Head Checks in Rail: Causes, Progression, and When They Become Dangerous

Head checks in rail are fine, closely spaced fatigue cracks on the gauge corner, typically 2 to 5 mm apart and up to 2 to 5 mm deep. They are not a defect row in 49 CFR 213.113, but its (d)(7) definition names them as an origin of detail fractures, and they can hide those fractures from ultrasonic testing.

This guide is for track supervisors and rail testing managers who find head checks on a curve and must decide whether to watch, grind, re-test or replace. Head checks are one form of rolling contact fatigue (RCF); for RCF as a whole, see our rolling contact fatigue in rail guide.

Contents: What they are · Causes · Progression · 213.113 · UT masking · Invalid searches · Measuring depth · Grinding · Worked example · When dangerous · Limits · FAQ

What are head checks in rail?

Head checks in rail are short surface cracks that run across the gauge corner at a shallow angle into the head, repeated at regular intervals along the rail. They form where the wheel flange root presses and slides on the corner of the high rail in curves.

The Australian Rail Track Corporation's Rail Defects Handbook (ARTC RC 2400, Issue A, March 2006) calls them gauge corner checking, like "fishscales", and gives three field numbers:

• the cracks typically occur at about 2 to 5 mm intervals along the rail;
• they can grow to 2 to 5 mm deep;
• they grow down into the head at about 10° to 30° to the rail surface, then spread across the head.

The FRA review Rolling Contact Fatigue: A Comprehensive Review (DOT/FRA/ORD-11/24, Magel, 2011) says the cracks appear most often on the outer rails of curves and only infrequently on low rails, often over a whole curve. Clusters usually point to a nearby track geometry fault.

North American railroads more often say gauge corner cracking. Some US transit manuals describe head checks as running about 45° across the corner seen from above, a different angle from the 10° to 30° dip into the head.

What causes head checks?

Head checks form when high wheel-rail contact stress combines with enough slip to shear the surface metal, cycle after cycle, until its ductility runs out. That combination is strongest at the gauge corner of the high rail in curves, where the wheel flange root contacts the rail and slides.

ORD-11/24 attributes gauge corner cracking to high contact stresses "coupled with sufficiently large shear stresses because of slip between the wheel and rail." Three conditions make head checks appear sooner:

• Harder rail. Hard rail wears slowly, so cracks have time to grow. ORD-11/24 cites a European field test in which head checks formed on new grade 1100 rail after one month of traffic, against about two years on grade 900A rail.
• New high rail in shallow curves. German railways (DB) saw head checks develop most rapidly there, the same review reports.
• Track geometry faults. A cluster of head checks on otherwise clean rail points to tight gauge or a crosslevel error.

Wear removes cracked metal, so a rail that wears fast can carry fewer head checks than a premium rail that barely wears, and lubrication that cuts wear can leave cracks in place. See gauge face wear in rail.

How do head checks progress, and how deep do they get?

Most head checks stay shallow, but the ones that keep growing either break out as spalls and shells or turn down into the head as a detail fracture. The depth of the deepest crack, not the number of cracks, decides which path matters.

Published measurements give a sense of scale:

The stages below combine those sources. The depth bands are approximate, not a code classification.

ARTC RC 2400 notes that the crack "initially grows into the rail head but then levels out", and once the checks spread across the head they usually break out as small wedges or spalls. That path roughens the surface and makes testing harder; see rail shelling, spalling and surface defects.

The dangerous path is a branch that turns down, where bending, residual and thermal stresses take over and can grow it across the head. ORD-11/24 lists the 2000 Hatfield derailment in the UK, which killed four people, as a broken rail started by surface RCF, and describes South African cases in which gauge corner cracking started transverse defects of about 80% of the head area.

Key takeaway: track the deepest crack; around 5 mm is where head checks tend to stop being a surface problem.

How do head checks fit 49 CFR 213.113?

Head checks are not a row in the 213.113 remedial action table; they appear in the rule only as an origin of detail fractures. The definition reads:

"Detail fracture means a progressive fracture originating at or near the surface of the rail head. These fractures should not be confused with transverse fissures, compound fissures, or other defects which have internal origins. Detail fractures may arise from shelled spots, head checks, or flaking." (49 CFR 213.113(d)(7))

So a checked curve with no transverse crack found carries no 213.113 remedial action. It still carries a grinding decision and a testing problem. A remedial action applies once a detail fracture is found and sized by percentage of head area.

The table's detail fracture bands, set by the 2014 Improving Rail Integrity rule, run from 5 to under 25% of head area (action C) to 100% (A, or E and H). Detail fracture development in rail covers how the fracture grows.

FRA cause code T207 (detail fracture from shelling or head check) accounted for 145 derailments in 2006 to 2008, ORD-11/24 reports.

Key takeaway: cite 213.113(d)(7) for head checks, and size any detail fracture they produce against the table.

Why are head checks dangerous for ultrasonic testing?

Head checks scatter and reflect the ultrasonic beam at the gauge corner, so they can hide the detail fracture growing beneath them. The hazard is what they hide.

ORD-11/24 states the problem directly:

"the presence of larger and more critical internal defects can be shadowed by smaller surface cracks during inspection." (FRA DOT/FRA/ORD-11/24, 2011)

Masking hits detail fractures hardest, because they start at the surface damage doing the masking. A transverse fissure starts inside the head, so it is less often directly under the checked corner.

No study we found isolates head checks, but data on shells and RCF show the scale:

• In TTCI's 1999 benchmark of six hi-rail test systems (TD-99-006, Garcia and Reiff), transverse defects under shells were found about 62% of the time, against 77% for those with no surface anomaly. Only 7 of the 40 transverse defects were under shells, so treat the gap as an indication, not a precise rate.
• In TD-18-027 (2018), drilled reference holes in five rails removed for low-confidence UT returned on average 67% of their expected response.

On a test car, heavy checking shows as loss of the 0° back-wall echo (loss of bottom) and weak gauge-side angle channels. The operator records a no-test and usually stops to hand test.

Is the problem a UT dead zone?

No. The limit is surface noise and gating, not a contact-probe dead zone. A dead zone belongs to a single-element contact probe; rail systems use wheel or sled probes with a delay path, dual elements and angled beams, so the head is covered.

ORD-11/24 notes that near-surface noise is so large that analysis systems generally ignore the first few millimeters, and that conventional probes can pick up surface-originated defects larger than about 4 mm. So UT can't size a 1 mm head check, but it can show that the corner is blocking the beam.

Key takeaway: head checks matter to UT because they hide detail fractures, and UT gates out the shallow zone where the checks live.

When do head checks make a rail flaw search invalid?

When the qualified operator judges that head checks stopped the equipment from inspecting a length of rail, that length doesn't count as tested, and the owner must re-search, slow trains to 25 mph or replace the rail before the deadline.

Under 49 CFR 213.238(b)(4), the qualified operator must determine that each valid search "has not been compromised due to environmental contamination, rail conditions, or equipment malfunction." Heavy head checking is a rail condition.

Under 49 CFR 213.237(g), a length of track over which a valid search could not be made "may not be considered as internally inspected." Under 213.237(h), before the time or tonnage limit expires, the owner must do one of three things:

1. conduct a valid search for internal defects;
2. reduce operating speed to a maximum of 25 mph until a valid search can be made; or
3. replace the rail that had not been inspected.

49 CFR 213.241(c)(6) requires the record to show the location of any track not tested under 213.237(g).

Part 213 doesn't mention grinding, and a grinding pass earns no 213.237 credit; only the re-search counts. The deadline under 213.237(c) for Class 4 and 5 track is 370 days or 30 MGT, whichever comes first. See 49 CFR 213.237 rail flaw detection requirements.

Key takeaway: a no-test on a checked curve starts a 213.237(h) clock that only a valid search, 25 mph or new rail can stop.

How is head-check depth measured?

Eddy current testing (ECT) and ACFM measure near-surface crack depth; UT does not. Magnetic particle and dye penetrant show that cracks exist but give no depth.

Sources: FRA DOT/FRA/ORD-11/24; TTCI TD-14-007 and TD-14-008; FHWA eddy current guidance.

Eddy current depth and the crack-angle assumption

ORD-11/24 describes grinder-mounted eddy current scanners that record head-check position and depth before and after grinding, while cautioning that it is "difficult with eddy current to know the actual depth of damage." The probe senses the crack's extent along its face and converts it to depth with an assumed angle. TTCI's evaluation of a Rohmann eddy current trolley on Norfolk Southern track (TD-14-007, 2014) put it this way:

"The Draisine relies on the user to enter an assumed crack angle, as the actual angle cannot be determined through nondestructive testing." (TTCI TD-14-007)

At the default 25°, readings differed from sectioned samples by 5% to 33%, an error TTCI put at up to 0.003 in. per degree of angle error; measured angles ran from 18° to 28°. On 100 ft of low rail in a 6° curve, depth readings fell from 0.14 in. before grinding to under 0.04 in. after five passes.

ECT also needs a steady standoff of about 1 to 2 mm (ORD-11/24), and the FHWA notes that irregular surfaces can produce defect-like signals, as on rough, spalled corners.

ACFM

Alternating current field measurement (ACFM) induces a current in a skin less than 0.1 mm deep, and "the deeper the crack, the larger the response" (ORD-11/24). It suits hand surveys and post-grind checks.

For a broader method comparison, see eddy current testing vs ultrasonic testing.

Key takeaway: use eddy current or ACFM to size the checks and UT to search beneath them; neither replaces the other.

How much grinding removes head checks?

Grinding should remove a little more metal than the deepest crack, as early as possible: preventive cycles take tenths of a millimeter, while corrective grinding takes as many passes as the measured depth demands.

ORD-11/24 reports Canadian Pacific's "magic wear rate", the combined natural and grinding wear that just removes new cracks, as about 0.025 mm (0.001 in.) per MGT in sharp curves, 0.016 mm per MGT in mild curves and 0.008 mm per MGT on tangent track, for premium rail with little or no cracking. On a 25 MGT cycle in sharp curves, that is about 0.6 mm of vertical wear per cycle, of which the grinder removes about 0.2 to 0.3 mm.

A production grinder removes about 0.006 to 0.008 in. per pass (TTCI). In TTCI's TD20-002 work, corrective grinding restored a valid UT test at 68 of 88 locations that had failed (77%), and at every location where at least 0.060 in. was removed, about 8 to 10 passes.

ORD-11/24 warns that heavy corrective grinding removes the work-hardened layer and may leave the deepest cracks, so check depth after grinding. It lists improving UT among grinding's purposes because "signals cannot pass through cracks."

Wear limits cap total grinding; CPR's example is 0.75 in. (19 mm) of allowable wear. Our page on rail grinding and ultrasonic testing covers scheduling the two together.

Key takeaway: measure, grind past the deepest crack, measure again, then re-test.

Worked example: from eddy current depths to a grinding pass and a UT re-test

This example turns eddy current readings on one checked curve into a grinding plan and a UT re-test decision. The method is illustrative; owner procedures set real margins.

Situation. A Class 4 freight line carries 36 MGT a year. The test car records a 500 ft no-test on the high rail of a 4° curve, with loss of bottom and weak gauge-side channels. The last valid search of that rail was 200 days earlier.

Step 1: the 213.237 deadline. At 36 MGT a year, 30 MGT accumulates in 30 ÷ 36 × 365 = 304 days. That comes before 370 days, so tonnage controls. The deadline is 304 days after the last valid search, which leaves 304 − 200 = 104 days to get a valid search, slow trains to 25 mph, or replace the rail.

Step 2: measure the checks. An eddy current trolley at the default 25° angle reports the deepest crack in each 100 ft block. The checks are spaced about 3 mm apart, inside the ARTC range.

Step 3: set the removal target. Sectioned angles in TD-14-007 ran from 18° to 28°, up to 7° from the 25° default. At up to 0.003 in. per degree, the angle error could add up to 7 × 0.003 = 0.021 in. The removal target is the reading plus 0.021 in.

Step 4: plan the visit. The grinder is booked for 8 passes through the curve in 42 days, inside the 104-day window. At 0.006 to 0.008 in. per pass, 8 passes remove 0.048 to 0.064 in.

Checks. Block 1: 0.020 + 0.021 = 0.041; 0.041 ÷ 0.008 = 5.13 → 6; 0.041 ÷ 0.006 = 6.83 → 7; 8 × 0.006 = 0.048 > 0.041, so nothing is left. Block 2: 0.056 ÷ 0.008 = 7.00 → 7; 0.056 ÷ 0.006 = 9.33 → 10; 0.056 − 0.048 = 0.008. Block 3: 0.071 ÷ 0.008 = 8.88 → 9; 0.071 ÷ 0.006 = 11.83 → 12; 0.071 − 0.048 = 0.023. Block 4: 0.086 ÷ 0.008 = 10.75 → 11; 0.086 ÷ 0.006 = 14.33 → 15; 0.086 − 0.048 = 0.038. Block 5: 0.066 ÷ 0.008 = 8.25 → 9; 0.066 ÷ 0.006 = 11.00 → 11; 0.066 − 0.048 = 0.018. The deepest target, 2.18 mm, is well short of the roughly 5 mm turn-down depth and inside the 2–5 mm developed range, so grinding is still the right tool.

Step 5: re-measure, then re-test. After the visit, the trolley re-runs the curve and all 500 ft are re-tested by UT.

• Block 1 is clean at either rate, so a valid search can be expected.
• Block 2 is clean at 0.008 in. per pass; at 0.006 in. it may keep 0.008 in., which needs 2 more passes (0.008 ÷ 0.006 = 1.33 → 2) if the re-check shows cracks.
• Blocks 3 and 5 may keep 0.018 to 0.023 in. and need 3 to 4 more passes (0.018 ÷ 0.006 = 3; 0.023 ÷ 0.006 = 3.83 → 4).
• Block 4 needs up to 7 more passes (0.038 ÷ 0.006 = 6.33 → 7) and is the most likely to stay a no-test.

The 8 passes remove 0.048 to 0.064 in., just around the 0.060 in. at which every failed location in TD20-002 recovered.

Step 6: decide before the deadline. The first visit lands on day 242 (200 + 42), leaving 62 days. If Blocks 3 to 5 can't be ground again and re-tested by day 304, the owner applies the 213.237(h)(2) 25 mph limit over them or replaces the rail. Meanwhile, hand test those blocks from the field side as well as the gauge side; any detail fracture found goes straight to 213.113.

Key takeaway: the deepest block sets the grinding plan, and the 213.237 deadline sets how long you have to finish it.

When do head checks become dangerous?

Head checks become dangerous when they stop the rail from being tested or when a crack turns down into the head. Either condition can let a detail fracture grow unseen.

Treat a checked curve as urgent when any of these apply:

• Repeated UT no-tests, or loss of bottom that hand testing can't clear.
• Measured depth in the 2 to 5 mm range ARTC gives for developed checking, and above all near the roughly 5 mm at which one study found RCF cracks tend to turn transverse.
• Spalls, pits or shells forming on the checked corner.
• Depth rising between grinding cycles, so removal is falling behind initiation.
• A 213.237 deadline approaching with no valid search over the checked rail.
• A detail fracture found in the zone; 213.113 then sets the action by size.

How heavy the checking looks is a weaker signal than depth, so measure before you decide.

Key takeaway: depth and testability decide the risk; the appearance of the corner does not.

What are the limits of head-check inspection?

No single method sees both the head checks and what lies beneath them, and every depth number carries an error band.

• UT can't size shallow head checks. It gates out the first few millimeters, and heavy checking masks it; a no-test is missing coverage, not a pass.
• Eddy current depth depends on an assumed angle. TD-14-007 found errors of 5% to 33%, and rough or spalled rail adds lift-off noise.
• Surface methods find nothing internal. Eddy current, ACFM, MT and PT can't detect a detail fracture below their reach.
• Grinding may leave the deepest cracks. Re-measure after every corrective visit.
• Published figures are site-specific. ARTC values are typical Australian figures, TTCI depths come from particular curves, and the 62% vs 77% figure is for shells in a 1999 sample of 40 defects.
• No public US standard grades head checks by depth. The bands here are illustrative.

How EMA Industries applies this

EMA Industries performs ultrasonic rail flaw testing for transit agencies, short lines and industrial track owners, based in Southern California and mobilizing nationwide. On checked curves, the report should mark each no-test by milepost and rail, so the owner can plan grinding and a re-test inside the 213.237(h) window. To plan testing around a grinding program, see our ultrasonic rail testing services or contact EMA Industries.

Related reading: Rolling contact fatigue in rail · Rail shelling and spalling · Detail fracture development · Rail grinding and UT · Gauge face wear in rail

Frequently asked questions

What is the difference between head checks and shelling?

Head checks start at the gauge corner surface as cracks about 2 to 5 mm apart. Shelling starts 2 to 8 mm below the gauge corner, per ARTC, and grows roughly parallel to it. Both can lead to a detail fracture under 213.113(d)(7).

Are head checks a reportable defect under 49 CFR 213.113?

Not on their own. The remedial action table has no row for head checks. They appear only in the detail fracture definition, 213.113(d)(7), so a remedial action applies when a detail fracture is found and sized by head area.

How deep do head checks get?

Most stay shallow: TTCI found 90% of 880 RCF cracks in rails removed for poor UT were no deeper than 0.040 in. (1 mm). ARTC gives 2 to 5 mm for developed checking, and cracks near 5 mm tend to turn down.

Why does a test car report a no-test on a curve with head checks?

The cracks scatter and reflect the ultrasound at the gauge corner, so the 0° channel loses its back-wall echo and the gauge-side angle channels lose sensitivity. The qualified operator can't call the search valid, so 213.237(g) and (h) apply until a valid re-search is made.

How many grinding passes does it take to remove head checks?

It depends on the deepest crack. At 0.006 to 0.008 in. per pass, a 0.050 in. target takes 7 to 9 passes. TTCI found that removing at least 0.060 in. restored a valid UT test at every failed location in one study. Re-measure and re-test after grinding.

Sources

• 49 CFR 213.113, Defective rails, eCFR (text read on Cornell LII, October 2026)
• 49 CFR 213.237, Inspection of rail, eCFR (text read on Cornell LII)
• 49 CFR 213.238, Qualified operator, eCFR
• 49 CFR 213.241, Inspection records, eCFR
• Track Safety Standards; Improving Rail Integrity, final rule, 79 FR 4234 (January 24, 2014), Federal Register via govinfo
• Magel, E., Rolling Contact Fatigue: A Comprehensive Review, DOT/FRA/ORD-11/24 (November 2011), FRA, via ROSA P
• Australian Rail Track Corporation, Rail Defects Handbook, RC 2400, Issue A (March 2006)
• TTCI TD-14-007, eddy current measurement of RCF crack depth (May 2014), MxV Rail (formerly TTCI)
• TTCI TD-14-008, RCF measurement technologies (May 2014), MxV Rail
• TTCI TD-18-027, RCF and ultrasonic test confidence (November 2018), MxV Rail
• TTCI TD-20-002, corrective grinding and ultrasonic test exceptions, MxV Rail
• TTCI TD-99-006, Garcia and Reiff, Benchmarking Results of North American Rail-Flaw Detection Technologies (February 1999), MxV Rail
• FHWA, Eddy Current Testing, Federal Highway Administration NDE Web Manual
• Ekberg, A. and Kabo, E., Chalmers University of Technology (2014), abstract only (depth at which RCF cracks turn transverse)