An Automated Reconstruction Approach of Mechanical Systems in Building Information Modeling (BIM) Using 2D Drawings

Operation and maintenance of commercial facilities rely on complex building information to locate assets and troubleshoot equipment failures. Although building information models (BIM) provide an integrated information repository for searching for facility information, currently more than 90% of existing buildings in the U.S. were still built with 2D drawings. Manually reconstructing BIM, especially the mechanical systems, has been proven to be very labor-intensive. Hence, previous research studies have investigated automated 3D reconstruction approaches using point clouds data obtained by laser scanner techniques or 2D photos. But due to the needs for a line of sight, these approaches are not applicable for constructed buildings. Researchers have also developed approaches that convert 2D drawings to spatial building models, but the previous studies are limited to generate architectural components such as the wall, doors, and windows. Thus, this paper investigated challenges and approaches to automatically generate models for mechanical systems in buildings using 2D drawings. We analyzed the contents and characteristics of the mechanical components that are represented in drawings and developed a set of classification and algorithms that support the automated recognition of the spatial information and metadata of the mechanical systems. A software framework was proposed to utilize the developed computational approach to recognize 2D drawings for BIM reconstruction. The results of the prototype demonstrated that more than 80% of ducts can be recognized from various drawings in DXF format.