Ensemble neural networks for structural damage identification using modal data

Damage in a structure is defined as changes to its geometric and material properties, leading to a reduction in the structural stiffness which negatively affects the performance of the structure. Reduction in the structural stiffness produces changes in the modal parameters such as the natural frequencies and mode shapes. Artificial neural networks (ANNs) have been applied extensively in recent years due to their excellent pattern recognition ability that is useful for structural damage identification purposes. In this paper, ANNs based damage identification techniques were developed and applied for damage localization and severity identification in I-beam structures using dynamic parameters. Experimental modal analysis and numerical simulations were applied to generate dynamic parameters of the first five flexural modes of structures. In damage identification using ANNs, five individual networks corresponding to mode 1 to mode 5 were trained, and then a method based on neural network ensemble was proposed to combine the outcomes of the individual neural networks into a single network. The ensemble network has the advantages of all the individual networks from different vibrational modes. The results showed that ensemble neural networks have a strong potential for structural damage identification.