The Science Behind Ultrasonic Testing: Physics Applied to Rail and Welds

The science behind ultrasonic testing rests on one fact: sound travels through steel at a known speed, about 5,890 m/s for longitudinal waves, and reflects wherever it meets a change in acoustic impedance. A crack's steel-to-air face reflects more than 99.99% of the energy back to the probe. Arrival time gives the depth, and amplitude indicates the reflector's size.

This guide works through that physics in the order an inspector uses it: how the transducer makes sound, which wave types travel in steel, how frequency sets wavelength and resolution, why impedance makes cracks visible, what the near field and beam spread do to the signal, why the signal weakens with distance, and how an A-scan turns all of this into a depth reading. Each section is applied to rail and to welds, with numbers you can check.

What is the science behind ultrasonic testing?

Ultrasonic testing (UT) is a volumetric nondestructive testing method that sends short pulses of high-frequency sound into a part and listens for echoes. Sound above about 20 kHz is ultrasonic; flaw detection in steel usually runs between roughly 0.5 and 10 MHz. Every result an inspector reports comes from a chain of six physical steps, and a weakness at any link limits what the test can find.

The rest of this article takes each link in turn. The useful habit for anyone reading a UT report is to ask which link limits the result. A missed transverse defect under shelling, for example, is usually a coupling and reflection problem (steps 2 and 4), not an instrument fault.

Key takeaway: UT measures time and amplitude; everything else an inspector reports is inferred from those two numbers and the physics below.

How does an ultrasonic transducer make and receive sound?

A UT transducer makes sound through the piezoelectric effect: certain ceramics change shape when a voltage is applied and produce a voltage when squeezed. The instrument's pulser hits the element with a short high-voltage spike. The element rings at its natural frequency, which is set mainly by its thickness (a thinner element rings at a higher frequency). A backing material behind the element damps the ringing so the pulse lasts only a few cycles.

Pulse length matters for two reasons. A short, well-damped pulse separates two reflectors that sit close together (axial resolution), and it shortens the time after the initial pulse during which the probe cannot "hear" anything. That blind zone just below the surface is called the dead zone; EMA's article on the dead zone in ultrasonic testing covers how it is measured. Heavily damped probes give a broader frequency band and better resolution, while lightly damped probes give more energy at one frequency and better penetration.

Three probe styles matter for rail and weld work:

Wheel probes deserve a note because they change the physics. The sound first travels through the fluid inside the tire, then crosses the tire wall and the spray film into the rail. One published rail-testing patent (US 5,777,891) describes its transducers as "typically centered at 2.25 Mhz" and housed in fluid-filled wheel probes. Because the beam enters steel from a liquid, the refraction rules in the angled-beam section below apply, and the shape of the rail head becomes part of the probe's optics.

The same element usually both sends and receives (pulse-echo). Dual-element probes split the jobs between a transmitter and a receiver, which shortens the dead zone for near-surface work. In pitch-catch work, one probe sends and a separate probe receives, which lets an inspector look at paths that a single probe cannot.

Key takeaway: the probe's frequency, element size and damping set the wavelength, near field, beam spread and dead zone before the beam ever reaches the steel.

Which wave types travel in steel, and how fast?

Steel carries four kinds of ultrasonic wave used in testing: longitudinal, shear, surface (Rayleigh) and guided waves. Each moves the steel's particles differently, travels at a different speed, and suits different flaws. Velocity is the most important number in UT because every depth reading is velocity multiplied by time.

Velocities are from Evident's material sound velocity table, which itself warns that the values are "only a general guide." The surface-wave figure is calculated from the shear velocity with a standard approximation and is rounded. Actual rail steel, weld metal and heat-affected zones differ by a percent or two, which is why procedures calibrate velocity on a reference block rather than typing in a textbook value.

Longitudinal waves

Longitudinal waves are the fastest and the only type that travels through liquids and gases as well as solids. That is why every couplant and every wheel-probe fluid carries longitudinal energy, even when the beam inside the steel is shear. A 0° longitudinal beam in rail runs from the running surface down through the head, the web and the base. It finds flaws that lie roughly parallel to the running surface, and it watches for a drop in the base echo, which signals that something in the path is blocking the sound.

Shear waves, in brief

Shear waves move about 55% as fast as longitudinal waves in steel, so at the same frequency their wavelength is shorter and they resolve smaller features. They cannot travel in liquids, because a liquid has no shear stiffness. In practice they are produced by mode conversion: a longitudinal wave hits the steel surface at an angle and part of its energy converts. EMA covers their generation, angles and weld applications in detail in shear wave ultrasonic testing. This article only uses the velocity.

Surface waves

Rayleigh waves travel along the surface and decay with depth, so most of their energy lies within about one wavelength of the surface. They follow gentle curves and stop at sharp corners. On rail they can, in principle, show surface-breaking cracks such as head checks, but surface roughness, couplant and wheel damage also reflect them. EMA's article on surface wave UT for rail discusses where that works.

Guided waves, and why "Lamb waves" is the wrong word for rail

Guided waves are waves whose shape is set by the boundaries of the part, so the whole cross-section vibrates together and the wave can travel a long way. Lamb waves are one specific family of guided waves: those in a flat plate with two free surfaces, with symmetric and antisymmetric modes. A rail is not a plate. Its head, web and base form a complex waveguide that carries many modes, each with its own speed at each frequency. Researchers model those modes with semi-analytical finite element (SAFE) methods; for example, Setshedi, Loveday, Long and Wilke (Ultrasonics, 2019) fitted SAFE dispersion curves to measured rail signals and saw three propagating modes. Calling rail guided waves "Lamb waves" is therefore imprecise. The web of a rail behaves somewhat like a plate at some frequencies, but the rail as a whole does not.

"Guided wave UT" and "long-range UT" are also the same method under two names, not competing techniques. EMA's guided wave vs long-range ultrasonic testing article explains the naming.

Worked example: velocity from elastic constants

Velocity is not arbitrary; it follows from stiffness and density. For a longitudinal wave in a large solid:

v_L = √[E(1 − ν) / (ρ(1 + ν)(1 − 2ν))], and for shear, v_S = √(G/ρ), where G = E / (2(1 + ν)).

• (1 − ν) / ((1 + ν)(1 − 2ν)) = 0.71 / (1.29 × 0.42) = 1.310
• v_L = √(207 × 10⁹ × 1.310 / 7,850) = 5,878 m/s
• G = 207 / 2.58 = 80.2 GPa, so v_S = √(80.2 × 10⁹ / 7,850) = 3,197 m/s

These land within 0.2% (longitudinal) and 1.6% (shear) of the tabulated 5,890 and 3,250 m/s. The gap is the point: small differences in composition, heat treatment and texture move the velocity, so a 1.6% velocity error becomes a 1.6% depth error unless the instrument is calibrated on the material or a block of matching velocity.

Key takeaway: know which wave type you are using, because depth = velocity × time, and shear is about 55% the speed of longitudinal.

How do frequency and wavelength set what UT can find?

Wavelength equals velocity divided by frequency (λ = v / f), and wavelength controls both the smallest flaw a beam can reliably show and how far the sound can penetrate. Higher frequency means a shorter wavelength, better sensitivity to small reflectors and better resolution, but more scattering and less penetration.

Worked example: wavelengths in steel

Using v_L = 5,890 m/s (0.2320 in/µs) and v_S = 3,250 m/s (0.1280 in/µs):

Check one row: 5,890 m/s ÷ 2,250,000 Hz = 0.002618 m = 2.62 mm, and 0.2320 in/µs ÷ 2.25 cycles/µs = 0.1031 in.

A common training rule of thumb says a discontinuity should be larger than about half a wavelength to have a reasonable chance of detection. Treat that as a teaching guide, not a code requirement: orientation, shape, surface condition and the gain used matter as much as size. By that rule, a 2.25 MHz longitudinal beam in steel is comfortable with reflectors above about 1.3 mm, and a 2.25 MHz shear beam with reflectors above about 0.7 mm.

Why rail testing doesn't simply use 10 MHz

Higher frequency costs penetration. Grain scattering rises steeply with frequency (see attenuation below), and a test car has to send sound from the running surface to the base of a 7-5/16 in. rail and back. The near field also grows with frequency, and the beam becomes narrower, so each probe covers less of the head. Rail systems therefore sit in the low-megahertz range; the 2.25 MHz figure in the patent above is one published example, and one published hand-pushed rail cart (the UDS2-77) lists 2.5 MHz probes for the head and a 4 MHz probe for the web. Joint bar research by MxV Rail found 0.5 MHz worked best for that specific geometry, which shows how strongly the part decides the frequency.

Frequency, bandwidth and resolution

A probe does not emit one pure frequency; it emits a band around its center frequency. A broadband (heavily damped) pulse is short in time, which improves resolution between two close reflectors and near the surface. A narrowband pulse carries more energy at its center frequency and penetrates better, at the cost of ringing longer. Procedures specify the nominal frequency, and good practice also checks the actual frequency and pulse shape when probes are purchased or after damage, because a worn or chipped element can drift.

Key takeaway: choose frequency for the part, not for the smallest possible flaw; in thick or coarse-grained steel, penetration wins over resolution.

What is acoustic impedance, and why do cracks reflect sound?

Acoustic impedance (Z) is a material's density multiplied by its sound velocity, and the difference in Z across a boundary decides how much sound reflects there. Steel and air differ so much that a crack, which is a thin layer of air inside steel, reflects almost all of the sound that hits it squarely. This is the single most important reason UT finds cracks.

Z = ρ × v, usually stated in megarayls (MRayl, 10⁶ kg/m²·s).

Velocities for steel, acrylic and water are Evident's; densities and the air values are standard handbook figures. The steel row uses 1020 steel, so treat the third significant figure as approximate for rail steel.

The reflection and transmission equations

At a flat boundary hit straight on (normal incidence), going from material 1 into material 2:

• Pressure reflection coefficient: R = (Z₂ − Z₁) / (Z₂ + Z₁)
• Fraction of energy reflected: R²
• Fraction of energy transmitted: 1 − R²

A negative R means the reflected pressure wave is inverted (phase reversed), which happens whenever sound goes from a high-impedance material into a low-impedance one, such as steel into air. Some instruments can display that phase, but most rail and weld work reads the rectified amplitude only.

Worked example: impedance and reflection at three boundaries

Check the middle row: −44.76 ÷ 47.72 = −0.938; squared, 0.880; so 12.0% of the energy crosses, and 10 × log₁₀(0.120) = −9.2 dB.

Three practical lessons come out of this table:

1. An open crack is an almost perfect reflector. Only about 4 parts in 100,000 of the energy cross a steel-air face. If the beam hits the crack squarely, the echo is strong.
2. Couplant exists because air is worse than anything. A water film lets 12% of the energy through a steel-water boundary; an air gap lets through almost none. That is why a dry spot, a lifted wheel probe or a skipped spray turns into a gap in coverage. Under 49 CFR 213.241(c)(6), track that was not tested has to be recorded as such; it is not a "false call." EMA's article on ultrasonic couplant for rail inspection covers coupling in the field.
3. Round trips cost energy twice. In immersion or wheel-probe testing, the sound crosses the fluid-steel boundary going in and again coming back. With the pressure transmission coefficients 2Z₂/(Z₁ + Z₂) for each crossing, water → steel → water gives 1.938 × 0.062 = 0.120 of the pressure, or −18.4 dB, before any flaw or attenuation is counted.

What reflects less than an open crack

Not every flaw is an air gap. A crack that is pressed shut by compressive stress, such as a fatigue crack in the rail head under the residual stresses of a worked running surface, can transmit part of the sound across its faces and reflect less than its size suggests. Slag and other nonmetallic inclusions have impedances between steel and air, so they reflect a smaller fraction. Rough, branched or tilted crack faces scatter sound in many directions, and only a fraction comes back to the probe. These are the physical reasons a small, well-oriented reflector can echo louder than a large, badly oriented one, and why amplitude alone does not size a flaw.

Key takeaway: reflection depends on the impedance contrast and on orientation; an air-filled crack hit squarely returns almost everything, a tight or tilted one much less.

What happens when sound crosses a boundary at an angle?

When sound crosses a boundary at an angle, it bends (refracts) according to Snell's law and can convert between longitudinal and shear modes. Angle-beam testing is built on this: a wedge or a fluid path sets the incident angle so that a chosen shear-wave angle, such as 70°, enters the steel.

Snell's law: sin θ₁ / v₁ = sin θ₂ / v₂.

Between the two critical angles, only a shear wave enters the steel, which gives a clean single-mode beam. With an acrylic wedge, incident angles of about 36.4°, 46.7° and 52.1° give 45°, 60° and 70° shear in steel. The geometry of skip distances, sound paths and weld coverage belongs to angle-beam technique, which EMA covers in angle beam testing in UT. The point here is the physics of the boundary.

Worked example: why worn rail moves the beam

Rail test cars send the beam from a fluid-filled wheel into the curved head, so the fluid-to-steel boundary is the rail's running surface. That surface tilts as the head wears. TTCI put it plainly in a 2016 research digest: on worn rail, "the wrong volume of material is inspected."

Assume a fluid velocity of 1,480 m/s (water-like; actual wheel fluids differ) and a design target of 70° shear in steel (3,250 m/s).

• Required incident angle in the fluid: sin θ₁ = (1,480 / 3,250) × sin 70° = 0.4554 × 0.9397 = 0.4279, so θ₁ = 25.3°
• If the worn surface tilts the boundary so the effective incidence falls by 2° (23.3°): sin θ₂ = sin 23.3° × 3,250 / 1,480 = 60.4° refracted, not 70°
• If the effective incidence rises by 2° (27.3°): it passes the 27.1° second critical angle, and no shear beam enters the steel at all

A 2° change in the running-surface angle can therefore swing a nominal 70° beam by about 10°, or switch it off. This is why rail systems use multiple probes and why profile wear compensation is an active research area. Contact wedges on flat plate don't suffer this, which is one reason structural weld UT is more repeatable than dynamic rail testing.

Key takeaway: refraction is very sensitive near the critical angles, so the shape of the entry surface is part of the beam.

What is the near field, and why does it matter?

The near field is the zone in front of a transducer where the sound pressure rises and falls through a series of maximums and minimums; it ends at the last on-axis maximum, a distance N from the probe face. Evident's beam characteristics tutorial gives the formula for a circular element as N = D²f / 4c (equivalently D² / 4λ), and notes that amplitude-based flaw evaluation is difficult inside it, while thickness gauging there "is not a problem."

Beyond N lies the far field, where the beam pressure falls off smoothly with distance and the beam spreads at a steady angle. N is also the farthest point at which an unfocused probe's beam can be focused. EMA's near field vs far field in UT article covers the measurement side.

Worked example: near field length in steel

N = D²f / (4v), using v = 5,890 m/s = 5.89 mm/µs and f in MHz (cycles per µs), with D in mm.

As a cross-check against published data, the same formula for a 1.0 in., 2.25 MHz probe in water (1,480 m/s) gives 9.65 in., which matches the nominal value Olympus/Evident publishes for that probe in water.

Two patterns stand out. Doubling the diameter quadruples N, and raising frequency raises N in proportion. A large, high-frequency probe can have a near field deeper than the flaw you are looking for.

What the near field means on rail and welds

On a contact probe, the near field starts at the steel surface, so shallow reflectors fall inside it and their amplitude is unreliable. On angle-beam probes, part of the near field is used up inside the wedge, which helps. In immersion and wheel probes, the fluid path can hold some or all of the near field, which is one advantage of the design. When a procedure sets a reference level on a hole at a shallow depth that falls inside the near field, the reference amplitude can be misleadingly high or low; this is one reason distance-amplitude correction is built on several holes at different depths. EMA's DAC curve in ultrasonic testing article covers that construction.

Key takeaway: N = D²f / 4v; inside it, depth readings are fine but amplitude comparisons are not.

How much does an ultrasonic beam spread?

In the far field the beam spreads at an angle set by wavelength and element size: sin(α/2) = K × v / (f × D). Larger elements and higher frequencies give narrower beams. K depends on which edge of the beam you mean; for the −6 dB edge of a pulse-echo beam, a commonly published value is 0.514. Treat K as convention-dependent and check the one your procedure or probe data sheet uses.

Worked example: beam spread and beam width in rail

The width column uses 2 × distance × tan(α/2) at 7.31 in., the height of a 136RE rail (7-5/16 in.). It is a far-field approximation that ignores the element's own width and the near-field shape, so read it as order of magnitude.

The rail application is the interesting part. A 136RE web is only about 11/16 in. thick, while a 0.5 in., 2.25 MHz beam is roughly 1.5 in. wide by the time it reaches the base. Much of the beam's edge strikes the underside of the head and the web fillets rather than going straight down the web. That creates geometry echoes, and it is one reason base and web coverage from the running surface is incomplete. Bolt holes, which sit in the web, give strong, characteristic echoes for the same reason.

Beam spread also makes small flaws look bigger. As the probe moves, the flaw stays in the beam for a distance roughly equal to the beam width plus the flaw size, so the "length" over which an echo appears overstates the flaw unless the inspector sizes it with a defined drop method, such as the 6 dB drop, which halves the amplitude.

Key takeaway: beam spread decides coverage and how much of the rail's geometry the beam sees; bigger and higher-frequency probes give tighter beams.

Why does the signal weaken with distance?

Ultrasonic signals weaken with distance for three separate reasons: beam spread, material attenuation (scattering plus absorption), and losses at boundaries. Beam spread spreads the same energy over a wider area. Attenuation removes energy as the wave travels. Boundary losses, covered in the impedance section, happen at every interface the wave crosses. A UT procedure has to account for all three.

Material attenuation is usually written as an exponential decay of amplitude, A = A₀e^(−αz), or in decibels per unit distance. It has two parts:

The fourth-power rule is why frequency choice is so sensitive in coarse-grained steel. Moving from 2.25 MHz to 5 MHz multiplies Rayleigh-regime scattering by (5 / 2.25)⁴ = 24.4. In fine-grained rolled rail and plate, this hardly matters over a few inches. In cast or coarse-grained material it dominates.

Where attenuation decides the result

• Cast austenitic manganese steel frogs and crossing diamonds. FRA research report RR 22-25 describes the material as "highly attenuative" and coarse-grained, with a mean grain size near 0.94 mm, and reports that revenue-service examination of these castings uses only visual inspection. Low frequency (trials at 500 kHz and 2.25 MHz) helps penetration at the cost of resolution.
• Thermite (aluminothermic) rail welds. The cast weld metal has a coarser, more columnar structure than rolled rail, which raises noise and attenuation through the weld. EMA's aluminothermic weld inspection article covers how those welds are examined.
• Structural welds over long sound paths. AWS D1.1 builds a simple attenuation allowance into its indication rating (c = 2 dB per inch of sound path beyond the first inch). Holloway (CINDE Journal, 2017) showed this approximation can be about 6 dB off at 60° and 70° with long sound paths, which is enough to change an indication's class.

Transfer correction

Reference blocks and the test part rarely couple identically. Surface roughness, curvature, paint and mill scale all change how much sound gets in. A transfer correction compares the same signal, often a back-wall echo or a pitch-catch path, on the block and on the part, and adjusts the gain by the difference. Rail running surfaces vary from freshly ground to heavily shelled, so coupling losses on rail can change from foot to foot. The right response is a procedure-defined correction, not a guess.

Key takeaway: not all lost signal is attenuation; separate beam spread, material loss and coupling loss before deciding a reflector is small.

How do you read an A-scan?

An A-scan is a graph of echo amplitude (vertical axis) against time or distance (horizontal axis) at one probe position. It is the basic display on every conventional UT instrument, and every other display is built from A-scans. Depth comes from where an echo sits on the horizontal axis, and relative reflector strength from its height.

The depth equation

For a pulse-echo straight beam: depth = v × t / 2, where t is the round-trip time. The division by two is the most commonly forgotten step.

For an angle beam: sound path = v_S × t / 2, depth = sound path × cos θ, and surface distance = sound path × sin θ. Those are given here only to read the A-scan; skip distances and multiple legs are covered in EMA's angle-beam article.

Worked example: time of flight in a 136RE rail

Read the second row as an inspector would. An echo at 1.16 in. depth that arrives well before the 63.0 µs base echo, and that comes with a drop in the base echo, is consistent with a reflector lying roughly parallel to the running surface, such as a horizontal split head (which typically lies at least 1/4 in. below the running surface). It is a suspect, not a defect, until it is verified. EMA's horizontal split head article covers that defect.

Amplitude in decibels

UT instruments measure amplitude as a ratio in decibels: dB = 20 × log₁₀(A₁ / A₂). A 6 dB change doubles or halves the amplitude (not the energy). An echo at 80% of full screen height compared with one at 20% is 20 × log₁₀(4) = 12 dB stronger. Gain settings, reference levels and the "6 dB drop" sizing method all use this scale. EMA collects the related formulas in ultrasonic testing formulas.

What B-scans and C-scans are, and what rail test cars actually show

A B-scan stacks A-scans along a line of probe positions, showing depth against position, like a side view through the part. Rail test car and cart software commonly presents data in this form, often as a "B-scan" style map of each channel's gated hits along the rail, with A-scans available for detail. A C-scan is a plan view: amplitude or depth across an area, built from encoded raster scanning or a phased array. C-scans are routine in immersion tanks, composite inspection and encoded weld scanning. They are not the standard display on revenue-service rail test cars, because a test car scans along fixed beam paths rather than rastering over the rail's area. EMA's UT fundamentals: A-scan, B-scan and C-scan article explains each display.

Key takeaway: depth = v × t / 2; the A-scan's time axis gives location, and its amplitude axis, read in dB, gives comparison, not size.

How does the physics apply to rail?

Rail testing applies the same physics to a fixed, complicated shape: a head, a thin web and a wide base, tested from the running surface while moving. Each probe angle exists because a different defect orientation needs the beam to strike it squarely.

These beam sets are typical industry practice, not a federal specification; one published system uses 0°, 38° and 70° probes, and specific systems differ. The 70° beam is used for transverse defects because a transverse crack stands roughly vertical across the head, and a steep shear beam meets it close to face-on. A straight 0° beam would run parallel to that crack and miss it. EMA's how ultrasonic rail flaw detection works article walks through the test car and cart process, and the transverse fissure and vertical split head articles cover those defects.

Four rail-specific physics problems

1. Moving coupling. A wheel probe at speed has to keep a fluid film under its tire. Rust, grease, sand, ice and corrugation all break the film. A drop in the base echo on the 0° channel is often the first sign of lost coupling, and the system or the operator has to decide whether it was coupling or a defect.
2. Surface damage at the entry point. Shelling, spalling and head checks are rough reflectors right where the sound enters. They scatter the beam before it reaches a defect underneath. That is the physical cause of "masking," and EMA's rail shelling and spalling article covers those conditions.
3. Geometry echoes. Bolt holes, joint gaps, weld collars and the fillets under the head all reflect sound. They appear at predictable times and positions, so experienced analysts and software use those signatures to separate geometry from defects. TTCI notes that "Not all indications are from defects."
4. Temperature. Sound velocity in steel changes slightly with temperature, but the larger effect is usually in wedge plastics and probe fluids, whose velocity changes more and shifts refracted angles. EMA's cold weather UT testing article covers field effects.

Key takeaway: each rail probe angle is a physics answer to a defect orientation, and the rail's surface condition decides whether the sound gets in at all.

How does the physics apply to rail welds and structural welds?

Welds add three complications: changed grain structure, extra geometry, and planar flaws at fusion faces that may not face the beam. The physics is the same; the reflectors and the noise change.

Rail welds

Rail strings are joined in the plant by flash-butt welding and in the field mostly by thermite welding. Thermite welds carry a collar of excess metal, a coarser cast structure, and a profile change at the base. TTCI digest TD-00-013 found that test cars miss most of the rail base, so field welds need hand UT of the base, and that slag and porosity accounted for 40% of the thermite weld service failures it examined. Flash-butt welds are fine-grained but can contain flat "cold weld" or lack-of-fusion regions that lie in the plane of the joint, roughly perpendicular to the rail axis. For those planes, a beam traveling along the rail meets the defect face-on, while a 0° beam from the top runs parallel to it.

Structural steel welds (AWS D1.1)

Structural weld UT uses contact wedges at 45°, 60° and 70° on flat or gently curved plate, with no moving fluid path, which makes the beam more repeatable than on rail. Lack of sidewall fusion lies on the bevel face, so the best detection comes when the beam meets that face close to perpendicular; as a rule of thumb, the probe angle is about 90° minus the bevel angle (a 30° bevel suggests a 60° probe). AWS D1.1 rates indications with d = a − b − c, where c is the 2 dB-per-inch attenuation term discussed above. The classes depend on thickness, probe angle and loading, so there are no fixed dB bands. EMA's AWS D1.1 weld inspection criteria article covers the acceptance side; check the edition named in your contract.

Key takeaway: in welds, orientation and grain structure decide detection more than flaw size does.

Which federal rules govern ultrasonic rail testing?

For the general railroad system, FRA's Track Safety Standards in 49 CFR Part 213 set when rail must be internally tested (213.237), who may operate the equipment (213.238), what to do about a defect (213.113), the optional continuous-testing alternative (213.240) and the records (213.241). The rules regulate outcomes and procedures, not the physics: they do not specify frequency, probe angles or display type.

Class 6 and higher track falls under Subpart G instead (213.339 requires internal testing at least twice a year). EMA explains the interval rules in 49 CFR 213.237 rail flaw detection inspection requirements and rail testing frequency requirements.

What counts as a valid search

The physics in this article matters legally through one definition. Section 213.237(j) defines it:

"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)

Lost coupling, a masked head or a misread geometry echo can all mean the equipment did not perform as intended over that stretch. Rail that was not validly searched doesn't count toward the interval, and before the deadline the owner must re-search it, run at no more than 25 mph, or replace the rail (213.237(h)). For continuous testing, 213.240(e)(4) adds: "A suspect location is not considered a defect under § 213.113(c) until it has been field-verified by a person qualified under § 213.238."

What the 2014 Rail Integrity rule actually did

FRA's final rule "Track Safety Standards; Improving Rail Integrity" (79 FR 4234, January 24, 2014, effective March 25, 2014) implemented section 403 of the Rail Safety Improvement Act of 2008. In FRA's words, it established "minimum qualification requirements for rail flaw detection equipment operators" and revised "requirements for effective rail inspection frequencies, rail flaw remedial actions, and rail inspection records." Specifically, it:

• replaced the old fixed intervals (40 MGT or annually on Class 4–5 track) with performance-based service failure rate limits and the 370-day/30 MGT outer limits
• defined a valid search and set the options for track not validly searched
• created the 213.238 qualified operator rule, under which the employer qualifies the operator
• added a 4-hour deferred-verification period and revised parts of the remedial action table
• added testing requirements for plug rail

It did not mandate any particular UT technology, and it did not create an FRA technician certification. Continuous rail testing (213.240) came later, in a 2020 rule.

Rail transit systems not connected to the general railroad system are excluded from Part 213 (213.3(b)(3)) and fall under FTA rules and state safety oversight; in California, that is the CPUC.

Key takeaway: 213.237 says when, 213.238 says who, 213.113 says what to do; the 36/84-hour windows belong only to 213.240.

What are the limits of ultrasonic testing?

Ultrasonic testing can only report reflectors that the beam reaches, that face it closely enough to echo back, and that stand out from noise. Every limit below follows from the physics above, and an engineer quoting UT results should know them.

Overall, TTCI's digest TD-16-030 (2016) described conventional ultrasonic inspection as finding "about 90 percent" of rail flaws, with the remainder escaping periodic inspection because of inherent limitations. That figure is an industry research estimate for conventional detector cars, not a guaranteed probability of detection for any one test. No federal rule requires 100% detection; the rules require valid searches, intervals and service failure rates.

Key takeaway: treat any UT result as "no reportable reflector found by these beams on this day," not as proof the steel is flawless.

Where do PAUT, guided waves and software fit today?

Phased array UT (PAUT), guided wave UT and automated signal analysis are real tools, but for revenue-service rail testing in North America the standard practice is still conventional multi-probe UT from test cars and carts, with hand verification.

• PAUT steers and focuses a beam electronically from many small elements and builds sectorial or C-scan images. It is established for weld inspection, and AWS D1.1:2020 added a PAUT annex (check the edition in your contract). In rail, it remains mainly a verification and research tool. TTCI tested a head-contact PAUT prototype at up to 20 mph in 2016 (TD-16-050) that found transverse defects under shells and weld and web defects but could not inspect the base flanges. EMA's phased array ultrasonic testing article explains the method.
• Guided wave (long-range) UT is established for pipelines. In rail it is mostly research, including the SAFE modeling work cited above and broken-rail monitoring. It screens for large changes in section; it does not replace an internal rail test under 213.237.
• Automated analysis software flags candidate indications on test car data so analysts can review them. Vendors make claims about pattern recognition and fewer false calls; treat those as vendor claims unless an independent report backs them. Under 213.237 and 213.240, equipment responses still have to be interpreted, and suspects verified, by qualified people.

Key takeaway: new methods change how data is gathered and displayed, not the physics; ask what each method has shown in independent rail testing before buying it.

How EMA Industries applies this

EMA Industries performs ultrasonic rail flaw testing for transit agencies, short lines and industrial track owners, along with rail weld inspection and AWS D1.1 structural weld UT, MT and PT. The physics in this article is what drives the field decisions: probe frequency and angle for each defect type, coupling checks on worn or contaminated rail, and hand UT of weld bases that test cars don't reach. EMA's NDT Level III services cover written procedures, calibration practice and technique review. EMA Industries is based in Southern California and mobilizes nationwide; to discuss a rail or weld inspection scope, contact EMA Industries.

Frequently asked questions

What is the basic principle of ultrasonic testing?

Ultrasonic testing sends short pulses of high-frequency sound into a part and measures the echoes. Because sound travels at a known velocity in steel (about 5,890 m/s for longitudinal waves), the echo's round-trip time gives the reflector's depth (depth = v × t / 2). Reflections happen wherever the acoustic impedance changes, so a crack, inclusion or back wall each return a measurable echo.

Why do cracks reflect ultrasound so strongly?

A crack is a thin gap of air inside steel, and steel's acoustic impedance (about 46 MRayl) is more than 100,000 times that of air. Using R = (Z₂ − Z₁) / (Z₂ + Z₁), more than 99.99% of the energy reflects at an open steel-air face. The echo is strong only if the beam strikes the crack roughly face-on; a crack lying parallel to the beam, or one pressed shut by stress, returns much less.

What frequency is used for ultrasonic rail testing?

Rail testing uses low-megahertz frequencies because the sound must cross the full rail height and back. One published test car patent describes wheel-probe transducers "typically centered at 2.25 Mhz," and one published hand-pushed cart lists 2.5 MHz and 4 MHz probes. At 2.25 MHz the longitudinal wavelength in steel is about 2.6 mm (0.10 in.), which balances resolution against penetration.

How often does FRA require rail to be tested ultrasonically?

Under 49 CFR 213.237, Class 4 and 5 track, and Class 3 track with regularly scheduled passenger trains or a hazmat route, must be internally inspected at least every 370 days or 30 MGT, whichever comes first. Other Class 3 track is tested annually or every 30 MGT, whichever is longer, with a 5-year cap. The rule does not require weekly or monthly testing, and 213.237(c) sets no interval for Class 1 and 2 track.

Does FRA certify ultrasonic rail flaw technicians?

FRA does not certify them. Under 49 CFR 213.238, the employer qualifies each operator of rail flaw detection equipment through a documented training program, written authorization, at least 160 hours of supervised experience and periodic reevaluation. In general NDT, ASNT's SNT-TC-1A is likewise a recommended practice under which the employer certifies its own personnel.

Sources

• 49 CFR 213.237 Inspection of rail (eCFR); text read on Cornell LII
• 49 CFR 213.238 Qualified operator (eCFR)
• 49 CFR 213.113 Defective rails (eCFR)
• 49 CFR 213.240 Continuous rail testing (eCFR); text read on Cornell LII
• 49 CFR 213.241 Inspection records (eCFR)
• FRA, Track Safety Standards; Improving Rail Integrity, final rule, 79 FR 4234 (January 24, 2014), govinfo PDF
• Evident, Material sound velocities
• Evident, Flaw detection tutorial 2.4: Beam characteristics
• Setshedi, Loveday, Long and Wilke, "Estimation of rail properties using semi-analytical finite element models and guided wave ultrasound measurements," Ultrasonics 96 (2019) 240–252
• US Patent 5,777,891, real-time ultrasonic rail testing system (wheel probes, 2.25 MHz)
• UDS2-77 hand-pushed rail flaw detector specification (manufacturer page; vendor claim)
• TTCI Technology Digest TD-16-030 (Witte and Poudel, 2016), phased array rail inspection
• TTCI Technology Digest TD-16-050 (2016), head-contact PAUT prototype
• TTCI Technology Digest TD-99-006 (Garcia and Reiff, 1999), hi-rail detection evaluation
• TTCI Technology Digest TD-00-013 (2000), thermite weld failures
• FRA RR 22-25, ultrasonic inspection of special trackwork (ROSA P dot/64412)
• FRA DOT/FRA/ORD-26/07, MAPOD for handheld rail UT (ROSA P dot/92816)
• TSB Canada rail investigation report R19M0018
• Holloway, P., AWS D1.1 attenuation factor, CINDE Journal, 2017