A new time-domain analysis method that uses second generation wavelet transform (SGWT) for weak fault feature extraction is proposed. To extrata incipient fault feature, a biorthogonal wavelet with the characteristics of impact is constructed by using SGWT. Processing detail signal of SGWT with a sliding window devised on the basis of rotating operation cycle, and extracting modulus maxi-mum from each window, fault features in time-domain are highlighted. To make fitrther analysis on the reason of the fault, wavelet package transform based on SGWT is used to process vibration data again. Calculating the energy of each frequency-band, the energy distribution features of die signal are at- tained. Then taking account of the fault features and the energy distribution, the reason of die fault is worked out. An early impact-rob fault caused by axis misalignment and rotor imbalance is successfully detected by using this method in an oil refinery.