Application of AVMD and high-dimensional bidirectionally coupled feedback multistable stochastic resonance for bearing fault diagnosis

Stochastic Resonance (SR) is a phenomenon where noise is harnessed to enhance a weak signal, a process fundamentally characterized by the transfer of energy from the background noise to the characteristic component of the signal. Owing to this intrinsic characteristic, SR is rendered exceptionally suitable for the detection of faint signals, finding extensive utility in the diagnostics of rolling bearing faults. However, existing SR based methods still face significant limitations in improving diagnostic performance. Building upon the theoretical foundation of the high-dimensional bidirectionally coupled FitzHugh–Nagumo (HDBCFHN) neuron model, this work formulates a High-dimensional Bidirectionally Coupled Feedback Multistable Stochastic Resonance (HBCFMSR) system, engineered to enhance the practical utility of multidimensional coupled stochastic resonance in real world contexts. To achieve superior system performance, a quantum genetic algorithm (QGA) is implemented for the precise optimization of the model’s parameters. The principal innovations of this research encompass:(1) Each neuron node in the FitzHugh–Nagumo model is replaced with a multistable underdamped SR system to enhance particle dynamics and diagnostic accuracy.(2) An unsaturated symmetric multistable potential function is designed by integrating a periodic cosine term with a decay function, expanding particle motion and mitigating output saturation.(3) Adaptive variational mode decomposition (AVMD) is utilized to preprocess fault signals, facilitating more effective extraction of fault-related features.(4) A feedback mechanism is incorporated into the high-dimensional bidirectionally coupled multistable stochastic resonance system (HBCMSR), further optimizing particle dynamics and diagnostic precision. A high-dimensional bidirectional coupled feedback multistable stochastic resonance system (HBCFMSR) has been formed. The efficacy and robustness of the proposed methodology were empirically substantiated through comprehensive experiments on the Mechanical Fault Prevention Technology (MFPT) and Paderborn datasets. The results confirm its outperformance across a range of varying operational conditions.