时频谱相似性度量的故障特征提取方法

针对齿轮或轴承在局部故障损伤状态下的振动信号,提出一种基于时频谱相似性度量的故障特征提取方法,用于齿轮或轴承相关故障的诊断。该方法首先利用比例因子可调的S变换对振动信号进行时频变换;然后在S变换时频谱中,选取一个显著的冲击特征,保持其频率不变,令其沿时间轴方向,从初始时间平移至终了时间,同时计算冲击特征与所遮掩时频区块之间的余弦相似度和相关系数;平移结束后获得余弦相似度和相关系数的曲线。仿真信号和齿轮、轴承故障振动信号的处理结果表明,余弦相似度曲线和相关系数曲线均可展现出故障冲击特征的周期性变化规律,且两者的频谱均能够提取出故障特征频率,实现齿轮或轴承相关故障的识别。

Fault feature extraction from time-frequency spectrum by using similarity measurement

A fault-related feature extraction method, which is based on the analysis of time-frequency spectrum by using similarity measurement, is proposed for gear or bearing fault diagnosis. In this approach, the time domain vibration signal is transformed into time-frequency domain by using S transform. Then, a noticeable impact feature is selected and keeping its frequency constant, translated from the initial time of the time-frequency spectrum along the time axis to the end time, while the cosine similarity coefficient and the correlation coefficient between the impact feature and its masked block in the spectrum are calculated. After that, a cosine similarity coefficient curve and a correlation coefficient curve are got. Finally, fault-related features can then be extracted clearly with the analysis of the frequency spectrum of the two time series coefficient data.