Research On The AMB's Indentification And Control Based On Neural Networks

The Active Magnetic Bearing (AMB) which is a new type of bearing with high performance, is an electrimechanical system which consists of rotor, bearing and position controller. It has the virtues of no friction, no fray, no lubrication, lasting life, and is recognized for its brilliant application foreground. The most important field of AMB research is the research of the AMB controller. The performance of the controller affects the dynamic performance of the AMB system and the control precision of the rotor, which are very important to the application of the AMB.At present, the control to the AMB is mostly based on the linear control theory, and its mathematic model is achieved by linearization and approximation. But AMB has the characteristics of instability, non-linearity and parameter incertitude. There will be much serious model error between the linearization model and the actual AMB when the rotor keeps away from the balance point, and it will be hard to express the behavior of the system and to control it in higher precision. This thesis tries to identify the nonlinear model of the AMB based on neural networks and to seek a control way based on neural networks and genetic algorithm to control the nonlinear AMB system.Neural network is proved superior in approximating the arbitrary nonlinear continuous function, without any experiential knowledge of the object. As a result, it has become a significant tool in dealing with nonlinear systems, and plays a more and more important role in system identification and control. In this thesis, a kind of neural network is designed as a mathematical model of the AMB, based on the knowledge of the mechanical analysis of the AMB rotor and the fundamental ofneural network. It is proved by simulation results that the designed network--theimproved Elman network in this thesis, can represents the feature of the input-output relationship of the actual AMB system, and the range of the errors is acceptable. Besides, by contrast, the improved Elman network is more suitable to identify the system than both BP and Elman networks. The research in this thesis probes for a new way to model the AMB system.In the automatic control field, PID control is one of the earliest control strategies which have been developed. Because of simple-description, high-dependability, etc, it has been used widely. But the practice proved that the traditional PID controller can't meet engineering requirement when it faces non-liner, time-delay or time-varying system. So if system asked for better control performance, a more advanced algorithm is needed to design PID controller. Based on the traditional PID control, the PID controllers which are based on BP neural network algorithm and genetic algorithm were discussed and analyzed in this thesis. Then these several control methods were simulated in MATLAB .The simulation indicates that the control characteristics of the PID controller based on genetic algorithm are much superior to the traditional PID controller and the PID controller based on BP algorithm.