Scholarlink

<p>To address the dual challenges of low-precision tunnel leakage detection and inadequate repair material performance, this study integrates deep learning with materials science to establish an intelligent closed-loop framework of "detection-repair-monitoring." An enhanced RT-DETR detection model incorporating multi-scale edge perception and adaptive enhancement modules achieves precise leakage localization, demonstrating superior performance metrics (Precision: 85.52%, mAP: 76.99%) on a validation dataset of 4,646 leakage images and enabling objective-level defect targeting. Leveraging detection results, a cement-based crystalline material (Material X) has been engineered to seal leaks through catalytic hydration mechanisms, generating dendritic CaCO₃/C–S–H gels that preferentially precipitate within microcracks and capillary pores for effective sealing. Within the optimized mass fraction range of 1.0–1.5%, Material X demonstrates remarkable performance enhancements: impermeability pressure increases by 375%, 28-day compressive strength improves by 42%, post-freeze–thaw crack propagation reduces by 68%, and repaired structures exhibit 31% higher strength compared to original undamaged concrete. Critically, Material X retains chemical reactivity in saturated aqueous environments, fundamentally overcoming the limitation of conventional repair materials that lose efficacy in continuously wet tunnel conditions. This interdisciplinary synergy creates a sophisticated closed-loop operation system that transforms tunnel maintenance from passive emergency repair to proactive, data-driven lifecycle management, providing a comprehensive solution for systematic leakage control throughout infrastructure service lifecycles.</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

Synergistic application of deep learning detection and permeable crystalline material for tunnel water leakage management

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Shaowei Wang,
  • Yu Qi,
  • Jiangbo Xu,
  • Xiong Wu,
  • Ke-ke Shen,
  • Xi Liu,
  • Zhifeng Wang,
  • Yalin Nan,
  • Nengyuan Chen,
  • Jisheng Li

摘要

To address the dual challenges of low-precision tunnel leakage detection and inadequate repair material performance, this study integrates deep learning with materials science to establish an intelligent closed-loop framework of "detection-repair-monitoring." An enhanced RT-DETR detection model incorporating multi-scale edge perception and adaptive enhancement modules achieves precise leakage localization, demonstrating superior performance metrics (Precision: 85.52%, mAP: 76.99%) on a validation dataset of 4,646 leakage images and enabling objective-level defect targeting. Leveraging detection results, a cement-based crystalline material (Material X) has been engineered to seal leaks through catalytic hydration mechanisms, generating dendritic CaCO₃/C–S–H gels that preferentially precipitate within microcracks and capillary pores for effective sealing. Within the optimized mass fraction range of 1.0–1.5%, Material X demonstrates remarkable performance enhancements: impermeability pressure increases by 375%, 28-day compressive strength improves by 42%, post-freeze–thaw crack propagation reduces by 68%, and repaired structures exhibit 31% higher strength compared to original undamaged concrete. Critically, Material X retains chemical reactivity in saturated aqueous environments, fundamentally overcoming the limitation of conventional repair materials that lose efficacy in continuously wet tunnel conditions. This interdisciplinary synergy creates a sophisticated closed-loop operation system that transforms tunnel maintenance from passive emergency repair to proactive, data-driven lifecycle management, providing a comprehensive solution for systematic leakage control throughout infrastructure service lifecycles.