Prediction of Hysteretic Effects in PZT Stack Actuators using a Hybrid Modeling Strategy

摘要

In this paper, concepts associated with the Preisach model and nonlinear mapping functions (neural networks) are coupled to model the hysteretic behavior of piezoceramic actuators. Preisach concepts are utilized in choosing the initial data points and calculating the final displacements having nonlocal memory. In a traditional Preisach model generalization is typically handled by interpolation functions. These functions can lead to significant errors unless the number of data points is considerably high. In this study the generalization of all first order reversal curves is provided by a single neural network. The goal of this work was to enable real-time implementation and learning with a "limited" number of variables. Finally, a novel on-line training approach was developed to account for errors caused by frequency dependency and large variations of the input of the actuator. Results show excellent agreement between simulated and experimental results.