Intelligent perception and early warning for deterioration of prestressed concrete box-girder bridges via the variation mining of mapping from vehicle load to structural displacement

Perception and early warning for state deterioration of in-service bridges under complex traffic flows is an important task to ensure the efficient operation of urban transportation lifelines. Variation hiding in the fuzzy mapping from structural input (load) to output (response) can reflect and quantify the state deterioration of bridges. Using a prestressed concrete box-girder bridge as a case study, this paper proposes an intelligent perception and early warning method of state deterioration via the variation mining of digital mapping from vehicle load to structural displacement. First, the load and response signals belonging to each passing vehicle are automatically extracted according to the probabilistic distribution characteristics of the vehicle weight and displacement. Second, Gaussian mixture model (GMM) clustering and Metropolis-Hastings (M-H) sampling are employed to reconstruct load and displacement with different sampling frequencies, enabling precise matching between vehicle load and structural displacement. Third, Bayesian linear regression (BLR) is used to establish a dynamic piecewise linear model of vehicle load-displacement scatters, which reflects time-varying digital mapping between vehicle load and displacement. Finally, a two-level early warning framework using control charts and hypothesis testing is proposed based on the slope variation of the piecewise linear model. The results show that the method can perceive the performance variation of the bridge during service and realize the early warning of state deterioration in different time scales.