误差影响下滚动轴承多重故障模态特征信号的盲源分离方法

针对旋转机械设备受到测量误差及系统误差等因素影响，导致滚动轴承多故障模态信号难以分离的缺陷，提出一种误差影响下滚动轴承多重故障模态特征信号的盲源分离方法。对故障信号进行白化预处理得到白化矩阵，进而计算白化矩阵的4阶累积量，并构建4阶累积量矩阵；将累积量矩阵对角化，并取前K个较大特征值对应的特征向量作为新累积量矩阵；利用总体最小二乘方法估计最小化新累积量矩阵与目标正交矩阵的误差函数，最大程度地联合近似对角化新累积量矩阵，实现多故障信号的分离估计；为进一步评估该方法的有效性，选用时域相关系数及时频域双谱估计两种评价方法对分离结果进行验证。结果表明，该方法分离出来的信号与源信号相关系数高，并且时频域双谱估计相似，是一种有效分离多重故障的方法。

A Blind Source Separation Method for Multi-fault Modal Characteristic Signals of Rolling Bearings with Error Influences

The multi fault modal signals of rolling bearing are difficult to be separated due to the measurement errors and system errors. For this reason, a blind source separation method for multi-fault modal characteristic signals of rolling bearings under the influence of errors is proposed. A whitening matrix is obtained by preprocessing the fault signal, and then the fourth-order cumulant of the whitening matrix is calculated, and a fourth-order cumulant matrix is established. The eigenvectors corresponding to the larger K eigenvalues are taken as the new cumulant matrix by diagonalizing the cumulant matrix. The total least squares method is used to minimize the error function between the cumulant matrix and the target orthogonal matrix, thus estimating the fault source signal. Furtherly, the time domain correlation coefficient and the time-frequency domain bispectrum estimation are introduced to verify the feasibility and effectiveness of the proposed method. The results show that the signal derived from the proposed method has high correlation coefficient with the source signal, and the time-frequency domain bispectrum estimation is similar, so it is an effective method to separate multiple faults.