Temperature-induced response reconstruction for the dynamic reliability assessment of bridge girders

Abstract

Assessing the dynamic reliability of bridge girders under data missing of long-term structural responses still remains a challenge. The conventional methods are often inadequate to analyze the structural dynamic behavior from the incomplete data, resulting in the inability to accurately assess the dynamic reliability of bridge girders. Therefore, this paper presents a new temperature-induced response reconstruction model for the dynamic reliability assessment of bridge girders. Considering that the strong spatial–temporal correlation exists between the air temperature and the structural response, the temperature-induced response reconstruction model, namely Conv-SCINet, is established in this paper by combining the three one-dimensional convolutional layer-based block (Conv-Block) and the sample convolution and interaction network (SCINet) layers to reconstruct the long-term missing structural responses. The corresponding state parameter (i.e., the variance of the normal distribution) of structural responses can be evaluated by Bayesian probability recursion to form the limit state function for the dynamic reliability assessment of in-service bridges. Finally, the validity and feasibility of the proposed method are demonstrated using a prestressed concrete bridge and a steel box-girder suspension bridge. The results show that the Conv-SCINet model has better reconstruction performance of long-term structural responses and higher evaluation accuracy of dynamic reliability for bridge girders compared to the four currently acknowledged high-precision models. In addition, the proposed method also confirms the feasibility of using the air temperature to reconstruct missing structural responses and further evaluating the dynamic reliability of in-service bridges.