Identification of Ultrasonic Motors Based on Neural Networks

To the problems that the ultrasonic motors has the complex nonlinearities,such as dead zone and hysteresis,and the traditional identification methods are hard to be used to identify such systems directly,a modified back-propagation neural-network is proposed based on the static and dynamic characteristics of the ultrasonic motor.The ultrasonic motor model is established.By introducing a hysteretic operator to construct an expanded input space,the multi-valued mapping of hysteresis is transformed into a one-to-one mapping.The neuron with varying slope and dead zone is proposed to describe the feature of the dead zone in the motors.For the training of the proposed neural network,the generalized gradient is applied to approximate the gradient at the non-smooth points.The experimental results of the training and the corresponding model validation show the effectiveness of the proposed modeling method.