Detecting anomalous behavior of railway switches under real operation conditions: workflow and implementation

Abstract

Railway switches are crucial assets since they enable trains to change tracks without stopping. Larger parts of the infrastructure are compromised when certain switches fail. Regular inspections, maintenance and repairs are required to increase switch reliability, making them costly assets. Monitoring systems help determining the condition of assets. Nowadays nearly thousand switches in the Netherlands are remotely monitored by Strukton Rail. The current version of this monitoring system has helped to identify degrading and failing switches, but it also generates false alarms. There is room for improvement in how the monitoring system supports asset managers in making decisions regarding the asset. Here, we present a workflow that exploits switch monitoring data under real operation conditions. The running workflow implements a machine-learning model for automatic detection of anomalous switch functioning. Models for predicting switch degradation and failure evolution, and for identifying failure types are under development and remain to be implemented.