基于移不变稀疏编码的单通道机械信号盲源分离  

Shift invariant sparse coding for blind source separation of single channel mechanical signal

作　　者：朱会杰[1] 王新晴[1] 芮挺[1] 李艳峰[1] 张红涛[2] 赵洋[1] 

机构地区：[1]解放军理工大学野战工程学院,江苏南京210007 [2]防空兵指挥学院,河南郑州450052

出　　处：《振动工程学报》2015年第4期625-632,共8页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(61472444)

摘　　要：针对特征反复出现的机械信号,提出了一种使用移不变稀疏编码的单通道盲源分离方法。移不变稀疏编码将原始信号看成多个基与系数的卷积,能够根据信号的统计分布,利用信号自身特征自适应地学习到匹配的基和稀疏的系数。在恒定工况下,不同的信号源具有不同的特征,同一信号源的特征结构相似,将学习到的不同特性的基分别重构即可得到相应的源信号。将该方案应用于仿真的齿轮故障和轴承故障振动信号盲源分离问题中,以及用来提取实测的液压泵压力脉动。结果显示,这种方法较其他方法有所改进,所需人工经验少、抗噪能力强、信号恢复精度高、鲁棒性好,适用于单通道机械信号盲源分离,为单通道信号盲源分离提供了一种新思路。For the single channel mechanical signal with repeated features,the method for blind source separation based on shift invariant sparse coding was proposed in this paper.In the literatures of shift invariant sparse coding,a signal is described as the convolutions of multi bases and their coefficients.According to statistical distribution of a signal,shift invariant sparse coding could adaptively learn its bases and the sparse coefficients from the structures of the signal itself.Under stable condition,different signal sources have different features,and the features from the same source are similar,thus the learned bases with different features could be used to reconstruct corresponding signal sources.This scheme was applied in the blind source separation of simulated vibration signals of faulty gear and bearing,as well as the extraction of pressure pulsation of hydraulic pump.The result showed that this algorithm has improved a lot compared to other algorithms,and this algorithm needs less expertise,has strong anti-interference ability,in addition,it is robust and could recover original signals more accurately.Therefore,this technique is appropriate to blind source separation for single channel mechanical signal,and provides a new way for single channel blind source separation.

关 键 词：信号处理 移不变稀疏编码 盲源分离 正交匹配追踪 字典学习 

分 类 号：TN911.7[电子电信—通信与信息系统] TH165.3[电子电信—信息与通信工程]