Comparisons of prediction models for time-dependent reliability forecasting of bridge structures based on SHM data

As a promising way to substitute the traditional load test, structural health monitoring of bridges has received wide concern and has been put in many applications. During the long- term monitoring period, the health monitoring systems of bridges have produced a huge amount of monitoring data containing structural safety feature information for every day, every month and even every year. It becomes an important and challenging problem that how to predict time- dependent reliability of bridge structures based on such a big amount of monitoring data. In this paper, to incorporate both historical monitoring data and real- time monitoring data in the prediction of time- dependent reliability, four predicting models for time- dependent reliability forecasting are investigated, which include a grey model ( GM), an Auto- Regressive Moving- Average (ARMA) model, a polynomial function model, and a Bayesian dynamic linear model(BDLM). By analyzing a practical bridge as a case study, the computational accuracy and shortcomings of the four models are compared carefully. Based on the identified best prediction model and real- time monitoring data, the reliability indices of bridge structures are solved and predicted real- timely with the first order and second moment (FOSM) method. The safety- based time- dependent reliability of the actual bridge is predicted to illustrate the application and feasibility of the chosen best prediction model in this paper.