A novel adaptive weak fault diagnosis method based on modulation periodic stochastic pooling networks

Stochastic resonance, known for its strong capability to amplify weak signals, has been widely applied in rotating machinery fault diagnosis. However, the increasing intelligence of mechanical equipment and the harsh service environment leads to new challenges for stochastic resonance method. Moreover, the adaptive stochastic resonance system relying on the signal-to-noise ratio (SNR) as the loss function requires extensive prior knowledge of the signal to be measured, limiting its application in engineering. Therefore, this paper presents a modulation periodic stochastic pooling networks (MPSPN) with integral modulation factor. By using the normalized least-mean-square (NLMS)algorithm, an adaptive bearing fault diagnosis method based on MPSPN under unknown faults is developed. The study first proposes a modulated periodic stochastic resonance (MPSR) model and investigates its stochastic resonance characteristics through the steady-state probability density. Then, it introduces a modulation signal detection index (IMBF) and derives an adaptive weight allocation scheme under NLMS optimization. Finally, the superiority of the MPSPN system is demonstrated through simulations and the analysis of bearing fault data obtained from two distinct experimental platforms. The results indicate that, in comparison to the conventional periodic stochastic resonance (PSR) system, the MPSPN system is capable of effectively diagnosing unknown faults in bearings and significantly improving the SNR of the diagnostic output.