The Damage Identification of Truss Bridge Model Based on Generalized Regression Neural Network

The major factors affecting the neural-network-based damage identification methods and the corresponding strategies were discussed in details. The combined parameters were adopted and used as the input of neural network after preprocessing. The elementary modal strain energy coefficient criterion was employed to select the components of the mode shapes at some nodes. The two-stage damage identification was carried out on a typical truss bridge model using generalized regression neural network(GRNN), in which the effects of different noise levels with incomplete measured degrees of freedoms and the number of natural frequency on the results of damage identification are studied. The numerical simulation rusults show that, even using incomplete and noise-polluted lower frequncies and the first modal shape at some nodes, good performance of damage localization and quantification of the truss bridge model can be obtained based on GRNN.