Adaptive stochastic resonance with period guidance for blind bearing fault detection

To address the limitations of stochastic resonance (SR) methods based on signal-to-noise ratio (SNR) in blind fault detection of rolling bearings, a periodicity-guided adaptive SR method is proposed. First, an adaptive weighted squared envelope correlation sparse deconvolution method is developed to estimate the period of weak impulsive features in the presence of strong noise by integrating an adaptive period estimation strategy with a weighted squared envelope correlation-based sparse iterative framework. Subsequently, a time-delay feedback underdamped SR system is constructed, employing an asymmetric potential function with independently tunable potential wells. The introduction of time delay feedback further regulates the SR effect and improves the signal amplification capability. The dynamical characteristics of the system and the influence of key parameters on SR performance are analyzed through numerical simulations. The estimated period is then used to adaptively optimize the SR system parameters, forming an unknown fault detection framework. Experimental results on bearing datasets demonstrate that the proposed method achieves blind fault detection without requiring prior signal knowledge, while effectively extract weak fault features under strong noise conditions and delivering superior detection performance. The framework consistently improves the output SNR across different datasets, achieving values of 3.51, 3.39, and −2.03 dB, respectively, indicating its effectiveness in periodicity estimation and weak feature enhancement.