Rolling bearings are critical components in rotating machinery, and their failures may lead to significant economic losses and safety hazards. However, early fault signals are often weak and masked by strong background noise, making accurate fault diagnosis extremely challenging. To address this issue, this paper proposes a fault diagnosis method for rolling bearings based on improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), an autoregressive (AR) model, and observer-based eigenvalue extraction, combined with a particle swarm optimization-based kernel extreme learning machine (PSO-KELM). Targeting rotating machinery with rolling bearings, the approach begins by applying ICEEMDAN as a preprocessing step to decompose non-stationary vibration signals into multiple intrinsic mode functions (IMFs), from which all essential fault-related information is extracted. The preprocessed vibration signal is then reconstructed. Subsequently, an AR model is used to establish a state-space representation for the observer, which processes the reconstructed signal and generates a residual output by comparing it with the actual mechanical signal. Features are then extracted from the residual signal, including its mean, variance, maximum and minimum values, kurtosis, waveform factor, pulse factor, and clearance factor. These features serve as inputs to the PSO-KELM classifier for fault diagnosis. To validate the method, real vibration data from electric motor bearings were employed in a case study, covering normal conditions and three typical fault types: outer race fault, inner race fault, and rolling element fault. The results demonstrate that the proposed method effectively enables fault feature extraction and accurate identification of bearing conditions.
Dong et al. (Tue,) studied this question.