The drill-and-blast method remains the primary excavation technique for highway tunnels in mountainous regions, where the shotcrete lining plays a crucial role in the New Austrian Tunneling Method. However, due to blasting disturbances and complex geological conditions, deformation of the shotcrete lining is often unavoidable, posing serious safety risks during construction. Therefore, the necessity of monitoring shotcrete lining deformation and the importance of advanced deformation monitoring methods are self-evident. Traditional single-point monitoring approaches, mainly based on total stations, fail to capture the full-section deformation. In contrast, 3D laser scanning technology (LiDAR) provides new opportunities for tunnel deformation monitoring. A key technical challenge is the automatic extraction of the shotcrete lining from multi-temporal point clouds of tunnel construction scenes and the accurate quantification of its deformation. This study proposes a deformation monitoring framework based on terrestrial laser scanning (TLS) and a cloth simulation filtering (CSF) algorithm. Multi-temporal point cloud data of tunnel construction scenes were acquired using a self-developed TLS system. The CSF algorithm was then employed to accurately extract regions of interest from complex tunnel point clouds, achieving a segmentation accuracy of 96.12%. A deformation monitoring algorithm called GDef was developed to compute the 3D deformation of shotcrete lining, with an error of less than 2.5 mm. The proposed method is validated through simulation experiments and field tests in a triple-lane tunnel under construction, demonstrating its capability to detect lining deformation within millimeter-level accuracy. This provides a practical and efficient solution for high-precision deformation monitoring under complex tunnel construction conditions.

Seepage failure of levees can cause landslides and other hazards. Three-dimensional (3D) laser-scanning technology has become a new method for collecting levee hazard data. In this study, the 3D characteristics of landslide point clouds were investigated through systematic indoor model tests, and a feature recognition algorithm applicable to levee landslides was proposed. The major outcomes of this study are as follows: 1) the formulation of an adaptive random sampling boundary extraction (A-R-B) algorithm, which integrates random sample consensus plane segmentation, adaptive distance threshold calculation, and boundary extraction for levee landslide disaster recognition; 2) through feasibility analyses and accuracy tests of the A-R-B algorithm, this study demonstrated the capacity of the proposed method to accurately recognise the features of levee landslides, with a relative accuracy of 1 cm and an absolute accuracy of 3.5 cm in the extraction process; 3) the testing of the A-R-B algorithm and optimal parameters for the recognised levee landslide features using the point clouds obtained from laboratory models.

High rock temperature (HRT) and its associated thermal hazards, alongside secondary mechanical risks such as swelling pressures induced in clay layers, pose severe threats to the construction safety of deep-buried tunnels. This study aims to quantitatively reveal the evolution laws of the surrounding rock temperature field under varying heat source conditions. A combined approach of physical model testing and numerical analysis was adopted. Utilizing an independently developed test system with a 1:13 geometric similarity ratio, the coupled rock-heat-ventilation environment was simulated. A transient conduction-convection 3D numerical model was established in COMSOL and verified against experimental data under benchmark conditions. The research confirms that under the influence of localized block heat sources, the temperature field in the far-field region follows a significant linear attenuation law rather than the traditional exponential distribution, with a prototype-equivalent gradient of approximately 0.69 °C/m. Furthermore, the study quantitatively identifies 8 m³ as the critical volume for heat source geometric saturation, beyond which the incremental temperature rise efficiency decreases by 25%. It is further revealed that the effective cooling depth of conventional ventilation is only approximately 0.35 m, indicating a significant “ventilation shielding effect” within the deep surrounding rock.

The advancement of finite element technology has facilitated the numerical simulation of interfacial behavior in self-healing concrete. Although the cohesive zone model (CZM) has been introduced in previous studies, most rely on empirically assigned parameters and lack systematic sensitivity analyses of the cohesive properties at the microcapsule-cement interface. This study focuses on self-developed polyvinyl alcohol - shelled microcapsules, which were integrated into a three-dimensional mesoscale finite element model using cohesive zone theory and a random packing algorithm to simulate uniaxial compression. For self-healing concrete containing 10% microcapsules (2.36–3.20 mm), cohesive parameters were calibrated through orthogonal testing and identified normal and tangential traction as the dominant factors affecting mechanical performance. Subsequent single-factor analysis identified 1.5 MPa (normal) and 6 MPa (tangential) traction as the optimal parameters, leading to a 21.7% increase in cracking load and a 14.2% reduction in crack area relative to the initially calibrated values. Additional analysis showed that smaller capsules formed thicker shells under identical spraying conditions, enhancing bonding with the cement matrix. Combined simulation and experimental results confirmed that both particle size and dosage negatively correlate with compressive strength. Optimal performance was achieved at 5%–10% dosage with 1.6–2.36 mm capsules, whereas excessive dosage (>30%) severely degraded structural integrity. These findings provide valuable insights for the modeling and design of self-healing concrete.

With advantages in robustness to variations in ambient illumination and active sensing, light detection and ranging has become a primary modality for environmental perception in underground coal mines characterized by inadequate illumination and high dust concentrations. However, raw point clouds lack semantic attributes, which limits their direct application in scene understanding for digital mines. Although fully supervised deep learning methods for point cloud semantic segmentation achieve high performance, they rely heavily on labor-intensive point-wise annotations, incurring high costs and potential human error. To address this, this paper proposes a weakly supervised semantic segmentation framework specifically designed for coal mine roadways. The framework establishes a comprehensive workflow spanning from data acquisition to sparse-supervised training. By leveraging sparse labels to drive model optimization, the method maintains segmentation performance while significantly reducing annotation costs, achieving a trade-off between engineering precision and deployment overhead. Experimental results demonstrate that using only 0.1% sparse annotations, the proposed method achieves an overall accuracy of 91.64% and a mean intersection over union of 78.83%. Compared to fully supervised models, it reduces the annotation time by a factor of 2–3 times improvement in annotation efficiency with a precision loss of approximately 7%. The segmented structural point clouds are utilized for roadway deformation monitoring and parametric geometric reconstruction. This supports a comprehensive digital twin framework that enables high-fidelity modeling and early warning of centimeter-scale deformation. This research confirms that the proposed technology provides a cost-effective and efficient path for scene understanding in intelligent mines, demonstrating practical value and potential for practical deployment.