基于改进字典学习的单通道振动信号盲源分离算法

单一传感器采集到的机械信号可能是多个激振源的叠加,难以进行有效分离。针对单通道盲源分离问题提出了基于改进字典学习的单通道振动信号盲源分离算法。首先利用移不变字典学习算法学习信号中的移不变基函数,重构基函数得到反映信号时频域特征的移不变分量,然后利用自适应模糊C均值聚类算法及局部最大值检测算法对得到的各个移不变分量的包络谱提取关键点并聚类,最后将聚类后的移不变分量进行叠加得到源信号的估计。仿真数据的实验表明,算法在噪声环境下具有一定的鲁棒性,同时将该算法应用到实测某型直升机振动信号分离中,进一步验证了该算法的实际价值。

Single channel blind source separation of vibration signals based on improved dictionary learning

Aiming at the problem of vibration signals collected by a single sensor possible to be a sum of several sources and hard to be separated,a method based on improved dictionary learning was proposed for single channel blind source separation of vibration signals.Firstly,the shift invariant dictionary learning algorithm was utilized to learn shift invariant base functions in signals.Then,shift invariant components(SIC)to reflect signals’features in time domain and frequency one were obtained by reconstructing base functions.An adaptive fuzzy C-means clustering algorithm and the local maximum detection method were utilized to extract key points on envelope spectrum of each SIC obtained and these points were clustered.Finally,the clustered SICs were superimposed,respectively to acquire estimations of source signals.The tests of simulation data demonstrated that the proposed method has certain robustness in the presence of noise;it is used to conduct a certain type helicopter vibration signals separation,and verify its actual application value.