Incipient rolling element bearing weak fault feature extraction based on adaptive second-order stochastic resonance incorporated by mode decomposition

Incipient bearing fault characteristic is extremely weak and interfered by strong noise, which makes the early fault warning work very difficult. Considering traditional characteristic extraction methods cannot identify the fault frequency effectively, a method is proposed in this paper based on the cooperation of complete ensemble EMD with adaptive noise (CEEMDAN) and improved adaptive underdamped stochastic resonance (AUSR). Specifically, the principles and shortcomings of classical mode decomposition methods EMD, EEMD and CEEMD are briefly introduced first. Aiming at these shortcomings, CEEMDAN is adopted to decompose target signal for the extraction of sensitive IMF. Then, a more general theoretical analysis of USR is conducted by taking damping factor into account. Furthermore, an AUSR method is proposed based on GA. Both the superiority of CEEMDAN compared with other mode decomposition methods and the effectiveness of proposed overall analysis scheme are demonstrated by different cases of simulation analysis. Subsequently, the proposed method is further applied on two cases of experimental signals for bearing weak fault characteristic frequency enhancement and extraction. The analyzed results show that the characteristic frequency can be significantly enhanced with the help of proposed method, which further demonstrates its effectiveness and superiority in engineering application.