Modeling And Control Of Dynamic Hysteresis Nonlinear System Based On RBF Neural Network

As a typical smart structure, piezoelectric actuator has the advantages of high resolution, fast response, high energy density. It has been widely used in the field of aeronautics and astronautics, biological sciences, precision positioning and measurement, active vibration control. However, due to its inherent hysteresis nonlinearity, which brings much difficulty to the control of piezoelectric actuator, and even leads to serious system oscillation. Especially the hysteresis nonlinearity is rate-dependent, which greatly increases the difficulty of modeling and control. It has important theoretical research significance and engineering application value to model and control the intelligent structure effectively. In this paper, the modeling and control of the piezoelectric actuator is studied based on the radial basis function neural network model, to deeply study the rate-dependent hysteresis nonlinearity. The main work is as follows:To receive and debug the related experimental equipment, set up the conditions and parameters of the device, and set up a complete set of experimental platform for the study of the hysteresis nonlinear system. The MPI model is studied and realized to research the piezoelectric actuator and the giant magnetostrictive actuator preliminarily, and to understand the hysteresis nonlinearity and the rate dependence of the smart structure preliminary. The dSPACE supporting software of development is studied to familiar with the complete experimental development process, and to master the ideas of modeling and the steps of control.The Hammerstein model of piezoelectric actuator is established by the RBF neural network model and the ARX model, to characterize the hysteresis characteristics and the rate-dependent of the piezoelectric actuator. A hysteresis operator based on the PI model is developed, to overcome the multi-value mapping of the piezoelectric actuator, which is used as an auxiliary input for the RBF neural network. The identification method of the hysteresis model is given, and it was compared with the measured data of piezoelectric actuator. The experimental results shows that the designed modeling method is easy identification, the model has the advantages of simple structure, strong ability of frequency generalization. The model of piezoelectric actuator is established based on MPI model and BP neural network model, modeling effect of the three kinds of models are compared and analyzed.The development process of the control system based on the dSPACE experimental platform is mastered, the method and principle of the S-Function is introduced. The complex control strategy is composed of feedforward control based on RBF neural network inverse model and feedback control based on PID controller, the tracking control experiment is designed. At the same time, the complex control strategy based on the MPI inverse model is realized, and the feedforward control strategy based on the RBF neural network inverse model is adopted, the experimental comparison and results analysis is conducted. The results show that the designed control strategy is effective, which can meet the engineering and research needs. Compared with the latter two control strategies, its control accuracy is higher and frequency generalization ability is stronger.