As circular economy principles become increasingly embedded in regulations, digital passports are being recognised as key enablers of circular construction. Although several digital passport concepts have been proposed, their implementation remains constrained by the lack of standardised, hierarchically structured data requirements. This paper presents a computational workflow that combines semantic topic modelling and unsupervised clustering to derive a common digital passport taxonomy from the state-of-the-art literature. To address inconsistencies in property names across data requirements, large language models are used to generate contextual property descriptions from the literature. Expert evaluation aligns moderately with cluster cohesion and diverges on hierarchical fit, indicating that statistically optimal taxonomies require manual moderation. The workflow can be used to semi-automatically generate and moderate taxonomies that structure disparate data requirements. This application can assist construction professionals in developing standardised digital passport data templates aligned with evolving regulations.
Existing Building Information Modeling (BIM) checking methods focus on specific design defect types, while failing to identify and correct multi-type property-related BIM design defects in one loop. Therefore, this paper proposes an integrated framework to identify and correct multi-type property-related design defects in BIM via Large Language Models (LLMs). First, a BIM-to-Text method with component-balanced chunking is introduced to bridge BIM data and LLMs. Then, a prompt learning based method incorporating rule injection and few-shot prompting is proposed for defect identification, and a Retrieval Augmented Generation (RAG) method is established for correction suggestions. A hallucination control strategy combining key-identifier validation and token-length thresholds is developed to improve reliability. Experiments show that the proposed approach outperforms rule checking, and the fine-tuned LLM improves identification accuracy by 15%, with a 94% rationality rate for correction suggestions. Hallucination control increases accuracy from 64% to 85%, eliminating 92.5% of hallucinations in one round.
Automated quality inspection of shield tunnel segment assembly remains challenging because boundary ambiguity affects joint localization and point-cloud unfolding may weaken native 3D spatial relationships. Accordingly, this paper addresses the research question: How can shield tunnel segment assembly quality be automatically and accurately inspected from 3D point clouds in complex tunnel environments? This paper proposes an approach comprising fine-grained point cloud semantic segmentation without unfolding and a 3D geometric inspection algorithm quantifying intra- and inter-ring dislocations and longitudinal and circumferential joint widths. Wuhan Rail Transit Xingang Line field tests yielded 89.55% mIoU and 95.25% OA for segment-level segmentation, with MAEs of 0.82–1.26 mm and an overall RMSE of 1.08 mm across four indicators. The workflow provides tunnel quality inspectors with a standardized approach to segment assembly assessment, reducing reliance on manual experience. Future research can extend the workflow through cross-project adaptation, efficient computation, and interactive measurement position selection under occlusion.
Digital technologies are increasingly used to mitigate construction project delays. Existing reviews emphasize delay causes, management practices, or single tools, but rarely examine digital technologies as an integrated system or map their evolution. This paper reviews 790 journal articles published between 2000 and 2025 using content analysis and scientometric mapping. The content analysis identifies five groups of delay causes, four delay-response stages, and five digital technology functions in schedule management. It also summarizes typical mitigation mechanisms, reported effects, and failure boundaries across these stages. Co-citation and keyword networks highlight digital twin, machine learning, natural language processing, and knowledge management as emerging themes, indicating a shift toward AI-enabled and feedback-oriented schedule control. This paper proposes integrated technical and scientometric frameworks that provide a structured basis for future research and practical application in digital delay mitigation.
Accurate excavator bucket pose estimation is fundamental for automated earthmoving and precision grading, yet reliable real-time feedback remains difficult to obtain on dynamic construction sites. Proprioceptive sensors suffer from cumulative drift caused by mechanical backlash and structural deformation, while external vision or radar systems require fixed infrastructure and are prone to occlusion. This paper presents a body-aware LiDAR framework that uses an onboard LiDAR sensor as an internal observer of the excavator to estimate bucket pose in real time. The method follows a predict-observe-update paradigm that integrates kinematic motion priors, semantic point-cloud estimation based on an enhanced PointNet++, and model-based ICP registration with adaptive gating for robust tracking under dust, soil occlusion, and sparse observations. Field experiments on a SANY SY19E excavator demonstrate 1.2 cm positional accuracy and 2.1° orientation accuracy at 9.7 Hz, while capping maximum errors under favorable scenarios at 1.9 cm and 7.7°, providing a practical onboard estimation solution for future closed-loop control of construction machinery.
Steel-structure joints contain dense components and complex local relations, which make component-level BIM extraction difficult. This paper addresses the problem of decomposing heterogeneous local steel-joint geometric inputs into component instances and organizing the extracted information into a reusable BIM-oriented component-topology representation. An extensible framework is proposed by combining multi-view component recognition, 2D3D surface mapping, component-graph construction, and topology-aware graph encoding. Experiments on a parametric steel-joint dataset show that the proposed recognition pipeline improves component-level extraction, while graph encoding further refines relationally ambiguous components. The representation is useful for steel-structure designers and BIM engineers who need component-level information for scan-to-BIM assistance, design retrieval, assembly reasoning, or graph-database enrichment. Future work can extend the framework with larger datasets, more input sources, and richer component relations, and can further connect the extracted graphs with BIM databases and engineering design tools.
Manual visual inspection remains used for post-earthquake damage evaluation of structural components; however, it is subjective, time-consuming, and depends on the experience of the inspectors. This paper proposes an automated, rapid, and non-contact framework for predicting the seismic Damage Index (DI) of Reinforced Concrete Shear Walls (RCSWs). A stage-dependent DI is defined based on the reduction in lateral load-carrying capacity in the pre- and post-peak regions of the wall force–deformation curve, with emphasis on the post-peak region where strength degradation becomes critical. To support post-earthquake assessment, cracking and crushing patterns are automatically processed, and morphological features are extracted to describe the spatial distribution of visible damage across the wall surface. Using 199 damage images from 55 quasi-statically tested RCSWs, several machine learning models are trained under different input scenarios. The results indicate that the proposed framework can accurately predict seismic DI using image-based features, with R2 values up to 0.89.
Autonomous excavation requires decision-making across geometric reasoning, motion planning, and safe execution under strict physical constraints. Existing approaches rely on hand-crafted rules that fail to generalize or end-to-end learning without safety guarantees. This paper proposes autonomous physics-validated excavation (APEX), a hierarchical framework integrating large language model (LLM)-based geometric decomposition, masked reinforcement learning (RL), and polytope-validated trajectory execution. The LLM decomposes target geometries into sequential phases, converting spatial constraints into action masks that guide RL exploration without task-specific fine-tuning. An offline trajectory library, validated via convex polytope analysis and Fundamental Equation of Earthmoving (FEE) soil resistance, ensures every executed action satisfies hydraulic, stability, and slip constraints. APEX achieves 93.2% success in grid simulation and 90.1% in the Vortex physics engine across leveling, trenching, and multi-area excavation. Removing polytope validation causes an 18.8 percentage-point drop in transfer success rate.
This review article examines the integration of artificial intelligence (AI) in architectural design, focusing on human–AI collaboration, interaction, design stages, input–output data exchange, and human controllability. Despite recent advances, existing reviews emphasize application domains, offering limited insight into how responsibility, decision-making, and controllability are distributed between designers and AI systems. Such insight is essential for developing decision support systems, enabling progression from early design stages toward construction documentation, and reducing the gap between computer-aided design and construction. To address this gap, a PRISMA-based review of 96 journal and conference papers was conducted. The findings identify four dominant collaboration modes, with Human-on-the-Loop being the most prevalent. AI applications remain concentrated in schematic design, with limited but emerging engagement in later phases. Image-based data dominates inputs and outputs despite limited precision and editability. The review shows that collaboration mode and information embedded in the design process shape human controllability over AI-generated outcomes.
Accurate prediction of subgrade compaction states remains challenging due to the coupled effects of operations, materials, moisture, and historical states. This paper proposes a physics-guided (PG) framework integrating self-attention and bidirectional gated recurrent units (Bi-GRU) for pass-by-pass compaction estimation. The current state Kn is predicted using multi-source features related to construction, equipment, and materials. To ensure engineering consistency, three physical constraints governing monotonic evolution, directional moisture effects, and gradation rankings are embedded into the loss function. Multi-scenario evaluations demonstrate that a state-dependent transition mechanism (Kn−1 to Kn) is successfully established, and the network topology is regularized via non-PDE priors to correct systematic biases. Field validation on the Beijing-Tianjin-Tanggu Highway in China confirms that continuous predictions coupled with SHAP interpretability precisely isolate controlling factors, shifting quality control from passive detection to active root-cause remediation.
Productivity in construction engineering, while strongly impacting the cost, schedule, and overall performance of construction projects, is difficult to predict because of the impact of dynamic and uncertain factors that involve both quantitative and qualitative variables. To address the limited interpretability of black-box models, an Optimized Fuzzy Operation Rain Forest (OFORF) model for concrete-pouring productivity prediction is developed and evaluated. The proposed OFORF integrates fuzzy logic (FL) and an operation rain forest (ORF). Moreover, the Artificial Satellite Search Algorithm (ASSA) is applied to optimize the model. A real-world dataset from a 17-story reinforced concrete building project is used to evaluate model performance. OFORF significantly outperformed the basic and optimized models as well as three interpretable tree-based models by achieving the highest Reference Index value (0.9519). Based on the results, OFORF is an accurate and interpretable tool for forecasting productivity and supporting related decision-making in complex construction environments.
