基于独立分量分析的欠定盲源分离方法

Underdetermined blind source separation method based on independent component analysis

针对欠定盲分离算法只能分离稀疏信号,对不稀疏信号分离效果不理想,而经典独立分量分析算法中扩展Infomax算法能分离超高斯信号及亚高斯信号,却只能应用于观测数不少于源数的超定盲源分离等问题,结合扩展In-fomax算法,提出欠定ICA算法,通过生成隐藏数据将欠定盲分离问题转化为超定盲分离问题,应用经典的扩展Infomax算法进行分析,并对实测齿轮箱混合故障信号进行分离,用包络阶次方法对分离出的信号进行分析,成功识别出齿轮箱不同故障特征。验证该算法在齿轮箱故障诊断中的有效性。

The underdetermined blind source separation （UBSS） algorithm at present can only separate sparse signals, but can＇t separate non-sparse signals successfully. Classical ICA algorithms, such as, extended Infomax, can separate both super-Gaussian and sub-Gaussian signals, but it is used only in an over-determined BSS （OBSS）. Combined with an extended Infomax, an underdetermined ICA algorithm was proposed here. By generating hidden data, an UBSS problem was transformed into an OBSS one, and then an extended Informax algorithm was used to analyze the signals. This method could separate both super-Gaussian and sub-Gaussian signals in an UBSS problem. Through analyzing transient signals of a gearbox by use of the underdetermined ICA combined with the order envelope spectral analysis, its fault features were fully detected and the effectiveness of the proposed method was verified.