Anomaly Detection and Warning in Bridge Structure Health Monitoring: Method Based on Autoencoder

The health monitoring of bridge structures is essential to ensure traffic safety and extend the life of bridges. Anomaly detection and early warning are important tasks for the health monitoring of bridge structures. It can help detect structural defects and potential failures early and take appropriate repair and maintenance measures to avoid serious accidents. How to monitor the health of bridge structures is a thorny issue, and traditional monitoring methods have limitations in detecting complex abnormal situations. Therefore, it is necessary to develop a method based on autoencoders to accurately detect abnormal situations. The method proposed in this article is based on autoencoders, which learn the normal behavior of bridge structures through unsupervised learning to construct an autoencoder network. The autoencoder is trained using monitoring data of normal behavior to learn feature representations of normal behavior. In the anomaly detection stage, new monitoring data is input into the autoencoder. This article calculates the reconstruction error, and then evaluates the degree of anomaly in the monitoring data based on the size of the reconstruction error. The experimental results show that the recall performance range of the proposed method is between 89% -95%, which can provide reliable warning signals.