Computer-Aided Civil and Infrastructure Engineering最新文献

This study introduces a novel integrated framework for real-time tunnel boring machine (TBM) thrust prediction, addressing critical limitations in handling non-stationarity, complex spatiotemporal dependencies, and abrupt disturbances. First, a real-time windowed multi-resolution analysis process, which performs decomposition strictly within each segmented sample window, is presented to explicitly disentangle the latent multi-scale dependencies embedded in the thrust data. This ensures strict causality (using only current/historical data), prevents information leakage, and enhances resolution adaptability by capturing local dynamics specific to each data segment, overcoming global averaging effects. Second, a novel synergistic prediction architecture, integrating a hybrid static model with dynamic online residual correction, is proposed. A specifically optimized CNN-LSTM-attention primary model learns complex long-term global patterns. Crucially, an efficient random Fourier features-based online module is dedicated solely to real-time learning of the primary model's residual dynamics, acting as a dynamic corrector rather than an independent predictor. This targeted residual correction significantly enhances robustness against non-stationarity and disturbances. These innovations form an integrated solution and systematically address real-time capability, local adaptability, complex pattern learning, and dynamic error correction. The results indicate that the presented method reduces the mean absolute percentage error from 2.84% to 1.89% and increased $�R^2$ from 0.901 to 0.953. The generalizability of the model was further confirmed through the application of diverse datasets obtained from various chainages along the route. The proposed machine learning–based model can provide guidance for operators in real-time TBM parameter adjustment during construction

Pavement defects pose serious threats to road safety and infrastructure longevity. Following a full-supervised manner, many existing detection methods rely heavily on extensive labeled data. In this paper, motivated by the inherent diversity and imbalance of real-world pavement images, we propose a reconstruction-based unsupervised pavement anomaly detection framework. It leverages a conditional guided blurring diffusion model to reconstruct abnormal images as defect-free, combined with domain-adaptive feature refinement and a defect-aware feature selection module for robust anomaly scoring. By integrating simplex noise within the conditional guiding framework, our approach effectively preserves normal pavement textures while removing defects, enabling precise localization without relying on pixel-level annotations. Extensive comparison and ablation experiments on the Pavementscape dataset demonstrate that our method outperforms other unsupervised anomaly detection techniques and remains competitive with fully supervised segmentation approaches. These results underscore the potential of our unsupervised, diffusion-driven pipeline to address the costly annotation bottleneck in large-scale pavement inspection, offering a scalable and highly accurate solution for real-world road maintenance.

Asphalt pavement is widely used in transportation systems due to its superior comfort, rapid construction, and convenient maintenance, while also being confronted with the problem of shear failure damage. Herein, two-dimensional virtual models of asphalt mixture specimens are constructed based on the discrete element method for the virtual biaxial compression test to elucidate the underlying mechanisms behind shear failure damage. The results demonstrate that the increase in volumetric strain due to shear dilation signifies the onset of the unjamming transition, whereas the emergence of shear failure zones and vertical cracks reflects its manifestation in asphalt mixtures. Confining pressure has an inhibitory effect on the development of the unjamming transition, whereas temperature promotes its progression. The emergence of heterogeneous structures and the evolution of the force chain network into a disordered and branched structure are manifestations of the unjamming transition in the displacement field and force chain system. The outcome offers novel insights into the prediction and understanding of shear failure behavior in asphalt mixtures, establishing a fundamental framework for analyzing failure evolution and its influencing factors.

Effective management of tunnel boring machine (TBM) jamming is crucial for ensuring safety and mitigating construction downtime. However, previous studies have primarily focused on predictive modeling based on numerical datasets, with limited consideration of field-based geological conditions and inadequate investigation of the fundamental mechanisms underlying jamming phenomena. This study utilized two ensemble learning algorithms, Random Forest and Extreme Gradient Boosting, to predict TBM jamming based on a field dataset from 39 tunneling projects. A data augmentation technique was employed to construct an expanded dataset. The predictive model trained on the augmented dataset demonstrated improved detection of TBM jamming compared to the model developed without data augmentation. The jamming mechanism was successfully characterized, revealing the individual effects of geological factors and their complex interactions. A distinct difference in predictive uncertainty between correct and incorrect predictions supports the model's reliability. Finally, a practical risk management system was proposed by incorporating the predictive model with probability thresholds and validated through field application.

The intrinsic brittleness of the cement matrix limits its synergy with steel reinforcement bars, constraining energy dissipation and crack control capacity of concrete. Enhancing the ductility of cementitious materials is, therefore, essential for improving structural resilience. A porous carbon material, for example, biochar, offers a sustainable alternative that can improve ductility and energy dissipation capacity, while simultaneously contributing to carbon sequestration. Despite promising experimental observation, the fracture mechanisms underlying this toughening effect remain insufficiently understood. This study addresses this knowledge gap by developing a multi-scale voxel-based modeling framework for biochar–cement composites, linking microscale mechanical heterogeneity to macroscale fracture behavior. The elastic modulus of biochar–cement paste was first quantified across nanoscale (∼nm and ∼µm) to mesoscale (∼mm and ∼cm) through nano- and micro-indentation, providing scale-bridged inputs for the model. The framework explicitly resolves aggregates, interfacial transition zones, and biochar particles within a concurrent multi-scale domain, enabling simulation of localized fracture while retaining computational efficiency. The simulation results were validated through a three-point bending test and digital image correlation. These findings demonstrated that biochar could alter the crack propagation by redistributing interfacial stress and promoting multi-layered crack deflection, which significantly enhanced the energy dissipation by up to 90%. This study elucidates the multi-scale mechanisms by which the pore architecture of biochar enhances ductility, providing a scalable framework for the design of high-ductile, sustainable cementitious materials.

A substantial number of existing bridges in high-seismicity countries like Italy were not designed for seismic actions, thus being particularly vulnerable to earthquake-induced motions. While deck isolation from piers is commonly employed to reduce base shear and seismic vibrations, it often fails to keep deck displacements within acceptable limits, thus preventing a large-scale application of this technology. Damping levels higher than those provided by common isolation devices require nonlinear analysis methods, including unconventional hysteresis models. Aiming at improving the seismic response of bridges, this study proposes a unified optimal design strategy for bridges adopting generalized non-linear rate-dependent (RD) and rate-independent (RI) control systems based on Seleemah–Constantinou and Vaiana–Rosati models, respectively. The resulting generalized nonlinear control systems are then optimized using a meta-heuristic algorithm by simultaneously considering multiple competing objectives to mitigate bridge deck displacement, acceleration, and transmitted force to the pier. The RD and RI control systems tend to yield a displacement-constrained and an acceleration-constrained design objective, respectively. In both cases, the optimal Pareto front shows a significant improvement over the base-isolated response in terms of isolator displacement with further reduction or minimal increase in the force transmitted to the pier. The results of this study contribute to the development of an effective seismic mitigation strategy for bridges where both base shear and deck displacement provide major constraints.

Shape design is the dominant task of arch dam construction, involving significant computational costs. Conventional approaches are largely manual and experience driven. Though surrogate-assisted methods accelerate the procedure, the reusable “optimization policy” is ignored. Inspired by the cyclical interactions between designers and experts in real-world engineering, a deep reinforcement learning (DRL) framework is proposed for automated and intelligent arch dam shape optimization. The framework models the arch dam design as a DRL task and employs the Soft Actor–Critic algorithm to train the agent, with Gaussian process surrogate models accelerating the procedure. A weight-vector-based transfer learning strategy is introduced to generalize the framework to solve multi-objective problems. The framework is implemented on a real-world arch dam, and the results demonstrate that the agent effectively learns an optimization policy and generates a high-quality Pareto front. The selected optimal shape achieved 12.5% and 25.87% reductions in dam volume and tensile volume, respectively, demonstrating enhanced economic efficiency and structural safety. The same methodology can be widely applied to other engineering structure designs and has the potential to drive transformative advances in the engineering community.

Accurate work-phase recognition is essential for advancing energy efficiency and intelligent control. However, significant challenges impede the advancement of work-phase recognition technology, including the complexity of sensor input signals, reliance on manual intervention for time-frequency feature selection, limited model generalization, and suboptimal recognition accuracy. To address these issues, this paper proposes a deep learning framework that combines a feature fusion method that integrates gated recurrent unit (GRU) network feature extraction and fractional calculus feature (FCF) enhancement with a Bayesian-optimized random forest (RF) classifier. A GRU network with an integrated attention mechanism effectively reduces the need for manual feature selection, whereas FCF enhancement expands the feature space through fractional integration and differentiation without additional sensors. Feature-level data fusion and Bayesian optimization improve the generalization capability of the RF model. The experimental results for two typical types of machinery demonstrated recognition accuracies of 99.38% and 99.45% for work-phase recognition, confirming the superior performance of the proposed framework.

The Japanese government has been promoting construction information modeling initiatives, following the example of building information modeling, which has been widely adopted in the construction sector of Western countries. However, progress in preparing and utilizing 3D data for existing structures, which are primarily subject to maintenance and management, has been limited. Previous studies have proposed methods to automatically generate 3D models by capturing the three-dimensional structure and dimensions of target objects from point clouds obtained through laser scanners. However, these studies do not address the interoperability of various data, and international standard data schemas such as Industry Foundation Classes (IFC) by the International Alliance of Interoperability and CityGML by the Open Geospatial Consortium (OGC) do not define schemas that encompass maintenance and management information, such as inspection results. Therefore, this study proposes a one-source, multi-use data schema capable of comprehensively managing both structural and maintenance information for bridges. The proposed data schema complies with international standards by integrating IFC-Bridge into CityGML, a standard developed by the geospatial information standardization organization OGC. A validation experiment was conducted using drawings, inspection records, and point clouds of bridges in Shizuoka City, demonstrating that the schema can be applied to 20 bridges of four types for three different use cases.

The cover image is based on the article Generative adversarial network based on domain adaptation for crack segmentation in shadow environments by Yingchao Zhang and Cheng Liu, https://doi.org/10.1111/mice.13451.