Fatigue evaluation of Orthotropic steel deck welds based on WIM data and UD-BP neural network

The vehicle load spectrum for urban bridges in China has yet to be established, and there is an urgent need to develop fatigue assessment methodologies for urban bridges based on this spectrum. Taking an urban bridge in Ningbo, China, as the engineering background, this paper utilizes traffic flow data obtained from a Weigh-In-Motion (WIM) system to categorize vehicle axle numbers into five types. It analyzes the characteristics of axle weight, gross weight, and wheelbase for typical vehicles. Based on the equivalent damage principle and Miner's linear damage criterion, a standard fatigue vehicle load spectrum for this urban bridge is derived. A refined finite element (FE) model of the steel bridge deck is established, and stress analysis is conducted under three typical lateral loading conditions to obtain stress-time history curves for fatigue-prone details. The uniform design (UD) method is employed to design sample points for axle weights of various vehicle types, and the Back-Propagation (BP) neural network method is used to train and predict the relationship between axle weights and fatigue stresses. The variability features of reliability indices for seven typical details of the steel bridge deck under vehicle loads are discussed. The research findings indicate that predicting steel bridge deck fatigue stresses using the UD-BP neural network offers significant computational efficiency advantages compared to FE analysis. Among the seven typical details, the curved cutout of the diaphragm in the steel bridge deck is more prone to fatigue failure under vehicle loads.