Scholarlink

<p>The mechanical properties of shale rocks are essential for effective extraction of unconventional shale gas and oil. Digital rock imaging through scanning electron microscopy (SEM) plays a pivotal role in characterizing these properties. However, accurately segmenting clay from grain minerals in SEM images is challenging due to overlapping grayscale values. This study introduces a workflow that employs a deep learning algorithm for SEM image segmentation, coupled with finite element method (FEM) simulations to model shale elastic modulus. The accuracy of these simulations is validated against meso-scale laboratory microindentation tests. The deep learning algorithm, U-Net, was utilized to effectively segment shale SEM images into four phases including mineral grains, clays, organic matter, and pyrite, achieving a mean accuracy of 0.91 and an Intersection over Union (IoU) of 0.73. The model demonstrated robust performance in segmenting unseen images, particularly excelling in organic matter, followed by pyrite, mineral grains, and clay. Finite element method (FEM) simulations using the 2D lattice approach estimated Young’s modulus of shale at 43.7 GPa, contrasting with the 38.2 GPa observed in micro-scale microindentation tests. Discrepancies between these simulation and experimental outcomes were thoroughly analyzed. This research, integrating deep learning with 2D finite element method (FEM) simulations, offers a novel approach for directly modeling the mechanical behavior of heterogeneous shales from micro-scale SEM imaging, and provides insights into the microstructure-mechanical relationships in rock physics.</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

Texture-Based Segmentation of SEM Images of Shale Rocks and Estimation of Meso-Scale Elastic Modulus by 2D FEM

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Chunxiao Li,
  • Yansong Liu,
  • Longsheng Li,
  • Zihang Wang,
  • Heng Li

摘要

The mechanical properties of shale rocks are essential for effective extraction of unconventional shale gas and oil. Digital rock imaging through scanning electron microscopy (SEM) plays a pivotal role in characterizing these properties. However, accurately segmenting clay from grain minerals in SEM images is challenging due to overlapping grayscale values. This study introduces a workflow that employs a deep learning algorithm for SEM image segmentation, coupled with finite element method (FEM) simulations to model shale elastic modulus. The accuracy of these simulations is validated against meso-scale laboratory microindentation tests. The deep learning algorithm, U-Net, was utilized to effectively segment shale SEM images into four phases including mineral grains, clays, organic matter, and pyrite, achieving a mean accuracy of 0.91 and an Intersection over Union (IoU) of 0.73. The model demonstrated robust performance in segmenting unseen images, particularly excelling in organic matter, followed by pyrite, mineral grains, and clay. Finite element method (FEM) simulations using the 2D lattice approach estimated Young’s modulus of shale at 43.7 GPa, contrasting with the 38.2 GPa observed in micro-scale microindentation tests. Discrepancies between these simulation and experimental outcomes were thoroughly analyzed. This research, integrating deep learning with 2D finite element method (FEM) simulations, offers a novel approach for directly modeling the mechanical behavior of heterogeneous shales from micro-scale SEM imaging, and provides insights into the microstructure-mechanical relationships in rock physics.