On-line reliability prediction of bridges based on Gaussian particle filter

Abstract:

To dynamically predict reliability of bridge members with real-time monitored information, with the long-term mass monitored data of health monitoring system, the data-based dynamic model including monitoring equation and state equation was built, and then the mixed Gaussian particle filter(MGPF) was introduced. With particle filter method, Bayesian method and dynamic model, the posteriori distribution parameters of state variable and one-step forward prediction distribution parameters of monitored data were predicted. Through resampling technique, with MGPF, the prediction precision of dynamic model can be increased. Based on the real-time monitoring data, the weights of resampled particles can be constantly updated. Therefore, the problem of particle degradation is solved. Finally based on the real-time predicted distribution parameters, with the first order second moment (FOSM) method, the on-line and dynamic reliability of bridge members is predicted.