Kurtosis-Based Constrained Independent Component Analysis and Its Application on Source Contribution Quantitative Estimation

Abstract

Aiming at finding the major vibration and noise sources of vehicles, a quantitative estimation method for source contribution using the kurtosis-based constrained independent component analysis (cICA) algorithm is proposed. First, the similarity between the ICs and the reference signals with given characteristics is described by a concise and effective closeness measurement function. Meanwhile, how to choose the reference signals and the choice of some other closeness measurements is discussed. Then, a widely used contrast function, namely, kurtosis, is modified by the closeness measurement to obtain an enhanced contrast function. The fixed-point iteration and deflation approach are employed to train the separating matrix. Then, the enhanced contrast function is therefore maximized and the kurtosis-based cICA algorithm is obtained. After that, the source contribution is quantitatively calculated by the reduced energy of the mixed signals in each extraction: the reduction of the energy in mixed signals corresponds to the contribution of the extracted IC. The correspondence relationship between the ICs and source signals can be obtained by prior knowledge. Finally, the effectiveness of the proposed algorithm is verified by numerical simulation and experiments. The results show that the proposed method has high accuracy in separating sources and quantitatively calculating the source contribution.