Identification And Speed Control Of Ultrasonic Motors Based On Immune Algorithm And Dynamic Time Delay Neural Networks

An ultrasonic motor (USM) is a newly developed motor in servo system in 1980's. It has many excellent characteristics, such as compactness in size, high torque at low speed, silence, and no electromagnetic interferences. So it has been used in many practical applications. The working principle of the USM is based on ultrasonic vibration force of the piezoelectric elements and mechanical frictional force. And the operation of USM is influenced by many factors and it is complicated to construct a precise model of the USM, which make it difficult for the traditional control method to fulfill the effective control of the USM. Recently the artificial intelligent methods based on neural networks and fuzzy systems have become the main approaches to perform USM control. However, the existing intelligent methods for the USM control have some shortcomings, such as complex structures, slow convergent speeds and low convergent precision, and so on. Solving these problems is the key to improve the control efficiency of the USM. So it is very important to carry out further research on the control method of the USM using intelligent control methods and computer technique combined with the existing control methods.Aiming at the mentioned problems above, this thesis proposes a novel dynamical time delay neural network (DTDNN) based on a modified immune algorithm for USM identification and speed control. The major contents are summarized as follows:(1) There is an introduction to USM including its development, characteristics, control strategy and the applications.(2) In this thesis, the immune system and algorithm is introduced, and a modified artificial immune algorithm is proposed based on the elitism strategy. The reason that genetic algorithm can easily fall into local optimum value and get premature is its diversity is insufficient during the evolution. While the immune algorithm do not have that problem since it has the calculations of affinity and density which increase the diversity and help it jump out of the optimum value.As well as we know, in the initial period of evolution, the algorithm has little possibility to fall into the local optimum value because of the high diversity. With an increase of the evolution generations, there will be more and more antibodies with high fitness values. If threshold is a constant, the algorithm can easily become premature and get into the local optimum since the diversity is getting lower and lower. If threshold is an increasing function of evolution generations, the antibody's density will be increased efficiently with the increase of the evolution generations and that the suppression will be more powerful to preserve high diversity. So the algorithm would have strong ability to control the reproducing process. So this thesis uses the immune algorithm with a dynamical threshold to avoid the local optimum value.In the immune algorithm, models of antibodies with long range and high average fitness may be destroyed by the mutation operators. The best individual in the current group may be lost in the next generation because of that. And this may happens all the time in the evolution. In the thesis, the elitism strategy of the genetic algorithm is used to the immune algorithm. The best antibody in the evolution (called the elitist antibody) is cloned to the next generation directly without cross-matching to avoid the lost of the best individual and the local optimum value.(3) The neural network and the dynamic time-delay neural network (DTDNN) is introduced. And in the thesis, the modified immune algorithm is used for the update of the time-delay, the threshold, and the weight to achieving the aim of online learning.(4) This thesis proposed a modified immune algorithm-based DTDNN identifier.Multilayer static networks transform the dynamic time-molding problem into a static space mold problem. That will definitely bring many problems. But the dynamic recurrent multilayer network which introduces dynamic links to memorize feedback information of the history influence does not have that kind of problems. It has great developmental potential in the fields of system modeling, identification and control.In this thesis, a dynamic time delay neural network is used to identify an USM, and a novel learning algorithm based on an improved artificial immune algorithm is proposed for training the network. Numerical results show that the proposed identifier can approximate the nonlinear input-output mappings of the USM quite well, and its identification precision is superior to that trained by the DTDNN with the conventional algorithm as the study algorithm.(5) This thesis proposed a modified immune algorithm-based DTDNN controller.The neural network is a kind of information processing system. It can constitute the highly nonlinear dynamics system, and has some characteristics, such as the ability of large-scale parallel processing, self-adaptive, self-organize, self-study, the distribute storage, and so on. So the control system using the neural network method has the stronger adaptive ability and is more robust.In this thesis, a novel controller is specially designed to control non-linear systems using the DTDNN controller, which is trained by the improved immune algorithm. The USM is still considered as an example of a highly nonlinear system to test the performance of the controller. Numerical results show that good control effectiveness is obtained when applying the proposed control scheme to various reference speeds. The precision using the proposed method are obviously superior to those obtained using feedback algorithm based DTDNN controllers. It suggests that the controller presented here exhibits very good robustness.The DTDNN control method based on the modified immune algorithm for the USM proposed in this thesis solves some problems existed in the rapidly convergence, precision, stability and robustness in a certain extent, which provides reference for the further study on the USM.