Doctoral Thesis: Automated progress monitoring and dimensional conformance control, Automated 4D as-built model generation, Robust statistics, Automated Planar and linear classification and segmentation of point cloud data, Principal components analysis, Minimum covariance determinant

Subject: Remote Sensing, Mathematics, Statistics, Computer Science, Engineering–Civil

 

Abstract

Construction project progress monitoring and structural dimensional compliance control are essential for decision makers to identify discrepancies between the planned and the as-built states of a project, and take timely measures where required. In practice, monitoring is performed manually, a time consuming, error-prone and labour intensive task, particularly in large scale projects. Thus, large projects are monitored unsystematically by collecting limited onsite data, restricting the project management team to identify delays, and rework on time. The correct determination of the project’s performance also relies heavily on the correctness and completeness of the collected data during the monitoring process. Hence, site supervisory personnel spend considerable time just to manually control the quality of the manually collected onsite data. Several research studies have aimed to use remote sensing technologies such as LiDAR and cameras to acquire 3D point clouds of building elements to improve the quality of the collected data; however, these studies assume the planned BIM as a priori knowledge to assign the point clouds to their corresponding structural element, which provide inaccurate basis for reporting the as-built status of a project, especially when the planned and the actual differ or the planned model is not available with sufficient detail. Here, using the most up-to-date pattern recognition, robust statistical analysis and mathematical modelling techniques, a new robust approach was formulated, independent from a pre-existing planned model, to automatically generate the as-built model of common structural elements with predominantly planar and linear surfaces directly from the acquired point cloud. In the context of four experiments, ten sets of point clouds, nine from actual construction sites, were collected to express, verify and validate the diverse applicability of the proposed system for automated progress monitoring, structural displacement analysis, and dimensional conformation control. It was demonstrated that the novel robust planar and linear point cloud classification and segmentation method, presented here, achieved an overall accuracy of better than 90.4% for all datasets, indicating its generic applicability for construction projects. It was also shown that the proposed system is capable of automatically generating as-built models of common structural elements with the 3-5mm desired construction grade accuracy.