Application of the CEEMDAN and stochastic resonance based unsaturated multi-stable symmetric stochastic feedback pooling network in bearing fault diagnosis

Stochastic resonance (SR) is increasingly vital in bearing diagnostics for its capacity to enhance faint signals via noise exploitation, effectively converting noise into usable signal power. Nonetheless, current SR techniques display significant shortcomings in improving diagnostic capabilities. To overcome these obstacles, this study introduces an enhanced CEEMDAN-UMSSR-SFPN system, which incorporates the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Stochastic Feedback Pooling Network (SFPN) via optimized this setup using a Quantum Genetic Algorithm (QGA) to optimize the system parameters, leading to significant improvements. The key contributions of this study encompass the following: (1) A novel unsaturated multi-stationary symmetric potential function (UMSSR) has been introduced. This function merges sinusoidal and exponential components to broaden the scope of particle movement and effectively tackles the issue of output saturation; (2) The preprocessing method used by CEEMDAN for fault detection further enhances its applicability when processing complex non-stationary fault signals, ensuring stable performance under different operating conditions; and (3) The proposed method is incorporated into the Stochastic Feedback Pooling Network (SFPN) to develop a CEEMDAN-UMSSR-SFPN system, aiming to improve the accuracy of particle dynamics analysis and fault diagnosis. Experimental validation using the PADERBORN and Mechanical Failures Prevention Technology (MFPT) bearing fault datasets verify that the proposed system consistently delivers superior diagnostic performance across diverse operating conditions.