Abstract Summary
As part of the ‘Smart Circular Bridge for a circular built environment’, Interreg NWE project, three pedestrian and bicycle bridges are currently being developed. Built from an innovative bio-composite material the bridges promise to be fully recyclable. The bridge also innovates by incorporating Structural Health Monitoring technology to assure the structural integrity of the bridge and feeding back information for future designs. The setup includes both fibre optical strain gauges (FBGs), embedded inside the composite material itself, and accelerometers at selected locations. The usage of accelerometers for traditional modal parameter estimation was previously published. In this contribution, we explore the further usage of accelerometers for event detection and event classification in favour of the more expensive FBG setup which will act as a reference. This paper explores a vibration-based event detection and isolation algorithm for the first Smart Circular Bridge installed in Almere. The bridge is instrumented with 3 tri-axial MEMS accelerometers that have been gathering data since April 2022. The bridge is located within the Floriade Expo, ensuring high usage of the bridge and a wide variety of loading cases (ex. multiple groups of varying numbers of people crossing the bridge). Rather than grouping the data in blocks of fixed intervals, it was investigated whether data could be split into individual events. The raw acceleration time series show that individual passages could be identified and separated even during busy days, except for the very busiest moments. When running the event detection on the entire monitoring period, an event database is created, saving the start and end time of the events as well as acceleration data statistics (max., rms). Next, the event database is used to extract bridge loading statistics such as occupancy by hour of the day and by day of the week, which in turn can serve crowd control or future maintenance decisions. This analysis shows a clear relationship with the opening hours of the Floriade, an increased occupancy at the weekends and the influence of the weather conditions. Finally, it is investigated whether unintended use of the bridge, e.g. heavy vehicles crossing or crowds jumping, can be isolated from the event database. As the event data is not labelled, different machine learning algorithms, such as the Gaussian Mixture Models and BIRCH, are used to cluster the collected events in discernible groups. The identified clusters are then labelled a-posteriori using the associated FBG data. Amongst others, the results show it is possible to isolate vehicles crossing the bridge from the other events solely from the diverging vibration statistics of the event (in particular in the longitudinal direction). Additionally, the clustering strategy is applied to the time domain data rather than the statistics to improve the methodology further.
