Bridge health monitoring through photogrammetry-based digital twins: a topological data analysis approach to missing bolts detection

Bridge infrastructure is a fundamental component of transportation networks, necessitating precise and efficient methods for detecting missing bolts to ensure structural integrity and prevent systemic failures. Despite various advancements in missing bolt detection techniques, the application of point cloud-based Digital Twins (DTs) and Topological Data Analysis (TDA) still needs to be explored. This study introduces a novel methodology for detecting missing bolts for bridge health monitoring by integrating DTs with Persistent Homology (PH), a key technique in TDA. The proposed integration presents a more robust and accurate technique to simultaneously identify and localize the missing bolts, compared to existing 2D vision methods. To this end, the essential concepts, and theoretical foundations of TDA and the PH algorithm are presented, highlighting its robustness in quantifying data shapes. Then, a point cloud-based DT of the structural joints is created through the photogrammetry 3D reconstruction pipeline. Subsequently, for the precise localization of structural bolts in the 3D point cloud, a method based on Convolutional Neural Networks (CNN) has been utilized. Following this, a specialized method leveraging PH is implemented to identify missing bolts. Different parameter investigations were carried out to evaluate the performance of the suggested approach. Finally, a case study of a real bridge was assessed to evaluate the proposed pipeline on bolted connections with bolts. The results confirm the reliability of the method in accurately detecting and locating missing bolts within the point cloud data, achieving high detection accuracy with a false positive rate of less than 10%.