Hierarchical weighted resonant separation: a new approach for bearing weak fault diagnosis under strong noise interference

Abstract

Diagnosing weak faults in rolling bearings under strong noise interference remains a prominent challenge in the field of condition monitoring. This paper proposes a novel hierarchical weighted resonant separation (HWRS) framework, aiming to achieve direct separation and diagnosis of fault source signals under strong noise conditions. The framework constructs a dual-layer ‘weighting-separation’ architecture: in the weighting layer, cyclic frequency identification and an adaptive weighting strategy are employed to lock in and enhance periodic impulse components; in the separation layer, a weighted rank-1 constraint and a physical resonance mechanism are introduced to achieve physically interpretable separation of fault sources. Furthermore, a comprehensive evaluation system integrating qualitative and quantitative metrics was established to systematically assess performance. Simulation results indicate that the HWRS method consistently achieves precise separation of fault impulse waveforms when the signal-to-noise ratio is above −18 dB. Moreover, experimental validation based on the BJTU-RAO strong-noise dataset and the PU real compound fault dataset demonstrates that the proposed method can successfully separate and diagnose both single and compound fault features from deeply buried raw signals. The HWRS framework proposed in this study provides a new approach for the direct and interpretable separation and diagnosis of weak bearing faults in strong noise environments, demonstrating significant potential for engineering applications.