Automatic as-built BIM reconstruction for MEP systems from point cloud

• Main Author: Xie, Yuan
• Other Authors: Cai Yiyu
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; As-built BIM reconstruction; MEP system; Scan-to-BIM
• Online Access: https://hdl.handle.net/10356/174798
• Institution: Nanyang Technological University

Building Information Modeling (BIM) is gradually recognized and promoted as the new standard practice in the construction industry as well as the built environment. Mechanical, electrical, and plumbing (MEP), as a system requiring regular maintenance, takes an important place in the building operation and maintenance; its failure can cause significant impacts operationally, economically, and even environmentally. As-built BIM is necessary to enable efficient facility management for buildings without updated BIM. Modeling of the as-built BIM is currently practiced in a very manual and laborious way, requiring a considerable amount of time and effort even for a skilled modeler. This research proposes methodologies for the automatic reconstruction of as-built BIM of MEP systems from registered point clouds. The target MEP components include pipes with circular cross-sections and ducts with rectangular cross-sections. A novel neural network called PipeNet is developed to detect pipes and predict pipe parameters. Its novel dynamic scale manager can automatically and dynamically select a set of neighbor points suitable for the classification and parameter estimation. Ducts are detected by their 90-degree edges as a stable feature using a neural network called EdgePointNet. Bottom-up classification mechanism is embedded in the architecture design for both neural networks so that small targets in large scenes can be robustly detected. Then, a series of geometric processing algorithms are developed to reconstruct the pipe and duct models. Lastly, a graph-based geometric-constrained connectivity analysis is proposed to reconstruct the piping and duct systems faithfully to the input data with enhanced completeness guided by domain-knowledge-enabled interpolation. The final models of the systems are refined with coherence and converted to the Industry Foundation Classes format which is neutrally acceptable in the BIM industry. Compared to existing works, our solutions requires no additional data other than the unstructured point cloud with XYZ fields, no pre-processing to the input registered point clouds, and no prior knowledge of the targets' directions or dimensions. The integration of the developed solutions into a comprehensive scan-to-BIM system is described with modifications to coordinate with structural BIM and deploy in a progressive scan-to-BIM workflow. The solutions are thoroughly validated on both synthetic and actual scan data, and the results demonstrate its robustness, fast speed, and high accuracy. At last, contributions, limitations, and future works are summarized and discussed.