中文摘要：

针对大型风力机主轴承易发生故障且特征信号难以提取的问题和传统盲分离算法计算量大、收敛性较差的缺点,提出一种改进二阶统计量的盲源分离算法;利用信号的非平稳性,将传感器数据分成不重叠的时间窗,用广义时滞协方差矩阵代替标准协方差矩阵,然后估计每个窗内的时滞协方差矩阵平均值来提高算法的稳健性和精确度。且将该算法成功应用于某风场大型风力机主轴承故障信号的提取中。分析结果表明,该算法可有效分离大型风力机主轴承与其他部件的振动信号,与其他算法相比具有分离精度高、可靠性好等优点,对风力机主轴承的故障诊断十分有效。

英文摘要：

For the problem that the main bearing of wind turbine is likely to break down and the characteristic signals are difficult to extract, and the disadvantage of the traditional blind source separation algorithm that it needs a large amount of calculation and has poor convergence, the paper put forward an improved second-order statistics of blind source separation algorithm. By making use of the non-stationarity of the signals, divide the sensor data into non-overlapping time window, use generalized time delay covariance matrix instead of standard covariance matrix, and then estimate the average of tine delay covariance matrix of each window to improve the robustness and accuracy of the algorithm. Besides, the algorithm has been successfully applied to the faulty signal extraction for the main bearing of large wind turbine in a wind field. Analysis results showed that the algorithm can effectively separate the vibration signals of the main bearing of large wind turbine from those of other components. Compared with other algorithms, it has the advantage of high separation accuracy, good reliability and so on. It is quite effective for fault diagnosis for the main bearing of wind turbine.