Abstract:  Under variable working conditions, the components of gear fault vibration signals had complex time-frequency characteristics that crossing multiple time scales or overlapping time-frequency domains. The traditional decomposition method could not decompose this kind of signals based on local time-scale characteristics. A new MMD method was introduced. The MMD method innovatively regarded the mono-component signals as a linear combination of eigenvectors with different weight coefficients. By iteratively optimizing the weight coefficients, the corresponding component signals were obtained. The problems that large amount of data reduced the decomposition efficiency when MMD analyzed the actual vibration signals with high sampling rate was solved. And the MMD method was applied to analyse gear fault vibration signals under variable rotating speed working conditions.The results show that the method may effectively decompose to the overlapping fault component signals in time-frequency domain, and has obvious advantages compare to the traditional method based on time-scale characteristics.Combined with order analysis, the gear fault feature information may be extracted clearly and accurately.

Key words: multichannel multicomponent decomposition（MMD）, time-frequency concentration measure, variable rotating speed working condition, gear, fault feature extraction

摘要： 齿轮故障振动信号在非稳态工况下，其分量可能存在跨时间尺度或不同分量重叠的复杂时频特征，传统的以局部时间尺度特征为依据的分解方法无法分解，为此，引入一种新的多通道多分量分解(MMD)方法。MMD方法创新性地将单分量信号看成具有不同权重系数的特征向量线性组合，通过迭代优化出权重系数，便可获得相应的分量信号。解决了MMD分析高采样率的实际振动信号时大数据量会导致其分解效率降低的问题，并将MMD方法应用于变转速工况下齿轮故障振动信号的分析，结果表明，该方法可以有效分解出在时频域发生重叠的故障分量信号，较传统的以时间尺度特征为依据的分解方法具有明显优势，结合阶次分析可以清晰准确地提取出齿轮故障特征信息。

关键词: 多通道多分量分解, 时频聚集性度量, 变转速工况, 齿轮, 故障特征提取