A Method to Generate Realistic Synthetic Point Clouds of Damaged Single-Span Masonry Arch Bridges

Masonry bridges form an important component of the transport infrastructure in the UK and Europe. Structural issues in these bridges are often identified by the presence of visual surface defects, such as cracks. To automate this process, recent research efforts have focused on identifying defects from images, neglecting the potential information that can be gained from investigating accompanying geometric distortions. To explore the feasibility of using geometric distortions to identify cracks, a large synthetic point cloud dataset of single-span masonry arch bridges is developed in this study. For each bridge, a macro-scale nonlinear 3D finite element model is first created. Using the wide range of geometric and material parameters that characterize masonry arch bridges, finite element model generation process is randomized and automated. Batches of finite element calculations are subsequently performed to extract distorted bridge geometries arising from a range of commonly encountered differential settlement profiles. The distorted geometries are then imported into a 3D graphics environment, where they are converted into point clouds by considering realistic scanning geometries and measurement noise. This step closely simulates the laser scanning of bridges as a part of routine inspections and enables the generation of a large and high-fidelity point cloud dataset. With a simple example, it is demonstrated that by investigating geometric surface features, such as mean surface curvatures, cracks can be identified. Future work will utilize the dataset developed in this study to train neural networks to identify defect locations, quantify their magnitude and diagnose causes of structural damage, directly from point clouds.