Scholarlink

<p>This study investigates the efficacy of Recurrent Neural Networks (RNN) and deep learning models in real-time structural health monitoring (SHM) of large infrastructure. Using 33&#xa0;years of historical displacement data from a concrete arch dam, we developed predictive models to capture structural displacements at daily, bi-monthly, and monthly intervals. The daily prediction model achieved remarkable accuracy, with a Mean Absolute Error (MAE) of 0.25&#xa0;mm and a Root Mean Square Error (RMSE) of 0.65&#xa0;mm, making it highly effective for short-term monitoring and immediate issue detection. The bi-monthly model, shows moderate accuracy with an MAE of 4.43&#xa0;mm and RMSE of 5.98&#xa0;mm, offer a practical solution for medium-term trend analysis and resource-efficient monitoring. Meanwhile, the monthly model provides broader trend analysis despite lower accuracy, with an MAE of 7.53&#xa0;mm and RMSE of 10.00&#xa0;mm. These findings underscore the significance of real-time SHM in ensuring the safety and reliability of critical infrastructure and demonstrate the potential of RNNs and deep learning techniques in advancing SHM practices.</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

Structural health monitoring of arch dams with deep learning: a comparative study of recurrent neural networks in daily, bi-monthly, and monthly predictions

  • 期刊论文

  • 发表日期：

    2025年4月13日

查看全文
  • Kiarash Baharan,
  • Hassan Mirzabozorg,
  • Amir Masoud Babadi

摘要

This study investigates the efficacy of Recurrent Neural Networks (RNN) and deep learning models in real-time structural health monitoring (SHM) of large infrastructure. Using 33 years of historical displacement data from a concrete arch dam, we developed predictive models to capture structural displacements at daily, bi-monthly, and monthly intervals. The daily prediction model achieved remarkable accuracy, with a Mean Absolute Error (MAE) of 0.25 mm and a Root Mean Square Error (RMSE) of 0.65 mm, making it highly effective for short-term monitoring and immediate issue detection. The bi-monthly model, shows moderate accuracy with an MAE of 4.43 mm and RMSE of 5.98 mm, offer a practical solution for medium-term trend analysis and resource-efficient monitoring. Meanwhile, the monthly model provides broader trend analysis despite lower accuracy, with an MAE of 7.53 mm and RMSE of 10.00 mm. These findings underscore the significance of real-time SHM in ensuring the safety and reliability of critical infrastructure and demonstrate the potential of RNNs and deep learning techniques in advancing SHM practices.