基于改进的K-SVD和VMD的轴承故障特征提取方法

Feature extraction via improved K-SVD and VMD for bearing fault diagnosis

轴承作为旋转机械的核心零部件,在工作过程中易受其他部件影响,造成多部件耦合振动,产生的故障信号呈现非线性、非平稳特征,使得与故障信息相关的周期性冲击成分混入大量的背景噪声。传统K-奇异值分解(K-SVD)在字典学习过程中易受噪声干扰,难以确定初始化字典和迭代次数,导致稀疏表示效果较差,无法有效地提取故障特征。针对以上问题,提出了基于改进的K-SVD和变分模态分解(VMD)的轴承故障特征提取方法。通过VMD对隐藏的故障特征进行提取,根据原始数据构造与故障冲击成分高度匹配的初始化字典,选用包络谱峭度作为K-SVD中迭代次数的判断准则,通过包络分析诊断故障类型。该方法成功应用于两个案例中,与传统K-SVD相比,在稀疏表示效果、故障提取能力和运行时间上均有优势。

As the core component of the rotating machinery,bearings are affected by other components during the working process.The multi-component coupling is constantly vibrated,which makes fault signals appear nonlinear and nonstationary.The periodic shock components associated with the fault information are mixed with a large amount of background noise.Unluckily the traditional K-SVD is susceptible to interfering by noise in the process of dictionary learning.In addition,it is difficult to determine initialization dictionaries and the number of iterations,which cannot effectively extract fault features.A sparse representation framework based on the improved K-SVD and VMD is proposed for bearing fault diagnosis,with it adopted to extract the hidden fault characteristics.Then,the envelope spectral kurtosis is selected as the selection criterion for the iteration number in the process of K-SVD.The initial dictionary is learned from original signal,which was highly matched with the fault impact component.Finally,fault types can be identified with envelope analysis.Two cases prove that the proposed method can successfully extract the fault feature,which outperforms the traditional K-SVD in terms of the sparse representation effect,fault extraction ability and running time.