Anomaly detection in railway infrastructure

Abstract

This thesis elaborates the topic of artificial intelligence used for anomaly detection in railway infrastructure. In Switzerland, railways are the backbone of the public transport system and an important factor in the economy and society. Therefore, it is important to detect unwanted anomalies in railway infrastructure as fast as possible and before they result in an incident, whereby even a minor interruption can evolve into a major disturbance. Some of the challenges arising from this field can be met with artificial intelligence, especially with machine learning techniques and knowledge engineering. Railway infrastructures offer a wide range of potential in anomaly detection, since they are complex systems. Especially sub-systems, which are exposed to forces (e.g. acceleration or deceleration, rotating or moving) which can result in material wear and maintenance effort, provide promising use cases for anomaly detection. During the research, one particular problem was identified: The arc ignition in the pantograph-catenary system during train operation. Frequent arc ignition will accelerate attrition, respectively, the faster loss of material results in more frequent maintenance or malfunctioning, which eventually leads to a higher idle time or an interruption in operations. The literature review has shown that little effort was expended to solve the problem of arc ignition detection in the pantograph-catenary system....