RailCheck: A WSN-Based System for Condition Monitoring of Railway Infrastructure Jan Sramota, Amund Skavhaug Department of Mechanical and Industrial Engineering Norwegian University of Science and Technology Trondheim, Norway Email: {jan.sramota, amund.skavhaug}@ntnu.no Abstract—Contemporary tools used to monitor railway points and crossings are ineffective. Routine inspections of these crit- ical parts are still being performed manually by specially- trained inspectors. This creates higher expenditure and makes infrastructure difficult to maintain. With the expected further expansion of the railway network, this exerts increased pressure on infrastructure managers to ensure safe and predictable traffic. Hence there is a need for inexpensive and reliable condition- based maintenance systems. This paper describes an autonomous, near-real-time system built to this effect. It is based on acceler- ation measurements of train-track interaction, when the train is present. Using a wireless sensor network (WSN), data are aggregated over the Internet of Things (IoT) low-power wide- area network (LPWAN) structure into the Internet, where the big-data post-processing is performed. The performance and suitability of this system were evaluated on tracks in real traffic conditions and were found to be potentially beneficial for this sector. The system was built over a three-year period as part of the DESTinationRAIL H2020 EU-project. Keywords—IoT, condition monitoring, condition-based main- tenance, LPWAN, WSN, LTE-R, IQRF, railway I. I NTRODUCTION Worldwide railway networks have more than 1.15 million km of rail tracks [1], with further expansion planned. This inevitably requires methods and tools for the effective health monitoring of these large transportation networks. Rail tracks without points and crossings are today regularly inspected by equipped maintenance trains that use camera or laser-based systems to automatically evaluate their condition. These tools, especially the laser-based solutions, can operate at very high speeds of up to 450 kmh −1 , and are an effective solution to gathering a large number of very precise datasets that can be further processed with relative ease. A support decision system can autonomously conclude where to increase inspection intervals, set a schedule for maintenance, and decide which rails needs to be fully replaced, having reached the end of their life. Before these methods existed, railway inspectors were dependent on many local measurements which were per- formed manually and one at a time. Decisions to replace tracks were frequently made at inappropriate moments, sometimes long before it was necessary or much too late in the life-cycle of the track. Contemporary tools for monitoring the geometric quality of tracks allows more frequent inspection, predictive maintenance control, and use of the existing infrastructure with high efficiency throughout the whole life-cycle. The situation at railway points and crossings (P&C), on the other hand, is significantly different. Existing tools for moni- toring degradation processes on P&C are not fully applicable on these parts. Furthermore, track-stiffness-monitoring vehi- cles cannot be used for these sections either. While spending on the maintenance and renewal of rail track without P&C is constantly optimized, expenditure on P&C has remained rather rudimentary over the decades. Degradation processes are still solely inspected manually, which, in combination with the large number of these parts, leads to high spending and ineffective maintenance. In some countries [2] about 25% of all maintenance costs are still being allocated to these critical parts. To resolve this disproportion new systems must be developed. One such system, RailCheck, has recently been introduced by Norwegian University of Science and Technology as part of the DESTinationRAIL [3] project. II. STATE OF THE ART Systems addressing this problematic are usually train or infrastructure-based or a hybrid of both. Train-based solutions use sensors mounted on the train’s suspension to evaluate responses from the tracks. The advan- tage of this method is its ability to filter out, to a certain extent, the response of its own, suspension. The disadvantages are the large amount of captured data and difficult recognition of the current position over the specific P&C. The train- based solution may be the most interesting future method, since modern trains are often already equipped from stock by very precise vibration sensors for the train’s own self- diagnosis. If these data become accessible they might offer a very cost-effective and elegant way of gathering large numbers of representative datasets. Infrastructure-based solutions [4] are currently the most feasible approach to railway condition-based maintenance. Stationary sensors mounted directly on the rails detect vibra- tions from passing trains at specific preselected P&C, and data are transmitted through existing GSM/WLAN infrastructure. The main disadvantage is the higher cost due to the large numbers of sensors in the network. The hybrid solution [5] is usually an attempt to use station- ary sensors mounted on the rails and a gateway located on the train. The main difficulty is the transmission of data between the fast-moving train and the sensors.