Improved Complete Ensemble Robust Local Mean Decomposition With Adaptive Noise for Slewing Bearings Performance Degradation Assessment

Signal de-noising is one of challenging tasks in the slewing bearing performance degradation assessment due to the strong background noise and early weak fault characteristics. Adaptive digital driven methods like local mean decomposition (LMD) and robust local mean decomposition (RLMD) are a promising approach to implement signal-noise separation. However, the mode mixing problem limits its practical applications. Based on noise-assisted approach, complete ensemble robust local mean decomposition with adaptive noise (CERLMDAN) is used to solve this problem. Then, statistic detection of kernel principle component analysis (KPCA) is employed to select fault components for reconstruction and de-noising. After that, square prediction error (SPE) is utilized for performance degradation assessment model establishment. In the decomposition process of CERLMDAN, moth-flame optimization (MFO) is proposed to optimize the noise amplitude and ensemble trials for the improvement of signal decomposition ability. Meanwhile, a comparison is conducted between ensemble empirical mode decomposition (EEMD), variational mode decomposition (VMD), LMD, RLMD, CERLMDAN and MFO-CERLMDAN. The effectiveness of the proposed method is validated using numerical as well as experimental signals obtained through a slewing bearing highly accelerated life test. The results illustrate that MFO-CERLMDAN gets better result in signal de-noising, and SPE based on MFO-CERLMDAN-KPCA can assess the performance degradation of slewing bearing effectively.