Scholarlink

<p>Addressing the lack of robust early-warning indicators for jointed rock mass instability, this study develops an integrated multi-physics monitoring system—combining acoustic emission (AE), infrared radiation (IR), and digital image correlation (DIC)—to characterize signal evolution during shear failure. Quantitative analysis confirms that AE signals exhibit superior consistency with underlying failure mechanisms compared to IR and DIC. On this basis, a novel deep learning framework is proposed for the multi-step temporal prediction of AE signals. Unlike traditional single-parameter methods, this model leverages a multi-head attention mechanism to fuse heterogeneous features from multi-source monitoring data. Furthermore, an automated failure stage identification method based on wavelet basis functions is incorporated. The resulting predictive framework, comprising signal monitoring, trend prediction, and stage identification modules, demonstrates significant superiority in warning accuracy. This research provides a practical, intelligent solution for risk control in geotechnical engineering.</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

Deep Learning Framework for Early Warning of Shear Failure in Jointed Rock Masses

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Zhigang Yao,
  • Yong Fang,
  • Luo Hu,
  • Jianfeng Wang,
  • Kaicheng Ying,
  • Yapeng Fu,
  • Jian Cui

摘要

Addressing the lack of robust early-warning indicators for jointed rock mass instability, this study develops an integrated multi-physics monitoring system—combining acoustic emission (AE), infrared radiation (IR), and digital image correlation (DIC)—to characterize signal evolution during shear failure. Quantitative analysis confirms that AE signals exhibit superior consistency with underlying failure mechanisms compared to IR and DIC. On this basis, a novel deep learning framework is proposed for the multi-step temporal prediction of AE signals. Unlike traditional single-parameter methods, this model leverages a multi-head attention mechanism to fuse heterogeneous features from multi-source monitoring data. Furthermore, an automated failure stage identification method based on wavelet basis functions is incorporated. The resulting predictive framework, comprising signal monitoring, trend prediction, and stage identification modules, demonstrates significant superiority in warning accuracy. This research provides a practical, intelligent solution for risk control in geotechnical engineering.