Arc detection in pantograph‐catenary systems by the use of support vector machines‐based classification

Abstract

Predictive Maintenance, Prognostics and Reliability Centered Maintenance approach are becoming more and more important in the railway sector to reduce costs of operation and to increase reliability and safety. In fact, these are fundamental to optimize the maintenance process, to define new measures and algorithms which locate faults, to monitor health conditions of subsystems and to estimate residual life of components. In some cases it's possible to use existing sensors and existing processing hardware to extract new information from the existing available data. It's clear that this is usually the best option because the benefit can be achieved with little or no cost at all. This paper describes the result of a study performed with the aim of detecting arcing events without the need of additional equipment mounted on board the train. A set of data relative to voltage and current collected on high speed trains along with a set of measurements coming from photosensors are available. The data are processed by the use of an advanced classification technique, Support Vector Machines, with the aim of extracting important information such as the time coordinate related to anomalies in the overhead contact line and the status of the pantograph contact strip.