Deflection prediction model for bridge static load test based on artificial neural networks

Bridge load test is crucial for ensuring the safety of vehicles, but traditional methods are often time-consuming, labour-intensive, and expensive. Therefore, it is imperative to propose a method that enables rapid and accurate detection and assessment of bridge conditions. In this study, we focus on the half-through steel-concrete arch bridge of the Daning River Extra Large Bridge and present a smart prediction model based on an artificial neural network (ANN). Our model combines historical deflection data obtained from static load tests on existing bridges with finite element model. Additionally, it incorporates the Attention mechanism and Skip Connections. The research findings reveal that the Tanh-type activation function provides the best fit, with a maximum absolute error of 0.144 mm. By integrating the Attention mechanism and Skip Connections, the model achieves a maximum absolute error of 0.521 mm, with 75% of predicted values having an error smaller than 0.103 mm and 95% of predicted values having an error smaller than 0.196 mm. Field testing further validates the accuracy and feasibility of the proposed bridge deflection prediction model. The results of this study have significant implications in terms of reducing the time and cost associated with health detection for high-speed railway bridges.