Ultrasonic Testing

Ultrasonic Testing Mistakes and How to Avoid Them

Updated

Ultrasonic Testing Mistakes

The ultrasonic testing mistakes that cause missed or false calls in rail and weld UT fall into eight groups: calibration, couplant, probe choice, gain, echo reading, coverage, no-test handling and records. FRA's 2026 study ORD-26/07 found that, with one small exception, no two operators produced statistically similar calibration results, so the setup is where most errors start.

This guide is for rail test operators, weld UT technicians and the track or QA engineers who review their work. Each mistake below is tied to a published source or a rule, with its consequence and its fix. It includes a worked example of how a wrong velocity setting moves a rail head defect by more than an inch and a half.

Why do ultrasonic testing mistakes matter more than equipment limits?

Most ultrasonic testing mistakes happen before the first scan or after the last one, in setup and paperwork, not in the instrument. A flaw detector reports exactly what its settings tell it to. If the velocity, zero, angle or gain is wrong, the screen still shows clean, confident echoes in the wrong place or at the wrong height.

FRA's research supports this. In Application of MAPOD to Handheld Ultrasonic Rail Flaw Inspection (DOT/FRA/ORD-26/07, July 2026), ENSCO's team ran human factors tests on handheld rail UT calibration and reported:

"With a small exception, no two operators that participated produced statistically similar calibration results. Further, a single operator, using the same equipment on six occasions, produced inconsistent calibration results."

— FRA, DOT/FRA/ORD-26/07 (2026), Executive Summary

The same report states that "calibration errors can cause excessive false positives and false negatives". It also notes that the North American railroad industry has no standardized calibration procedure for handheld rail UT, and each organization follows its own.

The rest of this article takes each mistake group in turn. The table below is the short version.

Mistake groupTypical consequenceFirst fix
Calibration (velocity, zero, angle, DAC/TCG)Defect plotted at the wrong depth or position; wrong amplitudeVerify on a known reflector before and after the shift
Couplant and surface prepLost signal read as "no defect"Monitor the 0° back-wall echo; record untested rail
Wrong probe or frequencyDefect plane missed; noise from grain or wearMap the defect plane to the beam angle
Gain and reference errorsIndications below the gate or noise above itFix the reference reflector, then add scanning gain as written
Misread echoesFalse calls on bolt holes, welds, mode conversionLocate the reflector on the drawing before calling it
Coverage gapsBase, gage corner or weld zones never insonifiedWrite the coverage plan by zone and beam
No-test handlingTrack counted as tested when it wasn'tApply 213.237(g) and (h)
Poor recordsA good test that can't be provenRecord the six 213.241(c) items

Key takeaway: treat setup as part of the test, because the instrument can't tell you it was set up wrong.

How do calibration errors cause missed and false calls?

A calibration error shifts every reading taken afterwards, so one mistake at the start of a shift affects every indication of that shift. ORD-26/07 lists four factors typically set during handheld rail UT calibration: zero (the time spent in the wedge), the velocity of sound in steel, the refracted angle of the wedge, and gain to bring a known reflector to 80 percent of full screen height. The report says calibration is typically done at the start of the day's inspection and after any component is changed, usually on an IIW Type 2 block.

Each factor has its own failure mode:

Calibration factorWhat goes wrongConsequenceFix
VelocityLongitudinal velocity left in when a shear-wave angle probe is fitted, or a block of a different material usedEvery sound path is scaled by the ratio of the two velocitiesSet velocity on two known distances (for example the IIW radius and a second echo) for each probe
Zero (probe delay)Wedge delay not reset after a wedge change or as the wedge wearsConstant offset in sound path, largest in relative terms for shallow reflectorsRe-zero after every wedge or probe change and at the procedure's check interval
Refracted angleLabel angle assumed instead of measured; worn or cold wedgeDepth and surface distance both wrong, with no warning on screenMeasure the exit point and angle on the block; recheck when temperature changes
DAC or TCGCurve built on a block whose surface, material or hole size doesn't match the job; points taken in the near fieldAmplitude evaluation wrong at every depthBuild the curve on the reflector type the procedure names, with any transfer correction it gives

The angle and velocity errors are connected. FHWA's December 2024 memo on acoustic dissimilarities in steel weld UT warns that a reference block that isn't acoustically equivalent to the part shifts the refracted angle, and at 70° can push it toward mode conversion to a surface wave. The memo's warning is blunt: "the inspector has no indication this is happening and believes the weld is defect free."

The calibration process itself, including blocks, check intervals and what to ask for in an RFP, is covered in our ultrasonic testing calibration guide for rail inspection. Building and checking a distance amplitude curve is covered in the DAC curve guide.

Key takeaway: verify velocity, zero, angle and reference on a known reflector at the start and end of each shift, and after any change to the probe, wedge or cable.

Worked example: how a velocity error becomes a depth error

A velocity error scales the sound path, and the instrument converts that wrong sound path into a wrong depth and a wrong surface distance. The worst case is a common setup slip on single-channel flaw detectors: the operator calibrates a 0° probe with the longitudinal velocity, then fits a 70° shear-wave probe and doesn't change the velocity.

Inputs (steel only; wedge delay already removed):

  • Rail: 136RE, transverse defect 0.75 in. below the running surface
  • Probe: 70° refracted shear wave
  • Shear velocity in steel: 0.1280 in./µs (about 3,250 m/s, Evident)
  • Longitudinal velocity in steel: 0.2320 in./µs (about 5,890 m/s, Evident)

Step 1, the true geometry. 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, what the instrument displays with the wrong velocity. Displayed sound path = 0.2320 × 34.26 ÷ 2 = 3.975 in. The instrument then computes depth = 3.975 × cos 70° = 1.359 in. and surface distance = 3.975 × sin 70° = 3.735 in.

SettingSound path (in.)Depth shown (in.)Surface distance shown (in.)Error in surface distance
Correct shear velocity, 0.1280 in./µs2.1930.7502.061none
Longitudinal velocity left in, 0.2320 in./µs3.9751.3593.735+1.67 in.
Shear velocity 3,230 instead of 3,250 m/s2.1790.7452.048−0.01 in.
Correct velocity, probe delay set 2 µs too small2.3210.7942.181+0.12 in.
Wedge actually refracting 67°, instrument set to 70°1.9190.6571.804 (true 1.767)+0.04 in.

What it means. The longitudinal-velocity error multiplies every sound path by 0.2320 ÷ 0.1280 = 1.81. The defect is plotted 0.61 in. too deep and 1.67 in. too far ahead of the probe. A depth of 1.36 in. still lies inside the rail head (the 136RE head is close to 2 in. deep; check the section drawing), so nothing on screen looks impossible. A hand verifier who marks the rail from that reading looks for the defect in the wrong place, and a sizing or disposition call based on the depth is wrong.

The small errors matter less on their own. A 20 m/s velocity difference moves this reflector by about 0.01 in., while a 2 µs zero error or a 3° angle error moves it by a few hundredths to a tenth of an inch. They add up, though, and they grow with sound path.

The fix. On the 0° channel, the back-wall echo is a free velocity check: the base of a new 136RE rail should read 7.31 in. (63.04 µs at 0.2320 in./µs), and a 3% velocity error would put it at 7.53 in. Angle-beam channels have no back wall to check against, so the velocity, zero and angle must be proven on a block before scanning and rechecked afterwards. For more beam geometry, see our shear wave ultrasonic testing guide.

Key takeaway: a 0° base echo checks the 0° setup, but nothing on the screen checks an angle-beam setup except a known reflector.

How do couplant and surface condition cause untested rail?

Couplant failures don't produce false indications; they produce silence, and silence on an A-scan is easily read as sound rail. A steel-to-air boundary reflects about 99.996% of the incident energy at normal incidence, so even a thin air gap stops the beam. A clear screen can mean good rail or no sound in the rail at all.

The common field causes are:

  • Dry or interrupted couplant: an empty spray tank, blocked nozzles, or a gradient where the water runs off.
  • Rail surface condition: heavy lubricant, rust scale, debris, ice, or corrugation that lifts a sled or wheel probe.
  • Profile mismatch: ORD-26/07 notes that a flat probe face "may not make good contact with a rail that has heavy gage face wear".
  • Weld surfaces: spatter, loose scale and rough caps on structural welds, which procedures commonly require removed before contact UT.

Vehicle systems watch for this through the 0° back-wall echo. A test car patent (US 5,777,891) describes a "Lack-Of-Expected Response (LER) alarm" raised when the expected base echo disappears. The fix is to treat every loss of back-wall echo as a possible coupling loss until it is proven otherwise, and to record the stretch as untested if the signal can't be restored.

Losing coupling is not a false call. Under 49 CFR 213.241(c)(6), the location of track not validly tested must be recorded, which is covered in the no-test section below. Couplant selection, cold-weather mixes and wheel fill are covered in our ultrasonic couplant guide for rail inspection, and temperature effects in our cold weather UT guide.

Key takeaway: a quiet screen is a question, not an answer, until the back-wall echo proves the probe is coupled.

What goes wrong when the probe or frequency is wrong?

A probe finds a crack only when its beam meets the crack face at a useful angle, so the wrong probe can pass straight by a large defect. Choose the probe by the crack's plane against the beam direction, not by wave type.

Defect or zoneCrack planeProbe that respondsCommon mistake
Transverse defect in rail head (TF, DF)Across the rail70° shear wave, forward and reverseRelying on the 0° channel, which runs parallel to the crack
Horizontal split head, head-web separationHorizontal0°Expecting the 70° beam to catch it
Vertical split headVertical, along the railSide-looking probe; 0° shows only loss of back wallTreating a 0° loss of back wall as a sizing tool
Bolt hole cracksRadiating from the holeTypically about 37–45° beams, forward and reverseSkipping joints, which 213.237(f) requires the equipment to cover
Structural groove weldsFusion faces and root45°, 60° or 70° per the code's chartOne angle used for all thicknesses

Typical rail practice is 0°, about 37–45° and 70° probes plus side-lookers, though no public source names one standard North American channel set. Wear changes the picture. TTCI's 2016 digest TD-16-030 explains that the beam direction inside a worn rail differs from a new one, so "flaws may be missed during dynamic testing on worn rail because the wrong volume of material is inspected."

Frequency is a trade-off between resolution and penetration. Wheel-probe transducers are "typically centered at 2.25 Mhz" according to US 5,777,891. Higher frequency improves resolution, but grain scattering rises steeply with frequency in the Rayleigh regime, roughly with its fourth power. That is one reason FRA's RR 22-25 found cast austenitic manganese frogs so hard to test: the steel is "highly attenuative" and coarse-grained, and in-service examination of these components is visual only. For structural welds, the search unit frequency and size are set by the code edition named in the contract; check that edition rather than a remembered number.

Near-surface flaws are a separate issue. A single-element contact probe has a dead zone just under the surface, which duals, delay paths and angle beams avoid. That is a physics limit rather than an operator mistake, and it is explained in our dead zone ultrasonic testing guide. For how wear and section size move beams and gates, see rail profile types and ultrasonic inspection.

Key takeaway: list the defect planes you must find, then prove each one has a beam aimed at it.

How do gain and reference errors hide defects?

Gain errors decide whether a real reflector crosses the detection gate, so a few decibels can turn a hit into a miss. A 6 dB drop halves the echo amplitude, not the energy. If the reference reflector is set at 80% full screen height and the scan is then run 6 dB too low, an equal reflector shows at 40%, and with a 50% gate it never alarms. A 4 dB loss brings that 80% echo to about 50%, right at the gate.

ORD-26/07 measured the threshold trade-off in 512 simulated handheld scans. At a flaw of 25% head area, the mean probability of detection was 99.9% with a detection threshold of 0.10, 95.8% at 0.25 and 86.7% at 0.40. The report adds that "too low a detection threshold can result in an unacceptable number of false positives". There is no free setting; the threshold must be chosen and written down, not adjusted by feel.

The common gain and reference mistakes are:

  • Wrong reference reflector: a back-wall echo or a large hole used where the procedure names a smaller reflector, which sets sensitivity too low.
  • Transfer loss ignored: rail or plate with rough or coated surfaces returns less than the smooth block. Apply only the transfer correction the procedure gives; don't invent one.
  • Scanning gain confused with evaluation gain: the code or procedure adds scanning gain above the reference to find indications, then evaluates at reference.
  • Gain changed and not restored: ORD-26/07 notes that operators typically raise gain during operation when looking for flaw edges. Return to the recorded setting before scanning resumes.
  • Settings copied between instruments: dB scales differ, so rebuild the reference on the new unit.

Code systems also differ. ASME practice evaluates against a DAC curve, while AWS D1.1 rates indications as d = a − b − c against a single reference hole. Mixing the two approaches produces wrong ratings. Acceptance levels themselves belong to the code, and are compared in our guide to ultrasonic testing acceptance criteria standards.

Key takeaway: write down the reference reflector, the reference gain, the scanning gain and the gate level, and check all four at the end of the shift.

Why are mode-converted and geometric echoes misread?

Most false calls come from real reflectors that aren't defects, such as bolt holes, rail ends, weld reinforcement and mode-converted waves, plotted where a defect could be. TTCI's TD-16-030 (2016) put it plainly: "Not all indications are from defects, e.g., bolt holes and some welds have distinct ultrasonic signatures that must be distinguished from defects".

The geometric reflectors to rule out in rail are:

  • Bolt holes and rail ends at joints, which return strong angle-beam echoes and interrupt the 0° base echo.
  • Rail welds: TTCI's follow-up digest TD-16-050 notes that "un-ground flashing surrounding the weld can cause reflections that must be distinguished from true defects", and that a healthy flash-butt weld gave a geometry indication in the lower head fillet.
  • Head and web fillets and worn gage corners, which redirect angle beams.
  • Weld cap and root in structural welds, which give echoes at the end of the first leg.

Mode conversion adds a second trap. When a shear wave strikes a boundary at an angle, part of its energy can convert to a longitudinal wave, which travels about 1.81 times faster in steel. An echo carried partly by the faster wave arrives early, and the instrument plots it along the shear beam at a position that doesn't match any real reflector. In coarse-grained austenitic welds, TWI reports "high ultrasonic scattering associated with mode conversion effects, beam distortion".

The fix is the same for every geometric echo:

  1. Compute where the reflector would be from the sound path, angle and section drawing.
  2. Check it against known features: a hole, a weld, a fillet or the end of a leg.
  3. Confirm with a second angle or the opposite direction. TD-16-030 notes that agreement between probes "reduces the likelihood of false alarm".
  4. Hand-verify anything that can't be placed on the drawing.

Rail false-call causes and their costs are covered in depth in our guide to false calls in rail UT, and reading the screen itself in how to read a flaw detector.

Key takeaway: never call an indication you can't place on the section drawing, and never dismiss one for the same reason.

Where do coverage and scan-pattern gaps come from?

A coverage gap is a zone that no beam reaches, and a defect in it can't be found however carefully the rest is tested. Coverage mistakes are easy to miss because the report still shows a complete run.

Rail examples come from the published record:

  • The base. TTCI's thermite weld survey (TD-00-013, July 2000) notes that the ultrasonic rail flaw detection process "only scans the head and web of the rail, and not the majority of the base". FRA's 2014 rail integrity rule preamble describes the same reach: the head and web, "as well as the portion of the base directly beneath the web". In the TTCI survey, slag inclusions and gas porosity made up 40% of reported thermite weld failures, and 56% of those formed mainly at the web-base fillet. Weld and base-zone coverage needs a separate hand technique; see our aluminothermic weld inspection guide.
  • Under shells. In TTCI's 1999 benchmark of six hi-rail systems (TD-99-006), the hit rate for transverse defects under shells was about 62%, against 77% for those with no surface anomaly. The shell sample was only seven defects, so treat the gap as a direction, not a precise figure.
  • Joints. 49 CFR 213.237(f) requires equipment "capable of detecting defects between joint bars, in the area enclosed by joint bars". Skipping joints to save time leaves the zone the rule names.
  • Gage-face wear. Fixed beams on worn heads aim at the wrong volume, as TD-16-030 found.

Weld examples follow the code's scan rules. Structural weld codes typically require scanning from both sides of the weld axis where possible, with more than one angle on thicker sections and an extra scan for transverse cracks. Common gaps are scanning from one side only because of access, indexing more than the procedure's overlap allows, and using too short a scan zone to reach the root on the second leg.

The fix is a written coverage plan. List each zone (head, web, base, fillets, weld root, fusion faces) against the beam that covers it. Any zone with no beam goes in the report as not examined.

Key takeaway: if the coverage plan can't name the beam for a zone, the report must say that zone wasn't tested.

What must happen when a valid search can't be made?

Under 49 CFR 213.237(g), track over which a valid search couldn't be made doesn't count as internally inspected, and 213.237(h) gives the owner three options before the deadline. The decision that a search wasn't valid belongs to the qualified operator.

The rule text reads:

"If the person assigned to operate the rail defect detection equipment (i.e., the qualified operator) determines that a valid search for internal defects could not be made over a particular length of track, that particular length of track may not be considered as internally inspected under paragraphs (a) and (c) of this section."

— 49 CFR 213.237(g)

Before the time or tonnage limit expires, 213.237(h) requires the owner to do one of three things:

Option under 213.237(h)What it means
(1) Conduct a valid searchRetest the stretch, for example after cleaning, drying or hand testing
(2) Reduce operating speed to a maximum of 25 mphApplies to trains over that rail until a valid search can be made; it has nothing to do with test-vehicle speed
(3) Replace the rail that wasn't inspectedRemove the untested rail from service

A valid search is defined in 213.237(j) as "a continuous inspection for internal rail defects where the equipment performs as intended and equipment responses are interpreted by a qualified operator". Under 49 CFR 213.238(b)(4), the qualified operator must determine that each valid search "is continuous throughout the area inspected and has not been compromised due to environmental contamination, rail conditions, or equipment malfunction".

The mistakes here are procedural:

  • Not flagging the no-test. Lost coupling, heavy shelling or an equipment fault is noted in the operator's head but never recorded.
  • Calling a no-test a false call. It is untested rail, not a cleared indication.
  • Letting the deadline pass. The (h) options must be completed before the 213.237(a) or (c) limit, not at the next scheduled test.
  • Assuming masking automatically voids a search. The judgment is the qualified operator's, under 213.238(b)(4) and 213.237(g); 213.237(h) only lists what happens next.

Intervals and the valid-search rule are explained in full in our guide to 49 CFR 213.237 rail flaw detection requirements.

Key takeaway: a no-test is a regulated finding with a deadline, so record its limits the moment it happens.

Which record-keeping mistakes break compliance?

A test that isn't recorded the way 49 CFR 213.241 requires can't be shown to FRA as a test. For internal rail inspections, 213.241(c) lists six items, and 213.241(d) sets how long they are kept.

213.241(c) itemCommon mistake
(1) Date of inspectionOnly the shift start date recorded for a multi-day run
(2) Track inspected, including beginning and end pointsMileposts missing, or track name with no limits
(3) Location and type of defects found under 213.113Defect type left as "indication"
(4) Size of defects, if not removed before the next train movementSize omitted because a remedial action was "planned"
(5) Initial remedial action taken and its dateAction recorded with no date
(6) Location of any track not tested under 213.237(g)No-tests left out entirely

Under 213.241(d), the owner keeps each rail inspection record for at least two years after the inspection and for one year after initial remedial action. Electronic records must meet 213.241(j), which covers unique author identity, no changes to completed records, storage started within 72 hours, and amendments stored separately. Raw instrument files usually don't meet those conditions on their own, so a stored A-scan is supporting data, not the record.

For weld UT, the record is set by the code and contract rather than Part 213. A useful report names the procedure, instrument, probes, reference block, reference and scanning gain, scanned zones, indications with their ratings and the technician's level. Without the setup values, nobody can repeat or audit the test.

Key takeaway: if a reviewer can't rebuild the setup from the report, the report is incomplete.

What good practice can't fix

Avoiding these ultrasonic testing mistakes improves reliability, but it doesn't make UT certain. Several limits remain even with a perfect setup:

  • No method finds every flaw. TTCI's TD-16-030 states that conventional ultrasonic inspection detects "about 90 percent" of rail flaws, leaving about 10 percent undetected during periodic inspection. That figure is from 2016 and depends on system and flaw mix.
  • POD figures are system-specific. ORD-26/07's values come from simulated handheld scans on modeled rail profiles. They don't transfer to test cars or other procedures.
  • Some materials resist UT. Cast manganese frogs are inspected visually in service (RR 22-25), and austenitic welds need special techniques.
  • Masking is physical. A shell blocks the beam from what lies below it, whatever the setting.
  • Codes set the acceptance line, not the test. Finding an indication doesn't decide whether it is rejectable; that comes from 213.113 for rail in track or from the governing code for welds.
  • This article isn't a procedure. Settings, check intervals and reference reflectors come from the written procedure your employer or owner approves, signed off by a Level III.

How EMA Industries applies this

EMA Industries performs ultrasonic rail flaw testing for transit agencies, short lines and industrial track owners, rail weld inspection, and AWS D1.1 structural weld UT, from Southern California with nationwide mobilization. EMA's NDT Level III services cover the parts of this article that sit behind the scan: written procedures, calibration and reference checks, coverage plans and report review. To discuss a rail test, a weld inspection or a procedure review, see our ultrasonic examination services or contact EMA Industries.

Related reading: Ultrasonic rail inspection checklist · Near field vs far field in UT · AWS D1.1 weld inspection criteria

Frequently asked questions

What is the most common ultrasonic testing mistake?

Calibration error is the most widely documented. FRA's ORD-26/07 study found that operators calibrating the same handheld rail equipment produced statistically different results, and one operator was inconsistent across six attempts. Verifying velocity, zero, angle and reference gain on a known reflector before and after each shift is the main defense.

How much depth error does a wrong velocity setting cause?

The depth error is proportional to the velocity error. Leaving the longitudinal velocity set while using a 70° shear-wave probe multiplies every sound path by about 1.81. In the 136RE example above, a transverse defect 0.75 in. deep reads 1.36 in. deep and 1.67 in. too far ahead of the probe.

Is losing couplant contact a false call?

It isn't a false call. A false call is an indication reported as a defect that turns out not to be one. Lost coupling means no sound reached the rail, so that length of track wasn't validly tested. It must be recorded under 49 CFR 213.241(c)(6) and handled under 213.237(h).

What happens if part of the rail couldn't be tested?

Under 49 CFR 213.237(g), that stretch doesn't count as inspected. Before the time or tonnage limit expires, the owner must retest it, reduce train speed to 25 mph or less until it is retested, or replace the untested rail.

Why do bolt holes and welds cause false indications?

Bolt holes, rail ends, weld flashing and fillets are real reflectors, so they return echoes that can sit where a defect could be. TTCI notes that these features have distinct signatures that must be told apart from defects. Locating the reflector on the section drawing and confirming it with a second beam angle resolves most of them.

Sources

← All articles