Computer-Aided Civil and Infrastructure Engineering最新文献

The role of transportation system in ensuring equitable access to essential services and promptly recovering it post-disaster is critical to community resilience. This research introduces a framework aiming to strengthen transportation systems against external shocks, with an emphasis on geographical equity. To evaluate equity and address multiple network design objectives, we develop a two-level consolidated resilience index that measures network performance and community equity, employing a data-driven analytic hierarchy process for objective metric weighting, surpassing traditional expert scoring methods. Furthermore, we have implemented an equity-weighted Shapley value method to prioritize candidate links prior to investment. Finally, we have established a multi-objective bi-level program that integrates traffic distribution and travel behavior analysis. Our findings reveal that integrating equity considerations into candidate links selection phase significantly enhances fairness outcomes. The results also underscore the inseparable relationship between pursuing fairness and efficiency. This framework could potentially extend to other transportation systems’ investment strategies during the preparation phase, contributing to broader applications in resilience planning.

Accurate 3D segmentation of hydro power plant (HPP) components from point cloud data is essential for building high-fidelity digital twin systems that enable automation in construction, monitoring, and maintenance. However, existing point cloud segmentation methods suffer from high annotation costs. To address these challenges, a novel fully automated segmentation framework is proposed that assigns 3D semantic labels directly from unannotated point cloud data using only a textual prompt, without prior training on HPP-specific data. Experiments on six real-world HPP scenarios demonstrate that it achieves superior performance compared to state-of-the-art zero-shot baselines, with an average positive ratio of 72.56% and negative ratio of 20.45%, while significantly reducing the human effort and time required for segmentation. This study advances automation in construction by providing a practical, annotation-free solution for large-scale, fine-grained 3D segmentation of complex HPP environments, laying the foundation for efficient, intelligent digital twin creation and automated decision support in hydropower engineering.

Inspecting the backfill grout behind segment linings using ground-penetrating radar (GPR) is essential for the maintenance of shield tunnels. However, reinforcing rebars embedded in the segment linings generate strong clutter in GPR data, which obscures the detection of defect signals within the backfill grout. In addition, acquiring sufficient and consistent GPR data to train deep learning models is challenging due to restricted site access and variability in tunnel environments. To address these limitations, this study proposed a simulation-driven deep learning network for clutter elimination and defect detection in GPR images. A training database was constructed exclusively through finite-difference time-domain numerical simulations to model segment linings containing backfill grout defects. This configuration provides a standardized and well-controlled dataset for training and evaluating the network. Several architectures within the encoder–decoder framework, including U-Net 3+, were employed to develop models for eliminating rebar clutter. The performance of the reconstructed GPR B-scans was assessed using image quality and quantitative metrics, with U-Net 3+ demonstrating the highest accuracy. The findings confirm that realistic GPR signal characteristics can be learned and generalized through simulation-based data without relying on extensive field data. Finally, GPR B-scans collected from a full-scale tunnel lining segment were reconstructed using the proposed network to verify its practical applicability. This study demonstrates the potential feasibility of transfer learning from simulation-only data to real-world engineering applications, enabling more effective backfill grout inspection and supporting efficient maintenance.

With the rapid development of transportation infrastructure, highway interchanges have become critical nodes in the network, where congestion frequently occurs. Existing research on congestion mitigation in such regions is limited, often focusing on individual bottlenecks, which may overlook interactions between diverging and merging areas. Some other research adopted a macroscopic traffic flow perspective, without microscopically mitigating right-of-way conflicts caused by lane-changing demands of vehicles. To address this gap, this study proposes a collaborative multi-lane scheduling strategy for connected and automated vehicles under a cloud control system. By jointly optimizing vehicle passing sequences in both diverging and merging zones, the proposed method aims to improve overall traffic efficiency. Key contributions include a rolling traversal mechanism for global scheduling, a discretionary lane-changing strategy for enhanced lane utilization, and a double-checked trajectory planning approach that balances efficiency and comfort. This framework offers a scalable solution to alleviate congestion at complex highway interchanges under high traffic demand.

Heavy goods vehicles (HGVs) have a significant impact on road and bridge infrastructure, with overloaded vehicles accelerating structural deterioration and increasing safety risks. Bridge weigh-in-motion (B-WIM) systems estimate gross vehicle weight (GVW) using strain measurements, but inaccuracies in axle configuration recognition can reduce reliability. This study presents a low-cost computer vision (CV) extension for existing B-WIM installations that verifies strain-inferred axle configurations using traffic camera images and flags GVW estimates as reliable or unreliable. Experiments on a data set of over 30,000 HGV records show that by combining convolutional neural networks with strain-based heuristics, GVW reliability can improve from 96.7% to 99.89%, effectively excluding nearly all erroneous measurements. The approach operates without interrupting ongoing B-WIM operations and can be applied retrospectively to historical data. Limitations include the inability to detect raised axles (RAs), which the method excludes as unreliable. This method provides a practical, high-precision enhancement for structural health monitoring of bridges.

Accurately classifying damage levels from structural inspection images is critical for automated infrastructure assessment. Although deep neural networks achieve impressive performance, their black-box nature limits explainability, and prior studies using Grad-CAM often yield coarse or inaccurate saliency maps. To overcome these limitations, this paper introduces XIDLE-Net, a multitask model that simultaneously performs damage classification and saliency map prediction to enhance explainability in structural damage assessment. Combining a Swin Transformer encoder with a convolutional neural network decoder, XIDLE-Net is trained with dual supervision using damage labels and inspector gaze-derived attention maps, enhancing both classification accuracy and model explainability. Experimental results show that XIDLE-Net outperforms state-of-the-art methods in both classification and saliency explainability, achieving 78.1% accuracy, 94.3% area under the curve (AUC), and a 39.7% improvement in saliency prediction over ResNet-50 with Grad-CAM. To our knowledge, this is one of the first investigations to employ large-scale inspector gaze data for supervision and to quantitatively evaluate Grad-CAM in structural image classification. The results highlight the promise of human gaze data for advancing explainable vision-based structural health monitoring.

Three-dimensional (3D) site models form the digital foundation for modern construction management. However, creating these models from multi-source imagery presents two key challenges: accurately georeferencing camera poses during wide-view acquisition and precisely aligning multiple point clouds that possess non-uniform accuracy. This paper proposes a two-stage framework to address these challenges. The first stage performs local-to-world registration by integrating ground control points, detected via an enhanced HA-YOLOv8, as early-stage constraints in the 3D reconstruction process. The second stage, inter-model alignment, introduces a novel edge-aware method that utilizes refined structural edge features to merge local models. The framework was validated using images from crane cameras on a high-rise project, achieving a final modeling accuracy of 0.121 m for the main structures, resulting from precise registrations with low translation (0.102 m) and rotation (0.051°) errors. This approach provides a robust solution for generating high-fidelity 3D site models, supporting advanced digital construction applications.

While machine learning (ML) has advanced image-based damage detection, a critical gap remains: the automated translation of detected damage into standardized condition ratings used in structural assessments. Most existing approaches stop at semantic segmentation, overlooking the damage rating step essential for practical inspections. This paper presents a semiautomated system that bridges this gap by linking multi-label damage segmentation with condition rating prediction. Our contributions are: (1) a data-driven label taxonomy for damage segmentation, derived from statistical and semantic analysis of 2.2 million inspection records, and designed to support downstream condition rating; (2) a pipeline for converting textual inspection records into structured training data for automated condition rating, and a set of custom bidirectional long short-term memory (LSTM) models achieving up to $��99��\%�$ F1-score on this task; and (3) a reference system architecture integrating image segmentation and text-based damage rating within an interactive 3D inspection interface. The system demonstrates how integrating damage detection and condition rating within an interactive 3D interface can streamline inspection documentation and enhance decision support for concrete structures. Developed in compliance with German inspection standards and designed for adaptability, the system architecture offers a transferable framework for embedding ML-based automation into digital inspection workflows, ensuring that all components, from damage detection to condition rating, are aligned in an end-to-end process.

Current trenching trajectory planning methods for autonomous excavators primarily focus on a single cycle, lacking overall optimization of the entire task. This study proposes an offline method for overall optimization of trenching trajectories via reachability and excavation force maps. This method can generate a complete set of trajectories that covers the entire target trench area and satisfies the excavation force requirements without relying on expert demonstration data, which is conducive to improving the overall trenching efficiency. First, a base-position selection method that considers both force and pose requirements is proposed. Second, rules for generating key points of trenching trajectories are established. Subsequently, an overall optimization method for all trajectories covering the entire trench area is proposed. Finally, a prototype and an experimental scenario were built, and trenching experiments on four trench shapes are conducted. The results demonstrate that the planning process is completed within a few minutes, and the trench profiles obtained from the trenching experiments closely match the target shapes.

This paper introduces a real-time framework designed to optimize intersection signal timing and vehicles’ trajectories across a network of intersections in a mixed environment of human-driven and automated fleets. The network-level optimization model is decomposed into intersection-level sub-models, whose decisions are coordinated through information exchange, aiming to push them toward the network model's optimal solutions. At each intersection, a bi-level framework addresses both the signal timing and trajectory optimization models. A specialized greedy heuristic algorithm is developed for the lower-level problem where optimal connected and automated vehicles (CAVs) trajectories are constructed for a given signal timing plan. At the upper level, all the feasible signal timing plans are created, and the system selects the most effective one to implement. The study integrates the entire solution process into a receding horizon framework to ensure efficient handling throughout the study period. A case study demonstrated the system's capability to adjust signals and trajectories effectively under various traffic demands and CAV market shares. Results showed a reduction in overall arterial delay correlating with higher proportions of CAVs. The proposed system delivered solutions in less than 70 ms, which is significantly faster than the half-second solving time steps, ensuring decisions were made quicker than in real-time.