Estimation of Bridge Girder Cumulative Displacement for Component Operational Warning Using Bayesian Neural Networks

Abstract

The main girders of suspension bridges experience significant deformation due to temperature variations, wind dynamics, and vehicle loads, causing movement at the girder ends and friction among components such as bearings, expansion joints, and viscous dampers. Early warning of the component anomaly is vital for preventive maintenance. This paper develops a two-stage framework for predicting girder end displacement to facilitate anomaly detection. First, a Bayesian neural network is employed to predict girder end cumulative displacement, accounting for uncertainties inherent in the prediction process. Second, an anomaly detection algorithm utilizing a Mahalanobis distance–based approach is implemented to provide warnings to operations based on both measured and predicted data. The effectiveness of the proposed approach is validated using data collected from multiple loads and displacement responses of a suspension bridge. The analysis reveals that the GEV distribution is highly proficient in capturing the underlying pattern of the cumulative displacement indicator, enabling the establishment of an appropriate threshold. This method proves successful in identifying anomalies in critical components such as viscous dampers, enhancing predictive and preventive maintenance practices and contributing to the longevity and safety of bridge infrastructure.