Aiming at the phenomenon of mode mixing in the extraction of fault information from the vibration signal of a high speed elevator rolling guide shoe,by the method of singular value decomposition (SVD) optimizing local mean decomposition (LMD), a feature extraction method based on self-adaptive sharpening wavelet decomposition (SSWD) optimizing LMD was proposed.First of all, the low pass filter, high pass filter, wavelet basis function and scale function were constructed.The original signal was decomposed into a high-frequency detailed feature signal and a low-frequency approximate signal by the multi-resolution filtering characteristics of wavelet decomposition (WD).Then, signal enhancement was done on the high frequency detailed feature components, and the enhanced high frequency detailed characteristic signal and the low frequency approximate signal were reconstructed.Finally, the LMD method was used to extract the fault features’ PF component of the rolling guide shoe from the reconstructed signals.The instantaneous Teager energy waveform of the PF component was obtained for comparative analysis.Through the actual working condition signal processing and analysing, the experimental results show that, compared with the SVD optimizing LMD method, the method completely extracts the fault characteristic components of the vibration signal of the rolling guide shoe, and avoids the phenomenon of modal confusion.