from break detection, where a transmitter and receiver simply check that the signal still arrives. Rail is a multi-mode waveguide, and distance must be worked out with the selected mode's group velocity, not the bulk velocity. Our page on long-range ultrasonic rail inspection works through that calculation.

Key takeaway: guided waves can tell you a rail has broken, not that it passed its internal search.

What do machine vision and track imaging systems actually inspect?

Machine vision records the visible surface of the track at speed (joint bars, fasteners, ties, switch parts and rail surface damage) and is in field use as an aid to the visual inspector, but it cannot see internal rail defects. Visual inspection is its own duty under 49 CFR 213.233, which says "mechanical, electrical, and other track inspection devices may be used to supplement visual inspection".

FRA's own program shows the realistic role. Its Research Results Machine Vision Data Products for the Automated Track Inspection Program (RR 21-20, Baillargeon and Stuart, September 2021, ROSA P) describes algorithms that flag possible defects and extract high-resolution images of switches, frogs and other areas for FRA inspectors to assess. The images inform a person; they don't replace the inspection. Earlier FRA and ENSCO work on video inspection of joint bars (Research Results RR06-03, 2006) found the crack algorithm still needed refinement against misses and false detections from ballast, vegetation, grease and mud.

Part 213 has exactly one rule that mandates an automated inspection system, and it is a useful model for judging any technology. 49 CFR 213.234 requires automated inspection of concrete-tie track for rail seat deterioration on certain Class 3–5 main track, "as a supplement to visual inspection". It sets a performance standard: accuracy within 1/8 in., a sampling interval of no more than 5 ft, documented calibration, an exception report, and field verification of each exception within 48 hours by a 213.7-designated person. No equivalent performance standard exists for internal rail inspection by imaging, because imaging can't do it.

Fine cracks are also harder to image than they look. ORD-11/24 cites Canadian work (NRCC, 2004) finding RCF cracks an order of magnitude harder for vision systems than spalls. Our computer vision track inspection page covers camera systems in more detail.

Key takeaway: machine vision supports 213.233 and, for concrete ties, satisfies 213.234; it has no role in 213.237.

Can AI classify rail defects automatically?

AI models can sort ultrasonic B-scan images and camera images by defect type with useful accuracy in published studies, but no rule lets software replace the qualified operator, and published accuracies apply only to each study's own data. Treat AI as an aid to the analyst.

The research is real. Hu, Li, Jing and Aela (International Journal of Steel Structures 25, 2025, doi 10.1007/s13296-024-00927-3) trained a two-stage classifier on rail UT B-scan images and reported 88.56% accuracy, analyzing one image sheet in 0.45 s. That figure means roughly 1 in 9 images classified wrongly on their test set. It is not a probability of detection, and it doesn't transfer to another system's probes, rail or traffic.

The regulatory structure puts a person at the decision point. Section 213.237(i) requires a qualified operator; 213.238 makes interpreting equipment responses an authorized duty of that operator. Under continuous testing, 213.240 requires the data to be reviewed by a qualified interpreter and every suspect to be field-verified, and 213.240(e)(4) says a suspect "is not considered a defect under § 213.113(c) until it has been field-verified by a person qualified under § 213.238."

Thresholds involve a trade-off. FRA's handheld UT reliability study (DOT/FRA/ORD-26/07, ENSCO, 2026, ROSA P) found that a lower detection threshold raises detection, and also that "too low a detection threshold can result in an unacceptable number of false positives". An automated classifier faces the same choice; ask the vendor where it is set and how misses were measured. False calls are covered on our false calls in rail UT page.

Key takeaway: AI can speed the review, but a 213.238-qualified person still owns the call.

Why are radiography and acoustic emission rare in rail?

Radiography (RT) and acoustic emission (AE) are niche in rail work: RT is impractical in track and used mainly in shops and research, and AE monitors active crack growth but cannot size a flaw or give a valid search. The original version of this page presented both as routine rail methods; they aren't.

Radiography. RT needs access to both sides of the part, a controlled radiation area and time, all hard to get on an operating railroad. Field radiography also has staffing rules: under 10 CFR 34.41, "Radiography may not be performed if only one qualified individual is present." Where RT does appear, it is before the part reaches track. FRA's RR 22-25 (2022) on special trackwork reports that manufacturers radiograph cast manganese crossing diamonds before shipment, while in-service examination is visual only. RT of thermite welds was studied in North America in 1970 (AREA Bulletin 71), with the conclusion that more study was needed on its effectiveness, and practice still varies by country.

Acoustic emission. AE sensors listen for stress waves released by a growing crack. In rail, every passing wheel is also a source of stress waves; the passive air-coupled system above uses that very signal. Bruzelius and Mba titled their 2004 paper "An initial investigation on the potential applicability of Acoustic Emission to rail track fault detection" (NDT&E International 37, doi 10.1016/j.ndteint.2004.02.001), and the field basis has not moved far beyond that. ASTM's AE standard for gas cylinders, E1419/E1419M-26, states the general principle: "Other nondestructive test (NDT) methods must be used to evaluate the significance of AE sources".

Key takeaway: RT and AE answer questions rail testing rarely asks in track.

Worked example: how would a new technique fit a 213.237 program?

The safe pattern is to add the new technique as a supplement that improves the valid search, never as a substitute for it. This example uses a hypothetical Class 4 freight segment with no scheduled passenger trains and no hazmat-route status.

Inputs:

• Segment: 40 miles, Class 4, 24 MGT a year
• Last valid UT search: day 0
• Service failures in the last 12 months: 2

Step 1: the deadline. 30 MGT at 24 MGT a year is reached after 30 ÷ 24 × 365 = 456 days. The calendar limit of 370 days comes first, so the next search is due by day 370 (213.237(c)).

Step 2: the failure-rate check. The Class 4–5 limit is 0.10 service failures per mile per year: 40 × 0.10 = 4 allowed. Two failures give 2 ÷ 40 = 0.05, inside the limit (213.237(a)).

Step 3: the problem. On the day-0 run, the qualified operator records 1.6 miles across three curves where RCF prevented a valid search (213.237(g)). That rail is not inspected. The owner must re-search it, run at 25 mph or replace it before day 370 (213.237(h)).

Step 4: where the new techniques come in.

Step 5: size the grind. The eddy current scan reports a maximum estimated depth of 0.045 in. at the default 25° crack angle. Allowing for the 33% upper error found in TD-14-007: 0.045 × 1.33 = 0.060 in. TTCI's TD20-002 found corrective grinding restored a valid UT test in 100% of cases where at least 0.060 in. was removed. At 0.006–0.008 in. a pass (a midpoint of 0.007 in.), that is 0.060 ÷ 0.007 = 8.6, so plan 9 passes.

Step 6: schedule. Grinding on day 40 and the UT re-search on day 55 leave 315 days of margin before day 370. If part of the re-search is still invalid, the 213.237(h) options apply to that part.

Step 7: a vendor offers to replace the annual search. Suppose a vendor proposes a passive air-coupled monitor on revenue trains instead of the annual UT run. Test it against the rule:

If any answer is no or unproven, the system can run as a supplement while the UT search continues. Using it in place of the search would need an FRA waiver under 213.17, filed under Part 211.

Key takeaway: new tools can make the 370-day search easier to pass; only a valid search satisfies it.

What are the limits of advanced rail NDE?

Every advanced technique on this page has a gap that the conventional search, grinding or a hand test still has to cover.

• Base flanges. Head-contact methods, including vehicle PAUT, can't inspect the base flanges (TD-16-050). Test cars miss most of the base, so field welds need hand UT of the base.
• Test-track results. TTC, FAST and rig results are under controlled conditions with known defects. Revenue service adds contamination, lubrication, wear and weather.
• Evidence is method-specific. A detection rate, accuracy or ROC curve applies to the system, defect set and conditions tested. Don't carry it to another product.
• Surface methods see the surface. Eddy current and ACFM size RCF; they say nothing about a transverse defect deeper in the head.
• Joints and special trackwork. Few emerging systems demonstrate coverage inside joint bars, which 213.237(f) requires. Cast manganese frogs are examined visually in service (RR 22-25).
• No rail-specific standard. No public rail-in-track standard or field POD was found for rail guided waves (search, October 2026), and the EMAT and air-coupled results above are rig or test-track data.
• People still decide. Every route to compliance runs through a 213.238-qualified person who judges validity and verifies suspects.

How EMA Industries applies this

EMA Industries performs ultrasonic rail flaw testing and rail weld inspection for transit agencies, short lines and industrial track owners, based in Southern California and mobilizing nationwide. When an owner is weighing a new inspection technology, the useful first step is to map it against the 213.237 conditions above and keep the valid search in place until the evidence is in. To scope a testing program or review a proposal, see our ultrasonic rail testing services or contact EMA Industries.

Related reading: 49 CFR 213.237 rail flaw detection requirements · Continuous vs stop-and-verify rail testing · Guided wave ultrasonic testing · Rail flaw detection methods

Frequently asked questions

Which advanced rail inspection techniques are actually in field use?

Eddy current RCF scanning and machine vision imaging are in field use, and guided-wave broken-rail detection is in service on at least one heavy-haul line. Vehicle PAUT, EMAT and air-coupled UT have reached prototypes and test-track trials. None of these replaces the multi-probe ultrasonic search used for 213.237.

Does FRA approve specific rail flaw detection technologies?

FRA doesn't publish a list of approved devices for 213.237. The rule sets conditions instead: a valid search, equipment that detects defects between joint bars, and a qualified operator under 213.238. An owner who wants to depart from those conditions can petition for a waiver under 213.17.

Can phased array UT replace conventional rail flaw testing?

Not on current evidence. TTCI's prototype matched or beat a conventional truck on test-track defects at up to 20 mph in 2016, but it couldn't inspect the base flanges and hadn't been compared in revenue service. For now PAUT serves verification and weld work.

Is eddy current testing a valid search under 213.237?

It isn't, because eddy current inspects only a shallow surface layer and 213.237 requires a search for internal defects. Its value is sizing RCF so the grind is deep enough for the UT retest to succeed.

How accurate is AI rail defect classification?

Published studies report accuracies such as 88.56% on one B-scan dataset (Hu et al., 2025), but each figure applies only to that study's data and system. Under 213.240, suspects still need field verification by a 213.238-qualified person before they count as defects.

Sources

• 49 CFR 213.237, Inspection of rail, eCFR (text checked on Cornell LII, October 2026)
• 49 CFR 213.238, Qualified operator, eCFR
• 49 CFR 213.240, Continuous rail testing, eCFR
• 49 CFR 213.233, Track inspections, eCFR
• 49 CFR 213.234, Automated inspection of track constructed with concrete crossties, eCFR
• 49 CFR 213.17, Waivers, eCFR
• 49 CFR 213.113, Defective rails, eCFR
• 10 CFR 34.41, Conducting industrial radiographic operations, eCFR
• Magel, Rolling Contact Fatigue: A Comprehensive Review, DOT/FRA/ORD-11/24, FRA, 2011
• Coccia et al., On-line High-Speed Rail Defect Detection: Part II, DOT/FRA/ORD-12/02, FRA, 2012
• Baillargeon and Stuart, Machine Vision Data Products for the Automated Track Inspection Program, RR 21-20, FRA, 2021
• Bruzek, Einbinder and Tunna, DOT/FRA/ORD-26/07, handheld rail UT MAPOD study, FRA, 2026
• Alers et al., Development of a Prototype EMAT System for Inspection of Rails, FRA/ORD-80/45, 1980 (TRID record)
• Witte and Poudel, TTCI TD-16-050, Characterizing Rail Flaws Using Phased Array Ultrasound, 2016
• TTCI TD-16-030, Witte and Poudel, 2016
• Poudel, TTCI TD21-027, thermite weld inspection, 2021
• Baillargeon, Tournay and Kerchof, TTCI TD-14-007, eddy current RCF measurement, 2014
• Cummings, TTCI TD-18-027, RCF and UT confidence, 2018
• TTCI TD20-002, corrective grinding and valid UT tests, 2020
• Mariani, Nguyen, Zhu and Lanza di Scalea, Field test performance of noncontact ultrasonic rail inspection system, J. Transp. Eng. Part A, 2017
• Datta and Lanza di Scalea, High-speed ultrasonic rail inspection system during field tests, Experimental Mechanics, 2023
• Kenderian et al., laser-air hybrid rail inspection, Insight, 2005
• Papaelias, Lugg, Roberts and Davis, high-speed ACFM rail inspection, NDT&E International, 2009
• Ramatlo, Loveday, Long and Wilke, guided-wave broken rail detection, MATEC Web of Conferences, 2021
• Loveday, Long and Ramatlo, guided-wave monitoring of operational rail, Structural Health Monitoring, 2020
• Hu, Li, Jing and Aela, rail flaw B-scan hierarchical classification, Int. J. Steel Structures, 2025
• Bruzelius and Mba, acoustic emission for rail track fault detection, NDT&E International, 2004
• RIFLEX project report (EMAT rail head inspection), CORDIS, 2016
• ASTM E2700-26, E1774-17(2022) and E1419/E1419M-26, store pages