Structural damage detection and identification using neural networks

Abstract

A novel methodology is presented for on-line damage identification of discrete structural systems. The damage characteristic (location and severity) of the system first can be detected and then identified from the change of its dynamic properties (eigenvalues and mode shapes) through a backward-propa gation neural network. The neural network is constructed by three multilayer subnets that perform the tasks of input pattern generation, damage location identification, and damage severity determination, respectively. The methodology is demonstrated on two spring-mass systems. The effectiveness and limitations of the methodology are discussed. Nomenclature C = damping matrix of the discrete structural system di = dynamic residual vector K = stiffness matrix of the discrete structural system M = mass matrix of the discrete structural system \//» vdj = generalized eigenvalue and eigenvector of damaged system