Bayesian Dynamic Prediction And Monitoring Of Bridge Dynamic Responses

Bridge health monitoring systems have produced a huge amount of dynamic response data in the long-term service periods,and these data can reflect bridge health status to some extent.Time-variant prediction of the dynamic response is helpful to bridge risk warning and preventive maintenance decision,and meantime provides theoretical foundation for bridge dynamic reliability prediction.Considering the complex characteristics of the dynamic response data,such as tendency,randomness and so forth,how to build the reasonable and effective dynamic response prediction method has become one of the key and urgent problems in the field of bridge health monitoring.In this thesis,the time-variant prediction methods and abnormal monitoring mechanisms of bridge dynamic responses are thoroughly studied with the Bayesian dynamic updating and prediction theory.The detailed contents are as follows.（1）The Bayesian dynamic linear prediction method and the monitoring mechanism of bridge damage factors are proposed.The autoregressive（AR）model is built with bridge dynamic response data,and the AR model coefficients are adopted to construct the damage factors.Based on the initial damage factor data,the single Bayesian dynamic linear model（SBDLM）is built,and the corresponding probability recursion processes are given with Bayes method.The single Bayesian factor,the P_t/P_t-1 index and the cumulative Bayesian factor are applied to monitor the SBDLM,further,dynamic prediction and monitoring analysis of bridge damage factors can be achieved.（2）The Bayesian dynamic coupling linear prediction method and the monitoring mechanism of bridge extreme stresses are provided.Moving average method is used to decouple the coupling extreme stresses of the existing bridges,and dynamic coupling linear models of the decoupled extreme stresses are built,which are transformed into a few dynamic linear models.The dynamic linear models and the corresponding weights are dynamically updated and recursed with Bayes method,further,the dynamic probability recursion of dynamic coupling linear models about the decoupled extreme stresses is made.Meantime,the model is dynamically monitored.The coupling extreme stresses of the existing bridge can be obtained by adding the predicted decoupling extreme stresses,which can achieve the dynamic coupling prediction and monitoring of bridge extreme stresses.（3）The generalized monitoring mechanisms of Bayesian dynamic model are summarized.Model monitoring objectives are analyzed.The existing one-step prediction model and the alternative model are adopted to construct the single Bayesian factor,the P_t/P_t-1 index and the cumulative Bayesian factor.A relatively complete model monitoring processes are provided with the constant threshold value of monitoring indices,model prediction precision and so forth,which can monitor the outliers and model prediction performance.The above research results will provide the new application methods for bridge health monitoring data processing,and provide guidance for bridge preventive maintenance decision-making.