In-service measurements on tramway networks
Current condition monitoring systems in tram and railway networks rely mostly on reactive and preventive maintenance strategies, with in-service measurements used to resolve issues as they develop over time. The URITMIS project attempts to make the transition toward predictive maintenance by using large datasets, machine learning techniques, and digital twins of rail infrastructure to detect irregularities and defects at various stages of degradation. This technique will allow for automatic failure-type identification and categorization of safety/operational impacts, revolutionizing maintenance planning using data-driven foresight.
Data acquisition and signal processing
The initial source of data on the vibro-acoustic parameters of the track is the long-term monitoring of the tram infrastructure, which the leader and existing members of the project team are already conducting on the tram network of the city of Zagreb.
The acquisition system collects data in real time on the levels and characteristics of wheel vibrations of the tram vehicle that travels on the tram track network of the City of Zagreb. Analysis of vibration levels must be selective in order to reduce processing time – only those locations with elevated vibration levels are analyzed. All collected data will be processed online and sent to a remote service so that the operator can easily access them.
Track irregularities detection
There is a lack of data adjustment algorithms that would be developed within the framework of this project to be able to correlate the vibration data gathered by monitoring with actual anomalies on the track, estimate their advancement over time, and base maintenance on that. The innovativeness of this approach to data collection and processing is manifested in:
- determine the linear and point deformations of the tracks and their progression trend based on noise and vibration measurements, signal analysis and machine learning
- analysis of tramway vehicle vibrations and noise at the contact of wheels and rails,
- machine learning for detecting track damage in different driving modes of a conventional tram vehicle
- determining the transfer functions of rails and wheels to calculate the indirect roughness of the rail and wheels to
Implementation into GIS system
GIS system for predictive maintenance of the tram infrastructure is the end result, which, based on the collected vibro-acoustic parameters, performs automatic detection of certain types of damage and their tracking in the time domain.
Using the advanced vibro-acoustic signal analysis proposed in this project application together with an integrated system for data collection, positioning and data processing, with outputs such as combined wheel-rail contact roughness, detection of line and point damage is possible. The data collection will be carried out at several test locations that have been determined on the basis of research conducted by the research team on the tram networks in Zagreb and Osijek.
Machine learning
Using machine learning, the detection of infrastructure damage such as rail cracks, rail corrugation and fastening system degradation can be improved and applied to the operating conditions of conventional tram vehicles.
Machine learning models will be calibrated with the help of directly collected track parameters such as:
- track geometry
- local defects and defects along the route
- vehicle defects such as flat spots on the wheels
The end result of the methodology will the automation of the collection and processing of data on the state of the track and finally information on which locations need to be intervened, in what time period and to what extent.
