基于ICA的特征提取实验研究及特征独立性评价

Experimental Study of Feature Extraction Based on Independent Component Analysis and Evaluation of Feature Independence

在Adams环境下设计了悬臂梁激振仿真系统并在实验室搭建了悬臂梁激振系统,通过对系统多种方式的激振,采集了梁上各测点振动信号.使用MSNR,GOSA和JADE这3种ICA算法分别对信号进行处理;对记录的一组振动信号进行加噪处理;用独立性评价指数衡量了ICA分离模态频率的可靠性.实验结果表明:ICA算法能够有效地提取系统结构模态;3种ICA算法的效果随加噪程度的增大而变差.研究为解决具体工程问题提供了参考.

A cantilever excitation simulation system was designed in the Adams environment and a cantilever excitation system was built in laboratory.Vibration signals of measuring points of the cantilever were collected by loading various types of excitation on these two systems.The signals were separately processed by means of three kinds of ICA algorithm of MSNR,GOSA and JADE.Some noise processing was interfused into a group of recorded vibration signals and the independent performance index was used to measure the reliability of the modal frequency separated by ICA.The results show that ICA algorithm can separate the independence source signal effectively,namely it can extraction the system＇s modal.The effect of three ICA algorithms deteriorate with the increase of noise scale.And this research can provides valuable reference to the practical engineering problems.