Research In Fault Feature Extraction Based On Frequency Slice Wavelet Transform

Feature extraction is the key to fault diagnosis and feature recognition. To extract hidden features from fault signal, a time-frequency analysis method based on frequency slice wavelet transform(FSWT) are researched in this thesis.The deficiency of traditional time-frequency method such as Fourier transform(FT), short time Fourier transform(STFT), Wigner-Ville distribution(WVD), wavelet transform(WT) and the S transform(ST) are illustrated by a series of tests. On the basis of that, FSWT and its inverse transform and a selection method of scale factor and frequency slice function are studied systematically. The simulation proved that the proposed method which overcomes the drawbacks of traditional method can control the time-frequency resolution and can extract and reconstruct any time-frequency regions of signal.The fault diagnosis method using FSWT for looseness feature extraction is proposed. Using FSWT to sum up signal feature to diagnosis. Firstly, the characteristics in time-frequency domain is extracted, reconstructed and analyzed for valid fault feature. The application of proposed method achieves an ideal effect for generator units and a fan unit.To extract the main components of fault signal, two kind of preprocessing methods are investigated. The numerical experiment shows that the singular value decomposition(SVD) have better effect and higher signal to noise ratio than the principal component analysis(PCA).A time-frequency analytical method is proposed for fault signal preprocessing and optimal band selection. The main component is extracted by SVD and the feature is reveled after FK method. Reconstruct the selected optimal band with FSWT for diagnosis. Practice shows the proposed method by the combination of SVD, FK and FSWT can satisfactorily extract the main components and enhance the efficiency of fault diagnosis.