CEEMD-WVD多尺度时频图像的滚动轴承故障诊断

Rolling Bearing Fault Diagnosis based on CEEMD-WVD Multi-scale Time-frequency Image

针对一般EMD-WVD方法在提取时频图像信息不充分的问题,提出一种基于CEEMD-WVD多尺度时频图像的滚动轴承故障诊断方法。该方法采用互补集合经验模态分解(CEEMD)方法对故障振动信号进行分解,自适应地获得不同频段的固有模态函数(IMF)分量;选取前几个高频信号IMF模态分量,运用Wigner-Ville分布(WVD)对各IMF分量分别做时频分析,进一步转化成对应的多尺度的时频图像;然后提取各尺度时频图像的局部二进制(LBP)纹理特征,并利用其特征训练SVM分类器;最后用训练好的分类器对不同的轴承故障振动信号进行故障识别。实验结果表明,该方法有较强的自适应性且能生成高分辨率图像,故障识别率高,在凯斯西储大学(CWRU)的滚动轴承数据库上进行5类故障的实验,诊断正确率为99.75%。

Aiming at the problem that the general EMD-WVD method has insufficient information for extracting time-frequency image,a new fault diagnosis method for rolling bearing based on CEEMD-WVD multi-scale time-frequency image is proposed.The complementary ensemble empirical mode decomposition(CEEMD)method was adopted in order to decompose the fault vibration signal and adaptively obtained the intrinsic mode function(IMF)components of different frequency bands.By selecting IMF modal components of the first few high frequency signals,time-frequency analysis was performed on each IMF component using Wigner-Ville distribution(WVD),and further converted it into a corresponding multi-scale time-frequency image.Then the local binary(LBP)texture features of each time-frequency image were extracted,and the SVM classifier was trained with these features.Finally,the trained classifier was used to recognize faults of different bearing vibration signals.The experimental results show that the method has strong adaptability and can generate high-resolution images,and the fault recognition rate is high.The five types of faults are tested on the rolling bearing database of Case Western Reserve University(CWRU),and the diagnostic accuracy rate is up to 99.75%.