Automated Track Inspection: Geometry Cars, Rail Flaw Testing and the FRA Rules

Automated track inspection is not one system. It is three: track geometry cars that measure gage and alignment (required by 49 CFR 213.333 only on Class 6 and faster track), rail flaw test vehicles that search for internal defects under 49 CFR 213.237, and camera and sensor systems that supplement the visual inspection required by 49 CFR 213.233. None of them replaces the inspector, and only one of them finds a crack inside the rail.

Most writing about automated track inspection blurs that distinction. This guide separates the three jobs, gives the rule and the real operating limits for each, and works through what happens when a geometry car and a rail flaw car each report a problem on the same stretch of Class 4 track.

What is automated track inspection, and what are its three different jobs?

Automated track inspection is any inspection where a vehicle-mounted or fixed system measures the track and produces an exception list, instead of a person judging the condition by eye or hand. In US practice the term covers three jobs that use different physics, run at different speeds and answer to different sections of 49 CFR Part 213.

A fourth system, the gage restraint measurement system (GRMS), applies a lateral load and measures how far the gage widens under it. It is required only on Class 8 and 9 track under 213.333(h), and a track owner may elect to use it on lower classes under 49 CFR 213.110. FRA's own inspection cars carry it alongside geometry and machine vision.

Key takeaway: when someone says "automated track inspection", ask which of the three jobs they mean. The rule, the speed and the limits change with the answer.

Which FRA rules apply to automated track inspection?

No FRA rule requires a railroad to automate inspection in general. Part 213 names automated systems in exactly three places, and each one is narrow.

Rail flaw testing sits under a different set of sections. 213.237 sets the interval and the valid-search condition, 213.238 qualifies the operator, 213.240 allows continuous (non-stop) testing with later verification, and 213.113 says what must be done about a defect. None of these names a method. The word "ultrasonic" does not appear anywhere in Part 213; the rule requires "internal rail inspections" with equipment that can detect defects between joint bars, and ultrasonic testing is the industry's way of meeting it.

Two consequences follow. First, a claim that "FRA mandates automated track inspection" is wrong; the only mandated automated system on Class 1–5 track is the concrete-tie rail seat inspection in 213.234, and only for the larger railroads. Second, every automated exception is an indication until a person confirms it: a 213.7 designated person for geometry and rail seat exceptions (213.333(f), 213.234(e)), and a 213.238 qualified operator for rail flaw indications (213.240(d)(3)).

FRA said the same thing in its October 2024 proposed rule on geometry inspection:

"FRA notes that TGMS is not a substitute for visual track inspections, which inspect for numerous conditions aside from track geometry and remain essential to ensuring railroad safety." — FRA, Track Safety Standards; Track Geometry Measurement System Inspections, NPRM, 89 FR 84845, October 24, 2024

Key takeaway: Part 213 permits automation almost everywhere and requires it in two narrow cases; nowhere does it let a machine's exception stand without a person's verification.

What does a track geometry car measure?

A track geometry measurement system measures the shape of the track against the limits in Subpart C of Part 213, and reports every place where the track fails the limit for its class. The parameters are the ones a track inspector would otherwise check with a gage bar, a string line and a level, measured continuously along the rail instead of at spot locations.

The 213.333(b) design requirements tell you what a qualifying system has to be. Measurements must be taken no more than 3 ft from the contact point of wheels carrying at least 10 kips of vertical load, so the track is measured under load rather than unloaded. Sampling is distance-based, preferably every foot and never more than 2 ft. Repeat runs over the same site at the same speed may not differ by more than 1/8 in. Within 24 hours the system must produce a continuous plot and an exception report, and the report must carry enough location detail "which enable field forces to easily locate indicated exceptions" (213.333(e)).

Speed is the geometry car's advantage. The sensors are optical and inertial, nothing has to touch the rail head with a coupled probe, so the system runs at whatever speed its carrier runs. FRA's 2024 proposed rule describes autonomous systems (ATGMS) as "mounted to on-track equipment (in some cases revenue trains)" with inspections "conducted with minimal direct human involvement (e.g., uncrewed operations)". A sensor pack under a revenue freight car measures geometry at that train's speed, every time the train passes.

Key takeaway: a geometry car measures gage, alignment, elevation and surface under load at track speed, to 1/8 in. repeatability; it measures nothing inside the rail.

What is FRA's own Automated Track Inspection Program?

FRA runs its own fleet of inspection cars under the Automated Track Inspection Program (ATIP), and uses them to survey the railroads it regulates. In the October 2024 proposed rule FRA wrote that it "runs a fleet of track inspection cars under its ATIP, conducting compliance surveys on over 150,000 miles of track annually" (89 FR 84845). The cars measure track geometry and gage restraint and carry machine-vision cameras; FRA's research summary RR 21-20 says the program "uses machine vision inspection technology along with Track Geometry Measurement and Gage Restraint Measurement Systems" to evaluate track (FRA RR 21-20, September 2021).

The machine-vision side is newer. Under RR 21-20, FRA and ENSCO developed algorithms that flag possible defects such as fasteners in the images collected during ATIP surveys, and software that automatically extracts high-resolution images of switches, frogs, geometry defects and track with reduced lateral strength for FRA's own inspectors to review. The report calls the first iteration "promising" and a basis for automated evaluation of switch components. It describes a tool for FRA inspectors, not a certified replacement for them.

The same NPRM records that, as of 2024, "every Class I railroad uses some form of TGMS to measure track geometry", and lists the technologies now in use on US track: TGMS, ground penetrating radar, track imaging systems, ultrasonic rail flaw detection systems, machine learning-based component inspection, vertical track deflection systems and LiDAR 3-D scanning. That list is FRA's description of what railroads use; it is not a list of what any rule requires.

The figures above come from the Federal Register notice and the ROSA P research record; FRA's own program page could not be read for this article.

Key takeaway: FRA's ATIP cars survey more than 150,000 track-miles a year for compliance with geometry, gage restraint and visible component conditions; they do not test rail internally.

Can automated geometry measurement replace visual inspections?

Not under the current rule, and not without a waiver. Since 2018 FRA has let individual railroads test whether an automated geometry system plus fewer visual inspections is as safe as the 213.233 schedule, and the history of those programs is the clearest record of what automation can and cannot be credited with.

Two points stand out from that record. The proposed rule would add a geometry inspection requirement; it does not reduce the 213.233 visual schedule. And the only railroads running fewer visual inspections are doing so under a waiver or test program, each with conditions, extra metrics and an expiry date. As of November 2026 no final rule on 213.236 had been found on eCFR, and the outcome of the AAR petition was not confirmed for this article; check the dockets before quoting the current status.

The safety case the railroads make is real but has to be read with its qualifiers. In the Fifth Circuit case, the court recited BNSF's own observation that manual inspections had detected 0.01 defects per 100 miles against 4.54 using the automated technology. That figure is BNSF's count of geometry defects on its test territory, reported in litigation; it says nothing about internal rail defects, which neither method in that comparison was looking for.

Key takeaway: automated geometry measurement has earned waivers, a court win and a proposed rule, but as of late 2026 it has not replaced visual inspection for any railroad outside a waiver or test program.

How is automated rail flaw testing different from geometry measurement?

Automated rail flaw testing searches the inside of the rail with ultrasonic beams carried in fluid-filled wheels or sleds that roll along the rail head, and it is the only one of the three inspection jobs that can satisfy 213.237. FRA's 2014 rail integrity rule describes the coverage: the beams "normally scan the entire rail head and web, as well as the portion of the base directly beneath the web" (79 FR 4234, January 24, 2014, p. 4237). The base flanges outside the web are not reached.

The speeds are nothing like a geometry car's. The same 2014 preamble lists four test processes: a portable process "pushing a test device over the rail at a walking pace"; a start/stop (stop-and-verify) process that "tests at a slow speed (normally not exceeding 20 mph)" and halts at each indication to hand-verify it; a chase-car process where a test vehicle runs ahead and a verification crew follows; and a continuous process where the vehicle runs non-stop, data is analyzed centrally and suspects are verified later. A 1999 Colorado State University study for the Mountain-Plains Consortium put the practical number on stop-and-verify testing: test cars are "capable of speeds up to 30 mph in a non-stop testing mode", but because they stop to hand-verify indications they "typically average 6 to 8 mph" (Jeffrey and Peterson, MPC 99-106, August 1999). Vendors now rate hi-rail and rail-bound units at 30 to 45 mph; those are ceilings, not test averages.

Continuous testing under 213.240 is the closest rail flaw testing comes to the geometry-car model. The vehicle does not stop; every suspect goes on a report, a 213.238-qualified person field-verifies it within 84 hours, or within 36 hours if the suspected defect would call for remedial action A, A2 or B, and a suspect "is not considered a defect under § 213.113(c) until it has been field-verified" (213.240(e)(4)). Even here the automation ends at the report. FRA's 2020 rule that created 213.240 reported railroads' experience that priority-one and priority-two suspects were confirmed as defects about 95 to 99 and 65 to 70 percent of the time, while the low-priority suspects, confirmed less than 2 percent of the time, "account for approximately 85 percent of the field-verified defects" (85 FR 63362, October 7, 2020). The hand verification is where most of the defects are actually found.

The intervals are set by 213.237(c): on Class 4 and 5 track, and on Class 3 track with scheduled passenger trains or a hazardous materials route, no more than 370 days or 30 million gross tons between internal inspections, whichever comes first. Our guide to 49 CFR 213.237 covers the valid-search rule, the service failure rate and plug rail; the vehicle choice is compared in hi-rail truck vs inspection cart.

Key takeaway: rail flaw testing is automated in its data collection, not in its decisions; a qualified operator still reads every search and a person still confirms every defect, at speeds a geometry car would never run.

What do cameras and machine vision actually inspect?

Camera-based systems inspect what is visible from above the track: fasteners, ties, rail seats, joint bars and the running surface. Their legal standing on Class 1–5 track is set by one sentence in 213.233(b), which allows "mechanical, electrical, and other track inspection devices" to "supplement visual inspection". They do not shorten or replace the 213.233 schedule.

The one place Part 213 requires an automated vision-type system is 213.234, the concrete-tie rule. It applies to Class I and II railroads, intercity passenger and commuter railroads, on Class 4 and 5 concrete-tie main track and Class 3 concrete-tie main track with scheduled passenger service. The system must find rail seat deterioration, the loss of the tie surface under the rail that lets the rail cant and the gage widen, and the rule sets the only performance standard for an automated system anywhere in Part 213:

• accuracy within 1/8 in.
• a distance-based sampling interval of 5 ft or less
• documented calibration showing that recorded values represent the real deterioration
• an exception report listing every exception to 213.109(d)(4), with areas between 3/8 in. and 1/2 in. flagged as an "alert"
• field verification of each exception by a 213.7 person within 48 hours
• records kept for two years, and annual training for the people who handle the exceptions

The frequency is twice a calendar year, at least 160 days apart, where tonnage exceeds 40 MGT; once a year at 40 MGT or less; and once a year by automated or walking inspection on exclusively passenger track. Tangent sections of 600 ft or less are excluded.

Beyond rail seats, machine vision for fasteners and joint bars is still described by FRA as a developing tool. RR 21-20 covers the ATIP fastener and switch algorithms discussed above. An earlier FRA project on a video system for joint bar inspection (RR06-03, 2006; summarized from its abstract, as the FRA-hosted PDF could not be opened) reported that in field demonstrations operators rejected about 60 percent of detections as false alarms and the system missed about 15 percent of cracks, with ballast, vegetation, grease and mud named as the conditions to work on. Camera systems need the component to be visible and clean, and they hand a person a list to check, not a decision.

Key takeaway: vision systems inspect what can be seen, under one mandatory rule (213.234, concrete ties) and one permissive sentence (213.233(b)); every exception still gets a human field check.

Worked example: a geometry exception and a UT indication on the same Class 4 segment

Take a Class 4 freight main line, maximum 60 mph for freight and 80 mph for passenger trains under 49 CFR 213.9, with 136 lb rail. In one week a geometry car and a hi-rail ultrasonic test truck both run over it. Each reports one problem. The two problems take completely different paths.

Part 1: the geometry exception

The geometry car's exception report lists wide gage of 57 5/8 in. (4 ft 9 5/8 in.) at milepost 112.4. The 213.53 limit for Class 4 and 5 track is 4 ft 9½ in., or 57.5 in., so the reading exceeds the Class 4 limit by 57.625 − 57.5 = 0.125 in. The Class 2 and 3 limit is 4 ft 9¾ in., or 57.75 in., so the same gage is within the Class 3 limit by 0.125 in.

What happens next depends on why the car was running:

• If this is a voluntary geometry run on ordinary Class 4 track (cant deficiency of 5 in. or less), 213.333 does not apply and Part 213 sets no verification clock for the exception; the railroad's own procedure does. This gap is exactly what the 2024 proposed rule's one-hour remediation requirement was written to close.
• If the segment is operated at more than 5 in. of cant deficiency, the run is a required 213.333 inspection and the owner must field-verify and start remedial action "within two days after the inspection" (213.333(f)).

Either way, once a 213.7 designated person confirms the gage with a hand gage, the track at that location no longer meets Class 4, and under 213.9(b) it "is reclassified to the next lowest class of track for which it does meet all of the requirements". That is Class 3: 40 mph freight, 60 mph passenger, until the gage is corrected. The geometry car found a shape problem and the remedy is a speed class, not a rail defect action.

Part 2: the ultrasonic indication

The ultrasonic truck, running stop-and-verify, gets a response on a 70° forward-looking channel at milepost 118.9. The operator stops and hand-tests the spot, as the stop-and-verify process requires. The hand test confirms a detail fracture sized at 30 percent of the existing rail head area.

The chain here is 213.113, not 213.333:

1. Verification. Under 213.113(b) an indication must be verified within four hours, or immediately if it would require remedial action A, A2 or B. A 30 percent transverse-type defect falls in the 25 to under 60 percent band, which carries remedial action D, so the four-hour clock applies and is met on the spot.
2. Marking. The rail is marked with a highly visible mark on both sides of the web and base (213.237(e)).
3. Remedial action D. Joint bars, bolted only through the outermost holes, are applied within 7 days of the decision to keep the track in use. On Class 3 through 5 track, speed over the defect is limited to 30 mph until the bars are on, or less if the 213.7(a) designated person with a year of supervisory experience so decides; after barring, 50 mph or the class maximum, whichever is lower, which on this Class 4 line means 50 mph. If the bars are not applied within the 7 days, the limit drops to 10 mph until they are (note D, 79 FR 4234, p. 4257).
4. Record. The 213.241 record lists the location, type and size of the defect, the initial remedial action and its date.

Had the same truck been running a continuous test under 213.240, the indication would have gone on the suspect list with an 84-hour verification window (36 hours only if an A, A2 or B defect were suspected), and until verified it would have been protected by the most restrictive remedial action for the suspected size.

What the example shows

The geometry car passed milepost 118.9 and recorded nothing, because a detail fracture does not change the track's shape. The ultrasonic truck passed milepost 112.4 and recorded nothing, because an ultrasonic wheel does not measure gage. Both exceptions were confirmed by a person before anything was restricted. Neither system "inspected the track"; each inspected one thing about it.

What automated track inspection does not cover

Key takeaway: every one of these limits is why the rules pair each automated output with a human verification step; treat any system described as "fully automated, no human intervention" as a description of its data collection only.

How to specify automated inspection in a contract or RFP

Transit agencies, short lines and industrial track owners rarely buy "automated track inspection" as one service. They buy geometry runs, rail flaw testing and sometimes imaging from different providers, and the contract reads better when each is tied to its rule and its deliverable.

Our rail inspection RFP guide covers scope, pricing structure and evaluation; how often must rail be tested covers the interval arithmetic.

Claims to drop from your next presentation

The three articles this page replaces made the same mistakes most vendor material makes. Each is listed here with the fact that corrects it.

How EMA Industries applies this

EMA Industries does the rail flaw part of this picture. Our capabilities include hyrail ultrasonic rail flaw detection, manual rail scanning and weld examination, with an ASNT NDT Level III supporting the technical program, for transit agencies, short lines and industrial track owners in Southern California and on projects throughout the United States. Geometry measurement and visual inspection are separate programs that the track owner runs or contracts on their own rules; what our testing adds is the internal search those programs cannot make, delivered as a 213.241-style record of each defect's location, type and size and of any stretch that could not be validly tested, so that the owner's 213.113 remedial actions start from verified facts. To plan internal testing alongside an existing geometry and visual program, see our ultrasonic rail testing services or contact EMA Industries.

Related reading: Regular railway track inspections under 49 CFR 213.233 · 49 CFR 213.237 rail flaw detection requirements · FRA track safety standards · Continuous vs stop-and-verify rail testing · Automated ultrasonic systems · Railway track inspections

Frequently asked questions

Does a track geometry car find cracks inside the rail?

A geometry car measures the track's shape: gage, alignment, crosslevel, profile and warp. An internal crack such as a transverse fissure changes none of those until the rail breaks, so a geometry car passes over it without any reading. Internal defects are found only by an internal search under 49 CFR 213.237, which in practice means ultrasonic testing.

How fast does an ultrasonic rail test car travel?

Stop-and-verify testing normally runs at 20 mph or less and, because the vehicle stops to hand-check each indication, averages about 6 to 8 mph over a day (FRA's 2014 rule preamble; MPC 99-106). Continuous testing under 213.240 runs without stopping and verifies suspects within 84 hours, or 36 hours for the most serious suspected defects. Vendor ratings of 30 to 45 mph describe the vehicle's capability, not its testing average.

Does FRA require railroads to use automated track inspection?

Only in two narrow cases: automated rail seat inspection on specified concrete-tie track (213.234) and a qualifying track geometry measurement system on Class 6 and higher track or where cant deficiency exceeds 5 in. (213.333). A rule proposed in October 2024 would add geometry inspections three times a year on busier Class 1–5 track; as of November 2026 no final rule had been found.

Can automated inspection replace the visual inspections in 49 CFR 213.233?

Not under the rule as written. 213.233(b) lets devices supplement visual inspection, and FRA wrote in 2024 that a geometry system "is not a substitute for visual track inspections". A few railroads run fewer visual inspections under waivers or test programs with conditions attached, and BNSF won a 2023 court ruling over FRA's refusal to expand one, but each of those is an exception granted case by case.

What happens to an exception flagged by an automated system?

It stays an indication until a person verifies it. Geometry exceptions under 213.333 must be field-verified within two days; concrete-tie rail seat exceptions under 213.234 within 48 hours; rail flaw indications under 213.113(b) within four hours, or 84 and 36 hours under 213.240 continuous testing. Only after verification does a remedial action or a speed restriction follow.

Who is allowed to interpret an automated rail flaw test?

A qualified operator under 49 CFR 213.238: trained under the provider's documented program, with at least 160 hours of supervised rail flaw detection experience, written employer authorization and demonstrated proficiency on the specific equipment. A valid search under 213.237 exists only when the equipment performs as intended and that operator interprets the responses.

Sources

• 49 CFR 213.233, Visual track inspections, eCFR
• 49 CFR 213.234, Automated inspection of track constructed with concrete crossties, eCFR
• 49 CFR 213.237, Inspection of rail, eCFR
• 49 CFR 213.238, Qualified operator, eCFR
• 49 CFR 213.240, Continuous rail testing, eCFR
• 49 CFR 213.241, Inspection records, eCFR
• 49 CFR 213.113, Defective rails, eCFR
• 49 CFR 213.333, Automated vehicle-based inspection systems, eCFR
• 49 CFR 213.53, Gage and 213.55, Track alinement, eCFR
• 49 CFR 213.9, Classes of track: operating speed limits, eCFR
• 49 CFR 213.110, Gage restraint measurement systems, eCFR
• FRA, Track Safety Standards; Track Geometry Measurement System Inspections, notice of proposed rulemaking, 89 FR 84845, October 24, 2024, docket FRA-2024-0032
• FRA, Track Safety Standards; Improving Rail Integrity, final rule, 79 FR 4234, January 24, 2014
• FRA, Rail Integrity and Track Safety Standards, final rule, 85 FR 63362, October 7, 2020
• FRA, Notice of Petition for Extension of Waiver of Compliance (BNSF, docket FRA-2020-0064), Federal Register document 2025-20285, scheduled November 19, 2025
• BNSF Railway Co. v. Federal Railroad Administration, No. 22-60217 (5th Cir., March 15, 2023), opinion
• Baillargeon, J. and Stuart, C., Machine Vision Data Products for the Automated Track Inspection Program, FRA Research Results RR 21-20, September 2021, ROSA P record
• FRA, Video System for Joint Bar Inspection, Research Results RR06-03, March 2006 (abstract only; the FRA-hosted PDF could not be opened)
• Jeffrey, B. D. and Peterson, M. L., Assessment of Rail Flaw Inspection Data, MPC 99-106, Colorado State University for the Mountain-Plains Consortium, August 1999
• Sherrock and Neubecker, Unmanned Aircraft System Applications in International Railroads, DOT/FRA/ORD-18/04, February 2018, ROSA P record; Stuart and Doran, Automated Track Centerline Following for Drone Flight Automation, FRA RR 22-15, August 2022, ROSA P record
• FRA Research Results RR 22-25 on in-service examination of cast manganese frogs (cited from the project's verified-facts record; not re-read for this article)
• 14 CFR 107.31, Visual line of sight aircraft operation, Cornell LII
• Cornell LII text of 49 CFR Part 213 (alternate copy, read November 5, 2026)