Industries / Transport and vehicles / Railcar manufacturing
Testing and certification
Testing and certification proves, on test tracks, rigs and computer models, that a new locomotive or car is safe and does what its maker promised, and wins the regulator's permit to run it.
5
Milestones achieved, from 1829 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1829
A written test picks Rocket
The directors of the Liverpool and Manchester Railway, about to open their line, held public trials at Rainhill, near Liverpool, from 6 October 1829, with a £500 prize. Their written rules had each engine haul three times its own weight for ten trips of 1¾ miles each way, a 35-mile journey at an average of at least 10 mph, then refuel and do it again. Of the five locomotives that started, only the Stephensons' Rocket finished, and the railway gave the Stephensons the contract to build its locomotives, by Grace's Guide's account.
LessonA test written down before anything is built lets a buyer choose between designs on measured results.
Credit
the railway's directors; judges John Kennedy, John Urpeth Rastrick and Nicholas Wood; George and Robert Stephenson
2019
One permit across Europe
EU Directive 2016/797, part of the Fourth Railway Package, made the European Union Agency for Railways (ERA) the body that authorizes a vehicle whose area of use covers one or more member states, through a single online "one-stop shop". Member states had to apply the new rules by 16 June 2019, or a year later if they asked for more time. The agency must decide within four months of receiving a complete file, and later vehicles that match an authorized type get their permit on the applicant's declaration of conformity, without further checks.
Credit
the European Parliament and Council; ERA
2021
Cracks halt a new fleet
On 8 May 2021 Hitachi's Class 800, 801 and 802 intercity trains, the first of which entered service in Britain in 2017, were withdrawn after cracks were found in the bolster, the part of the car body that passes its weight down to the bogie. Britain's rail regulator, the Office of Rail and Road (ORR), found that fatigue cracks near the yaw damper brackets came from loads greater than the original design allowed, although the design met the applicable standards. Those standards set the accelerations a designer must allow for but not the strength of parts, and required no fatigue test unless the computer stress analysis found "critical uncertainties", which it did not.
LessonMeeting the standard does not prove a car body will survive the loads it meets in service.
Credit
ORR (review); Hitachi Rail
2022
Freight testing gets a campus
For 40 years TTCI, the research arm of the Association of American Railroads (AAR), ran the Federal Railroad Administration's test center at Pueblo, Colorado. When that contract passed to ENSCO in 2022, TTCI renamed itself MxV Rail and built its own campus on a former army depot nearby, with a 5.8-mile high-speed loop, a 2.8-mile loop for heavy-tonnage wear tests and labs whose "primary role", by its own account, is certifying new hardware built to AAR standards.
Credit
MxV Rail; the AAR
2025
A model holds up Acela
Amtrak ordered 28 high-speed trainsets from Alstom in August 2016 for service from 2021. US rules (49 CFR Part 213, Appendix D) require a computer model showing that a trainset runs safely before it can go on to further testing at high speed. Alstom began series production before its model was validated. In 2023 Amtrak's Office of Inspector General put the program more than three years behind schedule, and Amtrak officials told it that model validation was "the primary driver" of the delay. The trains, named NextGen Acela, entered passenger service on 27 August 2025.
LessonValidate the computer model before building the fleet, or every change it forces must be retrofitted to trains already built.
Credit
Amtrak Office of Inspector General (report); Alstom; Amtrak
Watch for
India's own test track
India's North Western Railway is building a 59 km broad-gauge test track between Gudha and Thathana Mithri in Rajasthan for the Research Design and Standards Organisation (RDSO), India's rail standards body, to test rolling stock, track, bridges, signaling and semi-high-speed trains, at a cost of about Rs 9.67 billion. A 4.5 km twisted track for static tests of rolling stock was finished by August 2026; a 31.5 km high-speed stretch and a 3 km accelerated-testing loop were due by September 2026 and a 20 km curve-testing loop by December 2026.
Achieved whenRDSO reports type tests of new rolling stock run on the track at its design speed.
Claimed by
North Western Railway, reported by Indian Infrastructure, August 2026
Open
01
Open dynamics model, benchmarked
An open-source computer model of the Manchester Benchmark passenger vehicle, the reference vehicle railway simulation programs are compared on, that runs the benchmark's published track cases and reports the same outputs the commercial programs did, such as wheel-rail forces. Publish the code and the side-by-side results.
The briefCredit · unclaimed02
Defects found after delivery
A public, coded list of design faults found in new locomotives and passenger cars after delivery, built from reports that regulators, investigators and auditors have published in several countries. For each case record the fleet, builder, part, how and when the fault was found, how long after entry into service, and which test, if any, should have caught it.
The briefCredit · unclaimed
Open 01 · the brief
Open dynamics model, benchmarked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source computer model of the Manchester Benchmark passenger vehicle, the reference vehicle railway simulation programs are compared on, that runs the benchmark's published track cases and reports the same outputs the commercial programs did, such as wheel-rail forces. Publish the code and the side-by-side results.
Why it is still open
Before a new US trainset can go on to high-speed testing, a computer model must show it runs safely, and a model that would not validate delayed Amtrak's NextGen Acela by years. The programs used for this work, such as MxV Rail's NUCARS, are licensed commercial software. The 1997–98 Manchester exercise compared five of them: NUCARS, VAMPIRE, GENSYS, ADAMS/Rail-MEDYNA and SIMPACK. A 2023 benchmark for switches and crossings again compared mostly commercial codes. Open-source wheel-rail contact codes exist, such as Vtech's CONTACT under the Apache 2.0 license. Two public projects come close. OpenRailVehicleDynamics, posted on GitHub in July 2026 and written by AI coding agents, compares its results with SIMPACK on two of its own vehicles, but its license (the Business Source License) is not an open-source license and it does not run the Manchester cases. TTB-2D, a MATLAB program from NTNU under the GPL, includes the Manchester benchmark vehicle's properties but models only up-and-down motion of trains on bridges, not the sideways running the benchmark tests. This research found no open-source whole-vehicle program shown against the Manchester results.
oversight.govmxvrail.comresearch.chalmers.segithub.comgithub.comgithub.com
Built with what exists
CONTACT or the open Kik-Piotrowski contact model for the wheel-rail contact; a general open-source multibody solver such as Project Chrono or MBDyn, or Python with SciPy; an AI coding agent to draft model code that a person checks against the benchmark definitions (see the Note "Readable robot programs").
First step
Borrow the benchmark definitions (S. Iwnicki, ed., The Manchester Benchmarks for Rail Vehicle Simulation, 1999) from a library and model one wheelset on straight track with CONTACT.
Done looks like
Every vehicle case in the benchmark run, each output plotted against the published results of the five commercial programs, and a published count of how many outputs fall inside the spread of those programs.
Open 02 · the brief
Defects found after delivery
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, coded list of design faults found in new locomotives and passenger cars after delivery, built from reports that regulators, investigators and auditors have published in several countries. For each case record the fleet, builder, part, how and when the fault was found, how long after entry into service, and which test, if any, should have caught it.
Why it is still open
The ORR's review of the Class 800 cracks concluded that, since the design met the standards, the industry as a whole should check whether its standards account for the loads trains really meet. Amtrak's inspector general found that design changes discovered during model validation would have to be retrofitted to trainsets already built. The Community of Metros, a benchmarking group run from Imperial College London, studied how often new metro cars fail against their time in service, for its member railways to compare. Searches of Zenodo and Hugging Face for this brief found no public dataset of such cases.
Built with what exists
Published reports from the ORR, Britain's Rail Accident Investigation Branch, the US National Transportation Safety Board, Canada's Transportation Safety Board and Amtrak's inspector general; a spreadsheet with a written codebook; a large language model to draft the coding of each report for a person to check (see the Note "Seven more flagged").
First step
Code the Class 800 and NextGen Acela reports and five more, and write the codebook from what they have in common.
Done looks like
150 cases from at least four countries, coded by two people working separately, with their agreement reported as Cohen's kappa, released under an open license with a table of which kinds of test missed which kinds of fault.