摘要
为了提高轴承在强干扰背景下的故障诊断精度,提出了一种基于频谱包络分割EWT算法的轴承故障特征提取方法。首先,针对传统EWT算法频段冗余分割导致的模态相似、信号失真等问题,基于三次B样条包络线极点进行了频段分割,有效提取了信号在不同频段的模态分量;然后,使用裕度因子分析了模态分量的敏感度,并分离出了高敏感模态分量,计算了高敏感模态分量的排列熵,组成了特征向量;最后,使用聚类法对频谱包络EWT特征、传统EWT特征、小波信息熵特征进行了分析,其中频谱包络EWT特征不存在类间交叉现象,且类内聚集度较高;将上述3种故障特征输入到支持向量机中进行了模式识别实验。研究结果表明:小波信息熵特征的诊断准确率为93.75%,经典EWT特征的诊断准确率为87.50%,频谱包络EWT特征的诊断准确率为98.75%;这表明频谱包络EWT特征的质量最好,能够在强干扰背景下有效提高轴承的诊断准确率。
In order to improve the accuracy of bearing fault diagnosis under strong interference,an improved empirical wavelet transform(EWT)algorithm based on spectral envelope segmentation was proposed.Firstly,aiming at the problems of modal similarity and signal distortion caused by redundant frequency band segmentation of traditional EWT algorithm,the frequency band was segmented based on the pole of cubic B-spline envelope,and the modal components of signals in different frequency bands were effectively extracted.Then,the sensitivity of modal components was analyzed by using the margin factor,and the highly sensitive modal components were separated.The arrangement entropy of highly sensitive modal components was calculated to form the eigenvector.Finally,the clustering method was used to analyze the spectrum envelope EWT feature,the traditional EWT feature and the wavelet information entropy feature.The spectrum envelope EWT feature did not possess the phenomenon of cross between classes,but the cohesion degree within classes was high.The above three fault features were input into support vector machine for pattern recognition.The research results show that the diagnostic accuracy of wavelet information entropy feature is 93.75%,that of classical EWT feature is 87.50%,and that of spectral envelope EWT feature is 98.75%,which show that the quality of spectrum envelope EWT feature is the best,which can effectively improve the diagnostic accuracy of bearing under the background of strong interference.
作者
龙雄辉
胡蓉
苏丹
LONG Xiong-hui;HU Rong;SU Dan(Guangzhou Railway Polytechnic,Guangzhou 510430,China;Fujian Provincial Key Laboratory of Big Data Mining,Fujian University of Technology,Fuzhou 350108,China)
出处
《机电工程》
CAS
北大核心
2022年第11期1567-1574,共8页
Journal of Mechanical & Electrical Engineering
基金
广东省教育厅普通高校重点科研项目(2021ZDZX1139)。
关键词
轴承振动信号分析
故障特征冲击分量
特征向量提取
经验小波变换
裕度因子
敏感模态选择
排列熵
rolling vibration signal analysis
shock component of fault signature
feature vector extraction
empirical wavelet transform(EWT)
clearance factor
sensitive modal selection
permutation entropy(PE)
作者简介
龙雄辉(1974-),男,湖南益阳人,硕士,副教授,主要从事机械制造、CAD/CAM、信息系统方面的研究。E-mail:longxionghui@126.com。
