Vibration resonance-assisted stochastic resonance for incipient bearing fault detection

Rolling bearings are critical components in rotating machinery, and timely fault monitoring is essential to prevent catastrophic failure. However, impulses from incipient faults are often submerged in heavy background noise, making it difficult to identify weak signatures reliably. To address this, this paper proposes a novel adaptive stochastic resonance (SR) framework that integrates vibration resonance (VR) modulation with a cascaded dual-feedback architecture. To overcome the limitations of fixed SR configurations, an adaptive preprocessing strategy based on non-negative matrix factorization is introduced. The decomposition rank is determined automatically via cross-validation, while the most informative components are selected using a spectral kurtosis-entropy index to extract features of early faults. The nonlinear dynamical response of the proposed VR-assisted SR system is investigated through numerical simulations using the fourth-order Runge-Kutta scheme, with system parameters optimized by a quantum genetic algorithm. Experimental validation on real vibration signals from a rotating machinery test rig demonstrates that the proposed approach achieves superior enhancement of fault signatures and signal-to-noise ratio improvement compared to conventional SR systems and the FK method. These findings indicate that coupling adaptive preprocessing with vibration-assisted SR provides an effective framework for detecting incipient bearing faults.