Automatic Drive-By Bridge Damage Detection via a Clustering Algorithm

Over the past two decades, the drive-by bridge inspection method as an active field of research has been proven to be effective by many studies. Meanwhile, with the recent growth and advancement in Machine Learning (ML) techniques, data-driven based Structural Health Monitoring (SHM) systems have piqued the interest of many scholars, as they have the potential to provide a fast and accurate solution to damage detection problems. Although some efforts have been made, the integration of ML techniques with drive-by methods still faces obstacles. For example, many data-driven drive-by approaches are based on supervised learning models requiring labels for different damage cases, while the damage labels are rarely available in practice. Additionally, their performance relies on a few extracted features from ML approaches, which are often not applicable to different bridges. Given this background, a novel automatic damage detection algorithm for the indirect SHM framework is proposed. It employs a cluster-based ML model and proposes a new damage index to indicate the damage and to update the database. The vehicle accelerations collected from a healthy bridge as labeled data are used to train the model. Using only raw vehicle accelerations as inputs, the proposed model can indicate the damage in real time and update the database automatically. Laboratory experiments are performed to validate the proposed methodology by employing a steel beam and a scale truck model. The results demonstrate the model’s feasibility and robustness, and suggest the potential of achieving automatic, efficient, and practical SHM systems in the future.