ABSTRACT
This study focuses on the system identification and damage detection of reinforced concrete bridges. The well-known autoregressive exogenous (ARX) algorithm is used to find a data set of autoregressive coefficients. Based on this data set, the eigenvalue realization algorithm (ERA) is used for the modal identification of the structures. ARX-ERA are used because the experiments includes the acquisition of output data, i.e. acceleration, based on a forced input, which is produced by an exciter. This provides more accurate modal results than output only methods. The innovation of this study is the use of these results in a new neural network framework that identifies the possible damaged elements and the type of damage. This is achieved using an improved version of the substructural identification method. Instead of a single submatrix scaling factor the new method further subdivides the element stiffness matrix in four additional submatrices according the force type: KN axial force submatrix, KMy,Vz, KMz,Vy, KT; each with its own submatrix scaling factor. The method is validated with experiments on different reinforced concrete bridges. The computational effort required is mitigated by the used of parallel computing on GPUs.
