Computer-Aided Civil and Infrastructure Engineering最新文献

Concrete mesostructure modeling is critical for simulation-driven material design, as it directly influences the accuracy of mechanical and durability analyses. However, conventional methods such as computed tomography–based reconstruction and aggregate sequential placement suffer from low efficiency and limited controllability. To address these challenges, this study proposes a regressor-guided conditional diffusion framework for 2D concrete mesostructure generation. By transforming continuous physical parameters into dynamic gradient fields, a separately trained regressor guides the denoising process to ensure strong compliance with aggregate gradation and particle size constraints. Experimental results demonstrate that this method significantly outperforms traditional conditional embedding models in terms of constraint accuracy, reducing aggregate volume fraction and particle size distribution errors by 12.6% and 31.6%, respectively. Additionally, the proposed framework achieves over two orders of magnitude greater efficiency than existing numerical techniques. Its modular design and fine-grained control capabilities establish a scalable, physically consistent solution for high-fidelity mesostructure generation in concrete and beyond.

The concept of contraflow left-turn lane (CLL) design has been proposed for nearly 10 years, which provides a novel approach to alleviate traffic congestion in urban areas, especially for those signalized intersections with heavy left-turn traffic. While putting forward the practical application of CLL design to area-wide signalized intersections, whether the control scheme of large-scale signalized intersections with CLLs could be more sustainable remains an open question. This paper analyzes spatiotemporal characteristics of the CLL system and proposes analytical traffic and emission models based on the finite capacity queuing model, in which the special queuing behavior brought by the pre-signal is explicitly considered. A sustainable traffic signal control framework is built to optimize signal timings of intersections with CLLs. A real-world case study based on a road network in Yangon is conducted, and the results illustrate the proposed method's efficiency and sustainability in managing signalized intersections with CLLs in road networks.

To accommodate environment-induced time-window constraints, environment-sensitive activities are arranged within allowable time windows while maintaining technological precedence and other logical relationships on the project. This research advances classical time–cost trade-off (TCT) analysis by incorporating time-window constraints in project planning, creating a more complex optimization problem that becomes computationally prohibitive for real-world applications. To overcome this challenge, an integrated project planning framework combining project time and cost into a single objective function is formalized. The optimization solution employs time-window scheduling algorithms to simulate method combinations through enumerated simulation. A reward function is defined to evaluate alternatives based on their impact on project cost and duration. In addition, a sample size reduction technique is utilized to maintain computational efficiency of random sampling without sacrificing accuracy. The methodology's practical application is demonstrated through a case study of a river-crossing bridge project in remote northern Canada, which is planned to validate its effectiveness in planning real-world infrastructure projects under stringent environment-induced time-window constraints.

Existing pavement performance prediction models often struggle to capture complex spatiotemporal dependencies in road networks due to reliance on empirical rules and scenario-based calibration. This study proposes pavement graph network (PaveGNet), a spatiotemporal graph network framework designed to model fine-grained pavement distress evolution. It constructs a multi-node graph encoding topological correlations and time-based state transitions, while integrating exogenous factors such as temperature, precipitation, traffic, and maintenance works. Experiments demonstrate that PaveGNet performs effectively in predicting fine-grained indicators of distress evolution. The prediction error for distress evolution rate was significantly reduced, from 9.005% with the baseline spatial–temporal graph convolutional network model to 2.670%. Ablation experiments were conducted to verify the contribution of temporal interdependence, spatial correlation, and external variables in the proposed PaveGNet framework. The results demonstrate that all three components play essential roles in prediction, with external variables showing the most significant impact. To further assess the modular robustness, parts of the spatial and temporal learning modules were independently replaced. The results indicate that the prediction of distress evolution rate relies heavily on the originally designed learning components. Overall, this framework provides a more realistic and scalable solution for the spatiotemporal prediction of pavement distress evolution in road networks.

Point cloud-based construction quality assessment and quality control (QA/QC) are playing an increasingly important role in large-scale complex building projects. However, this approach faces several challenges, such as the laborious and time-intensive process of manual point cloud segmentation, the high cost of point cloud labeling, and the lack of sufficient training data for deep learning-based automatic segmentation methods. To address these issues, this paper proposed a method for detecting construction deviations in large-scale complex building structures by utilizing synthetic point clouds for segmentation. The method automatically generated labeled synthetic point clouds with Gaussian noise using BIM and a virtual engine, significantly augmenting the limited amount of real point cloud data to train the semantic segmentation model, enabling the achievement of 94.2% overall accuracy (OA) and 81.1% mean intersection over union (M_IoU). Furthermore, a point cloud instance segmentation method according to density-based spatial clustering of applications with noise (DBSCAN) and voxel-vs-BIM was proposed to independently compare each instance object of different building components with its corresponding BIM model, assessing the construction accuracy of each component based on root mean square error metric and the level of accuracy specification. For components with an LOA3 accuracy level, further deviation analysis was conducted. Taking the structural construction deviation detection of beams, columns, and concrete thick shells in the core area of the Shanghai Grand Opera House as a case, the proposed method significantly improved the efficiency of QA/QC.

Pavement performance impacts transportation mobility, safety, and comfort, with timely maintenance relying on field survey data. However, frequent surveys are costly and impractical. This study examines how survey timing and frequency influence maintenance events and pavement deterioration to optimize management strategies. Using the pavement condition index as a performance indicator, historical data and maintenance records from the long-term pavement performance database were analyzed, and regional deterioration curves were derived using shifting factor methods. The proposed adaptive stochastic deterioration modeling method incorporates survey timing through sequential probabilistic events, capturing the stochastic nature of pavement deterioration. Mixture density networks predicted survey outcome distributions, while Monte Carlo simulations analyzed system behavior. Results show that longer survey intervals increase uncertainty, while uniform schedules reveal local cost minima, highlighting survey frequency ranges of over-survey and under-survey. Non-uniform schedules have been validated to offer potential for greater cost-effectiveness. These contributions provide road agencies with a new perspective for optimizing management strategies.

Using fiber-reinforced polymer reinforcement in concrete (FRP-concrete) to form composite-concrete structures is increasingly important in civil and infrastructure applications because of its high strength, durability, and corrosion resistance. However, one of the main challenges of the FRP-concrete structure is the brittle failure mode, which might result in sudden structural collapse. It is critical to closely monitor the strain of composite reinforcement for potential structural damage. This study develops a deep learning-based framework for the reinforcement condition identification in terms of non-contact strain using the crack image data from FRP-concrete beam system. The framework contained two stages: first, using crack pattern and strain data obtained from simulation to train the Kolmogorov–Arnold networks (KANs)-based convolutional neural network strain prediction model, which added a new KAN channel in convolution to enhance performance. Subsequently, the actual crack patterns from you only look once v11 segmentation results are used to generate the input of the prediction model. A three-bending test of an FRP-concrete beam is presented to validate this method. The developed framework achieves an average R² of 0.713, compared to the actual sensor data in reinforcement strain prediction. The results indicate that the intelligent framework has superior performance in strain prediction, addressing challenges in FRP-concrete structure applications.

A leak in a water distribution network (WDN) creates a localized pressure depression whose effect diffuses along the network topology through headloss coupling. This spatial diffusion is well approximated by graph Laplacian smoothing, so we use a graph convolutional network as a learnable surrogate of hydraulic information propagation on the graph. The method casts single-snapshot leakage analysis as graph-level classification of normal versus abnormal states using topology-aware convolution. Localization then leverages gradient-based node importance—sensitivities of the leak logit to node features—mapped to pipes via linear interpolation to produce leakage-risk contour maps. The framework couples pressure and demand features, accommodates partial monitoring by varying sensor density, and evaluates noise robustness by injecting Gaussian noise into inputs. In simulated WDNs, the approach maintains high accuracy across sensor-density settings and exhibits a predictable trend under increasing noise, where false negative rate rises faster than false positive rate as signal-to-noise ratio decreases. On the L-Town benchmark, we use real pressure measurements and nodal demand data consistent with metered consumption (generally <10% deviation), providing a realistic testbed. A simplified network representation accelerates convergence while preserving localization fidelity, enabling millisecond-level inference suitable for operational deployment. Together, these results support a physically grounded, computationally efficient pathway for leak detection and localization in smart water networks.

An ensemble-learning-embedded neural network for vortex-induced vibration (VIV) early warning was proposed. The new model consists of a core module based on ensemble learning and peripheral modules. The core module identifies abstract features of VIV, while the peripheral modules handle feature extraction and weight control. The core module of the new model is trained entirely using augmented datasets. Consequently, compared to other models with equivalent parameter counts, the new model can be trained using significantly fewer datasets. Displacement records from three VIV events at a cable-stayed bridge under construction were used to train and test the model. The new model demonstrated superior performance during testing. After retraining with data from a single VIV event at another cable-stayed bridge in the construction phase, the new model successfully achieved VIV early warning for the new bridge. The new model demonstrates significant potential for providing early warning of VIV due to its lower data requirements.

Shield tunnel tail grouting fills the annular gap between excavated soil and tunnel lining, supporting soil stability and controlling ground settlements during construction. However, existing grout performance prediction methods are limited by labor-intensive empirical testing, insufficient datasets, and inadequate modeling of liquid-to-solid phase transitions, resulting in consolidation deformation, ground loss, suboptimal formulations, and increased settlements. To address these issues, this study develops an explainable intelligent system for multi-performance grout optimization, integrating innovative experimental testing with advanced machine learning. A unique database is constructed from liquid-state (e.g., density, bleeding rate, fluidity) and solid-state (e.g., unconfined compressive strength (UCS), compressed deformation) properties, augmented via a physics-constrained generative adversarial network for realistic datasets. Four algorithms (artificial neural network (ANN), random forest (RF), extreme gradient boosting (XGB), support vector regression (SVR)) are ensemble-optimized using Bayesian techniques and K-fold cross-validation, achieving high predictive accuracy. SHapley Additive exPlanations (SHAP) analysis enhances interpretability, identifying water–binder ratio as dominant for liquid properties and cement–fly ash ratio for strength. Laboratory experiments and field applications validate the system, highlighting its efficiency in grout optimization, accurate prediction of consolidation-induced settlements, and improved deformation control, thereby enabling better settlement management, protection of adjacent structures, and informed decision-making in shield tunneling projects.