噪声背景下提取调制信息的改进奇异值分解技术

An Improved Method of Detecting Modulated Gear Fault Characteristic Based on Singularity Value Decomposition

在介绍现有奇异值分离技术基本原理及其在故障诊断中的应用的基础上,研究了利用信号时间序列重构的吸引子轨迹矩阵奇异值分布特征与信号特征的关系,引入自相关函数定量计算重构矩阵的延时步长,改进了现有算法,使得吸引子轨迹矩阵的重构更加合理。研究表明该方法能在强噪声背景下提取出所需的调制信号,并成功用于齿轮箱调制故障信号的提取。

In this paper, the fundamental principle of singularity value decomposition and its application in fault diagnosis are introduced. Based on the relations between the time series and the singular value distribution of the Singularity Value Decomposition about track matrix of attractor reconstructed by time series, the autocorrelation function is introduced, which improves the method in existence. The improved method is more reasonable in track matrix of attractor reconstructed and more exact in signal detecting. The study indicates that the improved method is successful in detecting the modulated fault signals in the gearbox even in high noise background, which provides a new idea on gear fault diagnosis.