PREDICTING POTENTIAL RAILWAY OPERATIONS DISRUPTIONS CAUSED BY CRITICAL COMPONENT FAILURE USING ECHO STATE NEURAL NETWORKS AND AUTOMATICALLY COLLECTED DIAGNOSTIC DATA

摘要

European passenger rail systems are massively interconnected and operate with very high frequency. Theimpacts of single component failures on these types of systems can significantly affect technical andoperational reliability. Today advanced diagnostic tools with broad functionalities are being added tosystems and system components. These tools monitor, control the operation and support the maintenanceof the highly sophisticated and interconnected components. This paper presents an approach for using aset of diagnostic event data from a passenger train exterior door system to predict the occurrence ofevents that might evolve into operational disruptions that impact train operation and therefore railwayreliability. This approach uses a neural network algorithm with a dynamic temporal behavior (the echostate network) in combination with principle component analysis. The proposed approach exhibits aprediction accuracy of up to 99%.