Dynamic modeling of configuration-controllable phononic crystal using NARX neural networks

Configuration-controllable phononic crystals (CCPCs) have broad application prospects in engineering because of their adjustable vibration-reduction properties. Owing to the complicated constitutive relationship and nonlinear geometric deformation, it is difficult to accurately predict the dynamic characteristics of CCPCs using the finite element method (FEM) or theoretical methods. In this study, we employed a nonlinear autoregressive with exogenous input (NARX) artificial neural network (ANN) to identify the dynamic model of the CCPC under an impact load, using data from over 100 experiments and numerous accumulated samples. The corresponding experimental data for the CCPC were used to train the ANN and determine the rational ANN model. The identification results indicate that appropriate number of neurons and time-delay orders can effectively reduce the identification errors. Compared with the response predicted by the FEM, the identification model can describe the nonlinear characteristics emerging from phononic crystal (PC) experiments more accurately. This study provides an efficient and accurate online identification approach for PC-modeling.