Time-dependent reliability prediction of bridge components based on BO-LSTM and Gaussian-Copula theory

Abstract In bridge health monitoring systems, there exists the dynamic nonlinear dependency of failure among the multiple monitoring points of bridge components. To investigate the impacts of this dynamic nonlinear dependency on the time-varying reliability indices of bridges, this study focuses on relevant research.It proposes a time-varying reliability prediction method for bridge components based on the monitoring data, a Bayesian optimization long short-term memory network (BO-LSTM) and Gaussian-Copula theory. The specific research contents are as follows: (1) Bridge monitoring data are first subjected to filtering, and based on the filtered monitoring signals, a BO-LSTM model is established to dynamically predict the extreme stresses of the existing bridge; (2) A trivariate Gaussian Copula model, which accounts for the dynamic nonlinear dependency of failure among three control monitoring points, is constructed to characterize the nonlinear dependencies among the monitoring data; (3) the validity of the proposed model and method is validated using monitoring data from the Fumin Bridge in Tianjin City of China. The above research results will provide the theoretical foundation and application for bridge reliability prediction and assessment.