Identification of Key Geometrical Features for Automatic Structural Damage Detection from Point Clouds

Abstract

Digital technology to collect point cloud data has progressed dramatically in recent years along with its structural health monitoring applications. With these large data sets, the development of automated algorithms to detect structural damage from point clouds is critical to quickly and efficiently characterize structures and identify regions of interest. Lidar or laser scanners, popular equipment to collect point cloud data, can create highly detailed and geometry accurate three-dimensional point clouds from object surfaces, however the lack of discrete geometric information restrict their primary application to visualization or basic measurements (e.g., lineal, area, or volumetric computations). Various types of damage including cracking, concrete spalling, and loss of cross sectional areas in steel members due to corrosion can be detected by exploring local geometric variations of each vertex with respect to its neighboring vertices. To detect structural damage, an algorithm is developed to compute a series of geometrical features using discrete differential geometry theories and covariance geometrical features that can explicitly identify vertex variation. Application of such damage detection algorithm can have substantial effect on reducing field data collection time in comparison to traditional visual inspection along with more objective structural damage state characterization at a safe distance.