SHM-Oriented Hybrid Modeling for Stress Analysis of Steel Girder Bridge

Abstract

Stress monitoring is always a challenging task in bridge structural health monitoring (SHM) since the measured pointwise stress is not enough for fully reflecting structural conditions. Therefore, a novel hybrid modeling technique was proposed in this paper, which calculates stress distribution with the aid of finite-element (FE) submodels from limited measured data. However, unlike common FE analyses, the requirement of complete input information is avoided by an FE model-based partial least-squares regression (FEM-PLSR) method. First, the regression equations among the FE model, unknown structural input, and output were set up, into which measured displacements, rotations, and strains were fused simultaneously. By solving the regression equations, the boundary conditions of the FE submodel can be precisely estimated. Then, the stress distribution can be calculated through FE analyses under the assumption of known local vehicular loads. Numerical simulations of a continuous steel box girder bridge were carried out for verification. Corresponding results indicated that the accuracy of the calculated stress distributions was competitive to the widely used multiscale FE model, even if the structural input information outside the submodel was not directly measured. Furthermore, the proposed method was also proved insensitive to random measurement errors. Finally, a large-scale experiment was designed to validate the accuracy of the hybrid model under three loading conditions. The strain distribution over both space and time accorded well with the measurement. Thus, the present hybrid modeling offered a novel way to obtain unmeasured structural responses, revealing great potential in the field of bridge SHM.