Neural Network Control Of Magnetic Suspension System

Maglev technology is a kind of new mechanical-electrical integration technologies, there are many obvious advantages, such as frictionless, no attrition, no lubrication, long life, low power consumption, no noise, which causes the worldwide scientific community to pay special attention. The research of the experimental magnetic suspension system GML is such a typical mechatronic system. The controller in the magnetic suspension control system is an essential element and the quality of the controller performance is directly related to the application of magnetic products. Therefore, in this thesis, the magnetic suspension system controller, which is based on the magnetic suspension control system, is studied. The mainly contents of this thesis are as follows:construct the magnetic suspension system and approximate the PID controller by BP neural network to realize neural network control of the system.First of all, the magnetic suspension system structure and working principle are analyzed, and the system is the object of study. The PID digital controller has been achieved by Matlab (Simulink), and it is connected to an external system through the data acquisition card, carrying out the digital control that a computer as a control platform.Secondly, according to the advantages that it can be fully arbitrary to the approximation of any complex non-linear system and its self-learning and adaptive ability, this thesis applies a neural network to approximate the PID controller so that the controller is no longer a simple linear combination, but a complex nonlinear mapping. Then by self-learning and self-adjusting, the neural network results in better effect of conventional PID controller. In this thesis, a BP network with three layers is utilized. The input state of input layer neurons is related to the control input of the incremental PID controller, i.e. input deviation at the k sampling time input deviation at the k-1 sampling time and input deviation at the k-2 sampling time. The output state of output layer neurons corresponds to the control output incremental of control volume of the PID controller.Finally, the simulation is carried out by Matlab and the comparative analysis of the dynamic and static characteristics of the magnetic suspension system, both mathematical and physical models, is studied with the system controllers. The simulation results show that under the control of the neural network the adaptability of the system is effectively enhanced, the dynamic and static quality of system is improved.