Due to the weak characteristic information of vibration signal and strong external noise interference,the vibration signal during the operation of dry gas seal is difficult to reflect the tribological characteristics of dry gas seal.Aimed at the above problems and the difficulty of separating vibration signal from noise,a signal denoising method based on the combination of CEEMDAN (complete ensemble empirical mode analysis with adaptive noise) and ICA (independent component analysis) was proposed.The acceleration signal of dry gas seal collected by experiments was first decomposed by CEEMDAN to get IMF component.The corresponding independent component was obtained by ICA transformation and its fuzzy entropy value was calculated.The component whose fuzzy entropy value did not meet the condition was set to zero,and the component which met the condition was reconstructed to get the denoising signal.The algorithm was analyzed and verified by using the dry gas seal acceleration test signal.It is proves that the method proposed is more effective than other traditional noise reduction methods,which provides a new way for fault diagnosis of dry gas seal in the future.