Abstract
Digital transformation is a mainstream trend in the construction industry, affecting not only new projects but also existing structures. The Scan-to-Building Information Modelling (Scan-to-BIM) approach helps create as-built BIM models for existing structures, enhancing their digital capabilities and enabling various practical applications in the industry. However, in many projects requiring as-built BIM models, the final models produced through Scan-to-BIM often fall short, leading to outcomes that do not meet practical expectations. Ensuring the effectiveness and quality of as-built BIM models in real-world applications is a critical area of study. This research aims to provide comprehensive recommendations to improve or ensure the quality of as-built BIM models, particularly for large and complex built environments.Through a case study approach, we focus on three stages of the Scan-to-BIM process: point cloud collection, point cloud processing, and BIM modelling. Each case study employs quantitative methods to evaluate performance across different stages using metrics such as registration error (RE), scan overlap rate (SOR), point error (PE), coverage rate (CR), time consumption (TC), model accuracy (MA), and model-to-point overlap (MtPO). While quantifying the completeness of point cloud models has been a significant challenge in the academic field, this study addresses the issue by introducing a 'voxel centroid approach' combined with the Point Cloud Participation (PCP) rate method, measuring completeness as CR in point cloud models. Moreover, most current research on the quality of as-built BIM models is limited to relatively simple environments. In contrast, this study explores more diverse and complex scenarios, closely aligned with real-world industry applications. From this comprehensive and in-depth investigation of the complex built environment, we have derived several new findings that are more applicable to practical use, along with effective recommendations.
The case studies also demonstrate the practical application of various technologies, including terrestrial laser scanning, mobile laser scanning, and drones equipped with digital photogrammetry for point cloud acquisition. In digital photogrammetry, the Structure from Motion (SfM) method was employed for camera pose calibration, while Multi-View Stereo (MVS) generated dense point clouds. Point cloud registration was achieved using the Iterative Closest Point (ICP) algorithm and its variants, incorporating feature descriptors, Random Sample Consensus (RANSAC) for initial optimization, and kernel functions. As-built BIM models were created both manually in Revit and automatically using Faro As-Built Suite.
| Date of Award | 12 Jun 2025 |
|---|---|
| Original language | English |
| Supervisor | Stephen Rowlinson (Supervisor), Alan Patching (Supervisor) & James Birt (Supervisor) |