The potential for material reuse is restricted by the lack of methods for integrating reclaimed materials into facade design workflows. This paper presents a multi-agent pipeline using LLMs to support integrating reclaimed materials into facade design. Tested through a web application in two design modes: a collaborative mode, where human and LLM collaborate, and an independent mode, without human guidance; the users reported the collaborative mode to be more useful. The proposed AI-human collaboration framework improved the tasks of searching for materials, evaluating options, and visualizing design iterations, highlighting the value of AI-driven design frameworks in encouraging reuse in architecture.
Current BIM-based compliance checking still relies heavily on manually interpreting construction standards, as regulatory texts are complex and unstructured. This limits scalability, consistency, and automation. This paper presents an NLP-based pipeline using a fine-tuned T5 transformer model to automatically extract structured Information Requirements from German construction standards (DIN 4108-2, -3, and -4). A manually annotated dataset of 1,007 sentences is used for supervised training. The model achieves an F1-score of 92.86% on test data and 88.03% on unseen regulations. Extracted rules are integrated into a BIM compliance checking workflow using SPARQL and IFC-based models, enabling automated, machine-processable regulatory verification.
Traditional computer vision struggles with domain-specific few-shot building element detection in street-view imagery, creating a bottleneck for extracting the building information needed for data-driven retrofit matching. This paper proposes a few-shot framework that combines a pre-trained Vision Transformer with a Random Forest patch classifier and adaptive cropping to address domain shift and limited annotations. With only 21 labeled images per element, the method achieves mean detection accuracy greater than 0.95 across seven elements and enriches building stock data at address level, enabling more accurate matching of retrofit measures.
Automated symbol recognition in Piping and Instrumentation Diagrams (P&IDs) is challenging due to varying standards and complex structures where text, lines, and symbols are intermingled. This study proposes a non-learning-based framework applying extended template matching using the symbol legend. The framework extracts the Region of Interest (RoI) masks via the Convex Hull algorithm and performs matching using Intersection over Union (IoU) as a similarity metric to reflect spatial similarity. Subsequently, Intersection over Minimum (IoM)-based Non-Maximum Suppression (NMS) is introduced to suppress duplicate detections. Experiments on real-world industrial P&IDs achieved an F1 Score of 0.992, demonstrating its applicability to P&ID digitization.
Industry Foundation Classes (IFC) enable interoperability across the AECO sector, yet IFC2x3 and IFC4 remain constrained by monolithic structures, limited external data integration, and weak support for real-time and web-based workflows. This paper applies comparative schema analysis and domain-based evaluation across four lifecycle phases, design, construction, operation, and deconstruction, to critically assess these limitations. Emerging IFC5 concepts are examined against identified requirements, highlighting both improvements and unresolved challenges, including governance complexity and implementation maturity. Results indicate increased flexibility and potential, alongside new dependencies and trade-offs. In conclusion, a forward-looking assessment and early-adoption priorities for IFC5-enabled data-centric workflows are provided.
The data-driven nature of digital twins (DT) has led to their promotion as a solution to challenges in the facility management sector. However, existing conversations are largely technically oriented, offering limited insight into how DTs are created in practice. This study addresses this gap by examining the process of creating a Matterport-based DT. Using vignette-based ethnography and the Social Construction of Technology framework, the study analyses how this process unfolds. The findings show the DT creation as emergent, shaped by user interpretations, social and environmental conditions. Overall, this study contributes an empirically grounded sociological account of the DT creation process.
The built environment faces pressure to improve efficiency and sustainability throughout its lifecycle. Digital twins (DT), leveraging BIM, IoT sensor data, and AI, offer significant potential for proactive and intelligent asset management. However, deployment is hindered by the Interoperability Imperative—the need for heterogeneous systems to exchange and interpret data seamlessly. This paper develops a Maturity–Barrier Logic that frames DT evolution as constrained by interoperability-related barriers. It identifies a persistent implementation gap, with applications stagnating at the Digital Shadow and failing to achieve autonomous Cognitive Twins. A diagnostic framework is proposed to support the transition toward federated, system-of-systems DT.
Conventional office fit-outs are designed for fixed requirements, which limits their long-term adaptability. This paper introduces a workflow that integrates configurable floor plan graphs with contextual reasoning from large language models (LLMs) for flexible office layout exploration. Interior space is represented as a graph that encodes geometry, adjacency, and accessibility, and is parameterised by binary partition decisions. For each candidate configuration, an LLM infers program assignments across multiple scenarios, and solutions are evaluated in a multi-objective combinatorial optimisation. Results show configurations with feasible performance across scenarios, suggesting that LLM-based contextual program assignment can complement solver-based methods focused on quantitative optimisation.
Building Information Modelling provides the geometric and semantic foundation for Digital Twins (DT), but its effectiveness depends on the availability of use-case-specific information. In OpenBIM workflows, Industry Foundation Classes (IFC) models often contain missing or incomplete information, limiting reliable integration with DT platforms. This paper presents an envelope retrofit-oriented data quality control using Information Delivery Specifications (IDS). Minimum requirements were formalized into IDS and validated against an IFC model. Results show that while geometry was generally complete, critical semantics, particularly material and thermal properties, were frequently missing. IDS-based validation enables early detection of data gaps and improves IFC data readiness.
Enabling predictive maintenance (PM) of road infrastructure by understanding condition developments requires accounting for internal and external influences. However, the corresponding data are highly heterogeneous and distributed, limiting comprehensive large-scale analysis. To bridge this gap, an existing Digital Twin (DT) approach is enhanced with environmental context data. Knowledge graphs enable scalable querying across heterogeneous subsystems, while a federated database addresses data distribution. The framework is validated using real-world data from Germany. Linking historical climate records with bridge condition development demonstrates comprehensive querying. By enabling scalable cross-system data association, the proposed DT framework enhances the practical applicability of existing PM methods.
