Deeper insight in railway switch condition nowcasting

Abstract

Railway switches are a crucial part of the railway system but prone to failures. Nowadays a common approach to monitor switches with electrical engines is to measure the current consumption of the switch's engine. Since this measurement principle is valid for creating raw data, effort is put into improving the evaluation of the current graph for monitoring the condition of switches. This aims to improve data analysis techniques for a well-functioning and reliable railway system avoiding unnecessary train delays due to switch failures. In this work unsupervised data mining techniques are investigated as a basis for the prospective step to condition based prognostics with a deeper understanding of the characteristics of the current graph and their meaning. Therefore different features and various combinations are explored as well. Learning from historic data and gaining knowledge is essential for further improvement and is combined with domain knowledge. This work aims to detect anomalies and failures from the current graph to prospectively identify degradation, avoid potential false alerts and switch malfunctions. It also enables a step towards supporting maintenance with better and more efficient maintenance plans, more reliable and understandable switch condition monitoring and builds a fundament to predictive maintenance.