Scholarlink

<p>The microscopic structural damage of rock under the coupled effects of load and cyclic water invasion leads to a weakening of its macroscopic mechanical properties. Based on in-situ CT scans during cyclic water invasion tests on yellow sandstone subjected to loading, a convolutional module is integrated into the decoder part of an improved Swin-UNet model to establish a mapping between porosity damage variables and the degradation of mechanical parameters, thus enabling effective prediction of sandstone performance decline. The findings indicate that: (1) Incorporating convolutional modules into the Swin-UNet decoder allows precise segmentation of sandstone pores (Dice coefficient &gt; 0.74, mIoU &gt; 0.64) and quantitative calculation of porosity. (2) The compressive strength and elastic modulus of sandstone and their degradation rates exhibit a single-exponential declining trend with increasing numbers of cyclic water immersion cycles, reaching 37.35% and 28.86%, respectively, after 10 cycles. (3) A damage variable defined via porosity and modeled through a power function relationship effectively predicts the mechanical performance of sandstone under different cycle counts, with errors less than 9%. Based on deep learning, the dynamic evolution of sandstone porosity was quantified, and a cross-scale mapping relationship between mesoscopic pore damage variables and macroscopic mechanical parameter degradation was established. However, the extension of this method to the field scale still faces uncertainties such as rock mass heterogeneity, scale effects, and complex environmental conditions, which require further.</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

Pore segmentation and mechanical parameter prediction of water-infiltrated loaded sandstone based on improved Swin-UNet

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Ke Wang,
  • Yuxuan Wang,
  • Zhe Qin,
  • Yigui Chen,
  • Jixiao Chen

摘要

The microscopic structural damage of rock under the coupled effects of load and cyclic water invasion leads to a weakening of its macroscopic mechanical properties. Based on in-situ CT scans during cyclic water invasion tests on yellow sandstone subjected to loading, a convolutional module is integrated into the decoder part of an improved Swin-UNet model to establish a mapping between porosity damage variables and the degradation of mechanical parameters, thus enabling effective prediction of sandstone performance decline. The findings indicate that: (1) Incorporating convolutional modules into the Swin-UNet decoder allows precise segmentation of sandstone pores (Dice coefficient > 0.74, mIoU > 0.64) and quantitative calculation of porosity. (2) The compressive strength and elastic modulus of sandstone and their degradation rates exhibit a single-exponential declining trend with increasing numbers of cyclic water immersion cycles, reaching 37.35% and 28.86%, respectively, after 10 cycles. (3) A damage variable defined via porosity and modeled through a power function relationship effectively predicts the mechanical performance of sandstone under different cycle counts, with errors less than 9%. Based on deep learning, the dynamic evolution of sandstone porosity was quantified, and a cross-scale mapping relationship between mesoscopic pore damage variables and macroscopic mechanical parameter degradation was established. However, the extension of this method to the field scale still faces uncertainties such as rock mass heterogeneity, scale effects, and complex environmental conditions, which require further.