基于波动的充液管道边界条件识别的神经网络法

n identification approach for boundary conditions of fluid-filled piping systems has been developed in the present work. Based on the wave propagation theory, a few lowest natural frequencies corresponding to different boundary conditions of the piping system are obtained. Then these frequencies and boundary conditions are put into a neural network as input and output patterns. Therefore the network is trained with the backpropagation algorithm to form the relationship between the frequencies and boundary parameters. Due to the effective mapping ability of the neural network, unknown boundary conditions can be determined by the trained network. In this paper, the axial, bending and torsional vibrations are taken into account. In addition the effect of fluid-structure interaction is also introduced into the modeling. Numerical examples of a straight and a curved piping system, each filled with fluid are presented to demonstrate the feasibility of the proposed approach.