Generalized Regression Neural Network-based Damage Identification of Truss Bridge Model

:A two-step method of structural damage identification based on Generalized Regression Neural Network(GRNN) is proposed. Some major factors affecting the neural-network-based method and the corresponding strategies, such as constructing combined parameters, adding noise to expand samples,identifying step-by-step, choosing nodes by elementary modal strain energy coefficients, etc, are discussed in details. The numerical simulation is carried out on a typical truss bridge model using GRNN, in which the effects of different noise levels under the condition of incomplete measured degrees of freedom and the number of natural frequencies chosen on the results of structural damage identification are investigated. The results show that combined parameters can overcome the shortage of using sole natural frequencies or sole mode shapes for structural damage identification and the use of elementary modal strain energy coefficients provides certain basis for the choice of nodes. Moreover, GRNN has good performance for structural damage localization and quantification even getting only noise polluted lower-order frequencies and the first-order mode shape at a few nodes.