Neural network based modeling of a piezodisk dynamics

Piezoelectric phenomenon is commonly used in microsystems. Many sensors as well as actuators are based on this phenomenon. Because of the nonlinear character of the piezo phenomenon, exact measuring of fast dynamic systems IS difficult with piezoelectric sensors. Piezo-based actuators on the other hand need feedback for the exact motion. This has increased the size of the system as well as the power consumption, which are undesirable characteristics in microworld. In this paper a solution for the problem is determined by modeling. First, a third order transfer function is generated to model the piezoactuator at the operating point. The parameters of a grey box-model are implemented as dynamic, because of the nonlinearity of the piezo actuator. This is the way to capture the characters of the transfer function to fit the real actuator at each operating point. A multilayer perception neural network is used to model the behavior of the system. The training data for the network is measured at different operating points. The model is validated by test data at different operating points. The agreement with the model and the measurements is excellent.