Research On Fault Diagnosis Of Steam Turbine Generator Unit Vibration Based On Hybrid Neural Network

As steam turbine generator unit is one of the essential equipment of a power plant, its safe, reliable and high efficient running is of great meaning to social production, human safety and economic benefits. As the development of science and technology, the structure of steam turbine generator set is more and more complex, the parameters are higher and higher and the capacity is bigger and bigger. Thus, it would bring more serious impact once failure occurred.In this thesis, SOM neural network algorithm is selected with an overall consideration of the characteristics of a steam turbine generator set for the fault diagnosis of steam turbine generator unit vibration. In order to improve the fault diagnosis accuracy as well as avoid the happening of leak detection and false detection, the corresponding neural network is improved, the application of the hybrid neural network algorithm in the fault diagnosis of the generator unit is been researched and corresponding concrete model is established, and finally a LabVIEW-based system for fault diagnosis of steam turbine generator unit vibration is developed.Firstly, three common vibration faults of generator unit are simulated in a rotor test platform and the received experimental data are processed in order to make convenient for the subsequent research on fault diagnosis method.Secondly, the typical fault data samples received from the rotor test platform are used for training SOM neural network and then the SOM algorithm is developed and optimized with PSO algorithm. And then later a fault diagnosis model based on PSO-SOM-LVQ hybrid neural network is established after a deep research on the fault diagnosis link. This model improves the accuracy and reliability of fault diagnosis as has been verified by the experimental fault data achieved before.Lastly, based on the deep research of the above algorithms as well as the Lab VIEW virtual instrument development platform, a hybrid neural network-based software platform for vibration fault diagnosis of steam turbine generator unit is developed. This platform realizes functions such as real-time monitoring, fault diagnosis and so on.