Abstract
Natural rock mass discontinuities exhibit unique mechanical properties that define weak zones within rock masses, playing a decisive role in the structure, strength, and stability of various rock engineering applications including tunnel support, surrounding rock classification, and slope reinforcement. Consequently, the identification of individual discontinuities and well-developed dominant sets is of paramount importance. This method partitions the automatic identification of dominant discontinuity sets into three sequential steps: point cloud normal vector calculation, individual discontinuity segmentation, and dominant set clustering: 1. Normal vectors are computed using a robust random Hough transform-based approach; 2. An improved region growing algorithm is proposed to segment individual discontinuities, incorporating considerations of curvature, planarity, and roughness in seed point selection and region growing criteria, along with dynamic outlier detection. Furthermore, extreme segmentation scenarios are qualitatively evaluated based on the relationship between thresholds and discontinuity count, while simultaneously screening for optimal threshold ranges; 3. Finally, an S-K-means clustering algorithm is proposed to achieve dominant set clustering. The algorithm's identification accuracy was validated using a rock slope case study, with results demonstrating dip direction and dip angle errors ranging from 0.7° to 2.5°, and mean errors of 1.8° and 1.7°, respectively. This method transforms the conventional approach of directly clustering point clouds to identify dominant sets by first segmenting individual discontinuities prior to clustering, thereby refining the procedure for dominant discontinuity set identification, enhancing computational speed and robustness of discontinuity clustering, and accommodating various discontinuity data types, thus providing a more accurate and efficient methodology for intelligent identification of rock mass discontinuities.
Full Text
Preamble
Title: Application of Improved Regional Growth Algorithm to Identification of Rock Mass Discontinuities
Authors: SIMA Jinsong, XU Qiang, DONG Xiujun, DENG Bo, HE Qiulin, LI Haoliang, LIU Jie, LEI Wenquan
Affiliation: State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu University of Technology, Chengdu 610059
Abstract
Natural rock mass discontinuities possess unique mechanical properties that define weak zones within rock masses, playing a decisive role in the structural integrity, strength, and stability of various rock engineering projects including tunnel support, surrounding rock classification, and slope reinforcement. Consequently, the identification of individual discontinuity planes and well-developed dominant sets is of paramount importance.
This method divides automatic identification of dominant discontinuity sets into three sequential steps: point cloud normal vector calculation, individual discontinuity plane segmentation, and dominant set clustering. First, normal vectors are calculated using a robust random Hough transform method. Second, an improved region growing algorithm segments individual discontinuity planes by incorporating curvature, planarity, and roughness into seed point selection and region growing criteria, supplemented with dynamic outlier detection. Additionally, the relationship between threshold values and discontinuity plane count is utilized to qualitatively assess extreme segmentation scenarios and identify optimal threshold ranges. Third, an S-K-means clustering algorithm is proposed to achieve dominant set clustering.
The algorithm's identification accuracy was validated using a rock slope case study, with results demonstrating dip direction and dip angle errors ranging from 0.7° to 2.5°, and mean errors of 1.8° and 1.7°, respectively. This approach transforms conventional methods by first segmenting individual discontinuity planes before clustering, rather than directly clustering point clouds to identify dominant sets. This refinement enhances computational speed and robustness of discontinuity clustering while accommodating diverse discontinuity data types, thereby providing a more accurate and efficient method for intelligent identification of rock mass discontinuities.
Keywords: Individual Discontinuity Planes; Dominant Discontinuity Sets; Improved Region Growing Algorithm; S-K-means Clustering; Intelligent Identification