Analysis of Railway Track Irregularities with Convolutional Autoencoders and Clustering Algorithms

Abstract

Modern maintenance strategies for railway tracks rely more and more on data acquired with low-cost sensors installed on in-service trains. This quasi-continuous condition monitoring produces huge amounts of data, which require appropriate processing strategies. Deep learning has become a promising tool in analyzing large volumes of sensory data. In this work, we demonstrate the potential of artificial intelligence to analyze railway track defects. We combine traditional signal processing methods with deep convolutional autoencoders and clustering algorithms to find anomalies and their patterns. The methods are applied to real world data gathered with a multi-sensor prototype measurement system on a shunter locomotive operating on the industrial railway network of the inland harbor of Braunschweig (Germany). This work shows that deep learning methods can be applied to find patterns in railway track irregularities and opens a wide area of further improvements and developments.