A secondary optimization strategy in stochastic resonance modelling for the detection of unknown bearing faults

Early fault diagnosis is a hot topic in the field of fault diagnosis. The collected vibration signals containing weak fault information are difficult to extract fault features due to the presence of strong background noise. Stochastic resonance (SR) is a signal processing method that can utilize noise to improve signal-to-noise ratio. However, SR mostly requires prior knowledge, such as the difficult to obtain bearing fault frequencies. A weighted piecewise bistable stochastic pooling network weak feature detection method based on a secondary optimization strategy is proposed in the paper. In the first layer of optimization, system parameters of each network unit are determined in the process of adaptive fault feature search based on Gini index. In the second layer of optimization, independent and identically distributed Gaussian white noise is added to each unit of the stochastic pooling network to enhance and extract weak signal features, and the unknown bearing fault types can be identified. The proposed method is applied to three different experimental datasets of bearing faults, and the diagnostic results all prove that compared to the single-layer optimization strategy, the proposed method has stronger weak signal enhancement ability and is more helpful for detecting unknown faults.