Kernel Independent Component Analysis and its Application in Blind Separation of Mechanical Faults

Abstract

A nonlinear blind separation method of mechanical fault sources is proposed. In the proposed method, the signal is transformed from the low-dimensional nonlinear original space into a high-dimensional linear feature space by the kernel function, so that nonlinear mixture mechanical fault sources can be separated by the linear ICA method in a new feature space. The simulation result shows that the proposed method is superior to the traditional ICA method in processing nonlinear blind separation problem. Finally the proposed method is applied to the nonlinear blind separation of bearing faults. The experiment result further verifies the validity of the proposed method.