Year: 2026
Type: dissertation
Abstract: Realizing the full potential of digital twins in facility management requires prescriptive modeling that does more than describe current states — it must anticipate operational needs, recommend actions, and adapt as conditions evolve. However, this remains difficult due to fragmented data silos and the lack of flexible mechanisms to connect diverse information sources. Most existing methods rely on rigid schemas or predefined ontologies which constrain adaptability. This dissertation proposes a no-model, dynamic linkage framework that connects structured and unstructured datasets without enforcing strict predefined standards. The resulting framework supports prescriptive digital twins (PDTs) that evolve in parallel with the built environment. Built on this foundation is a system that leverages knowledge graphs and Generative Artificial Intelligence (GenAI) to bridge fragmented data sources and drive prescriptive analytics in facility management. The work begins by identifying conceptual and practical limitations in current prescriptive analytics within the built environment. A review of over 600 studies reveals that prescriptive modeling in Architecture, Engineering, and Construction (AEC) is often underdeveloped and rarely connected to operational data. In response, the dissertation proposes a graph-based digital twin architecture that integrates structured and unstructured datasets using adaptive, schema-agnostic representations. One instantiation of the system takes the form of a multi-agent platform powered by large language models and equipped with a user-facing interface for natural language querying. Through extensive evaluation across BIM models and facility work orders, the system demonstrates high query accuracy, semantic alignment with Industry Foundation Classes (IFC) standards, and significant improvement in retrieving actionable insights from unstructured maintenance data. In addition, by combining domain-specific fine-tuning with agentic reasoning, the research shows how GenAI-equipped knowledge graphs can support both interpretive and prescriptive tasks in facility operations. The result is a blueprint for constructing intelligent, user-centric digital twins capable of adapting to dynamic facility environments, and a methodological foundation for future work at the intersection of AI, construction informatics, and operational decision support.
Source: TSpace (University of Toronto)
Author: Babaei Ravandi, Amirhossein
Institution: University of Toronto
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7203958450