Research On The AMB's Control Based On Neural Networks

Active Magnetic Bearing (AMB) has the properties of no friction, no fray, dispense with lubricating, lasting life, all outstanding excellence above can't be seen in traditional roll bearing and sliding bearing. AMB will be applied to these fields far and wide: aviation and space flight, vacuum technology, tail-wagging engine, energy source, traffic, and so on. 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 usual method is to get the linearization model at the balance point of the AMB, then to design the traditional PID controller based on linear theory. 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. What's more, the unknown and time-varying load disturbances much affect the control effect. Thus, the control effect of the traditional PID controller is far away from satisfaction. This paper tries to identify the nonlinear model and to control it based on neural networks.This paper identifies the nonlinear model of the AMB based on the nonlinear map ability of the neural networks after thorough analysis on the theory of the neural networks. Because of the instability, the displacement of the rotor isn't convergent and we can't get the I/O data of the AMB system when the AMB is open loop controlled. Then, this paper proposes a closed loop identification method by adding random disturbance to the output of the closed loop controller in order to well excitate the AMB. The AMB can be well identified by this means.Making use of the high mapping ability of neural networks, this paper designs a BP neural network compensation controller. The output of the controller contains the non-linear combination of proportional, integral and differential signal, which reduces the nonlinear effect of the system. The simulation research indicates that the rotor can be suspended in a larger scale. At the same time, in the interest of improving the dynamic performance of the rotor suspending, especially enhancing the rejection ability of load disturbance, this paper proposes a CMAC neural network adaptive controller by combining a neural network feedforward controller and a PID feedbackcontroller. The simulation indicates that the control characteristics are much superior to the traditional PID controller.