Using neural networks for F.E. model updating of structures in operational conditions

In this paper, a finite element model updating method based on neural networks is presented. The main objective of the paper is to identify the dynamic properties of a structure from response data recorded during operating conditions, extending the use of results from operational modal analysis to the neural networks-based updating methodologies. The neural networks used in this study have a feed-forward architecture and their inputs are the modal parameters, that is natural frequencies, damping ratios, and mode shapes of a structure in its operative conditions, whereas their outputs are the physical properties of the considered structure. Typically, the first step of neural networks is their training and it will be shown that trained neural networks are successful when simulated cases are considered but they have some limits when experimental data are used. For this reason an algorithm based on not-trained neural networks has been developed. Both numerical and experimental analyses carried out on simple structures will be presented to demonstrate the accuracy of the proposed updating approach.