Incipient bearing fault diagnosis under strong background noise based on enhanced signal adaptive decomposition method

Abstract

Incipient diagnosis of weak faults in bearings is critical for ensuring reliable operation of rotating machinery, yet it remains challenging due to strong background noise and non-stationary characteristics. To overcome this drawback, a novel approach named improved complete ensemble robust local mean decomposition with adaptive noise (ICERLMDAN) is proposed to enhance incipient fault diagnosis. During the process of ICEEMDAN, the noise amplitude and ensemble trials are dynamically adjusted to suppress mode mixing and improve decomposition accuracy, while adaptive noise injection ensures robustness against interference. Subsequently, three fault-related component-selection strategies are employed for the improvement of signal reconstruction. Under the operating condition of strong background noise, component feature-based selection strategy is suitable to quantify the impulsiveness of product functions, prioritizing components rich in fault-related transient features. This dual optimization methodology enables effective isolation of weak fault signatures from complicated vibration signals. Experimental validation on XJTU-SY bearing datasets demonstrates that the proposed method outperforms conventional techniques, such as VMD, EEMD, LMD, RLMD, and CERLMDAN, by achieving higher signal-to-noise ratios and more precise fault frequency identification. The results highlight the superiority of noise resistance and capability to extract subtle fault indicators, making it a promising tool for predictive maintenance in industrial environments.