Scholarlink

<p>With the widespread application of bridge health monitoring systems in transportation infrastructure, how to accurately identify abnormal states from high-dimensional, nonlinear, and non-stationary monitoring data has become a key issue in current bridge operation. Aiming at the shortcomings of traditional methods in bridge data feature extraction and modal decomposition, the research proposes an intelligent detection framework that integrates improved variational modal decomposition and convolutional autoencoder neural network. By improving the variational mode decomposition algorithm through adaptive parameter optimization strategy and utilizing the multi-scale spatiotemporal feature automatic learning ability of convolutional autoencoder neural network, one-dimensional bridge data is converted into two-dimensional images through Gram angle difference field, enhancing the accuracy of anomaly detection. The experimental findings reveal that the convolutional autoencoder neural network model achieves the highest accuracy of 95.1 and 93.8% in detecting missing data and drift data, respectively. The signal-to-noise ratio is improved to 25.8&#xa0;dB, and the reconstruction error is as low as 0.018. The improved variational mode decomposition method outperforms traditional approaches in regard to root mean square error and correlation coefficient, and the stability of the fusion system is as high as 99.85%. The new method has excellent data decomposition and detection capabilities. The main contribution of the research is the proposal and validation of an intelligent anomaly detection fusion framework for bridge health monitoring. By adaptively optimizing VMD parameters, the proposed framework mitigates mode aliasing and enhances the robustness of signal decomposition in noisy environments. This leads to improved feature extraction and accurate detection of anomalies such as data loss and drift. This has practical significance for improving the ability to detect and analyze anomalies in bridge data.</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

An adaptive VMD–ConvAE framework for robust anomaly detection in bridge monitoring data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Xirui Wang,
  • Fei Huang,
  • Hailing Chen,
  • Xiaoli Xie

摘要

With the widespread application of bridge health monitoring systems in transportation infrastructure, how to accurately identify abnormal states from high-dimensional, nonlinear, and non-stationary monitoring data has become a key issue in current bridge operation. Aiming at the shortcomings of traditional methods in bridge data feature extraction and modal decomposition, the research proposes an intelligent detection framework that integrates improved variational modal decomposition and convolutional autoencoder neural network. By improving the variational mode decomposition algorithm through adaptive parameter optimization strategy and utilizing the multi-scale spatiotemporal feature automatic learning ability of convolutional autoencoder neural network, one-dimensional bridge data is converted into two-dimensional images through Gram angle difference field, enhancing the accuracy of anomaly detection. The experimental findings reveal that the convolutional autoencoder neural network model achieves the highest accuracy of 95.1 and 93.8% in detecting missing data and drift data, respectively. The signal-to-noise ratio is improved to 25.8 dB, and the reconstruction error is as low as 0.018. The improved variational mode decomposition method outperforms traditional approaches in regard to root mean square error and correlation coefficient, and the stability of the fusion system is as high as 99.85%. The new method has excellent data decomposition and detection capabilities. The main contribution of the research is the proposal and validation of an intelligent anomaly detection fusion framework for bridge health monitoring. By adaptively optimizing VMD parameters, the proposed framework mitigates mode aliasing and enhances the robustness of signal decomposition in noisy environments. This leads to improved feature extraction and accurate detection of anomalies such as data loss and drift. This has practical significance for improving the ability to detect and analyze anomalies in bridge data.