Big-data driven assessment of railway track and maintenance efficiency using Artificial Neural Networks

Abstract

• Big data analysis of railway track geometry using machine learning algorithms. • High-speed railway track quality assessment using artificial neural networks. • High-speed track maintenance (tamping) efficiency evaluation. • Techniques for railway track maintenance optimization. • Data taken from high-speed recording cars – not track based sensors. This study will address the opportunity of using a big data driven approach to providing a more specific description of track quality, mainly selecting segments of the track exhibiting higher settlement with the use of data analytics and machine learning. The focus will be on a high-speed line in the UK with data covering over 15 years of track geometry. Data sets describing track geometry quality have an enormous volume, which means that it is impractical to apply conventional methods to process it fully. The overall aim of this work was to apply an AI technique to analyse the big data. An Artificial Neural Network (ANN) was developed on features from the available data set and used to identify segments of the track where the condition has either improved or deteriorated in the period between two inspection runs. The model achieved an accuracy of ∼ 98 % during training and was able to reliably identify segments of the track which underwent significant changes in measurements. The ANN was used to perform an initial analysis of the geometry data, which revealed that maintenance works (mainly large-scale tamping) may include healthy portions of the track and potentially reduce its life span. Approximately 50 % of the tamped track was not found to improve significantly, based on several different geometric features. On the other hand, local works with a sprinter tamper were much more efficient at eliminating defects.