基于MBLMS算法的内燃机振动信号盲源分离

Blind Source Separation of Vibration Signals of IC Engine Based on MBLMS Algorithm

应用Gray’s变量准则作为代价函数,研究了基于源信号非高斯性的多通道盲最小均方差算法。针对实际应用,生成3路卷积混合信号进行盲源分离,分离信号与源信号相比,除幅值缩放和时间延迟外,其他信息基本得到了保留。在实测内燃机振动信号分离中,所得到的3个分离信号经过分析后确认为排气门落座、活塞撞击和进气阀落座产生的信号。仿真信号和实测内燃机振动信号的分离结果表明,多通道盲最小均方差是一种有效的盲源分离算法。

This paper describes the algorithm of Multi-channel Blind Least Mean Square (MBLMS) based on nongaussianity of sources by using Gray’s variable norm as the cost function. Aimed at practical application, three convolutive-mixture signals are generated to operate blind source separation. Compared with the sources, separated signals keep all the information except scaling and time delays. In the separation of actually measured IC engine vibration signals, three separated signals are respectively generated by exhaust valve, piston slap and inlet valve by analysis. The separated results of simulated signals and the actual IC engine vibration signals indicate that MBLMS could be an effective algorithm for blind source separation.