Computer-Aided Civil and Infrastructure Engineering最新文献

This study presents SupportGAN, a knowledge-guided framework based on a two-stage generative adversarial network, for preliminary conceptual plan-view layout of corner- and cross-supporting structures in foundation pits. Design drawings of support structures collected from authoritative design institutes were semantically processed and expanded through a structural knowledge-guided augmentation (SKGA) approach. The SupportGAN model was then trained with multiple hyperparameter configurations to achieve optimal performance. Additionally, SupportGAN was evaluated and compared with two mainstream GAN models (pix2pix and pix2pixHD), demonstrating its superior capabilities. The design results generated by SupportGAN were evaluated using visual assessment and quantitative metrics. A feasibility assessment was also performed to confirm the economic and mechanical viability of the generated layouts: A pixel-count proxy showed a 6.90% material gap versus engineer designs, and results of the finite element (FE) analysis on two cases indicated comparable structural performance (force difference $��\le �12.74�\%�$). The results indicate that SupportGAN's outputs exhibit significant similarities to those of expert engineers across various aspects, demonstrating its potential to aid designers in the preliminary conceptual layout design of corner- and cross-supporting structures.

Traffic assignment serves as an important component in modeling flow distribution across infrastructure networks and supporting intelligent traffic management and urban planning. Fast algorithms for solving the stochastic user equilibrium (SUE) model are essential for enhancing computational performance and scalability of traffic assignment models applied to complex infrastructure networks. We augment the gradient projection (GP) algorithm for the SUE models through Barzilai–Borwein (BB) step size adaptation. For further optimizing computational performance, we explore iteration strategies within the GP algorithm: Jacobi parallelization, Gauss–Seidel sequential updating, and successive over-relaxation (SOR) with dynamic relaxation. Global convergence of these iterative methods in solving the SUE problems is theoretically established, with convergence conditions for the SOR derived. The results demonstrate the BB step size's superior performance, compared to alternative step size methods, across all network scales, and it achieves the best stability when demand factors and the dispersion parameter increase.

This study introduces a new posture-hour-based scheduling framework that integrates ergonomic constraints into construction planning. It shifts the traditional focus from “labor-hours” to “posture-hours” as a primary resource unit. The novelty lies in (1) defining and formulating this new problem in practical construction settings and (2) developing algorithms that treat posture capacity as an allocatable resource. This approach proactively manages musculoskeletal disorder risks by accounting for each activity's posture-specific demands and workers' posture-holding capacities as scheduling limits. Validated via a benchmark case and a real-world concrete pouring project, applying the framework significantly reduces cumulative exposure to high-risk postures, compared to traditional methods. While it may slightly increase project duration, this trade-off better balances physical workload, prevents posture-induced health issues, and enhances long-term workforce wellbeing and sustainability. This interdisciplinary research provides a foundational framework for integrating occupational health considerations, labor productivity metrics, and construction planning methodologies.

Infrastructure performance deterioration models are a critical component in asset management. While many jurisdictions have begun collecting more reliable asset condition data, an effective data-sharing mechanism is still lacking that enables cross-jurisdictional knowledge transfer for developing more reliable deterioration models, particularly for jurisdictions with limited or even no historical data. To bridge this gap, this study proposes a parametric transfer learning framework for collaborative deterioration modeling across jurisdictions by integrating a stochastic process model with a hierarchical Bayesian approach. Transfer learning is realized in two aspects to capture both intra- and inter-jurisdictional heterogeneity: by incrementally updating the learned globally shared information when data from a new jurisdiction become available, and by supporting parameter estimation even when some covariates are partially missing. The proposed framework is quantitatively compared with a jurisdiction-specific modeling strategy in terms of model uncertainty through simulation studies. Furthermore, case studies using a real-world historical bridge condition database collected from nine jurisdictions with different inventory sizes in Canada are conducted to compare independent and collaborative modeling approaches in two aspects: their ability to capture inter-jurisdictional heterogeneity and their impact on model uncertainty on lifecycle decision making. Results confirm the effectiveness and significance of the proposed collaborative modeling approach in infrastructure asset management.

Loose defects in semi-rigid base layers can critically compromise the structural performance and long-term serviceability of asphalt pavements. However, accurate identification and quantitative assessment of such defects remain challenging due to the lack of reliable ground-penetrating radar (GPR) data sets and the limitations of existing detection methods. This paper presents an integrated framework that combines data simulation, deep learning–based segmentation, and 3D reconstruction to address these challenges. First, a high-fidelity synthetic data set was generated using a random medium-based forward modeling approach to represent varying looseness depths and configurations. Second, a modified YOLOv8-seg architecture was proposed, featuring a novel multi-scale feature fusion module (DN module) and a Weighted Intersection over Union (WIoU) loss function to enhance segmentation precision under noisy and complex GPR conditions. It achieved a mean average precision (mAP) of 97.25% and a real-time inference speed of 32.05 frames per second (FPS). Third, a 3D point cloud reconstruction approach based on inverse distance weighted (IDW) interpolation was introduced to restore the spatial morphology of the detected defect regions. Delaunay triangulation was then used to estimate the volumetric extent of the defects, achieving an overall estimation accuracy of 78.07%. Finally, the proposed framework was validated on a full-scale pavement model, confirming its effectiveness in defect detection, morphological recovery, and quantitative assessment. The findings provide a reliable computational tool for pavement condition evaluation and maintenance planning.

In slurry shield tunneling, relying solely on human experience to adjust active control parameters manually often has limitations and can lead to safety hazards. Therefore, the intelligent decision-making for parameters of slurry shield tunneling based on machine learning algorithms is of great significance. This paper established a hybrid algorithm that combines the Bayesian method with the long short-term memory neural network to predict passive response parameters of slurry shield tunneling, achieving accurate predictions of slurry pressure and specific energy. Additionally, a hybrid algorithm combining particle swarm optimization with genetic algorithms is studied for the multi-objective optimization control of active control parameters in slurry shield tunneling. The objective is to minimize both in the specific energy and the error of slurry pressure during excavation. To facilitate practical engineering applications, an intelligent decision-making software for shield machine active control parameters is designed. This software provides a visualization interface for parameter optimization of slurry shield tunneling.

Timely detection of ground movement is critical in geotechnical engineering to prevent failures that could result in loss of life or property damage. In tunneling applications, this task is commonly referred to as convergence monitoring, which has conventionally relied on geodetic instruments such as total stations, or, more recently, on advanced technologies such as mobile laser scanning. A new computer vision-based approach, requiring only a small number of targets, offers a faster and more cost-effective alternative for ground deformation monitoring. The method introduces interconnecting distance embedding (IDE), which captures changes in pairwise distances between markers before and after deformation. The pairwise distances serve as an embedding of the spatial configuration of the markers; however, a point registration algorithm is required to compare movement across different time steps. Compared with conventional registration techniques—such as mean squared error or least absolute deviations, which perform well for a rigid body but lose effectiveness when deformation alters point configurations—a new IDE-based algorithm was developed that enables more robust tracking. In addition, computer vision offers richer insights at lower cost, supporting both quasi-static and short-interval monitoring (e.g., hourly or daily). Furthermore, in the New Australian Tunneling Method, it further streamlines the integration of convergence monitoring into design workflows, thereby enhancing efficiency, flexibility, and reliability.

Routing an emergency response vehicle (ERV) is essential to sustain the society and save people in high risk. ERV must have top priority on highways to pass through without interference with other non-ERVs. Congested road traffic conditions, however, might not allow for such a priority. When non-ERVs fully occupy all lanes in a road segment, in principle, an ERV cannot pass through the segment until the congestion releases. On the other hand, the situation is often observed in the field that non-ERVs sidestep to make way for an ERV, and the ERV straddles lanes and runs along a lane delineation line. The present study devised an optimization model to allow such a situation. Whereas most ERV routing problems have been formulated for uncongested traffic conditions, the present study presents a robust formulation that can work for both congested and uncongested traffic conditions. The connected and autonomous vehicle (CAV) environment is prerequisite for the present modeling framework. The proposed optimization model is robust to derive an optimal trajectory for ERV in road segments of diverse traffic and geometric conditions.

Prediction of concrete chloride ingress under varying environmental conditions is computationally demanding, particularly when mesostructural effects are considered. Uncertainties in service history and material properties further limit conventional models. This study develops a data-driven dynamic mode decomposition framework for efficient prediction. It decomposes spatio-temporal chloride concentration data into eigenmodes with temporal coefficients for accurate reconstruction and extrapolation. Its performance is demonstrated under constant, annual cyclic, and multi-frequency boundary conditions. The reduced-order representation cuts data storage by over 99% and enhances computational efficiency by over 91%. Sensitivity analyses indicate higher accuracy when input data are collected after long-term chloride ingress and covers sufficient boundary cycles. Linear transformations of surface concentration fluctuations can be directly mapped to temporal coefficients of corresponding oscillatory modes. An analytical model expressing chloride profiles as an explicit function of depth and time is derived, applicable to all scenarios predictable by the proposed method.

Digital twin technologies are gaining global attention as foundational components for data-driven infrastructure management and urban simulation. However, significant challenges remain in integrating and utilizing geospatial datasets, including inconsistencies in spatial referencing methods, the absence of semantic object identification, and alignment issues due to positional inaccuracies. This study proposes a 4D spatial–temporal information infrastructure leveraging a voxel-based spatial referencing method known as Spatial IDs. The infrastructure enables unified spatial indexing across heterogeneous datasets and supports high-speed data retrieval, dynamic distribution, and automatic object identification from point cloud data. Key system components include a binary voxel-based search index, a data compression and distribution mechanism using LASzip format, and a semantic segmentation module that integrates map-based object annotations with point cloud geometry. Demonstration experiments involving city-scale digital twin construction, drone route interference detection, and object identification confirmed the infrastructure's efficiency and scalability. The proposed infrastructure reduced retrieval time to under one second and achieved high object identification accuracy (F1-score ≥ 0.99), illustrating its applicability in real-time geospatial analytics and intelligent urban systems.