An improved informative frequency band selection method for early fault detection of rolling element bearings

Detecting rolling element bearing (REB) fault symptoms in real-world industrial settings is quite challenging due to external noise and interference from other components, especially in the early stages of fault. Therefore, enhancing the fault symptoms is crucial in practical applications. REBs usually exhibit two key characteristics: impulsiveness and cyclostationarity. However, most traditional band selection methods, which focus on maximizing these factors, are often ineffective due to impulsive noise and cyclostationary-type signal from other components, such as gears. This limitation arises from two sources: the demodulation technique and the chosen indicator for band selection. To address this limitation, first, an improved envelope spectrum with adaptive thresholding strategy is developed to extract the most pertinent bearing fault signatures. Building on this improved demodulation approach, a novel Multi-Harmonics Energy Index (MHEI) is then introduced to quantify the fault harmonic energy relative to background noise. This index facilitates the identification of the optimal demodulation band containing the most relevant fault information. A genetic algorithm optimization with a gridline-based initial population selection scheme is employed to select this optimal band automatically. The robustness and accuracy of the proposed method are validated through experimental tests incorporating inner race, outer race, roller, and compound faults in the presence of various noise types. The results demonstrate the method’s effectiveness even under challenging conditions with significant non-related impulsive and cyclostationary noise.