Turning data driven condition now- and forecasting for railway switches into maintenance actions

Abstract

Railway switches are crucial for normal operation and during disruptions of the railroad system since they allow trains to use alternative routes. Switches moving parts are subject to high deterioration and prone to malfunctioning, representing a potential safety hazard. Thus frequent inspection, maintenance and renewal are required. Models to optimize the railroad system operation and reduce costs are possible on the basis of inspections vehicles, online condition monitoring, inspection standardization and data-based models. This paper presents a switch condition now- and forecasting model based on continuous monitored data (switch engine current during blades movement). The model is capable of identifying unusual behavior due to emerging failures without the need of manually set switch-specific thresholds. In this approach no labelled training data set of historic switch failures is required for training the model. Its output combined with maintenance information and the switch functional model sheds light on switch degradation modes, helping to optimize maintenance actions.