关键词: 独立分量分析, 局部均值分解, 降噪, 滚动轴承, 故障诊断

Abstract: The classical ICA could only be applied to overdetermined blind source separation problem, which meaned the source of the observed number should be not less than number of signal source,according to this feature,a new method of LMD combined with ICA was proposed. With the approach,collected single-channel vibration signals were first operated with LMD，each components then were rearranged to build a virtual channel noise based on cross-correlation criterion, the virtual channel noise with collected signals was input into ICA,using FastICA algorithm based on negative entropy realize the separation between source signals and noise signals was realized so as to achieve the noise reduction purpose. At last, spectrum analysis method was used to compare the two signals before and after noise reduction.The noise of the signals is filtered out in a certain degree，and the spectrum is less burr，and the fault characteristic frequency is more obvious，which is advantageous for the fault feature extraction,experimental analyses prove that the new denoising method proposed herein is valid.

Key words: independent component analysis(ICA), local mean decomposition(LMD), noise reduction, rolling bearing, fault diagnosis