采用PARAFAC的欠定盲分离中机械振源数估计方法

针对复杂机械系统振源数未知的欠定盲源分离(UBSS)问题,为提高欠定盲源分离的性能,提出一种基于平行因子分析(PARAFAC)和核一致诊断(CORCONDIA)的欠定盲源数估计算法.该算法利用二阶非平稳源分离的基本思想,将中心化传感器数据分成不重叠的数据块,计算各数据块的单一时延协方差矩阵并叠加成三阶张量,即平行因子模型.利用核一致诊断算法估计PARAFAC模型的最佳组分数,从而得到机械系统的振源数.仿真实验结果表明:该算法可从非平稳欠定混合信号中准确估计振源数目.将所提算法应用于多机振动源实验,结果进一步验证了该方法的有效性.

Mechanical Vibration Source Number Estimation of Underdetermined Blind Source Separation Using PARAFAC

Considering the existing problem of underdetermined blind separation(UBSS)in the complex mechanical system with unknown number of vibration sources, a new estimation algorithm of source number in the UBSS method based on parallel factor analysis(PARAFAC)and the core consistency diagnostic(CORCONDIA)is proposed to improve the performance of UBSS method. The main idea of this algorithm is that the centralized sensor data are firstly divided into some non-overlapping data blocks. Then single time-delay covariance matrices of each data block are calculated and stacked into a third-order tensor, which is constructed into the PARAFAC model. CORCONDIA is used to estimate the optimal number of components in PARAFAC model. Thus the obtained number of components is the number of vibration sources. The simulation results show that the proposed algorithm can accurately estimates the number of vibration sources from the underdetermined mixtures of non-stationary signal. It has been successfully applied to the test of multi-source mechanical vibration, and the experiment results further verify the effectiveness of the proposed algorithm.