Scholarlink

<p>The identification of rock fractures and their evolutionary behaviors under external loading is of fundamental significance for elucidating rock failure mechanisms. A primary difficulty is that fractures are predominantly embedded within the rock mass and therefore cannot be directly observed in situ. To address this, the present study introduces an approach to reconstruct fracture representations from acoustic emission (AE) data. Unsupervised machine learning algorithms are employed to identify individual fractures: the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is first applied to remove noise and locate critical clusters of microcracks, after which the Expectation–Maximization (EM) algorithm is used to separate microcracks within each cluster. Confidence ellipsoids are then computed to represent individual fractures, and fracture intersections are characterized by combining the inferred fracture distributions via an error-ellipsoid method. Furthermore, AE events recorded as discrete points are logically connected to delineate fractures as continuous geometries. Results demonstrate that the developed approach reliably recognizes fractures, with the generated ellipsoids accurately indicating their locations and orientations. This investigation advances understanding of fracture formation and the associated distribution of microcracks, and the proposed approach can be further used to investigate the temporal evolution of fractures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Nucleation Characterization of Rock Fracture Based on Unsupervised Machine Learning

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Xiaoxuan Bai,
  • Chaoyang Zhu,
  • Chunlai Wang

摘要

The identification of rock fractures and their evolutionary behaviors under external loading is of fundamental significance for elucidating rock failure mechanisms. A primary difficulty is that fractures are predominantly embedded within the rock mass and therefore cannot be directly observed in situ. To address this, the present study introduces an approach to reconstruct fracture representations from acoustic emission (AE) data. Unsupervised machine learning algorithms are employed to identify individual fractures: the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is first applied to remove noise and locate critical clusters of microcracks, after which the Expectation–Maximization (EM) algorithm is used to separate microcracks within each cluster. Confidence ellipsoids are then computed to represent individual fractures, and fracture intersections are characterized by combining the inferred fracture distributions via an error-ellipsoid method. Furthermore, AE events recorded as discrete points are logically connected to delineate fractures as continuous geometries. Results demonstrate that the developed approach reliably recognizes fractures, with the generated ellipsoids accurately indicating their locations and orientations. This investigation advances understanding of fracture formation and the associated distribution of microcracks, and the proposed approach can be further used to investigate the temporal evolution of fractures.