Research On Vibration Fault Diagnosis Of Units With Multiwavelet And Neural Network

Recently,with the development of national energy strategy,the hydropower installed capacity and the rising number of its share in the proportion of the grid are both rising.Therefore,the stable operation of the units,will affect the stability of power system directly.It is important that how to ensure the safe and stable operation of the unit of hydropower,monitoring the fault signal in advance and nipping in the bud.In the study of hydropower units fault,it was found that about 80%fault can be reflected in the vibration signal among all the units fault.So it is very important to process the real-time vibration signal when we are studying the unit fault.It will improve the diagnosis accuracy and efficiency greatly.In the process of signal processing,due to the multi wavelet with compact support,symmetry,orthogonality and high order disappeared moment and other excellent characteristics,which ensure the signal energy and reduce distortion.This paper uses multi-wavelet to denoise and feature extraction of the energy of the original signal.Then a sample data is collected,and it will be classify by the neural network.The neural network would be optimized by particle swarm algorithm,in order to convergence faster.Based on in-depth understanding of multi wavelet and neural network,the research on the following aspects:At first,the paper discusses the current situation of the development of the domestic and foreign status monitoring and the existing problems of fault diagnosis.Based on the study of the traditional signal analysis methods and fault diagnosis method,this paper puts forward the multi-wavelet and neural network combination for vibration fault diagnosis of units,which solve the problem of extracting the feature vectors and the convergence speed of neural network.Next,the paper studies the methods of vibration signal denoised by multi-wavelet,the multi-wavelet theory is introduced to the hydropower generating unit vibration fault signal denoising,the threshold method is used to reduce the noise,with two kinds of signals to verify the effectiveness of the method respectively,and compared with the db2 wavelet,to verify its superiority.Then the principle of neural network and particle swarm optimization algorithm are introduced,and the RBF neural network is optimized by PSO algorithm.The key parameters of neural network and particle swarm optimization are discussed in this paper.The feature extraction of vibration signals of hydroelectric generating units is carried out by using multi-wavelet transform,and the fault diagnosis is carried out by the RBF network which is optimized by particle swarm algorithm.The results show that this method can be very good for the fault diagnosis of the operation status of the unit,and it has the value of engineering application.Finally,the multi-wavelet decomposition and neural network to diagnose the fault of the measured signal for a domestic hydropower station combined.The results showed that the method is able to extract more complete eigenvectors,network convergence speed faster,high rate of correct diagnosis.The paper provides a new kind of method for fault diagnosis of hydropower unit.