Building Information Graphs (BIGs): remodeling building information for learning and applications

Abstract Despite significant advances in Building Information Modeling (BIM) and increased adoption, numerous challenges remain. Discipline-specific BIM software tools with file storage have unresolved interoperability issues and do not capture or express interdisciplinary design intent. This hobbles machines’ ability to process design information. The lack of suitable data representation hinders the application of machine learning and other data-centric applications in building design. We propose Building Information Graphs (BIGs) as an alternative modeling method. In BIGs, discipline-specific design models are compiled as subgraphs in which nodes and edges model objects and their relationships. Additional nodes and edges in a meta-graph link the building objects across subgraphs. Capturing both intradisciplinary and interdisciplinary relationships, BIGs provide a dimension of contextual data for capturing design intent and constraints. BIGs are designed for computation and applications. The explicit relationships enable advanced graph functionalities, such as across-domain change propagation and object-level version control. BIGs preserve multimodal design data (geometry, attributes, and topology) in a graph structure that can be embedded into high-dimensional vectors, in which learning algorithms can detect statistical patterns and support a wide range of downstream tasks, such as link prediction and graph generation. In this position article, we highlight three key challenges: encapsulating and formalizing object relationships, particularly design intent and constraints; designing graph learning techniques; and developing innovative domain applications that leverage graph structures and learning. BIGs represent a paradigm shift in design technologies that bridge artificial intelligence and building design to enable intelligent and generative design tools for architects, engineers, and contractors.