Data-driven condition now- and forecasting of railway switches for improvement in the quality of railway transportation

Abstract

Railway switches are crucial for guiding trains to tracks or platforms and allow trains to take alternative routes in case of disruption. Switches are a costly asset since its components and functions require frequent inspection, maintenance and renewal. The switch moving parts are subject to high deterioration and prone to malfunctioning, posing in the worst case a safety hazard if no action is taken. Nowadays online condition monitoring, inspection vehicles, standardization of both inspection and maintenance actions, as well as data-based models are tools supporting decision making for optimizing preventive and condition-based maintenance plans. This leads to asset life extension, cost reduction and an overall improvement in the quality of railway transportation. Strukton Rail uses POSS® system to monitor critical assets such as switch engines. Over 10,000 assets worldwide, most of them switches, are equipped with sensors and monitored by this system. For switches POSS® measures the engine power consumed during the switch blades movement. Switch malfunctioning can lead to irregularities in the power consumed during this movement. When these irregularities exceed certain thresholds derived from manually selected reference curves, POSS® gives an alarm indicating that the current state of the switch is different than expected. Maintenance experts asses the warning and corresponding measured power, and decide on the urgency of inspection. Ongoing research that does not rely on manual reference and threshold selection is improving state-of-the art condition monitoring. Current efforts developing novel data-driven methods focus on providing diagnostic and prognostic information to support decision making by maintenance experts. This contribution presents a data-based model that exploits measured power consumption and incorporates environmental conditions to detect relevant anomalies in switch behavior. The model is capable of reducing the number of false alarms in comparison to the condition monitoring currently in operation, providing a diagnosis on the current switch status (nowcast) and identifying some emerging failures of mechanical and/or electrical nature in an early stage (forecast). This information can be translated into traffic interference prevention and targeted maintenance for asset health enhancement.