Application of QGA-MCKD and stochastic feedback pooling network in variable-condition bearing diagnostics

Stochastic resonance (SR) has gained significant application in bearing diagnostics due to its ability to amplify weak signals through noise. However, existing SR methods have limitations in effectively enhancing diagnostic performance. To address these challenges, this paper proposes an improved QGA-MCKD-SFPN method, combining Quantum Genetic Algorithm (QGA) with Maximum Correlated Kurtosis Deconvolution (MCKD) and a Stochastic Feedback Pooling Network (SFPN) to enhance system performance. The main contributions include: (1) introducing a double Gaussian potential function constructed by combining power and Gaussian potential functions, facilitating adjustment of the system's potential landscape and alleviating output saturation by expanding particle movement range; (2) preprocessing fault signals with MCKD to enhance fault characteristics and employing QGA to optimize MCKD filter parameters; and (3) incorporating a feedback term into a traditional Stochastic Pooling Network (SPN) to construct the SFPN system, which improves particle dynamics and diagnostic accuracy. Validation using Mechanical Failures Prevention Technology (MFPT) and PADERBORN bearing fault datasets demonstrates that the proposed method consistently outperforms conventional SPN systems under various operating conditions.