Novel R-C-C fusion machine learning method for automatic damage detection of heritage buildings using 3D point cloud data

With the advancements in 3D laser scanning technology, a more efficient and accurate approach to crack detection can be achieved. This study presents a novel crack detection called by R-C-C method. The technique utilizes a 3D laser scanner to obtain point cloud data of a structure. The point cloud is then processed to calculate its roughness and establish a minimum roughness level to reduce the cloud’s density. The CANUPO classifier is trained using crack and joint samples as training data. The classifier can differentiate between the two classes and further trains to identify cracks in the point cloud. The results are cracks exported as a point cloud, providing a reliable and efficient method for crack detection. The proposed crack detection method is innovative and provides a more efficient and accurate approach than traditional manual inspection methods. About 93.10% of points were reduced through roughness analysis, which can minimize the analysis time, which is the biggest disadvantage of point cloud analysis. The first CANUPO of R-C-C removes points corresponding to most bricks through a machine learning classifier. However, the remaining points through the first CANUPO include points including joints as well as cracks. Based on these results, the second CANUPO conducted, which focused on separating joints and cracks through learning. The second CANUPO was able to locate the exact location of the crack after isolating the joint and crack. The study’s results demonstrate the proposed method's feasibility and effectiveness in detecting cracks in masonry structures.