中文摘要：

在形态学滤波的基础上，结合匹配跟踪算法（Matching Pursuit，MP）和盲源分离算法（Blind Source Separa—tion，BSS）各自的特点，提出了一种基于最优匹配跟踪信号分解的欠定盲源分离算法。利用MP算法将非线性信号通过投影分解，在分解过程中利用遗传算法寻找最优原子，有效提高了算法匹配的精度和效率。将所得到的匹配分量和滤波后的原始观察信号组成新的多维信号，解决了单通道信号盲分离的欠定问题。利用快速核独立分量分析（Fast Kernel Independent Component Analysis，FastKICA）算法实现信号的盲分离，并分析了分离的不同源信号对于故障的贡献率。将该方法用于仿真信号和实际的轴承试验的信号，试验结果表明算法能够很好地解决单通道信号的盲分离难题。

英文摘要：

Based on the morphological filter and the characteristics of Matching Pursuit and blind source separation , an underdetermined blind source separation method based on optimal matching persuit algorithm is proposed. Firstly, the collected signal was purified by morphological filter and the noise which affected the diagnosis effect was removed. Then, the matching pursuit （MP） method optimized by genetic algorithm was used to decompose the nonlinear signal into some atom signals in the process of projection separation, thus increasing the matching accuracy and efficiency of the algorithm. These atoms and original observed signals were constituted to construct new observed signals, and then the under-determined problem was transformed into a multi-channel positive definite one. Finally, the Fast kernel independent component analysis （FastKICA） algorithm was used to achieve the separation of the new observed signals, and the failure contribution of different source signals was analyzed. The proposed method can be applied to the simulated signals and real test signals of the rolling bearing. Test results indicate that the mentioned method can well solve the difficult problem of under-determined blind source separation.