Self-constructing recurrent fuzzy neural network for ultrasonic motor drive

The ultrasonic motor (USM) is a popular actuator and used in industry and academic research in the last decades because it has the advantages of high holding torque, good response characteristics, high torque density, silent operation, free of electromagnetic noise and compact size. However, the motor are time-varying and highly nonlinear system which the parameters varied with increasing temperature and changes in drive frequency, load torque and phase difference of two-phase voltages. The investigations are focused on three parts, first we construct complex dynamic model of traveling-wave ultrasonic motor (TWUSM) by MATLAB/ SIMULINK in this paper, a hybrid model which combines the strength of the equivalent circuit method and the finite element method is derived, On the other hand, a novel controller and driving system are presented. For the part of control, we present a self-constructing recurrent fuzzy neural network for the speed control of a TWUSM to track periodic reference trajectories. Two types of online learning algorithms are the structure learning and the parameter learning. The structure learning has the ability of identifying whether the fuzzy rules are generated or not, while the parameter learning algorithm used the supervised gradient decent method to adjust the connected weights in the consequent part. When the system is in steady state, fuzzy decision-making method is used to delete unimportant fuzzy rules automatically, so that to get the simplest structure of SCRFNN while maintaining the good control performance. Finally, simulation results show that the control effort is effective, as well as confirm the theoretical work.