Research On Fault Diagnosis Of Hydropower Generating Unit Vibration Based On Neural Networks And Wavelet Packet

High efficiency is important in the modern administration of hydropower. New technologies to promote improvements include "Condition—Based maintenance". New management systems for "unattended" stations should provide improved levels of automation, reliability, security and stability. Stable operation is an important aspect of equipment performance, which is assessed through vibration analysis. The neural network and wavelet packet provides a method means of diagnosing abnormal vibrations, and make a good effect.In this paper, discuss the theme and related background, and dissertate wavelet understructure and neural network theory. In allusion to practical, dissertate wavelet and neural network related algorithmic and application in detail. Main work and task was concluded by next follow aspects:First in this paper point that there is a big difference between the vibration of water turbo-generators and other rotating machines. The nature of the coupling caused by hydro,mechanical and electrical factors, no unique relationship between the specific faults and associated vibration features and additionally the super-imposing of various vibrations at a certain position of a unit make it difficult to ascribe the severity of vibration in accurate words. Then on the base of researching wavelet theory and neural networks, Using wavelet packet which can separate the real signal from noise signal effectively and attains the purpose of de-noise by using time-frequency localizing analysis, local feature abstract, time-change filter wave, restraining some frequency range and other characters of wavelet and emulate the fault diagnosis.Secondly, In allusion to hydro power generating unit fault diagnosis signal, according to the property of linear and energy conservation of wavelet packets, fault diagnosis method based on wavelet Transform is presented .Thus it proposes the BP neural network fault diagnosis algorithm based on the incompletion wavelet packet transform and energy normalization as preprocessors. After preprocessed, the sample signal will be sent to BP neural network to be trained which effectively reduces the numbers of the inputs and the hidden layer nodes, thereby reducing the size of the neural network, degrading its complexity and minimizing its training time. It identifies fault location with accuracy. Furthermore, in allusion to different fault byrecognizing make this network train successful under the circumstance that in the nature of less date. By instance analysis in Nan Ya-he hydro power station and achieve good effect .