The Architecture, Engineering, Construction, and Operations (AECO) industry faces challenges in seamless data exchange due to centralized, siloed systems that hinder collaboration throughout the building lifecycle. This research addresses these deficiencies by proposing a conceptual AECO data space prototype using a federated system architecture. By relying on state-of-the-art web technologies, the proposed implementation enables nested data spaces, metadata management, and role-based access control, ensuring transparency, scalability, and user-friendliness. Stakeholder feedback highlights the prototype's user-friendliness, role-based data access, and centralized data visibility while also revealing a lack of user awareness regarding decentralized data storage mechanisms.
Facility management involves processing structured sensor data and unstructured maintenance records to monitor building performance and support decision-making. This study presents a graph-enhanced, multi-agent system that integrates a structured data storage system, a knowledge graph, and a GPT-powered LLM execution framework. A web-based interface enables users to submit natural language queries and obtain actionable insights. A case study on the Galbraith Building at UofT demonstrates how the system links historical maintenance records with sensor data. The results highlight its potential for predictive and prescriptive maintenance, while also reducing manual data processing, decreasing reliance on technical expertise, and improving operational efficiency.
Ensuring the safe and uninterrupted operation of tunnels necessitates continuous monitoring and maintenance. Digital methods, including digital twins, have proven their effectiveness in reducing operating costs but require digital models that are often unavailable for existing tunnels.Therefore, this paper conceptualizes a deep learning-based approach to transform unstructured tunnel documentation into digital models through three core modules: (1) reconstructing the tunnel geometry from construction drawings, (2) analyzing inspection reports to create a damage model for structural condition assessment, and (3) analyzing point clouds, videos, and product data sheets to model the tunnel ventilation system.
A hydropower plant digital twin is a virtual model that integrates real-time data and simulations to optimize performance, monitoring, and maintenance. Its development enhances efficiency, reduces downtime, and enables predictive maintenance. Geometry generation is key, providing precise 3D models for simulations. However, creating geometries from laser scans is challenging due to complex structures, occlusions, and noisy data. This research presents a robust geometric segmentation approach, improving digital twin creation. It outlines a pipeline for point cloud segmentation, component relationships, and mesh generation, enhancing accuracy and efficiency for better decision-making and predictive capabilities in hydropower management.
Disassembly is an important strategy in achieving material circularity and closing the loop in material flow. While disassembly sequencing is heavily studied in manufacturing, few examples consider the constraints present in construction. To address this gap we propose a graph based method for disassembly sequencing with constraints on stability and internal stress. We test our algorithm on three frame structures of increasing complexity and element count using construction specific heuristics to determine source nodes and disassembly direction. Our algorithm can compute feasible disassembly sequences with sufficient speed to support applications in online robotic path planning.
Automatic fault detection and diagnosis (FDD) is essential for energy efficiency and indoor air quality. Unsupervised FDD methods address the need for labeled data but struggle to identify root causes. This paper introduces a Graph Deviation Network (GDN)-based method for detecting and diagnosing faults in time-series data. GDN models variable relationships and enhances explainability using attention weights. Applied to FCUs in a building case study, it determines fault extent and uses rule-based diagnosis to classify faults. Results show superior anomaly detection and sensor correlation modeling, providing users with insights into the root causes of detected faults.
Existing AI research in BIM often overlooks the multimodal nature of BIM data. This study proposes a graph representation learning approach to embed multimodal design data into high-dimensional vectors for supporting learning. Specifically, large text embedding models are utilized to encode object attributes, while geometries are embedded by a point cloud-based approach. The individual embeddings are concatenated and processed by GraphSAGE to leverage graph topology. An object classification experiment on a self-constructed graph dataset achieved 97.7% accuracy and an F1 score of 0.97 using attributes and graph topology. Future work will refine geometry embeddings and explore broader BIM applications.
The construction sector must accelerate progress towards net-zero targets. Current digital twin frameworks lack integrated carbon data, limiting their effectiveness for sustainability. Industry Foundation Classes (IFC), as an international open data standard, offer the potential to address this gap without exacerbating interoperability issues. This paper proposes extending the IFC schema to include carbon data, focusing on road infrastructure. This study maps the key entities for carbon management to a granular level and adds new carbon property sets to the key entities. It establishes a data foundation for computational carbon management of road infrastructure.
Recent research has increasingly focused on the automated checking of building permits, either in very limited application areas, using isolated standards, or using completely new approaches. This work presents a holistic approach to data-driven building permits, including BIM-based collision checking and calculation, combining the international standards IFC and an extension of the IDS format, as well as the national standard XPlanung. While the holistic approach offers the possibility of broad use and flexible extension with other approaches, the combination of open standards within a model checking process demonstrates their wide range of applications in automating and supporting checks.
