Modeling of micro-dispalcement stage based on diagonal recurrent neural network

A novel methodology is proposed in this paper for real-time modeling of a nanometer scale positioning stage driven by the piezoelectric ceramics. The precision of the stage is limited by the intrinsic nonlinear and hysteretic behaviors of the actuator. By integrating a second-order linear dynamics and a diagonal recurrent neural network, a nonlinear dynamic model is developed and experimentally validated. Stage positioning data are used to train the network. The error of the model is reduced by adjusting parameters on-line. The results of experiment show that the average error and the maximum error within the main journey of 80 mum are reduced to 80 nm and 110 nm, respectively. The positioning precision is improved compared with the traditional Preisach model.