A novel stochastic resonance model based on bistable stochastic pooling network and its application

Analysing the vibration and sound signals of machine components is the primary approach for machine condition monitoring and fault diagnosis. However, due to the special working operating conditions of rotating machinery, the collected signals often contain strong noise components generated by other parts of the machine and harsh environment. These noises severely affect the analysis and processing of the target signal. Stochastic resonance (SR) is an effective technique to extract and enhance periodic or ape-riodic signals submerged in noise. Consequently, SR has been widely used for fault diagnosis of rotating machinery. In this study, a bistable stochastic pooling network (BSPN) model based on the traditional SR model is proposed to improve the efficiency of weak fault diagnosis. The least mean square algorithm is used to perform linear weighted optimization on the output vector of random noise-optimized BSPN. At the same time, the optimal weight vector of the random stochastic pooling networks with any number of nodes is obtained. Subsequently, analog signals are used to examine the output signal-to-noise ratio (SNR) of the BSPN. Finally, the efficacy of BSPN system is validated through bearing data collected by two different experimental systems. The experimental results indicate that ordinary array system cannot avoid frequency conversion interference, so it is unable to extract extremely weak fault signals. On the contrary, the BSPN system can accurately detect the weak. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

The study presents a BSPN model based on the traditional SR model to improve the efficiency of weak fault diagnosis, optimizing optimal weight vector through linear weighted optimization on random noise-optimized BSPN output vector, and validating the efficacy of the BSPN system through bearing data collected by two different experimental systems.