Comparative analysis of multi-stage filtration methods for crack detection in masonry structures

This study investigates two advanced multi-stage filtration methods—C-C-D (CANUPO-CANUPO-Dip angle filtration) and D-C-C (Dip angle filtration-CANUPO-CANUPO)—to enhance the detection of cracks in the point cloud data of the Birkenhead Priory Church, a heritage masonry structure. These methods integrate machine learning-based classification (CANUPO) with dip angle filtration to effectively discriminate cracks from other structural elements, including bricks and joints. A significant aspect of this research is the optimization of key filtration parameters, specifically the Local Neighbor Radius (LNR) and dip angle, which were fine-tuned to achieve optimal accuracy. The optimal parameters—LNR of 0.020m and dip angle of 84.4°, demonstrated a marked improvement in filtration precision by minimizing irrelevant points while retaining crucial crack data. The initial point cloud consisted of 605 crack points, 24,178.4 brick points, and 1133.5 joint points. Application of the C-C-D method reduced the crack points to 175 (a 71 % reduction), the brick points to 6.4 (a 99.97 % reduction), and the joint points to 7.6 (a 99.3 % reduction). By contrast, the D-C-C method, adopting a more conservative filtration approach, reduced the crack points to 328 (a 45.8 % reduction), the brick points to 99.4 (a 99.6 % reduction), and the joint points to 43.8 (a 96.1 % reduction). Additionally, both methods were applied to a large-scale model of the Birkenhead Priory Church, confirming the strengths of each approach, where the C-C-D method demonstrated superior filtration, making it highly effective for applications requiring precision and minimal data retention. Meanwhile, the D-C-C method preserved more structural detail, especially in crack detection, making it better suited for scenarios where feature retention is essential. The comparative analysis highlights C-C-D's advantage in high-precision applications, while D-C-C excels in retaining finer details. These findings provide practical guidelines for method selection: C-C-D is ideal for high precision and efficiency, while D-C-C is suited for studies requiring detailed structural feature preservation. This research offers valuable insights for practitioners in structural health monitoring and heritage conservation, balancing precision with feature retention to maintain the integrity of historical masonry structures.