时频谱相似性度量的故障特征提取方法  

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

作　　者：郭远晶[1] 魏燕定[2] 金晓航[3] 林勇[1] GUO Yuanjing;WEI Yanding;JIN Xiaohang;LIN Yong(Zhijiang College,Zhejiang University of Technology,Shaoxing 312030,China;Zhejiang Province Key Laboratory of Advanced Manufacturing Technology,Zhejiang University,Hangzhou 310027,China;Key Laboratory of Special Purpose Equipment and Advanced Processing Technology of Ministry of Education,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学之江学院,浙江绍兴312030 [2]浙江大学浙江省先进制造技术重点研究实验室,杭州310027 [3]浙江工业大学特种装备制造与先进加工技术教育部重点实验室,杭州310023

出　　处：《振动与冲击》2020年第12期70-77,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金(51505424);浙江省自然科学基金(LQ17E050006,LY15E050019)。

摘　　要：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。A fault-related feature extraction method,which is based on the analysis of time-frequency spectrum by using similarity measurement,was proposed for gear or bearing fault diagnosis.In this approach,the time domain vibration signal was transformed into time-frequency domain by using the S transform.Then,a noticeable impact feature was selected and kept 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 were calculated.After that,a cosine similarity coefficient curve and a correlation coefficient curve were obtained.Finally,fault-related features can then be extracted clearly with the analysis of the frequency spectrum of the two time series coefficient data.

关 键 词：故障诊断 冲击特征 S变换 余弦相似度 相关系数 

分 类 号：TH165.3[机械工程—机械制造及自动化] TN911.7[电子电信—通信与信息系统]