It is an effective fault diagnosis method to collect vibration signals from fault rolling bearing of tufting machine. In actual engineering applications, the collected vibration signals usually contain a lot of noise signals which require signal denoising. Local mean decomposition (LMD) method is a new adaptive timefrequency analysis method and it has been successfully applied in fault diagnosis. In order to improve the performance of LMD, the cubic Hermite interpolation approach is applied to construct the local mean function and envelope estimation function instead of moving average algorithm. A new method of fault diagnosis was proposed, first of all, wavelet packet denoising process was applied to denosing, and then the improvedLMD method was applied to decompose signals into several product functions (PF). Finally, the most effective PF was selected to conduct the Hilbert envelope spectrum analysis and the fault features were extracted. The simulation test and the actual rolling bearing fault diagnosis test showed that the proposed method is effective．