Fault Diagnosis For Turbine Based On Wavelet Analysis And Neural Networks

Fault diagnosis of turbine is an important aspect of the fault diagnosis technology application. Among the incidence of many common faults, the vibration fault is account for more than 95%.Based on this consideration; I selected the subject on fault diagnosis technology that, in particular, to explore ways to predict and diagnose intelligently fault of rotor vibration.The present paper has mainly carried on research works theoretically of two aspects about signal processing based on the wavelet analysis and intelligence failure diagnosis based on the neural network. The main research content summary is as follows:(1)The selection of wavelet packet mother functions and their bases is discussed in this paper. There are vast application of wavelet transform and wavelet packet transform in the fault diagnosis fields. The transforms help us to obtain a number of feature information of fault signals. But in the face of a lot of mother function of wavelet transform and a number of bases after transform, we must select a proper mother function of wavelet transform and his bases, because not all mother functions and their bases are proper.(2) The key to training of a radial basis function (RBF) network is to determine the parameters of hidden layers of the network. There are a number of training methods of RBF networks. But the shortcomings of the methods are that the training speeds are too slow and the ability to classify is unstable. In view of these shortcomings, this article uses the advanced genetic algorithm--immunity genetic algorithm to optimize the hidden layer parameters of RBF neural network. At the same time, we seek the best hidden layer units based on construction method in the training process.(3) In this paper, "wavelet packet--energy" method is used to extract the characteristics of signals. Wavelet packet analysis can be effective in extracting the useful elements of turbine machine vibration signals as the basis for fault diagnosis. According to high-frequency vibration signals in the strong noise background, a new energy-based adaptive threshold selection algorithm is proposed. This method regarding the diagnosis frequency distribution range is broad when the signal has strong time variation and fault in the complex environment has the good application prospect. The experiments of fault diagnosis demonstrate that the method is valid.