Abstract: Aiming at the problem that the useful characteristics of the trackside acoustic bearing signal are weak and easy to be covered by strong noise, a method combining the minimum entropy deconvolution with the improved local mean decomposition is designed and implemented to achieve the purpose of signal noise reduction and fault diagnosis. The cubic Hermite interpolation was used to improve LMD and LMD decomposition accuracy.The collected strong noise signal was used for MED noise reduction, and then the improved LMD algorithm was used to decompose the multi-component signal into a single component signal. The kurtosis value of each component was calculated, and the component with the largest kurtosis value was selected. Finally, the envelope spectrum analysis was used to extract the fault characteristics of rolling bearing.The peak signal to noise ratio of the signal was calculated as the noise reduction index to reflect the noise reduction performance of the method.The experimental results show that the designed method can improve the signal-to-noise ratio by 5.13 dB, accurately locate and extract the location and signal characteristics of bearing defects, and has better noise reduction and information resolution ability.