基于盲源分离技术的故障特征信号分离方法

信号采集过程中，传感器测量到的信号是实际振动信号在此测量方向的投影值，由于其他不相干振源的影响，测量信号由多个振动信号成分组成。在分析多振源信号混合模型的基础上，采用盲源分离技术分离不同的振源信号，讨论分离结果的广义初等相等性质的影响，研究估计振源数目的方法和选取测量信号的方法，利用二阶特征矩阵联合近似对角化算法，从测量信号中分离故障特征源信号。该算法可减小信号采集不当造成的影响，有效提高特征信号的提取。

USING BLIND SOURCE SEPARATION TO RECOVER THE MECHANICAL FAULT FEATURE

The signal collected by accelerator is the projection of true vibration signal in the measured position. While the real vibration direction is unknown and there are multiple vibration sources, the sampling signal is mixed by the projection of the fault feature vibration and other non coherent vibrations. In this paper, the problem of recovering mechanical fault feature from mixture signals is considered. The blind source separation technology is introduced to resolve the problem. The influence of generalized equality property of separated signals for fault diagnosis is discussed. The paper gives the algorithm for determining the number of sources. The algorithm of how to choose fit samples for blind source separation from all samples is also given. As a suitable solution, the joint approximate diagonalization of second order eigen matrices algorithm (JADSOE) is presented for recovering the spectrum feature of mechanical fault from mixture signals. The spectrum of each vibration signal is successfully separated from mixture signals. The algorithm can diminish the effect of sample data collected at unreliably measured position.