基于K-奇异值分解字典学习的振动信号压缩感知方法

Compressed Sensing of Vibration Signals Based on K-Singular Value Decomposition Dictionary Learning

针对在齿轮箱的状态监测和故障诊断过程中传统的奈奎斯特采样定律采集到的振动信号数据量过大的问题,提出基于K-奇异值分解(K-SVD)字典学习的振动信号压缩感知(CS)方法,以实现对振动信号的高效压缩采样;在实验中分别将基于K-SVD训练生成的2种字典和离散余弦变换(DCT)固定字典用于信号的重构,并对其结果进行对比分析。实验结果表明,在相同压缩率时,与DCT固定字典相比,本文中所提出的方法能有效地提高重构信号的相似度。

In the process of condition monitoring and fault diagnosis of gearbox,a method of compressed sensing(CS)of vibration signals based on K-singular value decomposition(K-SVD)dictionary learning was proposed to solve the problem of excessive data of vibration signals collected by using traditional nyquist sampling law to realize efficient compression sampling of vibration signals.In the experiment,two kinds of dictionaries based on K-SVD training and discrete cosine transform(DCT)fixed dictionaries were used for signal reconstruction,and the results were compared and analyzed.The experimental results show that,compared with DCT fixed dictionary,the proposed method can effectively improve the similarity of reconstructed signals at the same compression rate.