On prognosis of wind turbine faults based on nonlinear mixed vibration signals: A PSO based EMD and KICA combined approach

Combining the advantages of empirical mode decomposition (EMD) and kernel independent component analysis (KICA), an underdetermined blind source separation method of nonlinear mixed wind turbine faults based on EMD-KICA in combination with PSO (Particle Swarm Optimization) is proposed. In the proposed method, the nonlinear mixed signals are firstly decomposed into a set of intrinsic mode function components with EMD, and these components and original observed signals are further combined to construct new observed signals. In this way, the underdetermined blind source separation problem is transformed to over-determined blind source separation problem. The new observed signals are transformed to determined signals through principal component analysis (PCA) and then the kernel independent component analysis method is adopted to carry out the blind source separation of the mixed signals, obtaining estimation of the blind signals. The simulation result verifies the effectiveness of the proposed solution to cope with the blind separation of nonlinear signals. In addition, the result is further confirmed by a case study carried out for fault signals of rolling bearings.