Scholarlink

<p>This study presents a vision-based approach for assessing the vibration serviceability and dynamic behaviour of a lightweight pedestrian bridge subjected to various human-induced loads. A non-target camera system is employed to capture structural responses under ambient conditions, walking, running, jumping, and cycling. A comprehensive data cleansing strategy, including outlier removal and signal alignment, is implemented to ensure data quality. The displacement data obtained are used to evaluate comfort indices in accordance with ISO 2631-1:1997, providing quantitative insights into occupant experience. Furthermore, a Bayesian-optimised bidirectional long short-term memory (BiLSTM) neural network is developed to predict comfort indices from time-series input data. The proposed deep learning framework demonstrates strong generalisation capability, effectively capturing the dynamic characteristics across multiple bridge regions. The integration of Bayesian hyperparameter optimisation enhances model reliability and prediction accuracy. Overall, the study highlights the potential of vision-based sensing and advanced machine learning techniques in structural health monitoring and real-time serviceability assessment of pedestrian bridges.</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

Vision-aided vibration analysis and comfort evaluation of a pedestrian bridge via Bayesian-optimised BiLSTM neural network

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Wai Kei Ao,
  • Yuhang Lu,
  • Miaomin Wang,
  • James Brownjohn

摘要

This study presents a vision-based approach for assessing the vibration serviceability and dynamic behaviour of a lightweight pedestrian bridge subjected to various human-induced loads. A non-target camera system is employed to capture structural responses under ambient conditions, walking, running, jumping, and cycling. A comprehensive data cleansing strategy, including outlier removal and signal alignment, is implemented to ensure data quality. The displacement data obtained are used to evaluate comfort indices in accordance with ISO 2631-1:1997, providing quantitative insights into occupant experience. Furthermore, a Bayesian-optimised bidirectional long short-term memory (BiLSTM) neural network is developed to predict comfort indices from time-series input data. The proposed deep learning framework demonstrates strong generalisation capability, effectively capturing the dynamic characteristics across multiple bridge regions. The integration of Bayesian hyperparameter optimisation enhances model reliability and prediction accuracy. Overall, the study highlights the potential of vision-based sensing and advanced machine learning techniques in structural health monitoring and real-time serviceability assessment of pedestrian bridges.