Changdong Li

Reliable extraction of surface-visible rock blocks from 3 dimensions (3D) point clouds is hindered by the combinatorial burden of global plane intersections and by weak traceability between reconstructed entities and measured points. This study proposes an edge-priority geometric-topology decoupling algorithm. First, robust normal estimation and local geometric descriptors partition the point cloud into planar and edge candidates. Second, edge candidates are classified as ridges or valleys, and ridge points are clustered into 3D topological skeletons that encode observed boundary connectivity. Third, the edge-removed planar candidates are segmented into finite discontinuity patches. Skeleton-constrained geometric association then assembles compatible patches into manifold-like surface-visible block complexes, whose topology is checked using a boundary-aware Euler relation. Finally, directional voxel filling estimates irregular-block volumes. Tests on the RockBench slope and an additional field slope demonstrate that the workflow retains point-to-trace-to-discontinuity-to-block correspondence while avoiding exhaustive global plane-intersection enumeration. The method reconstructs visible block entities conditioned by point-cloud coverage and observed edge connectivity.

Linear rock discontinuities play an important role in controlling the structural integrity, deformation behavior, and stability of rock masses. However, their automatic identification and quantitative characterization from three-dimensional point cloud data remain challenging because of fragmentation, noise interference, and complex spatial distributions. To address these issues, an automated framework for the identification and geometric information extraction of linear rock discontinuities from terrestrial laser scanning (TLS) point clouds is proposed. High-resolution point cloud data were collected from a rock slope at the Longtangou Pumped Storage Power Station, Tianjin, China. An improved GoogLeNet model was developed for automatic identification of linear discontinuities. Statistical filtering, skeleton extraction, and geometric refinement were subsequently applied to improve structural continuity. A semi-automatic segmentation approach combined with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and principal component analysis (PCA) was employed to extract the geometric information of individual discontinuities, including trace length, trend, and plunge. The results show that the proposed model achieves a classification accuracy exceeding 95%. A total of 460 independent linear discontinuities were identified and quantitatively characterized in the study area. Comparison with manual measurements indicates that most trace length errors are within 5%, whereas vector angular errors range from 0.15° to 5.89°. Orientation analysis reveals that high-angle and near-vertical discontinuities account for more than 60% of the total population and exhibit a dominant northwest-trending orientation. The proposed framework provides an efficient and reliable approach for automatic identification and geometric information extraction of linear rock discontinuities in rock engineering applications.

Automatic rock discontinuity extraction based on three-dimensional (3D) point clouds represents a crucial approach to intelligent geological mapping in modern engineering. Although existing ’planar feature-driven’ methods are supported by rigorous spatial geometric theories and demonstrate robust feasibility and reconstruction accuracy, their identification efficiency in complex engineering scenarios remains highly constrained. Specifically, these methods rely heavily on the precise computation of pointwise normal vectors and 3D curvatures, which leads to massive computational overhead and extreme noise sensitivity when handling large-scale point clouds. To circumvent the bottleneck of traditional normal vector dependency, this study introduces an efficient instance segmentation method for individual discontinuity without requiring the computation of point normal vectors and any other 3D features, termed the Voxel-based Edge-prioritized Differential Analysis (VEDA). This framework pioneeringly utilizes 3D spatial homogeneous transformations and an orthogonal voxel architecture to map unorganized point clouds into a 2.5-dimensional (2.5D) raster elevation matrix through dimensionality reduction. Subsequently, locally weighted scatterplot smoothing (LOWESS) is introduced for profile fitting, and an adaptive residual threshold (T_res=μ+1σ) is constructed to enable the highly sensitive screening of abrupt morphological changes. Ultimately, the complete reconstruction of individual discontinuities is achieved based on the 26-connectivity topological criterion. Cross-validation was conducted using an open-source benchmark dataset and a surveyed highway slope. Experimental results demonstrate that, under the optimal voxel resolution, VEDA successfully circumvents the computational costs associated with nearest-neighbor searches and singular value decomposition. The overall computational efficiency is improved by more than three times compared to traditional clustering algorithms (with a reduction in running time of approximately 69.28% from 103.01s to 31.64s). Regarding orientation extraction, the maximum errors for both dip direction and dip angle are strictly controlled within 8°. In individual-discontinuity instance segmentation tests under real, complex rock mass scenarios, the proposed method achieved the highest Precision, Recall, F1"-" score and Intersection over Union(IoU) among four methods, which comprehensively outperforms that of existing mainstream ’planar feature-driven’ algorithms. This study provides a highly robust and novel theoretical framework for the fully automatic, high-precision 3D digital investigation of complex rock masses.

The strength of the glacier–bed interface plays a critical control over glacier dynamics. Previous studies have examined the effects of factors such as temperature and interfacial roughness on the mechanical behavior of the ice–rock interface, whereas the impact of bedrock properties remains relatively underexplored. Using a centrifugal apparatus, we measured the adhesion strength of ice on bedrocks with varying properties. Rock porosity and surface wettability were identified as the primary controlling factors. Experimental results show that variations in bedrock porosity and wettability can lead to maximum differences in the ice adhesion strength of up to 1.5‐fold and 1.25‐fold, respectively. Micro‐rhizoidal ice structures observed via cryo‐electron microscopy at the ice–rock interface suggest an “ice‐root effect” as the underlying mechanism responsible for the enhanced ice adhesion on high‐porosity rock substrates. Analyses demonstrate that the ice‐root effect is governed by the contact area between ice roots and pore walls. On this basis, we proposed a parameter termed the ice‐root contact area coefficient, f _{A , ir} , to quantitatively characterize the ice‐root effect. Experimental data further reveal a strong linear correlation between ice adhesion and f _{A , ir} . Regarding rock wettability, a strong linear correlation was observed between ice adhesion and the static water contact angle (WAC) θ ₀ of bedrock, indicating that θ ₀ can be regarded as the optimal parameter for describing the wettability effect. This study provides a quantitative framework for understanding and evaluating the influence of bedrock properties on the mechanical behavior of the ice–rock interface.

In cold mountain regions, climate warming is driving the melting of ice‐filled fractures, significantly influencing local hydrological cycles and geological activities. However, thermal‐hydraulic (TH) coupling responses within these fractures during warming remain poorly understood. To address this, we conducted numerous visualized experiments of TH processes in ice‐filled fractures with diverse geometries and flow conditions. A novel volume‐based phase segmentation method was developed to track flow channel dynamics during ice melting. The results show that ice melting dynamically alters fracture permeability, which evolves through two distinct patterns: increase‐stabilization and increase‐decrease‐stabilization. We established constitutive relations between apparent permeability and ice content, identifying two characteristic evolutionary modes during ice melting: an early‐response type (deceleration followed by acceleration) and a late‐response type (acceleration followed by deceleration). Furthermore, we uncovered how fracture transmissivity is controlled by flow channel geometry during ice melting. Channel width evolution moderates transmissivity in a sub‐linear manner, while aperture changes enhance it following a super‐cubic scaling law. Finally, the study summarizes the intertwined and temporally asynchronous nature of TH coupling in ice‐filled fractures. These insights contribute to a deeper understanding of groundwater flow in frozen fractured aquifers and have important implications for water resource management in warming cold regions.