An accurate and low-cost vehicle-induced deflection prediction framework for long-span bridges using deep learning and monitoring data

Abstract

Accurate vehicle-induced deflection prediction is critical to bridge structural management and maintenance. However, most existing studies use finite element models and weigh-in-motion (WIM) system to predict the deflection of bridges, the cost of building finite element models is expensive. An accurate and low-cost vehicle-induced deflection prediction framework is developed in this study. In this framework, the calculation method of the spatio-temporal distribution matrix of the vehicle loads is proposed. A fusion attention mechanism bidirectional long short-term memory (ATT-BiLSTM) model is used to predict vehicle-induced deflection. Taking a suspension bridge in China as an example, the accuracy of short- and long-term vehicle-induced deflection prediction is tested, and different deep learning models are used for comparison. The results show that the proposed framework can predict short-term vehicle-induced deflection with a mean absolute error (MAE) of less than 2 mm, which can also accurately predict the extrema of vehicle-induced deflection within a certain time window. The proposed framework is independent of finite element models and lowers the cost of bridge deflection prediction.