Abstract
This dissertation thoroughly investigates the properties of wavelet transform and proposes methods to enhance the performance of the wavelet-based monitoring scheme. The theoretical background of Fourier and wavelet transform is introduced at the beginning. Meanwhile, an in-depth discussion on the limitation of Fourier-based monitoring methods is presented. For a mechanical system that conventional methods cannot yield satisfactory monitoring on system defects, a wavelet-based monitoring scheme is performed and its superior applicability is verified. The spectrum mean-square widths of wavelets are examined regarding to their lengths, and then a wavelet basis function that possesses adequate frequency resolution can be selected. To enhance the discrimination on defects and increase signal-processing efficiency of a wavelet-based scheme, this dissertation proposes a method to estimate the fundamental period of a mechanical system by manipulating wavelet coefficients. Signal segmentation may be achieved by this method. This method is advantageous because a mechanical fundamental period may be accurately estimated even though measurement does not use a considerably long window to acquire data. The computational efficiency is evaluated according to the algorithm of the proposed method afterward. A detail comparison with the conventional correlation method in computational efficiency is presented. In addition, a new criterion is proposed to improve the capability of discrimination on frequency in the dyadic wavelet transform. The effect of spectral leakage on an energy plot may be reduced by this criterion, so that the problem of incorrect judgement on the dominant level may be resolved. Finally, the applicability of these proposed methods is verified by using experimental and simulated data acquired from mechanical systems.