Finite element model updating considering boundary conditions using neural networks

Abstract

A novel technique to evaluate the bridge boundary condition using neural networks is proposed. It can be used to establish a more accurate finite element (FE) model considering the behaviors of boundary conditions. In the proposed method, the aging and constraining effect of the boundary condition is represented by an artificial rotational spring at each support. A relationship between the responses of the bridge and the rotational spring constant is analytically investigated. This relationship can be used to estimate the rotational spring constant of the bridge using neural networks. The proposed method was verified through laboratory tests and field tests on a steel girder bridge. The proposed method can estimate the bridge boundary conditions directly from the actual behaviors of bridge supports, and this can effectively reduce the uncertainty of boundary conditions in FE model updating.