The chapter focuses on the applications of neurocomputing to the analysis of identification problems in structural dynamics, the main attention is paid to back-propagation neural networks. The analysed problems relate to (a) application of dynamic response to parameter identification of structural elements with defects modelled as a local change of stiffness or material loss; (b) updating of FEM models of beams, including the identification of material parameters and parameters describing possible defect; (c) identification of circular void or supplementary mass in vibrating plates; (d) identification of a damage in frame structures using both eigenfrequencies and elements of eigenvectors as input data. In the examples involving the experimental measurements the application of a random noise to increase the not sufficient number of data is proposed. The presented results have proved the proposed method capable of carrying out the appointed task and indicated good prospects of neurocomputing application to dynamics of structures.

Introduction

Artificial neural networks (ANNs) have features associated with their biological origin (Haykin, 1999). They are massively parallel and can process not only crisp but also noisy and incomplete data. Moreover, ANNs can be used for both approximation and classification purposes (Paez, 1993). Thanks to their generalization features they can be applied to the analysis of problems which obey certain rules learned by the networks during the training process. Such features enable us to apply ANNs in both direct...