What happened
KLD Labs makes a wayside machine vision system that photographs the trucks of passing railcars. The Federal Railroad Administration and the Association of American Railroads sponsored Transportation Technology Center, Inc. (TTCI) to test it in revenue service. It ran at CSX's Hague site south of Waycross, Georgia, from December 2016 to December 2018. Its algorithms were retrained and refined with images that railroad experts reviewed. The August 2020 final report states that the system captured measurement data with more than 97 percent reliability. It measured axle spacing on all but three of 1,176 axles. At the Hague site, bolster height was measured in 73.41% to 95.48% of annotated samples, by spring type. Broken spring detection caught one of two known broken springs and had a 1.5–8.5 percent false positive rate. A broken spring and a spring out of its nest were found only by manual review. Uptime averaged 80% from March to July 2018, against a 99% target, mostly because of air conditioner failures. The project stopped short of methods to broadcast alerts. No derailment or safety outcome is reported.
Sources
-
Evaluation report, by an evaluator that the makers or users commissioned.
Ferguson Stewart, M. (FRA), M. Witte and A. Meddah (TTCI). In-Service Performance of a Truck Component Inspection System. Final Report DOT/FRA/ORD-20/31, Transportation Technology Center, Inc. for the Federal Railroad Administration, Office of Research, Development and Technology, August 2020. Section 4, Conclusion, p. 23.
Measurement performance (consistently above 97%) was more reliable than detection performance. In general, this work suggests that MV is reliable and well-suited to make precise measurements of select components on the railcar.
Details
AI system
- Workflow
-
Cameras on both sides of the track imaged each passing truck, with AxleScan on the axle ends and TruckScan on the springs and side frame. KLD calibrated the cameras with precision targets placed at known locations. Railroad experts reviewed the images in TTCI's manual viewer, and the algorithms were retrained with images from that review.
Follow-up
- Conditions for replication
-
The equipment hut stays cooled and powered, since most downtime came from air conditioner failures. The algorithms are refined for the truck types in service, since those developed on uniform test cars needed refinement at first.
Entry details
- Sector
-
Railway safety / freight car inspection.
- Purpose
-
KLD Labs built the truck component inspection system to photograph the trucks of passing railcars, measure their components and detect defects automatically.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). KLD truck component inspection system measures railcar parts at CSX (AIBD-0955). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-0955.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)