Rolling Bearing Fault Detection Based on Improved Piecewise Unsaturated Bistable Stochastic Resonance Method

Abstract

Stochastic resonance (SR) has been widely studied and used for the fault detection of rolling bearings due to its good ability to detect weak signals. However, the output saturation of the SR system restrains its enhanced extraction performance of weak signal features to some extent. Therefore, this article proposes an improved piecewise unsaturated bistable SR method, which can effectively improve the ability to detect weak signals and is applied to rolling bearing fault detection. An improved piecewise unsaturated bistable potential model is constructed, which can independently adjust the structure of the potential model, the segmented points, and the slope of the potential walls, thereby better alleviating the output saturation and obtaining a greater output signal-to-noise ratio (SNR). Meanwhile, the constructed modified SNR index and particle swarm optimization (PSO) algorithm are used to automatically choose the optimal parameters of the proposed SR system. Finally, the performance of the proposed SR method is analyzed with the help of the rolling bearing vibration data. By comparing with some improved bistable SR methods, the proposed SR method has better extraction ability for weak periodic signals and can obtain a higher spectral peak at the fault frequency.