Operational-level maintenance is essential to ensuring the normal functions of systems and equipment in the industry. The current practice for decision-making for operational maintenance is still expert-led and heavily relies on experts’ implicit knowledge from their experience. This leads to the urgent need for better knowledge management, which can facilitate more efficient and effective operational maintenance. To fulfill this need, knowledge graphs (KGs) have been proposed by previous studies, due to their ability to digitize knowledge and their structural and semantic features to facilitate various decision-making approaches. Although multiple KG-based methods have been proposed and implemented, several questions about maintenance-related knowledge graphs remain unclear, including 1) What methods are employed to exploit the capabilities of KGs in maintenance decision-making, 2) In which scenarios KGs are applied, and how they are utilized, and 3) What are the challenges and opportunities of implementing KGs in practice. To address these questions, this study made a comprehensive systematic review. A total of 270 papers were retrieved from Scopus, and 76 papers were closely examined. The review finds that 1) KG-based querying, reasoning, and enhancement of other algorithms are common methods in decision-making problems; 2) most KGs are applied in fault diagnosis, but KGs also contribute to maintenance action identification and organization; and 3) most applications of KGs in industrial maintenance decision-making are limited to simple tasks and more exploration of efficiently using and sharing KGs is needed. This study summarizes the state-of-the-art works in knowledge graphs for operational maintenance, which can facilitate further research.

Ground Penetrating Radar (GPR) is a widely adopted non-destructive sensing technology for subsurface inspection in civil and transportation engineering. Despite its potential for pavement condition assessment, the large-scale application of GPR in automated inspection has two key challenges: the scarcity of annotated real-world datasets and the lack of deep learning models designed for the unique characteristics of 3-Dimensional (3D) GPR data. This study addresses these limitations by firstly introducing a cost-effective data preparation pipeline that integrates orthomosaic Red Green Blue (RGB) imagery with 3D GPR scans to generate annotated 3D GPR datasets. The proposed method uses the aligned segments of RGB and GPR data, using pavement surface images as a reference to transfer labels of surface-visible defects to corresponding GPR segments, enabling efficient large-scale annotation in a real-world dataset collected on a highway section under operation. In addition to the dataset contribution, we propose a specialised 3D Convolutional Neural Network (CNN) architecture incorporating residual connections, mixed convolutional kernel sizes, and both depthwise and channelwise attention mechanisms to enhance feature representation and defect classification. The model is evaluated on binary classification tasks for detecting patch and crack defects in pavement structures. Experimental results demonstrate that the proposed network outperforms baseline architectures across multiple evaluation metrics. Ablation studies further confirm the effectiveness of the designed architectural components. This work contributes a scalable and practical method for real-world dataset generation, along with a novel deep learning framework.

IntroductionAccurate prediction of the bearing capacity of double shear-bolted connections in structural steel is essential for ensuring safety and efficiency in structural design. This study explores the application of ten machine learning algorithms to enhance prediction accuracy while addressing the interpretability challenges often associated with such models.MethodsModels were tuned with 10-fold crossvalidation and assessed using RMSE, R2 and a20 accuracy index. A comprehensive sensitivity analysis evaluates the influence of input parameters, while advanced interpretability techniques, such as partial dependence plots, accumulated local effects, and Shapley additive explanations, are employed alongside parametric studies to elucidate the decision-making processes of the models.ResultsThese methods facilitate the identification of critical variables that influence bearing capacity predictions at both local and global scales.DiscussionThe study demonstrates that machine learning can be a trustworthy and data-driven complement to conventional mechanics-based approaches, when coupled with rigorous interpretability, advancing both safety and efficiency in steelconnection design. The findings highlight the potential of interpretable machine learning approaches to not only improve predictive precision but also provide actionable insights into complex model behaviours, ultimately advancing structural engineering practices and promoting data-driven design methodologies.

The rapid advancements in digital twins technology and intelligent transportation systems are reshaping the landscape of modern road infrastructure and autonomous mobility. As smart cities evolve, the integration of digital twins with road networks and vehicle technologies becomes increasingly critical for improving efficiency, safety, and sustainability. This study presents a systematic review of research on road digital twins, smart vehicles, and autonomous vehicle technologies. The scientometric analysis was conducted on the data acquired from the Scopus database using VOSviewer to generate insights into research trends, citation patterns, and the interrelationships between key topics in the field. The results reveal a growing emphasis on intelligent systems, vehicle-to-everything communications, multi-agent systems within road digital twins, and smart vehicle technologies. However, gaps remain in integrating digital twins with road infrastructure and applying virtual and mixed reality for simulation and human-machine interaction purposes. This paper identifies the emerging trends, gaps, and prospects for research in these areas, focusing on enhancing the integration of smart or autonomous vehicles with smart city infrastructure. The findings shape future research directions and development priorities in road digital twins systems.

Abstract Building information modeling (BIM) is a collaborative process for creating and managing digital representations of physical and functional characteristics of facilities, central to the digital transformation of the architecture, engineering, and construction (AEC) industry. Despite OpenBIM standards like industry foundation classes (IFC), cross-platform interoperability remains a significant challenge to AEC digital transformation. Using a mixed-methods approach, this study examines BIM interoperability issues and hidden costs through interviews with 15 Chinese AEC professionals and a survey of 216 respondents. Key findings include: (1) only 14.75% regularly use IFC, with limited IFC understanding; (2) over half experience attribute loss and multiple concurrent interoperability issues; (3) only 16.13% can quantify interoperability costs; and (4) larger firms and those with diverse tools face more frequent problems and hidden costs. Although the AEC industry recognizes these issues, willingness to invest in solutions remains low. This study contributes to the body of knowledge by providing the first large-scale empirical benchmarks on OpenBIM interoperability practices in China, proposing a four-category hidden cost taxonomy, and developing a multilevel associative framework linking institutional, organizational, and technical factors to cost manifestations. This paper offers evidence-based recommendations for stakeholders.

The construction industry has a profound impact on the environment, consuming significant amounts of energy, water, and raw materials, while also generating waste and greenhouse gas emissions. To mitigate these effects, it is crucial to adopt sustainable construction practices, which are often supported by environmental certification. A key component of these certifications is Life Cycle Assessment (LCA), a process that is frequently hampered by labor-intensive and time-consuming manual data handling. This study introduces an enhanced BIM-based decision support system that integrates LCA processes with advanced technologies, including Large Language Models (LLMs). The system automates the extraction, transfer, and analysis of environmental data from BIM models using tools like Revit, Dynamo, and LCA databases. The open-source model “Meta-Llama-3.1-8B” is utilized to analyze LCA results, recommend sustainable material alternatives, and facilitate automated scenario comparisons through trade-off analyses. These features enable real-time feedback, dynamic decision-making, and improved optimization of material selection for sustainability. The TR building at UPC serves as a case study to validate the system. The findings indicate significant efficiency gains, enhanced decision-making capabilities, and reduced environmental impacts. By incorporating LLMs, this study addresses existing gaps in BIM-LCA integration and offers a scalable and adaptable framework for sustainable construction, thereby facilitating real-time environmental assessments and supporting certification requirements with greater precision.

Digital twins (DTs) allow the digitalization of road infrastructure inspection, though this is hindered by limited annotated data. This work exploits the relational nature of continuous asset condition monitoring to reformulate image-based defect detection as image difference classification (IDC) to reduce data reliance. This was evaluated in a case study on low-resource traffic sign inspection with different IDC classifiers using a newly-curated, high quality dataset. Results indicate that the instruction-based classifier outperforms encoder-based ones and gains from comparison with reference images. This shows that IDC can be an effective task modeling for tackling data constraints in infrastructure inspection and DT asset condition updating.

Occlusions challenge the scan-to-BIM pipeline; while point cloud completion methods show impressive results on synthetic benchmarks, performance on real-world scans lags significantly. This paper investigates if this is caused by bounding box information leakage during input normalisation, letting models exploit implicit context unavailable in practice. This leakage is rectified by normalising partial inputs to create an unbounded completion task. Modifications are introduced to aid the model in learning object bounds and point confidences. Experiments demonstrate existing models rely heavily on leaked bounds, whereas the proposed adjustments significantly improve performance in realistic settings, achieving a 25% and 40% reduction in Earth Mover's Distance and Chamfer Distance. Findings establish that addressing boundary leakage is critical for real-world applicability, providing scan-to-BIM practitioners with methods that generalise across domains and enhance downstream tasks such as meshing. This paper establishes that future research must focus on completion methods that operate without implicit spatial priors.

BIM and semantic graphs are increasingly used in the AEC industry. Converting indoor point clouds into standards-compliant IFC BIM and graph representations remains non-trivial and often requires expert knowledge and manual post-processing steps. To address this, an LLM-enabled multi-agent framework is proposed to automate Scan-to-BIM and Scan-to-Graph from point clouds, with support for natural-language, intent-driven reconstruction. The framework integrates query interpretation, point-cloud inference, topology-aware IFC generation, semantic graph construction, and multi-stage validation into a single end-to-end pipeline, explicitly handling of wall-wall junctions and wall-slab alignment. The framework is evaluated on three datasets captured with different sensing modalities (mobile laser scanning, RGB-D, and synthetic). On the TUMCMS test set, a laser-scanned point cloud dataset captured in office environments at the Technical University of Munich, room areas of generated IFC files achieve mean absolute deviations of 1.4m(2) (Llama-based) and 1.51m(2) (Qwenbased), corresponding to mean relative deviations of 4.70% and 5.10% versus manually created BIM. The results demonstrate robust and generalized performance of the framework for wall and space reconstruction across datasets.

Existing studies on question-answering (QA) of construction laws and regulations (CLR) heavily rely on rigid linguistic rules or rudimentary deep-learning models. This study integrates the domain-specific knowledge graph and large language models (LLMs) for CLR QA, using China as a case. It involves (i) developing a Chinese CLR (CCLR) knowledge graph with 3071 triples, (ii) devising a three-stage integration between the CCLR knowledge graph and LLMs, (iii) establishing a 6359-question QA dataset, and (iv) comparing the QA performance of 7 LLMs with and without the CCLR knowledge graph. Experimental results show that incorporating the CCLR knowledge graph improves the average QA performance of the seven LLMs by 19.7%, with gains ranging from 12.6% to 26.6% across 8 CCLR subdomains. Effect analysis further reveals that external CLR knowledge contributes the largest share of improvement (58.4%), followed by graph-based knowledge organization (18.5%). This study reveals the necessity of CLR-specific knowledge incorporation into general-purpose LLMs, demonstrates the superiority of knowledge graph over the knowledge bases built on isolated documents, and offers a large-scale QA dataset.

Digital twins have promising potential to enhance infrastructure operations and maintenance by enabling real-time monitoring and predictive maintenance. However, for professionals in the infrastructure industry with limited digital skills, existing digital twin systems can be complex and unintuitive, challenging to learn and use, requiring extensive prior training, and contributing to low adoption rates. This paper proposes ChatTwin, a natural language interface for infrastructure digital twins, designed to support immediate usability. By enabling users to interact with digital twins directly through natural language, ChatTwin allows users to complete tasks, access information, and explore functionality with minimal onboarding. ChatTwin was benchmarked with over 300 crowd-sourced prompts, and a user study (N = 24) was conducted. Results demonstrate that ChatTwin effectively interprets user input, improves task efficiency, reduces cognitive load, and enhances user experience, suggesting that it can make digital twins more accessible to professionals with lower technological familiarity, particularly during early adoption.

Digital Twin technologies are increasingly used in infrastructure and the built environment to create dynamic, data-driven models of physical assets and processes. This review analyses recent advancements across sectors such as tunnels, bridges, roads, buildings, construction management, and urban planning, covering all life-cycle phases from design to operation. Integrating Digital Twins with Building Information Modelling, Internet of Things sensors, and Artificial Intelligence enhances real-time monitoring, decision-making, and asset performance. Key methods include monitoring, modelling, and simulation, which improve resource use and proactive maintenance. However, adoption faces challenges such as poor data interoperability, high costs, and technical complexity in merging multiple technologies. Ethical and governance issues around data privacy and security also persist. The review identifies future research needs in improving interoperability, expanding predictive analytics, and assessing large-scale impacts. It highlights Digital Twins' potential to improve resilience, efficiency, and sustainability, stressing the need for policy support and stakeholder collaboration.

Tower-based surveillance (TBS) systems provide continuous, high-resolution, and real-time monitoring over wide spatial extents. However, accurate geolocation of TBS imagery remains largely underexplored due to large viewpoint discrepancies, nonlinear radiometric distortions, and limited spatial overlap with satellite imagery. To address these challenges, we propose TowerSatLoc, a high-precision and fully automated correspondence-based geolocation framework that registers TBS imagery to geo-referenced satellite images under complex conditions. The framework first applies a bird's eye view (BEV) transformation using intrinsic and extrinsic camera parameters to reduce viewpoint discrepancies. Within this process, an adaptive projection factor (APF) and an automatic orientation estimation method are introduced to improve geometric consistency. To handle severe cross-domain differences, semantic representations extracted from DINOv2 are fused with multiscale geometric features from VGG19 and refined through hierarchical residual matching and confidence-aware resampling. The framework further incorporates robustness-enhancing mechanisms to improve localization reliability under extreme viewpoint variations and uncertain tower geocoordinates. Extensive experiments on a newly constructed dataset, TowSatData, demonstrate the effectiveness and robustness of TowerSatLoc across diverse real-world scenarios, providing a scalable and generalizable solution for correspondence-driven cross-view geolocation.

Information Quality (IQ) frameworks, such as Wang’s and Strong’s (1996), treat IQ dimensions as discrete categories as seen above the sea surface. Yet scant research has examined the actual causal structure among these dimensions beneath the water, a phenomenon we termed the “iceberg effect.” This study aims to examine the iceberg effect in Building Information Modeling (BIM) by using Industry Foundation Classes (IFC)-based data exchange as an example. It does so by conducting a systematic review of 25 studies, a secondary analysis of 18 practitioner interviews, and a Decision-Making Trial and Evaluation Laboratory (DEMATEL)-Interpretive Structural Modelling (ISM) causal modelling with 10 domain experts. The results reveal that practitioners systematically over-report visible downstream symptoms (data loss 72%) while under-recognizing upstream root causes (classification errors 17%), leading to a perception-causality gap, i.e., the “iceberg effect.” The classification of dimension (D3) is the principal CAUSAL driver (D−R = +1.37), followed by data loss as a downstream effect (D−R = −0.57), forming a causal model chain as follows: Classification → Data Relations/Semantics → Data Loss/Geometry. This model indicates a causal asymmetry between IQ dimensions. It extends Wang’s and Strong’s framework by demonstrating that IQ improvement in standardized data exchange should follow the causal hierarchy rather than practitioner-reported frequency.

Creating geometric digital twins of buildings remains a labor-intensive process, often limited to the reconstruc tion of structural elements. Non-structural components and their spatial relationships with indoor spaces are rarely integrated into a unified digital representation. This paper proposes a novel semi-automated method for generating graph-based geometric digital models for digital twins from 3D point cloud and image data. The ap proach extracts spatial and object information from these data based on deep learning methods. Multiple 2D detectors are trained on different public and customized datasets to broaden class coverage, and their predictions are mapped into a unified label space, fused per view, projected into 3D space, and merged into object instances that correspond to the same physical element. A new graph schema is then introduced to represent indoor spaces and elements, capturing both hierarchical and spatial relationships. The schema links each entity to its geometric representation and supports temporal snapshots. All extracted information is structured and stored in a graph database using the proposed schema. The method is validated on two real-world datasets, one residential house and one institutional facility, capturing a broader and more differentiated range of object classes across different building types and topologies. The promising results indicate that the method has the potential to be generalized to a wider range of buildings.

Laser scanning is increasing used as an efficient and accurate way for quality assessment for prefabrication elements. However, previous studies have faced challenges in automating point cloud processing due to reliance on manual parameter tuning for individual cases. To address this gap, this study presents a method for automating the verticality assessment of prefabricated buildings by leveraging laser scanned point cloud data in combination with synthetic datasets. The experiments establish reference parameters and suggestions for adjustment patterns under varying conditions, offering practical guidance for implementation in real projects. The reference parameters were validated on real datasets collected using three terrestrial laser scanners, Leica BLK360, FARO Focus X330 and Leica RTC360, without additional tuning. Consistent results across scanners confirmed the robustness and transferability of the parameters, with the automated checks identifying the same defective wall panels as manual inspections. The findings demonstrate that the proposed method is both effective and accurate while significantly reducing the need for manual intervention. In addition to verticality assessment, the proposed approach provides a general framework that can be extended to other automated quality checks. It can be applied across different projects and adapted to assess other geometric criteria such as flatness, horizontal alignment, and overall geometric assessment.