Automated real-time damage detection strategy using raw dynamic measurements

Abstract Over the last decades, several techniques have been developed in the context of Structural Health Monitoring (SHM) programs. However, when it comes to novelty (or damage) detection, these methods are generally based on human decisions. Moreover, most strategies already published in this topic mainly focus on modal identification procedures and tracking their outputs i.e., structural modal parameters. Such approaches usually lead to high computational costs and can still be insensitive to minor changes in structural behavior, thus missing crucial damage scenarios in their initial manifestations. To circumvent these drawbacks, recent researches showed that the use of symbolic representations derived directly from raw time-domain data (e.g. acceleration measurements) could provide more damage-sensitive responses with lower computational effort. Indeed, good results were achieved by representing raw measurements in terms of their statistical distributions over time. Nevertheless, the lack of information regarding the frequency spectrum represents a decisive drawback. Therefore, this paper presents a novel symbolic object, which considers both time and frequency responses of structural dynamic measurements. The proposed methodology employs a k-medoids clustering over such objects within a moving time-window framework and uses a single-valued index to indicate whether a novelty is present in the acquired data. Numerical simulations and two practical studies – a 3D frame tested in laboratory and a motorway bridge – show that the proposed approach provide an unsupervised and adaptive scheme for SHM applications.