The UK railway network is subjected to an electrification process that aims to electrify most of the network by 2020. This upgrade will improve the capacity, reliability and efficiency of the transportation system by providing cleaner, quicker and more comfortable trains. During this process, railway infrastructures, such as tunnels, require to be adapted in order to provide the necessary clearance for the overhead line equipment, and consequently, a rigorous real-time health monitoring programme is needed to assure safety of workforce. Large amounts of data are generated by the real-time monitoring system, and automated data mining tools are then required to process this data accurately and quickly. Particularly, if an unexpected behaviour of the tunnel is identified, decision makers need to know: i) activities at the worksite at the time of movement occurring; ii) the predicted behaviour of the tunnel in the next few hours.
In this paper, we propose a data mining method which is able to automatically analyse the database of the real-time recorded displacements of the tunnel by detecting the unexpected tunnel behaviour. The proposed tool, first of all, relies on a step of data pre-processing, which is used to remove the measurement noise, followed by a feature definition and selection process, which aims to identify the unexpected critical behaviours of the tunnel. The most critical behaviours are then analysed by developing a change-point detection method, which detects precisely when the tunnel started to deviate from the predicted safe behaviour. Finally, an Artificial Neural Network (ANN) method is used to predict the future displacements of the tunnel by providing fast information to decision makers that can optimize the working schedule accordingly. -> Link to full text in repository