To investigate the pipeline-soil interaction mechanism and settlement deformation characteristics of the high-speed railway tunnels in marine soft soil under train-induced vibration loads, a 1∶40 scale model test was conducted based on the Zhanjiang Bay undersea Tunnel, the control engineering of the Guangzhou-Zhanjiang High-Speed Railway. Numerical simulations were also carried out, and the dynamic response distribution and deformation development of the tunnel structure and surrounding soil were further analyzed. The results showed that, under train-induced vibration loads, the acceleration response of the soil surrounding the tunnel was found to exhibit significant spatial nonuniformity, generally characterized by a distribution pattern of “stronger below and weaker above, and greater near the tunnel and smaller farther away”. In particular, the soil at and near the tunnel base was directly affected by the train load, and thus exhibited the most pronounced dynamic response. Affected by the combined effects of soil damping, wave dispersion, and energy dissipation at the tunnel-soil interface, the vibration energy was gradually attenuated with increasing propagation distance, and significant differences were observed in the vibration attenuation patterns along different propagation directions. Compared with lateral propagation, the vertical vibration was more significantly attenuated during its propagation into the overlying and underlying soils. After lateral vibration propagated into the middle and far field regions, the attenuation rate was relatively reduced, indicating a stronger capacity for sustained propagation. Under cyclic loading, the ground settlement was found to exhibit significant cumulative and staged characteristics. The overall settlement trough was characterized by a distribution pattern of larger settlement in the middle and smaller settlement on both sides. Settlement developed rapidly during the initial loading stage, and the settlement generated during the first 1 920 loading cycles accounted for 71.4% of the total settlement. As the number of cycles increased, the soil was gradually densified, the additional deformation induced by each individual load decreased, and the settlement growth rate was reduced. The settlement evolution was characterized by a transition from rapid accumulation to gradually decelerated development.

Addressing the current lack of risk loss quantification for specific tunnel collapse incidents at tunnel faces, a quantitative analysis of risk losses associated with tunnel face collapses in mountainous highway tunnels in Yunnan constructed using the drill-and-blast method was conducted. By establishing a risk quantification framework integrating occurrence probability and loss consequences, a quantitative calculation model covering personnel casualties, project delays, and economic losses was developed. Corresponding risk acceptance criteria were formulated based on the ALARP principle.The study focused on five high-risk processes: hazard clearance, mucking, steel arch installation, anchor rod placement and shotcrete application. Combining field survey data with the exceedance probability and loss magnitude of collapse risks at each process stage, quantitative analysis was performed using F-N curves for casualties, F-TD curves for schedule delays, and F-D curves for economic losses. Results indicated that personnel injury risk was significantly influenced by workers' spatiotemporal distribution, peaking during steel arch installation. Schedule delays in minor collapses were primarily driven by casualties and subsequent rectification, while major collapses were controlled by cleanup and repair workloads.Economic losses heavily depended on personnel compensation and indirect costs from schedule delays, often exceeding 70% of total losses in Southwest China, suggesting that even small-to-medium collapses may reach unacceptable risk thresholds.This study provides quantitative insights for risk assessment and management in mountain tunnel construction safety.

To address the problem of large deformation of surrounding rock in tunnels crossing fault fracture zones, the Heimaguan Tunnel in Gansu was taken as the engineering background. Based on the field monitoring data of surrounding rock pressure, crown settlement, and horizontal convergence obtained from 23 monitoring sections, the deformation evolution characteristics of the surrounding rock and the mechanical behavior of the supporting structures were systematically analyzed.The results showed that significant deformation occurred in the fault fracture zone section. The maximum unilateral horizontal convergence reached 787.7 mm, while the maximum crown settlement reached 472.6 mm, exhibiting typical characteristics of large deformation in fault fracture zones. The deformation and pressure of the surrounding rock exhibited significant spatial discreteness, with the coefficients of variation of crown settlement and horizontal convergence reaching 83.2% and 93.0%, respectively, indicating the highly heterogeneous nature of the surrounding rock within the fault fracture zone. During the construction stage, the load-sharing ratio of the secondary lining ranged from 5.18% to 38.46%, with an average 20.29%, indicating that the surrounding rock load was mainly borne by the primary support, while approximately 20% of the load was carried by the secondary lining. Under the coupled effects of high in-situ stress, bedding-controlled structure, and groundwater softening, the surrounding rock deformation and support stress exhibited pronounced spatial non-uniformity. The research results can be referred to for the control of large deformation and the support design of similar tunnels crossing fault fracture zones.

To address the drawbacks of cumbersome procedures and poor generalization ability in traditional crack detection methods, a CR-YOLO network for tunnel lining crack detection based on deep learning was proposed. Considering the slender morphological characteristics of cracks, the context guided block(CG Block)was incorporated into the network to fuse local and global information; by integrating channel and spatial attention mechanisms at both local and global levels, the model was enabled the model to fully capture the contextual information of crack regions. Meanwhile, a reparameterized generalized feature pyramid network(RepGFPN)module was added to improve the feature pyramid network(FPN)for object detection, which achieved more efficient fusion of multi-scale features and enhanced the capture of high-level semantic information and low-level spatial details. This method optimized the model performance under constrained computing resources and reduced inference latency without a significant increase in computational overhead. Experimental results on the self-collected dataset demonstrated that the A_P50 and A_P50-95 of CR-YOLO reached 90.2% and 66.4%, respectively, which represented an increase of 6.2% and 3.7% compared with the baseline model YOLOv10. The model outperformed other one-stage object detection networks(YOLOv3, YOLOv5, and YOLOv9)in terms of detection accuracy. Additionally, its inference speed reached 138.7 fps, enabling real-time detection of tunnel lining cracks.

To address the low efficiency of traditional geological logging and the excessive manual intervention required in numerical modeling, an integrated digital-numerical analysis method for discontinuous rock masses based on binocular vision was proposed. In this method, binocular photogrammetry, automated 3D/2D geometric Boolean operation modeling, and discontinuous deformation analysis(DDA)were integrated. By constructing a unified framework for geometric and mechanical data, an automated data workflow from on-site image acquisition and digital extraction of geological information to mechanical analysis of discrete blocks was achieved. To verify the reliability and engineering applicability of the proposed method, simulations of the progressive instability and failure processes of surrounding rocks induced by discontinuity cutting were carried out in two-dimensional and three-dimensional spaces, based on a tunnel in Southwest China and the Suocaopo Tunnel in Guizhou Province, respectively. The results showed that the exposed discontinuity information of rock masses was identified and converted into discontinuous block models for DDA simulations. Under given geometric and mechanical parameters of discontinuities, the sliding, separation, falling, and collision processes of rock blocks cut by complex joints were analyzed. An automated technical workflow for discontinuity identification, discontinuous modeling, and stability analysis of complex jointed rock masses was therefore established, which could provid a reference for the identification of potentially unstable blocks, rockfall risk assessment, and support design during tunnel construction.

To address the problem of poor edge localization in deep learning-based methods and low efficiency of traditional percolation algorithms for tunnel crack detection, a two-stage approach that integrates an improved DeepLabV3+ with a skeleton-guided percolation algorithm was proposed. In the first stage, an improved DeepLabV3+ model with a lightweight backbone network, incorporating a CBAM attention module, an optimized ASPP module, and a Dice loss function, was developed to achieve high-recall crack pre-segmentation.In the second stage, a skeleton-guided percolation growth strategy combined with morphological constraints was applied to refine crack edges and measure crack widths. A tunnel crack dataset containing 20 504 pixel-level annotated images was constructed, covering various tunnel lining regions(crown, haunch, sidewall)and surface conditions(dry, wet, stained). Experimental results on this dataset showed that the pre-segmentation module achieved an accuracy of 90.1% and a recall of 86.7%. The improved percolation algorithm increased the precision to 98.5% while maintaining high recall, and improved computational efficiency by approximately 20 times. Engineering validation demonstrated a detection rate exceeding 84% for cracks wider than 0.1 mm, with a mean absolute error of less than 0.3 mm.The proposed method effectively balanced detection accuracy and computational efficiency, providing a feasible solution for automated tunnel lining crack detection.

To address the water hazard risks faced by deep-buried tunnels crossing high confined water fault zones in river valleys, the regional engineering hydrogeological characteristics were systematically analyzed, the evolution law of the seepage field was revealed, surface directional drilling grouting was proposed and applied as a water hazard control measure, and its application effect was evaluated. The results showed that an obvious groundwater drawdown funnel was formed after tunnel excavation in high-pressure water-rich areas. Highly permeable fault zones altered seepage paths and led to groundwater convergence, while low-permeability surrounding rock blocked water flow and created local high-head zones, exposing the tunnel construction to both water inrush and high external water pressure. Surface directional drilling grouting with a total length of 748 m was applied to the high confined water fault zone, increasing the thickness of the pre-grouted ring to more than 11.9 m. The tunnel seepage discharge decreased from 7.0 m³/(d·m)to 1.1 m³/(d·m), and the external water pressure was reduced by 90.0%. Directional drilling grouting effectively sealed hydraulic channels, improved the integrity of surrounding rock, reduced rock permeability, and mitigated water hazard risks such as tunnel water inrush and high water pressure.

Seepage through concrete cracks was identified as a critical factor inducing tunnel lining defects. Its evolutionary process was accompanied by complex energy conversion and heat exchange. However, the microscopic thermodynamic behavior could hardly be captured by traditional methods. Therefore, the crack seepage mechanism was taken as the core research object. Infrared thermography was employed as a diagnostic tool for temperature fields, supplemented by a high-speed camera, and laboratory seepage tests under various working conditions were conducted. Specifically, the temperature evolution law and the intrinsic energy mechanisms in the crack area under isothermal water injection were revealed. During the first water injection, a three-stage temperature change(an initial rise, a subsequent decline, and a final stabilization)was observed in the crack, which was governed by the competitive trade-off between hydration reaction heat and evaporative heat absorption. With increasing water injections, the hydration heat was dissipated, and the temperature response transformed into a two-stage pattern of direct cooling followed by stabilization. It was thus confirmed that evaporative cooling became the dominant energy consumption mechanism during the seepage process. In addition, the thermal response rate and spatial extent were significantly affected by the crack width, and the apparent crack size in the infrared thermograms was found to be larger than the actual size, attributed to the amplification of crack geometric information by thermal diffusion effects. The energy coupling mechanism in the seepage process was clarified, and a theoretical basis was established for leakage diagnosis based on thermal signals.

Shield tunneling in cobble-boulder stratum was frequently accompanied by significant ground disturbance and a high risk of surface collapse. For the prediction and mitigation of this problem, empirical methods were primarily relied upon in current practice, leaving a notable gap in quantitative early-warning research on stratum instability. Based on the slurry shield tunneling project of the Beijing Subway Line 1 Branch, laboratory model tests were conducted to investigate the impact of cobble blocking in front of the cutterhead on surface settlement patterns and the evolutionary process of stratum collapse. The results indicated that, influenced by the large-particle granular skeleton and discontinuous medium characteristics, the cobble-boulder stratum exhibited a significant delayed surface settlement phenomenon. This evolutionary process could be broadly divided into four stages: slow development, rapid increase, temporary stability, and collapse failure. A ground loss ratio of approximately 7% to 10% was observed in the cobble-boulder stratum, significantly exceeding that of conventional sandy strata. Furthermore, this ratio continuously increased as the volume of stagnant boulders increased. When the boulder stagnation rate exceeded a critical threshold of 30%, surface collapse was highly likely to be triggered, typically occurring after the shield machine passed the corresponding monitoring point by a distance of approximately 0.5 D (where D is the tunnel diameter). The accumulation of large-diameter boulders in front of the cutterhead profoundly affected the shield's disturbance range; stagnant boulders exacerbated over-excavation, causing prominent cavities directly above the excavation face and in the surrounding stratum, and substantially reducing overall stability. These findings provided quantitative early-warning indicators and theoretical support for mitigating surface collapse induced by cobble blocking in similar engineering projects.