Fault diagnosis of rolling bearings based on weighted reconstruction and refined composite multiscale distance similarity entropy under time-varying operating conditions

Aiming at the problem of feature masking caused by the strong coupling between noise and periodic fault components when rolling bearings operate under complex time-varying conditions, this paper proposes a digital twin driven fault diagnosis method based on adaptive segmented weighted reconstruction and refined composite multiscale distance similarity entropy (RCMDSEn). First, a dynamic model of rolling bearings is established to generate typical fault characteristic signals under different operating conditions. Second, to address the limitation that fixed weights fail to adapt to time-varying processes in traditional intrinsic mode function (IMF) reconstruction, an adaptive threshold evaluation criterion is constructed. By dynamically adjusting the contribution proportions of the ratio of periodic modulation components caused by fault to the generalized interferences (PMGI) and kurtosis, a PMGIK fusion strategy is realized to achieve segmented optimization under variable rotational speeds. Finally, the improved RCMDSEn is utilized to quantify the multiscale features of the reconstructed signals, and Bayesian optimized extreme learning machine (Bayes-ELM) is integrated to realize fault pattern recognition. The effectiveness of the proposed method is validated using the Case Western Reserve University dataset and measured data. The results demonstrate that the developed digital twin and noise-aware diagnostic framework, integrating physics-guided virtual signal generation, adaptive physical-virtual denoising, and multiscale complexity quantification, effectively mitigates feature masking issues under time-varying operating conditions.