Abstract: In order to solve the problem of correctly identifying incipient fault for electromechanical equipment and improve classification ability, a novel method of incipient fault intelligent diagnosis based on lifting wavelet package transform (LWPT) and support vector machines (SVMs) ensemble, is proposed. Firstly, a biorthogonal wavelet with impact fault property is constructed via lifting scheme, and the LWPT is carried out to extract sensitive frequency-band features from orginal signals. Then the fault characteristic frequencis can be detected by envelope spectrum analysis of wavelet package coefficients (WPCs) of the most sensitive frequency-band. Secondly, with the distance evaluation technique, the optimal features are obtained from the statistical characteristics of original signals and WPCs. Finally, the optimal features are input into the SVMs ensemble to identify the different fault cases. This method was applied to incipient fault diagnosis of rolling element bearings. Testing results show that the proposed method can effectively extract the fault features, and has better classification performance than the single SVMs, with a high classification success rate.

Key words: Lifting wavelet package transform Feature extraction Support vector machines ensemble Incipient fault diagnosis