A modal parameters identification method based on recurrent neural networks

This paper presents a recurrent neural network-based approach for modal parameters identification of structure-unknown systems. The proposed approach involves two steps. The first step is to build a recurrent neural network to map the complex nonlinear relation between the excitations and responses of the structure-unknown system by off-line learning. The second step is to propose a procedure to determine the modal parameters of the system from the trained neural networks. The dynamic characteristics of the structure are directly evaluated from the weighting matrices of the trained recurrent neural network. Furthermore, an illustrative example demonstrates the feasibility of using the proposed method to identify modal parameters of structure-unknown systems. The method proposed can be used to research on fault diagnosis of engineer structure.