Digital Twins (DTs) couple BIM and IoT for sustainable building operation, yet many implementations remain siloed, hard to scale, and difficult to interpret. This paper presents a stepwise DT prototype for an environmental-education facility, integrating indoor/outdoor sensing (CO₂, PM, temperature, humidity, occupancy) with a Revit BIM model. A Dynamo–Python pipeline maps cloud time-series to shared BIM parameters, enabling in-model visualization, threshold alerts, and KPI reporting. Initial deployment delivers the sensing-and-display setup and a reusable BIM data schema ready for automated synchronization. The lasting contribution is a replicable BIM-centric implementation method for room- and zone-level monitoring in sustainable cultural buildings.
This study uses the XGBoost algorithm to predict accident types in pipeline construction projects through a multi-class classification approach. The model was trained on 1,184 real accident records containing information such as accident time, type, and required treatment. Results showed 83% overall accuracy, 0.82 macro-average F1 score, and 0.81 weighted F1 score. While some classes achieved high recall, categories with limited data performed lower. The findings indicate that XGBoost can support accident type prediction and risk analysis in construction projects using historical incident data.
Documenting information about materials in existing buildings is essential for circular construction and the reuse of elements. While multimodal Vision-Language Models (VLMs) can classify materials from images, they often lack the contextual semantic depth required for technical documentation. This study introduces a VLM-augmented Retrieval-Augmented Generation (RAG) framework that utilises ontology-constrained LLM reasoning to bridge this gap. By integrating domain-specific text retrieval and formal ontologies, the method enables the identification of building element layers and material types. The method is validated using a reference dataset of Dutch residential buildings and demonstrated by enriching semantic graphs with predicted material data.
Soviet-era apartment blocks in Eastern Europe are being demolished, generating mixed waste and limiting circular reuse. Existing assessments rely on generic archetypes and lack element-level material detail. This study develops an automated algorithm to detect standardised Soviet building series for scalable material stock analysis. A distance-based classification using apartment layouts, floor area, height, and staircases was validated on 337 buildings and applied to Estonia’s building stock. The model achieved 70% series-level accuracy, with dominant series represent 61% of the stock. This element-level quantification enables targeted deconstruction and higher material recovery for circular economy planning.
To address the shortage of skilled labor in the construction industry, the application of augmented reality on construction sites offers substantial potential to enhance coordination and communication among different trades, thereby reducing labor-intensive processes. This requires the reliable transfer of geometric and semantic information from Building Information Models, including all data relevant to construction execution, through suitable data exchange formats. This paper defines information requirements for the IFC schema for the BIM-to-field use case in the construction trades, discusses appropriate exchange formats for geometric and semantic data, and presents the development of an IFC-to-Unity converter.
Building Information Modeling (BIM) education in architecture, engineering, and construction remains largely tool‑centric, limiting students’ understanding of data quality, automation, and early performance reasoning. This paper presents a project‑based framework that treats BIM models as computational, data‑driven artifacts. Integrating programming fundamentals, API‑based model access, simplified energy and structural reasoning, and human‑in‑the‑loop LLM assistance, the approach fosters computational BIM literacy through learning‑by‑doing. Student feedback indicates improved awareness of modeling consequences, data quality, and performance implications. The framework addresses persistent shortcomings in BIM education and suggests reframing BIM instruction as data quality education through direct computation on models.
Construction projects involve conflicting objectives beyond time and cost, including quality, safety risk, resource stability, and environmental impact. Many-objective optimization (MaOO) helps address this complexity in planning, but it provides limited support for decision-making during execution, and selecting one solution from Pareto-sets often depends on human judgment. To overcome these gaps, this paper proposes an integrated planning and control framework combining Opposition-Based NSGA-III (OBNSGA-III) for six-objective optimization, an Entropy-based-VIKOR model for compromise solution selection, and a Dynamic-MaOO re-optimization mechanism for execution control. A case study based on 25-activities is used to demonstrate effective decision support from planning to execution.
Building maintenance and renovation decisions rely heavily on expert judgment and contextual factors, often requiring manual assessment of conditions, accumulated experience, and locally applicable standards or regulations. This study introduces a multi-agent AI framework designed to automate recommendations for building maintenance and renovation by coordinating task classification, regulatory analysis, and solution generation through LLMs. Specifically, given inputs including a building’s location, type, and reported issues, the framework first uses a task identification agent to classify the case as maintenance or renovation. A knowledge agent then retrieves, structures, and reasons over applicable building codes, standards, and maintenance guidelines specific to the issue. Building on this contextualized knowledge, a solution generation agent produces maintenance recommendations or renovation strategies, or evaluates proposed renovation plans for regulatory compliance. The proposed multi-agent architecture enables the automation of key reasoning steps that are traditionally performed by human experts, including task classification, regulation interpretation, and recommendation formulation.
Distributed Reinforcement Learning (RL) has emerged as a promising approach for coordination HVAC control across buildings in district energy systems. However, existing studies vary widely in system architecture, problem focus and practical feasibility. This paper presents a structured review of recent work on distributed RL for Building Energy System (BES). We introduce a framework that classifies approaches according to the level of distribution and the key problem dimensions addressed, which cover scalability, coordination, privacy, safety and training efficiency. Using a research map, we analyze trends, highlight unresolved challenges, and identify directions toward practically deployable, privacy-aware, and safe distributed HVAC control.
Large-scale linear construction projects involve complex interdependencies, extended sites, and numerous execution variants, while existing approaches lack personalized recommendation systems for trenching applications. This paper presents a simulation-based framework integrating AI-driven optimization methods to support data-informed planning under limited historical data availability. A bottom-up simulation model captures process-specific dynamics and enables synthetic data generation, forming the basis for proposing focused parameter ranges with promising execution variants. Parametric validation across 3,200 simulation runs confirms realistic system behavior. The results indicate that AI-integrated simulation frameworks represent a viable and transferable approach for optimization in data-scarce linear construction environments.
