Research On Hydrogenerator Units Shaft Systems Modeling Based On Nonlinear Dynamic Characteristics Of Bearing

The running stability and reliability of large water turbine generator set is extremely noticed by people. It is not only necessary for accurately design but also important for running in security, malfunction diagnose and maintain to do the dynamics analyse as accurate as possible. The rotor-bearing system of water turbine generator set is multi-degrees of freedom system with partial nonlinearity. So far, the research methods on dynamics characteristic of this kind of system mainly have transfer matrix method, integration time-domain method, finite element method. As a new type of computer which simulates the thinking of brain, Neural Network is a massive Self-adapt nonlinear calculation system. It has the potential to build the water turbine generator set rotor-bearing system model.In this thesis, a kind of Feed-forward Neural Network----RBFNN has been used to do the research on bearing modeling, which is the most nonlinear device in the rotor-bearing system of water turbine generator set. According to the sliding bearing working principle, several parameters which are relative to the bearing oil-film force are chosen to be the input of RBFNN after fixed. Then a nonlinear mapping relationship is established from bearing status parameters to oil-film force. Based on several basic learning arithmetic and the physiology characteristic of RBFNN, a online-learning arithmetic has been improved to fit the sliding bearing force. The result then has been used in the whole rotor-bearing system model based on Riccati transfer matrix method. The axes track at the guide-bearing has been turn out. It validates the feasibility of bearing modeling by RBFNN, and construct a foundation for whole rotor-bearing system modeling by neural network.