Research On The Magnetic Bearing Control Based On Neural Network Theory

This dissertation applies the neural network control theory into Active Magnetic Bearings system. Based on the analysis on the working principle and structure of Active Magnetic Bearing, the dissertation erects the model of signal-freedom-degree transfer function.Based on the analysis on the working principle and the learning algorithm of the neural network theory, the dissertation applies the self-learning ability of neural network into PID controller which is commonly used in the Active Magnetic Bearings system. Two kinds of neural network PID controllers are designed. One is PID controller based on single neuron, and it adjusts the proportional, integral and differential coefficients by on-line learning algorithms of neural network; The other is PID controller based on BP neural networks, it uses the nonlinear approximate ability and realizes the best parameters combination by learning the system performance criteria. Simulation results show that applying the PID controller of neural networks performs well in signal track and noise rejection theoretically.The PID controllers of single neuron are made up into C program in AMB with PC real-time control system. The rotor is suspended successfully both in static and dynamic stimulation state with its displacement error restrained within 10μm. This very experiment result indicates that the single neuron controllers designed have a good stability and dynamic ability.