Abstract: Aiming at the problem that the early fault signal of rolling bearing was very weak and the fault feature extraction was difficult, the frequency domain decomposition, signal reconstruction, noise reduction and unmixing of the early bearing fault signal were studied. A combined noise reduction method based on improved complete empirical mode decomposition (ICEEMD) and efficient fast independent component analysis (EFICA) were proposed. Firstly, ICEEMD algorithm was used to decompose the vibration signal to obtain a series of Intrinsic Mode Function (IMF). Secondly, according to the kurtosis criterion, the corresponding components were selected to complete the reconstruction of the virtual channel signal and the vibration shock signal. Finally, EFICA was used to perform blind source separation on the reconstructed signal, which effectively reduced the noise in the vibration signal, maximized the energy amplitude of the fault frequency, and facilitated the identification of fault characteristics; the experimental research on early failure of actual rolling bearings was designed. The results indicate that the method can obviously suppress the interference of noise and modulation frequency components, and highlight the characteristic frequency components of the fault. Comparing with the combined method of CEEMD and EFICA, the signal-to-noise ratio increases by 24.76%, which can identify fault information more accurately and meet the requirements for bearing diagnosis.

Key words: rolling bearing; ICEEMD; EFICA; fault diagnosis; combined noise reduction