Scholarlink

<p>Bimrock is a heterogeneous geotechnical material composed of rock fragments and a matrix with significantly different strength and stiffness, widely distributed in natural slopes and engineering geological formations. While previous studies have extensively examined the effects of block size and content on bimrock’s macro-mechanical behavior, the influence of block shape and its spatial distribution has received less attention. Quantifying these characteristics, especially through multi-index evaluation, remains challenging. Thus, this study introduces a DEM–machine learning framework to predict the biaxial peak strength of bimrock, considering variations in block shape and distribution. A position-optimized, non-overlapping placement strategy was implemented, increasing convergence from 74 to 98% at 55% block content and reducing average generation time per specimen by 37&#xa0;s at 60% block content. Subsequently, a smoothing factor was introduced to suppress sharp angles, which helped generate more realistic block geometries. Additionally, nine geometric and spatial descriptors, including newly defined shape indicators, were extracted and ranked via machine learning, from which four key indicators were selected for strength prediction. A dataset of 395 numerical samples was generated to train the models. Machine learning models demonstrated strong predictive capability, while micromechanical force chain analysis revealed how block geometry and spatial distribution influence peak strength. Region area variance (RAV) exhibited a negative correlation with peak strength, marginal length variance (MLV) exhibited a positive correlation, and centrifugation rate (CFR) and longest axis slope variance (LAS) showed relatively strong nonlinear relationships.</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

An improved 2D DEM-based method for random block generation in bimrock and peak strength prediction in biaxial tests using machine learning

  • 期刊论文

  • 发表日期：

    2025年11月22日

查看全文
  • Yiming He,
  • Shaorui Sun,
  • Jihong Wei,
  • Huilin Le,
  • Jiaxuan Liu,
  • Haotian Fan,
  • Chunlin Jin,
  • Jiahao Huang

摘要

Bimrock is a heterogeneous geotechnical material composed of rock fragments and a matrix with significantly different strength and stiffness, widely distributed in natural slopes and engineering geological formations. While previous studies have extensively examined the effects of block size and content on bimrock’s macro-mechanical behavior, the influence of block shape and its spatial distribution has received less attention. Quantifying these characteristics, especially through multi-index evaluation, remains challenging. Thus, this study introduces a DEM–machine learning framework to predict the biaxial peak strength of bimrock, considering variations in block shape and distribution. A position-optimized, non-overlapping placement strategy was implemented, increasing convergence from 74 to 98% at 55% block content and reducing average generation time per specimen by 37 s at 60% block content. Subsequently, a smoothing factor was introduced to suppress sharp angles, which helped generate more realistic block geometries. Additionally, nine geometric and spatial descriptors, including newly defined shape indicators, were extracted and ranked via machine learning, from which four key indicators were selected for strength prediction. A dataset of 395 numerical samples was generated to train the models. Machine learning models demonstrated strong predictive capability, while micromechanical force chain analysis revealed how block geometry and spatial distribution influence peak strength. Region area variance (RAV) exhibited a negative correlation with peak strength, marginal length variance (MLV) exhibited a positive correlation, and centrifugation rate (CFR) and longest axis slope variance (LAS) showed relatively strong nonlinear relationships.