A damage detection strategy on bridge external tendons through long-time monitoring

In recent years, Structural Health Monitoring (SHM) has gained a lot of attention, given the need to detect a structure damage at an early stage. A series of technological advances, especially in the world of new sensors, has allowed more structures to be equipped with an always increasing number of monitoring systems of different nature. The state of the art of monitoring systems involves the interaction and cooperation of elements such as low-cost sensors, efficient communication networks, data transfer and storage, often based on cloud architectures. If on the one hand the amount of data collected by the new SHM systems tends to be of considerable size, the search for damage passes through a process of information synthesis aimed at defining features able to describe the health of the monitored structure. Although many papers in literature are focused on the definition of an early warning through the most suitable damage feature, less attention has been paid till now to the challenge of implementing a fully automatic monitoring system that can serve as a robust and reliable tool for decision making. This paper presents a framework/architecture for a real-time data elaboration process, based on different alarm levels to track an ongoing and growing damage. Into details, data coming from a large number of MEMS accelerometers, installed on tensioning cables inside a box composite highway bridge, are continuously processed and analyzed at the sensor level. This is done on a microcontroller equipping each sensor, thanks to both fit-to-the-purpose algorithms that do not require huge computational effort and a strategy which can manage each sensor independently from the others. At a first stage, the proposed strategy is able to identify every kind of anomalies in the collected data; then, the benign phenomena, such as the occurrence of heavy though not extraordinary loading conditions, are identified and separated from those which clearly point at a damage, such as the breaking of the strands of prestressed cables. As these events occurred during the monitoring and have been recorded, we have a check about the capability of the chosen algorithms to perform this clustering. Different output examples are discussed in this paper in order to provide a significant case study where the effectiveness of a SHM system is discussed in a damage detection perspective