基于ADMM字典学习的滚动轴承振动信号稀疏分解

在稀疏分解过程中，字典模型构建的结果会直接影响稀疏分解的效果。为获得结构更好的字典，提出了一种基于交替方向乘子法(ADMM)的字典学习方法，在字典学习过程中采用交替方向乘子法逐个更新字典中原子，得到的字典具有良好的结构。将该字典学习方法应用到滚动轴承振动信号稀疏分解中，能获得更快的字典学习速度和更好的稀疏分解效果。与K-SVD字典学习方法相比较，证明了所提方法在轴承信号稀疏分解中的优越性。

Sparse Decomposition of Vibration Signals of Rolling Bearings Based on ADMM Dictionary Learning

In the processes of sparse decomposition, the effects of sparse decomposition would be directly affected by dictionary model construction.A dictionary learning method was proposed based on ADMM herein. In the processes of dictionary learning, the ADMM was used to update the atoms in the dictionary, which might obtain the dictionary with better structure. The method was applied to the sparse decomposition of the vibration signals of rolling bearings, shorter dictionary learning time and better sparse decomposition results might be obtained. Compared with the K-SVD dictionary learning method, the proposed method has the superiority in the sparse decomposition of bearing signals.