Files
Mertens_62042000_2026.pdf
Closed access - Adobe PDF
- 2.19 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Odometry is a safety-critical function in railway systems: it estimates train motion from onboard sensors and directly supports braking supervision, localization, and protection functions. In current industrial practice, odometry parameters are manually tuned by experts through costly and time-consuming rigorous test campaigns, often during the night on commercial tracks, trying to create loss of adhesion (slip/slide) with water and soap. This thesis investigates a data-driven approach to support and accelerate this configuration process for Alstom trains. The goal is to predict slip/slide-related odometry parameters from train configuration data and historical test outcomes, in order to provide a better initial odometry configuration, closer to final expert-validated parameters, while preserving safety constraints. In this context, Safety Integrity Level 4 (SIL 4) - the highest level of safety performance - compliance alone is not a sufficient objective. A trivially safe strategy based on excessive conservatism, such as strongly overestimating speed envelopes or systematically forcing restrictive operation, would degrade operational performance. For Alstom, odometry performance is a key product differentiator, so the engineering target is to satisfy safety and performance simultaneously, in line with European Train Control System (ETCS) / European Rail Traffic Management System (ERTMS) interoperability and safety principles used in European mainline rail systems. An initial trace-replay strategy was explored, where each candidate parameter set is evaluated by replaying recorded test traces. Although technically relevant, this approach proved difficult to scale because of heterogeneous data formats and high computational cost. The work therefore pivots to a train-configuration-centered formulation, using structured features that describe train type, wheel-slide-protection (WSP) behavior derived from the independent Traction Control Unit (TCU) and Braking Control Unit (BCU) settings, wheel sensor architecture, sensor configuration, and associated validated odometry settings. The expected contribution is a practical methodology for supporting expert decisions with data, reducing calibration effort, and improving repeatability of odometry configuration in an industrial context.