Feature Extraction for Track Section Status Classification Based on UGW Signals

• Published in: Sensors (Basel, Switzerland) Vol. 18; no. 4; p. 1225
• Main Authors: Yuan, Lei, Yang, Yuan, Hernández, Álvaro, Shi, Lin
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 17.04.2018; MDPI
• Subjects: Classification; deep learning algorithm; feature extraction; Monitoring; Monitoring systems; temporal and spatial dependencies; track status classification; Transportation networks; ultrasonic guided wave (UGW); deep learning algorithm; temporal and spatial dependencies; ultrasonic guided wave (UGW); feature extraction; track status classification
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s18041225

Track status classification is essential for the stability and safety of railway operations nowadays, when railway networks are becoming more and more complex and broad. In this situation, monitoring systems are already a key element in applications dedicated to evaluating the status of a certain track section, often determining whether it is free or occupied by a train. Different technologies have already been involved in the design of monitoring systems, including ultrasonic guided waves (UGW). This work proposes the use of the UGW signals captured by a track monitoring system to extract the features that are relevant for determining the corresponding track section status. For that purpose, three features of UGW signals have been considered: the root mean square value, the energy, and the main frequency components. Experimental results successfully validated how these features can be used to classify the track section status into free, occupied and broken. Furthermore, spatial and temporal dependencies among these features were analysed in order to show how they can improve the final classification performance. Finally, a preliminary high-level classification system based on deep learning networks has been envisaged for future works.