Abstract: A fault diagnosis method of improved particle filter and wavelet packet analysis was proposed in the application of turbine vibration. There was a sample degradation problem in the re-sampling stage of traditional particle filter. And a re-sampling algorithm which was a weight sorting and the survival of the fittest to obtain the improved particle filter was studied. The signal was filtered by the improved particle filter. Then wavelet packet analysis was used to extract the features from the noise reduction signal. Finally the fault diagnosis results were obtained by using SVM. It is shown that the fault identification rate of the noise reduction signal is significantly higher than that of original signal. No matter which kinds of signal are, the recognition rate of fault diagnosis using wavelet packet analysis is higher than that of FFT analysis. It shows the superiority of the improved particle filter and wavelet packet analysis in the stream vibration fault diagnosis.

Key words: improved particle filter, weight sorting, survival of the fittest, wavelet packet analysis, fault diagnosis of vibration