In-service railway track condition monitoring by analysis of axle box accelerations for small to mid-size infrastructure operators

Abstract

About 50% of life-cycle-cost of railway tracks are caused by maintenance actions which are nowadays typically conducted within corrective maintenance schemes. The accelerating digitalization and development of low-cost sensors provide the chance especially for small infrastructure operators to introduce cost-effective track condition monitoring on a daily basis utilizing embedded sensors on their in-service vehicles. This will allow the challenging step forward from reactive corrections to proactive preventive maintenance actions to significantly reduce maintenance cost. We present the overall framework and first results of a prototype implementation of the complete system for quasi-continuous condition monitoring regarding short-wavelength (a few centimetre to a few meter) defects of railway tracks such as rail corrugation from the embedded sensor to the visualized data analysis result. We gather georeferenced triaxial axle box accelerations in the frequency range from 0.8 Hz to 8000 Hz by a prototype measurement system on a shunter locomotive operating on the railway network of Braunschweig inland harbour in Germany (total track length about 15 km). The prototype implementation in operational environment provides data of up to now four months of shunting operation to develop and evaluate data analysis algorithms. The acceleration sensor data is combined with further relevant data such as the digital map of the railway infrastructure and other operational data to be prepared for data analysis by our prototype implementation of a land-side data management system. We present results of the land-side data analysis chain including the track-selective georeferencing by multi-sensor-fusion, the extraction of relevant features from the axle box acceleration data for pattern recognition and the further intelligent data analysis to provide spatiotemporal information about track conditions. The obtained results are finally visualized for the infrastructure operator by the data management web-frontend.