Ductwork systems, as critical components of mechanical, electrical, and plumbing (MEP) systems, significantly influence buildings' embodied carbon and operational efficiency. Prior studies in carbon-aware assessment for ductwork systems have been limited in efficiency problems and lacking automated analysis. Although large language models (LLMs) provide a solution, hallucination problems exist in generic LLMs. The proper representation of LLM operation inputs and LLM agent integration with the particular workflow remain challenges. Therefore, this paper proposes an automated openBIM-based approach for carbon-aware ductwork design assessment. It starts with developing an integrated ductwork KG (DKG) schema of carbon-aware ductwork assessment, which is followed by developing conversion algorithms to automatically transform the ductwork IFC design to the DKG file. Lastly, a KG-augmented LLM multiagent framework is developed to automatically implement carbon-aware assessment tasks and generate analysis. An office building was used to demonstrate, showing the high accuracy and time efficiency of the proposed method.

The digital transformation of bridge maintenance is critically hindered by the fragmentation of multi-source heterogeneous data and the lack of semantic interoperability across engineering documents. This paper proposes a human-in-the-loop automated construction and refinement framework for Bridge Maintenance Knowledge Graphs (BM-KG). The framework synergizes ontology-guided Large Language Models (LLMs) with enhanced Graph Neural Networks (GNNs) within a systematic "extraction-fusion-completion" pipeline. First, to guarantee high-precision extraction, an ontology-guided LLM module is developed, employing domain constraints and dynamic prompts to transform unstructured text into traceable triples. Second, to ensure rigorous knowledge fusion, a parent-child collaborative strategy is introduced, utilizing spatial hierarchy to uniquely align entities and resolve attribute conflicts. Finally, to facilitate credible knowledge completion, an Ontology-enhanced Hybrid GNN (OH-GNN) is proposed, embedding engineering logic into graph learning to reliably infer implicit relationships. Experiments on a large-scale professional bridge corpus demonstrate the framework's high-precision and superiority: it achieves an F1-score of 91.29% in knowledge extraction, significantly outperforming traditional supervised baselines; attains an entity alignment F1-score of 85.71% in knowledge fusion; and improves link prediction MRR to 0.476 while reducing the OCVR to 5.1% in completion tasks. Furthermore, holistic ablation studies confirm the contribution of each core component. The proposed framework provides an advanced informatics foundation for intelligent lifecycle management and decision support in bridge engineering.

Efficient resource scheduling and delivery are crucial for minimizing delays and costs in construction. Autonomous Mobile Robot (AMR) can enhance operational efficiency and safety. An integrated resource scheduling and grasping framework is proposed for construction robots, aiming to improve overall operational efficiency through fully automated task reception and execution. The proposed mathematical model first generates an initial scheduling plan based on construction tasks and spatial information. Then, the dual-delay strategy-driven swarm optimization algorithm outputs optimized task instructions, guiding the robot to execute tasks and perform precise grasping through the FFB6D-AF network. Experimental results demonstrate that the proposed TDSO algorithm outperforms baseline algorithms across multiple test cases and proposed FFB6D-AF network also improves pose estimation accuracy, with a 5.4% gain in ADD <2 cm. By integrating grasping time into scheduling optimization, the framework reduces total waiting time by 35.9% across three construction scenarios, demonstrating significant operational improvement.

Deep learning (DL)-based point cloud processing has been widely applied to 3D reconstruction in the construction industry. However, DL methods typically require large, fully annotated data sets for effective learning, which can be time-consuming and labor-intensive to produce. This paper proposes a transfer learning method to improve the effectiveness of point cloud tasks for construction scenes. This paper investigates the performance of pretrained models across various DL tasks, algorithms, and transfer learning configurations, specifically fine-tuning and partial transfer learning, and evaluates them in three representative construction scenarios: underground garage data set (UGD); construction site data set (SITE); and pipe system network data set (PSNet). The results demonstrate the following: (1) pretraining on data sets from diverse construction environments improves the mean intersection over union by 8.7%, 8.3%, and 45% on UGD, SITE, and PSNet, respectively, with backbone-only pretraining achieving the best performance; (2) a moderate pretrained sample size combined with a larger training sample size achieves better pretraining results; and (3) pretraining can also improve the accuracy of instance segmentation and point completion.

Automated semantic understanding of dense terrestrial laser scanning (TLS) point clouds is a prerequisite for Scan-to-BIM, digital twin maintenance, and as-built verifcation. Yet for operational industrial mechanical, electrical, and plumbing (MEP) facilities, this challenge remains largely unsolved: water-treatment TLS scans exhibit extreme geometric ambiguity, severe occlusion, and extreme class imbalance that architectural benchmarks such as S3DIS and ScanNet cannot adequately represent. We present Industrial3D, a terrestrial LiDAR dataset with 612.7 million expert-labeled points at 6 mm resolution from 20 room scenes, 13 dataset areas, and 7 operational water treatment facilities. At 6.6x the scale of the closest comparable MEP dataset, Industrial3D provides the largest industrial MEP testbed for within-domain scene understanding. We further establish a cross-paradigm benchmark of nine methods across fully supervised, weakly supervised, unsupervised, and foundation-model settings. The best supervised method reaches 55.74

Instance segmentation of point clouds plays a crucial role in the Scan-to-BIM process. However, current Scan-toBIM methods primarily focus on deep learning (DL)-based semantic segmentation and still rely on machine learning (ML) for subsequent instance segmentation, which requires extensive parameter tuning. Additionally, DL-based methods demand large, manually labeled scan datasets, which are time-consuming and labor-intensive. To address these challenges, this paper proposes a DL-based instance segmentation approach that leverages synthetic point clouds and examines their impact on DL-based instance segmentation. Comprehensive experiments reveal three key findings. First, synthetic point clouds substantially boost model performance: integrating synthetic data with real scans and a pre-trained model improves mean Average Precision at IoU = 50 (mAP50) and mean Average Precision at IoU = 25 (mAP25) by 27.57% and 25.69%, respectively. Second, DL-based instance segmentation surpasses ML clustering once training data covers more than three indoor areas, with synthetic data further enhancing accuracy and stability. Third, synthetic point clouds reduce labeling effort by 90% while achieving comparable segmentation quality, demonstrating strong scalability for real-world deployment.

HVAC systems need to collaboratively optimize energy consumption, air quality, and thermal comfort under dynamic occupancy to address building energy efficiency and indoor environmental quality. However, traditional control strategies often struggle to balance these objectives, especially in enclosed spaces with high occupancy density and sudden variations, such as classrooms, where CO₂ can rapidly exceed safety limits within a short period without effective ventilation. Therefore, this study proposes a physics-integrated hierarchical predictive control framework using reduced-order models calibrated with measured data and a personalized “center-edge-corner” zoning strategy for adaptive spatial load matching. A three-layer decision mechanism (emergency response, optimal control, energy-saving control) enables dynamic multi-zone regulation. Results show that the proposed framework achieves satisfactory thermal comfort, with average temperatures of 23.0 °C in the uniform scenario, 22.6 °C in the center-concentrated scenario, and 22.4 °C in the dynamic scenario. In terms of air quality, the proportion of CO₂ concentration exceeding 800 ppm is 0% in the uniform scenario. In the center-concentrated scenario, the proportion exceeding 1000 ppm is 1.4%, with a 13.2% energy reduction compared to the baseline strategy. In the dynamic scenario, the energy consumption of the proposed strategy is nearly identical to that of the fixed strategy, but the proportion of CO₂ concentration exceeding 800 ppm is reduced from 10.2% to 5.9%. CFD simulation results provide high-resolution quantitative visualizations and further confirm that the proposed framework effectively prevents local pollutant accumulation and overheating in high-density zones. This study provides an efficient and deployable solution for intelligent energy-saving retrofits of existing buildings.

Contrastive learning is effective for aligning paired views or modalities, but alignment beyond two modalities remains non-trivial and comparatively underexplored. Pairwise CLIP-style losses decompose multi-modal alignment into independent two-way comparisons and therefore do not explicitly model higher-order dependencies among multiple modalities. Recent beyond-pairwise objectives approach this problem from statistical or geometric perspectives, but arbitrary-modality alignment still lacks a principled criterion for defining what each modality should preserve and compress relative to the others. We revisit arbitrary-modality alignment through the Information Bottleneck principle. In multi-modal learning, sufficiency should preserve information predictable from the remaining modalities, while minimality should compress modality-specific information not supported by them. This naturally leads to a One-vs-All view, where each modality is characterized with respect to the remaining modalities. We propose OVA-IB, an Information Bottleneck framework for arbitrary-modality alignment. OVA-IB optimizes a tractable One-vs-All contrastive lower bound for sufficiency connected to a Dual Total Correlation-style objective, uses a parameter-free geometry-aware projection score, and derives a tractable upper-bound regularizer for minimality by bounding each representation's dependence on its own input with representation distributions induced by the remaining modalities. Experiments on classification, regression, modality-agnostic evaluation, and cross-modal retrieval benchmarks demonstrate strong and robust performance.

Three-dimensional (3D) object detection based on point clouds plays an important role in building and construction scene analysis. However, deep learning (DL)-based object detection heavily relies on large, manually labeled point cloud datasets, which are timeconsuming and labor-intensive to create. To address these challenges, this study proposes a building information modeling (BIM)-based synthetic point cloud generation method to enhance the accuracy of DL-based object detection for building indoor scenes. A total of 53 comparative experiments were conducted to evaluate object detection performance under varying algorithms, generation methods for synthetic point clouds, training datasets, and application scenarios. The results indicate that (1) incorporating synthetic point clouds into real point clouds with a pretrained model improves mAP50 by 22.84% and mAP25 by 22.79% compared to using only real-world point clouds; and (2) replacing real-world point clouds with synthetic samples reduces the time required for preparing training data by more than 90% while maintaining comparable performance.

Ensuring automatic safety inspection on dynamic construction sites requires reliable multi-entity visual interpretation and rule-based compliance checking, which conventional computer vision systems struggle to achieve due to limited semantic reasoning. Although modern vision-language models (VLMs) support end-to-end multimodal reasoning, generic VLMs lack fine-grained visual grounding for safety-critical inspection. This study proposes a transparent, object-centric VLM framework that (1) formalizes four visually grounded reasoning strategies-Enumerate Anchor Entities, Key-feature Identification, Sequential Inspection and Explicit Visual Grounding to guide VLMs toward evidence-based inspection; (2) auto-generates a self-instruct visual-grounding dataset for cold-start supervised fine-tuning, enabling internalization of inspection behaviors; (3) optimizes rule-level accuracy using Group Relative Policy Optimization with designed ComplianceReward function that reinforces reasoning trajectories leading to correct answers. Trained on Qwen3-VL-8B, the proposed agent achieved 0.916 violation recall and 0.904 overall accuracy in real-site evaluation, outperforming baseline models. The framework delivers scalable, interpretable, and domain-aligned multimodal reasoning for automated construction safety inspection.

The construction industry is progressively integrating Environmental, Social, and Governance (ESG) principles to foster sustainable development. These principles are vital for promoting responsible practices that address the environmental and social impacts of construction activities. However, existing ESG management approaches are often inefficient and costly, posing significant challenges to the industry. The complexity and fragmentation of ESG data management hinder the ability to monitor and report sustainability indicators in an authentic and effective manner, thereby limiting the industry’s progress in achieving its sustainability goals. Based on this situation, this paper proposes a comprehensive data management framework leveraging 4D Building Information Modelling (BIM) to enhance sustainability practices in construction projects. The proposed framework aims to streamline ESG data collection, processing, and analysis, ultimately improving the efficiency and effectiveness of ESG management in the construction sector. The study focuses on two main areas: (1) Identified a construction ESG knowledge graph for openBIM, (2) Proposed a construction ESG data collection and processing framework for 4D BIM, and (3) Developed smart contracts to serve the blockchain in framework. The framework simplifies the ESG data collection process in construction projects, enabling real-time monitoring and optimization of sustainability performance. We verified the feasibility of the proposed framework based on the Hong Kong OPark2 project, which demonstrated the availability of the designed data flow and technical components.

Crowdsourced street-view imagery from social media provides valuable real-time visual evidence of urban flooding and other crisis events, yet it often lacks reliable geographic metadata for emergency response. Existing image geo-localization approaches, also known as Visual Place Recognition (VPR) models, exhibit substantial performance degradation when applied to such imagery due to visual distortions and domain shifts inherent in cross-source scenarios. This paper presents VPR-AttLLM, a model-agnostic framework that integrates the semantic reasoning and geospatial knowledge of Large Language Models (LLMs) into established VPR pipelines through attention-guided descriptor enhancement. By leveraging LLMs to identify location-informative regions within the city context and suppress transient visual noise, VPR-AttLLM improves retrieval performance without requiring model retraining or additional data. To evaluate the framework, we conduct comprehensive testing across two morphologically distinct urban environments: San Francisco and Hong Kong. The evaluation utilizes established query sets, synthetic flooding scenarios, and real social media flood images integrated into the San Francisco benchmark, alongside a newly curated Hong Kong dataset. Integrating VPR-AttLLM with three state-of-the-art VPR models-CosPlace, EigenPlaces, and SALAD-consistently improves recall performance, yielding relative gains typically between 1%-3% and reaching up to 8% on the most challenging real flood imagery. Beyond measurable gains in retrieval accuracy, this study demonstrates a robust pipeline for LLM-guided multimodal fusion in visual retrieval systems. By embedding principles from urban perception theory into attention mechanisms, VPR-AttLLM bridges human-like spatial reasoning with modern VPR architectures. Its plug-and-play design, strong cross-source robustness, and interpretability highlight its potential for scalable urban monitoring and rapid geo-localization of crowdsourced crisis imagery in heavily urbanized environments.

Manual constructability checking of complex mechanical, electrical, and plumbing (MEP) designs is inefficient and error-prone. However, automating this process is hindered by the complexity of MEP drawings, the variety of design specifications, and the insufficient integration of checking algorithms with fragmented checking processes. Addressing these issues, this paper proposes an automated checking system that integrates large language models (LLMs) and graph technologies within a collaborative multi-agent framework. The system comprises three components: the MEP graph agent that constructs structured scene graphs from drawings; the MEP rule agent that converts textual rules into knowledge graphs; and the MEP checking agent that identifies violation via graph matching and reasoning. Validated on three real-world projects, the system achieved an average accuracy of 92% in violation detection. By offering an intuitive interface for issue identification, this approach significantly enhances coordination efficiency and reduces manual effort. This framework establishes a robust foundation for future AI-driven automated compliance checking in the construction domain.

Amid rising sustainability demands, Environmental, Social, and Governance (ESG) management is gaining traction in construction. Accurate, dynamic spatial information on construction progress is vital for quantifying environmental performance. However, dynamic environmental performance evaluation during construction remains underexplored. LiDAR enables accurate spatial data acquisition yet linking it to actionable environmental metrics poses challenges. This paper introduces a BIM-LiDAR integration framework featuring three contributions: (1) proposing a semantic knowledge graph for construction data and environmental performance metrics; (2) developing a BIM enrichment mechanism via multimodal large language model (MLLM) -based point cloud segmentation; and (3) establishing unified framework leveraging enriched BIM to quantify construction progress and environmental impacts. Tested on a construction project, it achieved 96.54% segmentation accuracy. Through precise geometric verification, it enabled the measurement and validation of a 5.77% improvement in the project's environmental performance across key milestones, transforming manual reporting into a credible, data-driven foundation for environmental management.

This paper presents a locally deployable emergency response system that empowers building-level decision-making by integrating large language models (LLM) with BIM-based knowledge graphs. To address hallucinations and the absence of localized context in LLM, the system combines expert knowledge retrieval and spatial reasoning through two modules. An expert module generates context-aware emergency guidance through vector-based retrieval and task-oriented prompting, while a spatial reasoning module uses BIM-based encoding of building topology and equipment metadata into a knowledge graph to support routing and equipment localization. The improved DeBERTa model identifies query keywords and matches them with knowledge graph, achieving an F1NER of 86.4% in emergency term recognition. Results demonstrate the system’s superiority over baseline agents, with RAG-based agent achieving broader response coverage and an F1SC of 78.6%. The proposed system has potential to support secure, context-specific, and actionable emergency response and preparedness in high-risk environments such as hospitals, factories, and transit hubs.

Automating compliance check for geometry-intensive regulations remains a significant technical bottleneck in Building Information Modeling (BIM), primarily due to the semantic disparity between high-level regulatory logic and structured IFC data. Existing methods, often reliant on static rule templates, struggle to traverse multihop reasoning chains or resolve latent spatial dependencies across multiple building entities. To address these challenges, a Spatial-Geometric Reasoning System for Building Information Modeling (SGR-BIM) is proposed as an integrative graph-driven reasoning framework. SGR-BIM dynamically constructs a cross-modal knowledge graph that aligns user intent, regulatory semantics, and BIM geometry, enabling interpretable reasoning without rigid hard-coding. Validated on 679 expert-verified queries from fire safety codes, the framework achieves 84.3% accuracy, representing an 8.6% improvement over enhanced-tool single-agent baselines. This research provides a graph-based semantic reasoning paradigm, enhancing the transparency and flexibility of automated geometric compliance check workflows in the Architecture, Engineering, and Construction (AEC) industry.