As the adoption of open Building Information Modeling (openBIM) gains traction, the industry faces a significant bottleneck: the complex, multifaceted nature of building lifecycle data. This complexity makes information retrieval a daunting task, requiring end-users to have a profound familiarity with underlying data structures like IFC schemas. Yet, many construction industry professionals lack the technical skills to formulate formal queries in languages like SQL. This technical gap impedes the broader adoption of openBIM, posing a barrier to the industry's advancement.
While the research community has begun exploring natural language-based searches to mitigate these challenges, the rapid advancements in deep learning-based natural language processing have added another layer of complexity. It has become increasingly difficult to navigate the balance between prediction accuracy and computational costs in domain-specific applications.
Our study addresses this issue by focusing on the state-of-the-art deep learning techniques for “neural semantic search”. To make the system adaptable across various applications and disciplines, we've introduced a semantic annotation scheme rooted in the IFC schema. Through a comparative approach, we've evaluated the efficacy of traditional and emergent deep learning architectures for this specific task. Our findings highlight the importance of domain-specific, context-aware text representation learning for effective information retrieval.
In summary, our project bridges the technical knowledge gap and enhances the intuitiveness of openBIM tools, allowing for a more accessible and efficient information retrieval process, thereby paving the way for broader industry adoption.
Our solution, rooted in the advancement of neural language models, fundamentally transforms the way stakeholders interact with BIM models by enabling intuitive, natural language-based queries. By doing away with the need for traditional templates or object tree-based searches, we streamline access to complex, multi-faceted building data.
The cornerstone of our approach is the semantic contextual understanding of user queries, anchored in the Industry Foundation Classes (IFC) schema—the backbone of openBIM. By putting IFC schema at the center of our system design, we ensure that our solution is intrinsically tied to the mature industry standards. It accommodates the depth and complexity that professional applications demand while making the system vastly more user-friendly.
Our proposed system enables a diverse group of stakeholders—from architects to contractors—to focus on leveraging BIM data for informed decision-making, rather than the cumbersome task of query formulation, or manual navigation within the complex layers of BIM data. Importantly, our system is designed to comprehend user intent in a context-aware manner (i.e. not solely based on keywords and synonyms) and return precise relevant information. For example, users can effortlessly retrieve specific information, such as the number of doors, types of materials used, or total area of particular flooring. This capability streamlines the process of quantity estimation, reducing both the margin of error and the likelihood of costly oversights.
In sum, our project not only makes BIM more accessible but also more efficient and effective, fundamentally altering the way the construction industry interacts with digital representations of the built environment.