Computer-Aided Civil and Infrastructure Engineering最新文献

{"title":"A Two-Level Semantic Segmentation Framework for Visual Corrosion Pattern Assessment in Steel Bridges","authors":"Sal Saad Al Deen Taher, Ji Dang","doi":"10.1016/j.cacaie.2026.100173","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100173","url":null,"abstract":"Corrosion assessment in steel bridges requires not only estimating the spatial extent of visible deterioration but also distinguishing visual corrosion appearances that may support inspection interpretation. Existing vision-based corrosion-segmentation studies predominantly formulate the task as binary corrosion-versus-background mapping, which supports damage localization but provides limited information regarding corrosion-pattern distribution. This study proposes a two-level semantic segmentation framework for visual corrosion assessment of steel bridge components. Level 1 performs binary pixel-wise segmentation to quantify visible corrosion coverage, whereas Level 2 assigns corrosion pixels to four inspection-oriented visual categories: Uniform, Crevice, Underfilm, and Other Localized corrosion. Field images were annotated using an operational visual taxonomy by the first author, and category-label reproducibility was subsequently evaluated through a blinded second-annotator audit of the categorized validation polygons. Transformer-based and CNN-based models were evaluated using common image subsets, input resolution, and evaluation metrics, while model-specific implementation and optimization settings were documented explicitly. Performance was assessed using region-overlap metrics, class-wise precision, recall, and F1-score, and Boundary-F1 to quantify edge agreement along irregular corrosion fronts. The highest-performing binary configuration achieved an mIoU of 0.843. The adopted formula-derived Level 2 configuration achieved a five-class mIoU of 0.605 and a mean corrosion-category F1-score of 0.678; a manual-weighting variant produced a marginally higher five-class mIoU of 0.607 but was not adopted as the primary formula-derived SegFormer configuration. Binary-merged evaluation showed that Level 2 retained substantial corrosion-coverage capability, although dedicated binary models remained superior for foreground–background segmentation. The framework provides complementary representations of corrosion coverage, visual-category distribution, and boundary quality. It is intended to support, rather than replace, expert inspection, field measurement, and engineering evaluation.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"60 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Physics-Informed Neural Network (DM-PINN-FP): Incorporating Cracking Preference Index for Asphalt Mixture Fatigue Prediction","authors":"Quan Liu, Huimin Li, Yangming Gao, Hanyu Zhang, Yuanyuan Li, Denise Lee, Jiantao Wu","doi":"10.1016/j.cacaie.2026.100171","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100171","url":null,"abstract":"Reliable fatigue prediction is critical to asphalt pavement design. However, most models fail to account for the dual nature of fatigue cracking, which comprises both cohesive and adhesive failure modes. To address this gap, this study presents a Dual Mode Physics-Informed Neural Network Fatigue Prediction (DM-PINN-FP) model tailored for asphalt mixtures. The adhesive failure cracking preference index (<ce:italic>β</ce:italic>) is quantitatively determined through pull-off tests. A Fatigue Cracking Sensitivity Index (FCSI), derived from this cracking preference index, is incorporated as a control parameter within the physics-informed network to represent the effects of aggregate type, temperature, and asphalt film thickness. Grounded in the viscoelastic continuum damage (VECD) theory, the overall damage is decoupled into adhesive and cohesive failure components, leading to coupled evolution equations for dual-mode damage. Building on these mechanisms, a dual-branch physics-informed prediction model is developed, integrating physical constraints and monotonicity conditions within a deep learning framework enhanced by multi-head attention and Long Short-Term Memory (LSTM) networks. Results show that the DM-PINN-FP model successfully mitigates cumulative error issues common in standard neural networks and substantially improves fatigue prediction accuracy. Further analysis reveals that temperature elevation predominately influences the cracking mode and accelerates stiffness degradation. Moreover, the proposed model effectively captures the impacts of different aggregate types and temperature conditions on damage progression. These findings validate the cracking preference index as a valuable link connecting mesoscopic material design parameters with macroscopic fatigue behaviors.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"56 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691819","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Perception-to-Control Computational Pipeline for Humanoid Robot-Based Construction Surface Finishing via Attention-Based Normal Refinement and Residual Reinforcement Learning","authors":"Tianyu Ren, Houtan Jebelli","doi":"10.1016/j.cacaie.2026.100168","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100168","url":null,"abstract":"Surface finishing in construction governs dimensional tolerance and visual quality but remains labor intensive because effective finishing requires perception of tolerance-scale deviation under occlusion, boundary-aware coverage planning near seams and openings, and stable long-stroke execution with regulated contact. This study presents a computer-aided perception–planning–control pipeline for autonomous humanoid surface finishing on vertical and slab-like surfaces. Multimodal LiDAR and RGB-D observations are fused into a surface map and refined with a normal-consistency attention network that stabilizes surface normals while preserving sharp transitions needed for boundary reasoning. From the refined representation, a boundary-aware planner constructs deviation and boundary distance fields and generates lane-based stroke trajectories with explicit clearance constraints around edges and openings. Execution is achieved using constrained whole-body control with balance stabilization and impedance regulation, augmented by bounded residual reinforcement learning that compensates systematic tracking bias without violating feasibility constraints. Evaluation in ROS-integrated Isaac Sim across diverse drywall and concrete scenes shows that the proposed refinement reduces task-region height error from 3.9 mm to 2.4 mm and normal error from 14.6° to 9.1°. With identical stroke plans, residual adaptation increases deviation reduction to 0.60 on drywall and 0.53 on concrete, improves task success to 93.3%, increases the minimum boundary margin to 27.9 mm, and reduces redundancy on concrete to 0.19 while maintaining 96.8% coverage. A preliminary physical test further shows that the humanoid can approach a real wall, estimate the local wall normal, align the finishing tool perpendicular to the wall plane, and initiate finishing contact.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"27 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Robust Sensor Fusion and Localization Correction for Tunnel Inspection Robots via a Probabilistic Baseline Model","authors":"Kaitian Wang, Mingxin Gao, Jianxin Cao, Yang Liu","doi":"10.1016/j.cacaie.2026.100172","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100172","url":null,"abstract":"Autonomous tunnel inspection robots are essential for intelligent infrastructure management and the construction of structural digital twins. However, in GPS-denied and slippery underground environments, non-systematic errors such as wheel slip frequently corrupt odometry data, severely degrading the spatial fidelity of structural defect mapping. To bridge the gap between robotic navigation and reliable infrastructure assessment, this paper proposes an unsupervised, data-driven anomaly detection and state correction framework based on a probabilistic baseline model. Initially, by analyzing the coupled motion response of the onboard sensor fusion system, a probabilistic baseline model is established via Kernel Density Estimation (KDE) to learn the nominal kinematic data distribution. A diagnostic factor is then derived to detect abnormal slip conditions in real-time. Subsequently, robust sensor fusion is achieved by integrating this probabilistic model into the observation equations of an Extended Kalman Filter (EKF), effectively rectifying the robot's motion state. The proposed framework is validated through numerical simulations and physical experiments in an operational subway tunnel. Results demonstrate that the proposed approach outperforms four mainstream unsupervised anomaly detection methods (autoencoder, variational autoencoder, isolation forest, and one-class SVM) in diagnosis accuracy, inference latency, and memory footprint, while maintaining exceptional robustness against severe noise interference. Moreover, field validation on an operational subway tunnel with real induced wheel slip confirms a slip diagnosis accuracy of 99.12%, with the longitudinal positioning error converging to within 0.2 m after correction.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"26 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691818","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Rapid Prediction of Railway Ballast Particle Motion Using an Acceleration-Based Continuous Convolutional Neural Network","authors":"Kai Liu, Yiming Du, Min Xue, Qiantao Ma, Ganzhong Liu, Rong Chen, Ping Wang","doi":"10.1016/j.cacaie.2026.100167","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100167","url":null,"abstract":"The discrete element method (DEM) can describe the mesoscopic motion and contact behavior of irregular railway ballast particles, but its low computational efficiency limits long-duration simulation of ballast-bed evolution under repeated loading. To address this limitation, this study proposes an acceleration-based continuous convolutional neural network (CCNN) for rapid ballast particle motion prediction. Different from position-based prediction methods, the proposed method reformulates the network output from particle position correction to acceleration correction, and updates particle velocity and position through kinematic integration, thereby reducing the error amplification caused by velocity back-calculation through positional differencing. A calibrated DEM model evaluated against laboratory angle-of-repose tests was used to generate 100,000 training samples, and the trained model was subsequently assessed using same-scenario and cross-scenario ballast deposition cases. For a 50 s prediction involving 433 ballast particles, the proposed method achieved a computational efficiency approximately 542 times higher than DEM, with a 9.88% relative error in the angle of repose and a 75.6% reduction in average centroid drift compared with the position-based method. In a cross-scenario case where the particle number increased by a factor of 2.5, the proposed method remained approximately 402 times faster than DEM and reduced the centroid drift error by 97.75%. These results indicate that the acceleration-based CCNN can substantially improve DEM-based ballast particle motion prediction efficiency while maintaining stable particle-position evolution and preliminary cross-scenario adaptability. The proposed method provides a technical basis for rapid DEM-based ballast particle motion prediction and future extension to ballast-bed structural optimization and evaluation of maintenance effectiveness.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"30 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691111","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancement of Intelligent GPR Detection based on Category-Aware Generative Framework of B-scan Images","authors":"Zuhao Wang, Guangyao Li, Fengzijun Pan, Zepeng Fan, Xiaopei Cai, Dawei Wang","doi":"10.1016/j.cacaie.2026.100170","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100170","url":null,"abstract":"Deep learning has advanced intelligent detection of ground penetrating radar (GPR) applied in transportation infrastructure inspection, yet its performance remains limited by the scarcity and class imbalance of labeled GPR B-scan images. To address this challenge, this study proposes GPR-ImaGen, a category-aware generative data augmentation framework for GPR B-scan images. Built on the pretrained rectified flow model Flux.1-dev, the framework generates defect category-consistent B-scan images under semantic guidance from engineering annotations, alleviating the scarcity and class imbalance of the training data for intelligent detection. To better adapt the model to the characteristics of GPR B-scan images and limited-data conditions, parameter-efficient fine-tuning, a robust loss function and a boundary-focused timestep sampling strategy are incorporated into the training process. The proposed framework achieves an FID of 10.496 and a Bscan-FID of 10.828. Extensive task-oriented experiments across multiple intelligent detection models and diverse limited-data settings show that the generated data consistently improve downstream detection performance. Among them, YOLOv11 achieves the highest detection performance (mAP@50 of 91.6%) at a 100% augmentation ratio, while mAP@50 increases from 73.6% to 81.9% under a data-scarce setting with only 25% of the real training data retained. These results indicate the practical potential of a category-aware generative framework for improving the robustness and reliability of intelligent GPR detection for transportation infrastructure inspection.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"37 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691051","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Lifecycle updating of rockfill dam deformation models via dependence-aware sequential Bayesian inference with posterior-guided continual surrogate learning","authors":"Zhihong Huang, Gang Ma, Yi Zhang, Guike Zhang, Zhitao Ai, Yifan Cui, Wei Zhou","doi":"10.1016/j.cacaie.2026.100169","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100169","url":null,"abstract":"Rockfill dams exhibit pronounced stage-dependent deformation during construction, impoundment, and long-term operation, requiring monitoring-updated deformation models for lifecycle prediction and digital twin-based safety assessment. However, existing updating methods are often deterministic and stage-isolated, while conventional probabilistic approaches insufficiently capture dependence among mechanically related constitutive parameters. For multi-stage lifecycle analysis, these methods also lack cross-stage adaptability and often require repeated finite-element simulations or full surrogate retraining, leading to high computational cost. This study proposes a dependence-aware sequential probabilistic framework with posterior-guided continual surrogate learning for dynamic updating of rockfill dam deformation models. The framework progressively assimilates monitoring data to track lifecycle parameter evolution, preserves empirical parameter dependence via Copula-enhanced prior modeling within a sequential Bayesian updating scheme, and adaptively refines the surrogate by concentrating new training samples in posterior-relevant regions while retaining historical deformation knowledge. Application to the 303 m Lianghekou core-wall rockfill dam shows that the proposed method reduces the global relative settlement prediction error from 37.4% to 9.7% and reduces the dominant computational cost of lifecycle surrogate updating by 45.8% compared with full surrogate retraining. The updated model further enables uncertainty-informed settlement prediction, spatiotemporal deformation reconstruction, and posterior-informed diagnosis of safety-related deformation characteristics, providing an interpretable and efficient approach for lifecycle deformation analysis and digital twin-based dam safety management.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"8 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Real-Time, Crowdsourcing-Enhanced Forecasting of Building Functionality During Urban Floods","authors":"Lei Xie, Peihui Lin, Naiyu Wang, Paolo Gardoni","doi":"10.1016/j.cacaie.2026.100163","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100163","url":null,"abstract":"Urban flood emergency response increasingly relies on infrastructure impact forecasts rather than hazard variables alone. However, real-time predictions are unreliable due to biased rainfall, incomplete flood knowledge, and sparse observations. Conventional open-loop forecasting propagates impacts without adjusting the system state, causing errors during critical decisions. This study presents CRAF (Crowdsourcing-Enhanced Real-Time Awareness and Forecasting), a physics-supervised, closed-loop framework that converts sparse human-sensed evidence into rolling impact forecasts. By coupling physics-based simulation learning with crowdsourced observations, CRAF infers system conditions from incomplete data and propagates them forward to produce multi-step, real-time predictions of zone-level building functionality loss without online retraining. This closed-loop design supports continuous state correction and forward prediction under weakly structured data with low-latency operation. Offline evaluation demonstrates stable performance across simulation-based storm scenarios. In real-event application during Typhoon Haikui (2023) in Fuzhou, China, CRAF reduces 0.5–2.5 h ahead forecast errors by 85%–95% relative to fixed rainfall-driven forecasting and by 75%–80% relative to updated rainfall-driven forecasting, while outperforming a conventional Ensemble Kalman Filter–based data-assimilation baseline and limiting computation to approximately 10 minutes per update cycle. Sensitivity analysis further shows that the framework remains stable under 10%–30% perturbation or missingness in accepted crowdsourced observations. These results show that impact-state alignment—rather than hazard refinement alone—is essential for reliable real-time decision support, providing a pathway toward operational digital twins for resilient urban infrastructure systems.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"19 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691112","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"PoYOSS: A framework for post-yield oedometer stress-strain prediction of sensitive clays","authors":"Mohammad Sadegh Farhadi, Tim T. Länsivaara","doi":"10.1016/j.cacaie.2026.100161","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100161","url":null,"abstract":"Accurate prediction of soil settlements is one of the most fundamental tasks in geotechnical engineering. Traditionally, it is based on soil sampling and oedometer tests in the laboratory, which requires considerable time and financial resources. In addition, it determines settlement properties at individual local points, which may well be a source of uncertainty, especially in heterogenous soils. This study presents a novel mathematical approach to predict the oedometer post-yield stress-strain, <ce:math altimg=\"si23.svg\"></ce:math>, behaviour of sensitive clays sampled in southern Finland. The proposed model comprises five sub-models: a) data collection, b) data pre-processing, c) deformation moduli determination, d) learning-based predictor, and e) conflict-resolution Game theory models. In the data collection and data pre-processing sub-models, results from constant-rate-of-strain (CRS) oedometer tests conducted for five testing sites, were compiled and processed to refine minor imperfections. The deformation constrained moduli were then approximated by developing a multi-objective optimization model coupled with the Game theory Condorcet’s social choice rule (SCR) model. As the learning-based predictor, artificial neural network (ANN) models were trained to predict the post-yield <ce:math altimg=\"si23.svg\"></ce:math> behaviour, and the best performing model was identified through the Game theory Condorcet’s SCR conflict-resolution model. The results indicated that the ANN model trained by the Bayesian Regularization (BR) method, using natural water content, <ce:math altimg=\"si55.svg\"></ce:math>, soil sensitivity, <ce:math altimg=\"si56.svg\"></ce:math>, and yield stress, <ce:math altimg=\"si108.svg\"></ce:math>, predicted the deformation curves most accurately. Further evaluation of different ANN training methods and input combinations, using the Game theory bargaining Borda scoring method, showed that <ce:math altimg=\"si55.svg\"></ce:math> and <ce:math altimg=\"si56.svg\"></ce:math> together can be the best combination of soil properties to predict the post-yield <ce:math altimg=\"si23.svg\"></ce:math> curves almost accurately.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"26 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691144","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Structure-Guided Graph Attention Learning for Traffic Conflict Prediction under Heterogeneous Expressway Geometries","authors":"Yongqing Hao, Bei Zhou, Zongzhi Li, Duo Li, Hongshuai Zhu, Ke Wang","doi":"10.1016/j.cacaie.2026.100165","DOIUrl":"https://doi.org/10.1016/j.cacaie.2026.100165","url":null,"abstract":"Accurate traffic conflict prediction on urban expressways remains challenging because similar observed traffic states may carry different risk implications across merging, weaving, and diverging segments. Existing data-driven models often learn shared correlations or impose fixed structures under geometric heterogeneity. To address this issue, this study develops Causal-GAT, a structure-guided graph attention framework for heterogeneous traffic conflict prediction. Within the framework, SAM-inferred structures are incorporated as segment-specific structural priors to organize observed kinematic relationships and guide graph attention learning. Structural Agnostic Modeling (SAM) is first applied to observed traffic variables to infer segment-specific proxy structural topologies. These priors are then injected into both feature reconstruction and graph attention so that message passing is guided by segment-specific structure in addition to data similarity. Experiments on the MAGIC UAV trajectory dataset from Shanghai’s Inner Ring Elevated Road cover three representative segment types within one expressway corridor and a four-level traffic conflict classification task defined by Time Spent in Conflict (TSC). Across these segments, Causal-GAT achieves the best Macro F1 among the baselines, with the largest relative improvement appearing in the off-ramp segment, where geometric and traffic conditions are more stochastic. The inferred proxy structural topologies and attribution results are also broadly consistent with traffic-flow differences among merging, weaving, and diverging environments, while indicating that proxy structural importance and model-based predictive contribution are related but not identical. Overall, the findings suggest that structure-guided attention helps mitigate spurious associations in heterogeneous geometric conditions and provide a more interpretable basis for segment-sensitive traffic safety analysis.","PeriodicalId":156,"journal":{"name":"Computer-Aided Civil and Infrastructure Engineering","volume":"20 1","pages":""},"PeriodicalIF":11.775,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148691052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}