Time Domain Substructural Post-earthquake Damage Diagnosis Methodology with Neural Networks

Abstract

An emulator neural network (ENN) and a parametric evaluation neural network (PENN) are constructed to facilitate a substructural parametric identification process for post-earthquake damage diagnosis of civil structures by the direct use of dynamic response measurements under base excitations. The rationality of the proposed methodology is explained, and the theory basis for the construction of two neural networks is described according to the discrete time solution of the state space equation of a substructure. An evaluation index called root-mean-square prediction difference vector (RMSPDV) is presented to evaluate the condition of a object substructure. Based on the trained ENN and PENN, the inter-storey stiffness parameters of the object substructure are identified with enough accuracy. The sensibility and the performance of the proposed methodology under different base excitations are examined using a multi-storey shear building structure by numerical simulations.