基于峭度的转子振动信号盲分离

推进器发动机在运行时存在多个不同振源信号相互重叠的情况,这些振动源信号难以用常规的方法分离开来,这对于试车时正确判断振动源十分不利.本文在独立分量分析(ICA)的基础上,建立了一种基于峭度的快速定点抽取算法,并给出了具体实现步骤.仿真分析验证了算法的有效性,并体现了该算法具有迭代次数非常少的显著优点(3～15次).进一步研究发现了多源信号盲分离的局限性和混合矩阵的病态情况对盲源分离效果的影响.在本文的最后,将所建立的算法应用到转子振动信号的盲分离中,成功地将各个源信号分离出来,表明所建立的算法对转子振动信号的盲源分离具有优良的性能.这为振动源信号的盲源分离打下了基础.

Blind Source Separation Based on Kurtosis with Applications to Rotor Vibration Signal Analysis

Multi-vibration source signals overlapping each other in running engine are difficult to separate by the ordinary methods,which thus obstrusts to exactly estimate vibration sources in running test.According to the discussion of Independent Component Analysis(ICA),a fast fixed-point extract algorithm based on the kurtosis is developed,where the iterative steps get fewer,usually 3 to 15.The morbid condition of mixed matrix and the influence on the separation result is discussed.The effectiveness of the algorithm for rotor vibration signals of engine is confirmed by the BSS test on the test-bed.