Research on bearing fault diagnosis based on secondary decomposition and stochastic configuration network

This To address the issue of insufficient feature extraction and low fault diagnosis accuracy in bearing vibration signals during fault occurrence, a bearing fault diagnosis method based on secondary decomposition and SCN is proposed. First, in the bearing signal decomposition and synthesis phase, the bearing vibration signal sequence is decomposed using the CEEMDAN. Then, the sequence with the maximum sample entropy value is decomposed using a hybrid optimization algorithm, combining PSO and GA, to optimize the VMD method. IMFs free of noise after decomposition are selected using an effective weighted kurtosis index and then reassembled into a new signal. Next, time-domain, frequency-domain, entropy, and hybrid features are extracted from the decomposed signal. Finally, during the bearing fault mode recognition phase, a vector-weighted averaging algorithm is introduced to optimize the key parameters of the Stochastic Configured Network, while integrating ensemble learning and regularization optimization strategies to improve the SCN. An ensemble INFO-RSCN model is proposed. The proposed algorithm and model are applied to experimental analysis of bearing fault signals, demonstrating their effectiveness and feasibility.