Intelligent BIM Searching via Deep Embedding of Geometric, Semantic, and Topological Features

Abstract

As a digital representation of buildings, building information models (BIMs) encapsulate geometric, semantic, and topological features (GSTFs), to express the visual and functional characteristics of building components and their connections to create building systems. However, searching for BIMs pays much attention to semantic features, while overlooking geometric and topological features, making it difficult to find and reuse rich knowledge in BIMs. Thus, this study proposes a novel approach to intelligent BIM searching by embedding GSTFs via deep learning (DL). First, algorithms for extracting GSTFs from BIMs and identifying required GSTFs from search queries are developed. Then, different GSTFs are embedded via DL models, creating vector-based representations of BIMs or search queries. Finally, similarity-based ranking is adopted to find BIMs highly related to the queries. Experiments show that the proposed approach demonstrates an efficiency of 780 times greater than manual retrieval methods and 4–6% more efficient than traditional methods. This study advances the field of BIM searching by providing a more comprehensive, accurate, and efficient method for finding and reusing rich knowledge in BIMs, ultimately contributing to better building design and knowledge management.

Keywords

• building information model (BIM)
• model search
• feature extraction
• similarity calculation
• deep learning
• feature embedding