Rock bridges are important structures for maintaining rock mass stability, but their shapes are not well known. The researchers propose a method for determining the shape of rock bridges based on experiments, discrete element methods and machine learning, which is applicable to complex joints with arbitrary spatial distribution. Numerical models are constructed using the discrete element method, and parameter matching is performed based on experimental results. The particles were clustered using the k-means algorithm with the maximum principal stress (σ₁) as an indicator and the selection of initial values was optimized. The density-based spatial clustering of applications with noise (DBSCAN) algorithm was used to delete the noise from the particles. Finally, the boundary lines of the particles were extracted by self-programming, and the shape of the rock bridges was determined. Twenty-four sets of simulations were used to analyze the effect of rock bridges on the specimens. The results show that the failure mode of the specimen changes from shear to tensile damage as the cohesive force of the rock bridges increases. The peak strength and peak strain of the specimens increased with the increase of cohesion in the rock bridge. Rock bridges are the fastest growing areas of stress in the specimen.

岩桥是维持岩体稳定性的重要结构，但其形状却鲜为人知。研究人员提出了一种基于实验、离散元方法和机器学习的确定岩桥形状的方法，该方法适用于任意空间分布的复杂节理。采用离散元方法构建数值模型，并根据实验结果进行参数匹配。以最大主应力（σ ₁）为指标，采用k-means算法对颗粒进行聚类，并优化初始值的选择。使用基于密度的噪声应用空间聚类（DBSCAN）算法来删除粒子中的噪声。最后，通过自编程提取颗粒边界线，确定岩桥的形状。使用二十四组模拟来分析岩桥对试件的影响。结果表明，随着岩桥内聚力的增大，试件的破坏模式由剪切破坏转变为拉伸破坏。试件的峰值强度和峰值应变随着岩桥内聚力的增加而增加。岩桥是样本中应力增长最快的区域。