NDT Methods for Rail Defects: Which Method Finds Each Defect

The NDT methods for rail defects are chosen by the plane of the crack. A 70° shear-wave probe finds transverse defects in the head, a 0° beam finds horizontal split heads, 37–45° channels find bolt-hole cracks, and eddy current measures surface cracking. Under 49 CFR 213.237, owners must test rail internally, but no method sees reliably beneath a shell.

This guide matches each defect type to the method that detects it, explains why that method works, and states where it fails. It is written for track supervisors, transit engineers and NDT technicians who have to decide what to specify, what a test result covers, and when a second method is needed.

For the full defect library and the UT capability review, see the rail defects guide for inspectors, what rail UT can and can't find and the UT vs other NDT methods by component comparison.

Which NDT methods for rail defects are in use, and what does each one sense?

Ultrasonic testing is the primary method for internal rail defects, and induction is a complementary one; eddy current, ACFM, magnetic particle, penetrant and visual inspection cover the surface. FRA described the field in its 2014 rail integrity rule:

"The main technologies utilized for non-destructive testing on U.S. railroads are the ultrasonic and induction methods. Ultrasonic technology is the primary technology used, and induction technology is currently used as a complementary system." — FRA, Track Safety Standards; Improving Rail Integrity, 79 FR 4234 (January 24, 2014), p. 4237

The methods sense different physical properties, so they reach different depths.

The frequency band comes from the same 2014 rule, which says rail UT "normally" runs at 2.25 to 5.0 MHz. The eddy current depth figure is FHWA's general figure for steel, and the ACFM skin depth is from FRA's 2011 rolling contact fatigue review (DOT/FRA/ORD-11/24).

Key takeaway: only the volumetric methods (UT and, for the head, induction) look inside the rail; everything else is a surface method.

Why does the crack plane decide the method?

A sound beam reflects best from a crack it meets face-on, so each probe angle is aimed at the plane a given defect grows in. A crack edge-on to the beam returns little energy, however large it is.

The 2014 rule describes the transducers as placed "at various angles with respect to the rail surface", and says the waves "normally scan the entire rail head and web, as well as the portion of the base directly beneath the web". No public source names one standard North American channel set, so the table is a typical arrangement. A 1999 Colorado State review of test-car data (MPC 99-106) lists 0°, 45° and 70° probes plus a side-looker, with the angled probes facing forward and backward.

The angle off normal to a vertical transverse crack is 90° minus the beam angle. A 70° beam meets it 20° off normal, which is close to face-on; a 0° beam travels parallel to it and sees nothing. A horizontal split head is the reverse case: the 0° beam hits it face-on.

Vertical splits and piped rail sit parallel to the 0° beam too. The 0° channel finds them indirectly, when the base echo drops out or arrives early, not by a direct flaw echo.

Key takeaway: specify channels by the defect plane you need to find, not by wave type.

Which method finds each rail defect?

For internal defects the answer is a specific UT channel; for surface damage it is eddy current, MT, PT or visual inspection, with UT only as a check of what lies underneath. The matrix below follows the defect definitions in 49 CFR 213.113(d), plus the surface conditions the rule doesn't list.

Two rows are not 213.113 table rows at all: shelling and corrugation. The remedial action table has no row for shelled rail, spalling, flaking or head checks, and 213.113(d)(9) says flattened rail does "not include corrugations".

How are transverse defects (TF and detail fracture) detected?

Transverse defects are found with 70° shear-wave channels firing forward and backward along the head. The crack plane is "substantially at a right angle to the length of the rail" (213.113(d)(15)), which a 70° beam meets close to face-on.

The two transverse types differ in origin, and the origin decides the blind spot. A transverse fissure starts "from a crystalline center or nucleus inside the head". A detail fracture is "a progressive fracture originating at or near the surface of the rail head", and "may arise from shelled spots, head checks, or flaking" (213.113(d)(7)).

FRA's 2011 rolling contact fatigue review adds that transverse defect growth "is not a rolling contact phenomenon"; bending, residual and thermal stress drive it. Rolling contact fatigue is where a detail fracture starts, not a fissure.

Size matters for detection. In TTCI's 1998 benchmark of six hi-rail systems on known defects (report TD-99-006, 1999), the systems found 66% of transverse defects of 20% head area or smaller and 96% of those 21% and larger.

Key takeaway: a TF is hidden by its small size early on; a DF is hidden by the surface damage it grows from.

How are vertical and horizontal split heads detected?

A horizontal split head is a 0° target; a vertical split head is a side-looker target, with the 0° channel and the eye as backups. The two run along the rail, so a 70° channel that looks along the rail meets them edge-on.

A horizontal split head starts "usually 1/4 inch or more below the running surface" and is "generally accompanied by a flat spot on the running surface" (213.113(d)(11)). The horizontal plane faces the 0° beam directly, which is why TD-99-006 recorded 92% detection for the two split heads it tested.

A vertical split head runs "through or near the middle of the head". The rule gives the visual clue: "A crack or rust streak may show under the head close to the web or pieces may be split off the side of the head" (213.113(d)(16)). TD-99-006 found its single VSH on every run, a sample too small for a rate.

Key takeaway: an inspector walking the track can sometimes see a VSH before the next test, because the rule's own definition names its visible signs.

How are bolt-hole cracks and head-web separations detected?

Bolt-hole cracks are found by 37–45° channels aimed through the web at the hole, because the crack grows on a path "inclined upward toward the rail head or inclined downward toward the base" (213.113(d)(1)). The holes sit behind the joint bars, so this is a UT job, not a visual one.

The rule makes coverage inside joints a requirement: "Inspection equipment shall be capable of detecting defects between joint bars, in the area enclosed by joint bars" (213.237(f)). That paragraph concerns the rail inside the joint; a cracked bar is a separate 213.121 matter.

Bolt-hole cracks were the weakest result in TD-99-006: 33% for two small cracks (0.25 and 0.38 in.), a warning rather than a rate. The hole itself returns a strong geometry echo that a trained operator has to separate from a crack signal.

A head and web separation is "a progressive fracture, longitudinally separating the head from the web of the rail at the head fillet area" (213.113(d)(10)). In practice it is typically picked up where the 0° and angled beams pass through the fillet into the web and the expected web or base echo is interrupted. Outside joint bars the crack may be visible at the fillet; inside them it is hidden.

Fully developed bolt-hole cracks "may continue horizontally along the head/web or base/web fillet" (213.113(d)(1)). At a joint, a long fillet crack can be either defect, so trace it toward the hole before naming it. FRA's accident cause codes separate the two (T201 bolt-hole crack; T210 and T211 head and web separation outside and within joint bars).

How are rail weld defects detected?

Rail welds are tested with the same 0°, 45° and 70° channels as plain rail, but the base of the weld needs a separate hand test. Test vehicles cover the head and web well and most of the base poorly.

213.113(d)(6) defines a defective weld by size and cause: discontinuities "exceeding 5 percent of the rail head area individually or 10 percent in the aggregate", from incomplete penetration, lack of fusion, slag or sand, under-bead or shrinkage cracking, or fatigue cracking. It adds that "Weld defects may originate in the rail head, web, or base".

That last clause is where the method gap sits. TTCI's thermite weld failure survey (TD-00-013, 2000) found the most common defect, slag and porosity, "often forming at the web/base fillet", and warned that "the ultrasonic rail flaw detection process only scans the head and web of the rail, and not the majority of the base." Its weld data from detector cars was "likely skewed" for that reason.

A later TTCI study (TD21-027, 2021) found that about 95% of thermite weld breaks at its test track were in the web, web-base fillet and base. It also stated that existing practices for characterizing thermite welds after installation "are limited to visual inspection". In its simulations, full coverage of the base needed a full-skip angle shot that used the rail, not the weld, as the reflector.

The weld's own shape adds false calls. In TD-00-013, nearly 25% of the 41 detector-car finds that were cut open turned out to be good welds. For a weld-by-weld plan, see aluminothermic weld inspection and flash-butt weld inspection.

Key takeaway: a passing test-car run over a thermite weld says little about its base.

How are head checks, RCF, shelling and corrugation handled?

Surface damage is measured with eddy current and checked visually; MT or PT can confirm surface cracking on a cleaned spot; UT is used to find what the damage is hiding. None of these conditions is a 213.113 table row on its own, but each can mask or start one that is.

Head checks and other RCF cracks

Eddy current is the established method for head-check depth, usually run before and after grinding. Its weakness is depth: TTCI's 2014 eddy current trial (TD-14-007) reported that the instrument "relies on the user to enter an assumed crack angle, as the actual angle cannot be determined through nondestructive testing". Depth errors ran from 5% to 33% against sectioned rail at the default 25° angle.

ACFM is a surface method with a useful property: in ORD-11/24's words, "the deeper the crack, the larger the response". Published rail ACFM work is lab and rig testing, not revenue-service practice. ASTM E2261/E2261M-17(2021) covers ACFM of welds for "surface breaking discontinuities" and "does not establish weld acceptance criteria".

MT and PT show where surface cracks are, on a cleaned and accessible surface, but they don't size depth. TTCI technology digest TD-14-008 (May 2014) said dye penetrant and magnetic particle inspection are "not capable of measuring any key characteristics below the surface". MT also needs magnetic steel; cast austenitic manganese frogs are essentially non-magnetic, so PT is the surface method there.

A squat is not an advanced head check. ORD-11/24 calls it "a surface defect most commonly associated with high-speed rail and areas of high tractive effort", a separate RCF defect near the head center whose branch cracks are often missed by conventional UT.

Shelling

Shelling starts below the surface at the gauge corner and becomes visible when it breaks out, so visual inspection finds the shell itself. The problem is what lies under it. The shell's crack faces reflect the beam before it reaches a detail fracture below.

In the 1992 Superior, Wisconsin derailment, the NTSB found detail fractures that had grown from shelling. It concluded that a series of overlapping cracks "prevented the penetration of the ultrasonic sound waves through the head of the rail" (NTSB letter R-94-1 and -2, 1994).

The only measured masking figure is TD-99-006's: about 62% detection for transverse defects under shells (7 defects) against 77% for those with no surface anomaly (33 defects). The figure applies to defects under shells, not to head checks.

Grinding is the usual fix: TTCI report TD20-002 found that corrective grinding restored a valid UT test at 68 of 88 exceptions (77%), and at all of them where 0.060 in. or more was removed. See the rail shelling and spalling guide and the head checks guide.

Corrugation

Corrugation is a repeating surface wave along the head, measured with profile trolleys or vehicle-mounted systems rather than flaw detectors. Part 213 sets no corrugation limit, and its effect on NDT is on coupling: FRA lists corrugation among the "poor, rail head surface conditions" that can "negatively influence the detection of rail flaws". The corrugation on rail guide covers measurement.

Key takeaway: treat shelling and heavy RCF as a coverage problem for UT, not only as surface damage.

How are base defects found?

Base defects are found mainly by eye during visual track inspections, and by hand UT where an owner specifies it, because test vehicles don't scan most of the base. A broken base is simply "any break in the base of the rail" (213.113(d)(2)).

FRA's 2014 rule says rail UT normally covers "the portion of the base directly beneath the web", not the flanges. The same preamble names the gap directly: "Due to the limitations of current technology to detect internal rail flaws beneath surface conditions and in the base flange area, FRA's research has been focusing on other rail flaw detection technologies."

That leaves the visual inspection under 49 CFR 213.233, made by a person designated under 213.7 "on foot or by traversing the track in a vehicle". The rule lets "mechanical, electrical, and other track inspection devices" supplement it, not replace it.

The gap matters in the accident data. FRA's accident database lists 53 main-track derailments caused by broken bases (code T202) in 2016–2025, against 110 from detail fractures (T207) and 57 from transverse or compound fissures (T220). Those counts were queried on October 24, 2026 and change with late filings.

Worked example: a transverse defect under a shell

Inputs (illustrative): 136RE rail; a transverse defect centered in the head 0.75 in. below the running surface; a gauge-corner shell 5.8 in. long and 0.40 in. deep, centered over the defect; shear velocity 0.1280 in./µs; wedge delay ignored. The shell size follows an example in TTCI's phased array study (TD-16-050, 2016), reading its third figure as depth; the defect depth is assumed.

Step 1: where the 70° probe must be. For a 70° beam, sound path = depth ÷ cos 70° = 0.75 ÷ 0.3420 = 2.193 in. Surface distance = depth × tan 70° = 0.75 × 2.7475 = 2.061 in. Round-trip time = 2 × 2.193 ÷ 0.1280 = 34.26 µs.

Step 2: where the beam crosses the shell. The beam reaches 0.40 in. depth at a surface distance of 0.40 × 2.7475 = 1.099 in. from its entry point. That is 2.061 − 1.099 = 0.962 in. short of the defect, measured along the rail.

Step 3: is that point under the shell? The shell runs 2.9 in. either side of the defect. The forward beam crosses the shell depth 0.962 in. before the defect, and the reverse beam 0.962 in. after it. Both points are inside the shell, so both 70° beams meet the shell's crack face before they reach the defect.

Step 4: what the other channels do. Where the 0° beam passes under the shell, it meets the horizontal shell face-on and the transverse defect edge-on, so it shows the shell and loss of bottom, not the defect. A side-looker or a hand probe placed on the field side of the head may get past the shell's edge; whether it does depends on the shell's width.

Step 5: what the rules say. If the qualified operator judges that a valid search couldn't be made, that rail "may not be considered as internally inspected" (213.237(g)). Before the interval runs out, the owner must make a valid search, cut speed to 25 mph until one can be made, or replace the rail (213.237(h)).

Result: at this geometry, the channel that would normally find this defect is the one the shell blocks. Removing the shell by grinding and re-testing is the route TD20-002 supports.

If a hand test finds and sizes the defect, the 213.113 table applies: a detail fracture of 60% to under 100% of head area takes A2 or E and H, and 100% takes A or E and H. Remedial action A assigns a person designated under 213.7 to visually supervise each operation over the defective rail; it is not a stop order.

What are the limits of NDT methods for rail defects?

Every method has a defect it can't see, and the test-vehicle methods share three gaps: the base, the rail under surface damage, and small defects. The limits below are the ones a specification or a test report should state.

• Surface damage blocks UT. Shells, head checks, engine burns and spalls reflect or scatter the beam. A no-test is untested rail, not sound rail.
• Most of the base isn't scanned by test vehicles (TD-00-013; FRA, 2014). Base and thermite weld foot coverage needs hand UT or visual inspection.
• Detection is not certain. The TD-99-006 trials found 72% of 49 known defects overall, and TTCI's 2016 phased array report put conventional UT at "about 90 percent" of flaws. Those are research figures for their own test conditions, not guarantees.
• Surface methods give no internal search. Eddy current, ACFM, MT and PT reach only the surface or just below it. None gives a 213.237 valid search, and none sizes a crack's depth without assumptions.
• Magnetic methods need magnetic steel. MT doesn't work on austenitic manganese castings, and it never finds internal flaws.
• Induction senses the head only, and a test contractor's representative quoted in the NTSB's 1994 letter called it more sensitive to surface conditions than UT.
• Newer methods are not yet routine. EMAT, laser UT, ACFM and guided waves were described as "under development" in a 2008 review of rail inspection (Papaelias, Roberts and Davis).
• Probe geometry assumes a known rail. Worn heads change the refracted angle, so a fixed probe can inspect "the wrong volume of material" (TTCI TD-16-030).

Key takeaway: write the test scope so that the base, welds and no-test lengths each have a named method and a named result.

How EMA Industries applies this

EMA Industries provides ultrasonic rail flaw testing for transit agencies, short lines and industrial track owners, along with rail weld inspection, MT and PT, and NDT Level III services. A scope built on the matrix above pairs the 0°, angled and 70° rail search with hand verification of suspects, hand UT of thermite weld bases, and a list of no-test lengths with the reason for each.

EMA is based in Southern California and mobilizes nationwide. To scope a rail test that matches methods to the defects on your line, contact EMA Industries.

Related rail defect guides

• Detecting rail defects: a complete guide for inspectors (the defect library)
• Ultrasonic rail testing capabilities
• Ultrasonic testing vs other NDT methods for railways
• Transverse fissure in rail and detail fracture development
• Vertical split head and horizontal split head
• Bolt hole cracks in rail
• Eddy current vs ultrasonic testing
• How often must rail be tested?

Frequently asked questions

Which NDT method is used to find internal rail defects?

Ultrasonic testing is the primary method, with induction as a complementary channel on some test cars, as FRA described in its 2014 rule. Internal transverse defects need 70° shear-wave channels, and horizontal defects need the 0° beam. Magnetic particle testing finds only surface and near-surface cracks, so it can't do this job.

What probe angle finds a transverse defect in the rail head?

A 70° shear-wave probe, fired forward and backward along the rail. The beam meets the transverse crack plane about 20° off normal, close to face-on. A 0° beam runs parallel to that plane and misses it.

Can eddy current replace ultrasonic rail testing?

No. Eddy current reaches only the surface layer, typically under 1–3 mm in steel, so it can't search the head, web or base for internal defects or give a 213.237 valid search. Its role is measuring head checks and gauge-corner cracking, mainly to plan and check grinding.

Why does shelling cause missed rail defects?

The shell's crack faces reflect the ultrasonic beam before it reaches the steel underneath, where detail fractures grow from shelled spots. In TTCI's 1998 trials, transverse defects under shells were found about 62% of the time, against 77% with no surface anomaly. Grinding the shell off and re-testing is the usual remedy.

Does a shelled rail need remedial action under 213.113?

Not by itself. The 213.113 remedial action table has no row for shelled rail; shelling appears only in the detail fracture definition, 213.113(d)(7). A shell that blocks the test can, however, mean the search was not valid under 213.237(g), which triggers the options in 213.237(h).

Sources

• 49 CFR 213.113, Defective rails: eCFR (text read on Cornell LII, October 24, 2026)
• 49 CFR 213.237, Inspection of rail: eCFR (read on Cornell LII)
• 49 CFR 213.233, Visual track inspections: eCFR
• FRA, Track Safety Standards; Improving Rail Integrity, final rule, 79 FR 4234 (January 24, 2014): govinfo PDF
• Magel, Rolling Contact Fatigue: A Comprehensive Review, DOT/FRA/ORD-11/24 (November 2011): ROSA P
• NTSB safety recommendation letter R-94-1 and -2 (Superior, Wisconsin, 1992): ntsb.gov PDF
• Garcia and Reiff, Benchmarking Results of North American Rail-Flaw Detection Technologies, TTCI TD-99-006 (February 1999): mxvrail.com PDF
• Kristan, Thermite Weld Failure Survey, TTCI TD-00-013 (July 2000): mxvrail.com PDF
• Baillargeon, Tournay and Kerchof, TTCI TD-14-007 (May 2014), eddy current RCF measurement, and TTCI TD-14-008 (May 2014), RCF measurement requirements: TD14-007 PDF, TD14-008 PDF
• Witte and Poudel, High-Speed Rail Flaw Detection Using Phased Array Ultrasonics, TTCI TD-16-030 (July 2016), and Characterizing Rail Flaws Using Phased Array Ultrasound, TD-16-050 (November 2016): TD16-030 PDF, TD16-050 PDF
• TTCI TD20-002, corrective grinding and UT exceptions: mxvrail.com PDF
• Poudel, TTCI TD21-027 (September 2021), thermite weld inspection after installation: mxvrail.com PDF
• Jeffrey and Peterson, Assessment of Rail Flaw Inspection Data, MPC 99-106 (August 1999): ugpti.org PDF
• FHWA, Eddy current testing: infotechnology.fhwa.dot.gov
• ASTM E2261/E2261M-17(2021), Examination of Welds Using the Alternating Current Field Measurement Technique: store.astm.org
• Papaelias, Roberts and Davis, "A review on non-destructive evaluation of rails: State-of-the-art and future development", Proc. IMechE Part F 222:367–384 (2008): doi.org/10.1243/09544097jrrt209
• FRA accident data (Form 54), main-track derailments 2016–2025 by cause code, queried October 24, 2026: data.transportation.gov