摘要
通过两组模拟信号对三种主流独立分量分析算法—JADE、Fast ICA、扩展Infomax算法的性能进行了对比分析,结果表明三种算法均无法完全分离超高斯源与亚高斯源形成的混合信号,Fast ICA算法对能量强弱差别大的混合信号失效。基于这一现象,提出了一种新的独立分量分析算法,以粒子群算法为优化工具,以分离矩阵为优化变量,最小化分离信号联合概率与边缘概率乘积的差值,并给出了具体的计算流程。仿真实验结果表明,该算法的性能显著优于上述三种独立分量分析算法。同时,新提出算法实施过程中不需要任何先验知识,相比其他三种ICA算法,更适合解决工程实际问题。最后,将该算法应用于对滚动轴承实验台实测信号的处理,通过对分离信号的分析实现了对滚动轴承故障类型的准确识别,进一步证明了算法的有效性。
Two sets of simulated signals were made to test the separation ability of three popular independent component analysis （ICA）algorithms including JADE,FastICA,and extended-Infomax.The results showed that the three ICA algorithms can＇t recover source signals from mixtures of super-Gaussian sources and sub-Gaussian ones precisely;FastICA fails in solving the separation problem of strong sources mixed with weak sources.A particle swarm optimized ICA algorithm minimizing the difference between joint probabilities and products of marginal probabilities of separated signals was proposed.The computing procedure was derived.Simulation tests showed that compared with the above three ICA algorithms,the proposed algorithm is the best;furthermore,the implementation of the proposed algorithm needs no prior knowledge,thus it is more suitable for solving practical engineering problems.Finally,the proposed algorithm was used to process the actual signals sampled from a rolling bearing test rig.The separated signals were analyzed to indentify the fault types of rolling bearings,the effectiveness of the proposed algorithm was verified.
出处
《振动与冲击》
EI
CSCD
北大核心
2015年第8期7-11,25,共6页
Journal of Vibration and Shock
基金
国家自然科学基金项目(51175049)
中央高校基金项目(CHD2011JC025)
