IDENTIFICATION OF NONLINEAR VIBRATION SYSTEM BY A NEURAL-NETWORK

This paper describes a convenient method of identification of nonlinea r vibration system by a neural network using time series data. The neu ral network with suitable selection of system inputs and outputs are a pplicable for system identification of nonlinear vibration by learning of response. Learning only time series data of impact response of pie cewise linear system, it is possible to predict both forced vibration response and impact response in time domain by the neural network. Als o, by learning only time series data of periodic response in coexistin g regions of chaotic and periodic response of Duffing system, it is po ssible to predict both chaotic and periodic response of forced vibrati on by the network, for given predicting conditions chosen nearby the l earned condition. The effectiveness of the method is demonstrated by n umerical simulations for piecewise linear system and Duffing system.