The separation method of nonlinear blind source for mechanical faults based on kernel parallel factor

The existing nonlinear blind source separation methods of multi-fault completely rely on matrix decomposition, however matrix decomposition needs to meet strict constraints to ensure the uniqueness of the decomposition. Based on the above deficiency, the parallel factor analysis (PARAFAC) is introduced into the blind source separation for mechanical faults, and combined with the high-dimensional mapping function of Kernel function. A nonlinear blind source separation method for mechanical faults based on the kernel parallel factor (KPARAFAC) is proposed. The proposed method makes full use of the advantage of the uniqueness of PARAFAC decomposition and non-linear mapping of the kernel function, and overcomes the shortcomings of the traditional nonlinear blind source method for mechanical faults. In the proposed method, the kernel function is used to map the nonlinear observation signal to the high-dimensional kernel feature space, and the P ARAF AC is used to blindly separate the new observation signal in the linear kernel feature space. The simulation results show that the proposed method is very effective, the proposed method is obviously superior to the traditional nonlinear blind source separation method for mechanical faults. Finally, the proposed method has been successfully applied to the blind separation of multiple faults of rolling bearing, and the experimental results further verify the effectiveness of the proposed method. The proposed method can effectively separate the characteristic frequency of each fault in the multi-fault signals of rolling bearing.