Improved mode shape expansion method for cable-stayed bridge using modal approach and artificial neural network

Abstract

The Structure Equivalent Reduction-Expansion Process (SEREP), which has been widely used to expand experimental mode shapes, has the limitation of low accuracy of expansion for experimental mode shapes that are poorly correlated with finite element (FE) mode shapes. To address this limitation, a novel mode shape expansion method using modal approach and artificial neural network (ANN) is proposed in this paper. The ANN replaced the least-squares method to optimize the modal coordinates and considered the natural frequency and experimental mode shape of the master DOFs as input data. The superiority of the proposed ANN method compared with the SEREP was verified using a numerical cable-stayed bridge model. The proposed method, which can use a large number of FE mode shapes and optimize modal coordinates based on the ANN, achieved high accuracy (modal assurance criterion > 0.9 and normalized mean absolute percent error < 5 %) in expanding experimental mode shapes that have poor correlation. In addition, using the proposed method, the number of required experimental data can be reduced, and additional processes such as optimal selection of FE mode shapes and FE model modification can be omitted.