Combined model: CEEMDAN and AVMD quadratic decomposition noise reduction and BiLSTM-BP classification model rolling bearing fault diagnosis

This paper proposes a noise reduction method based on CEEMDAN-IWSO-VMD, which enhances the feature extraction of rolling bearing fault signals and reduces noise interference. The process begins with complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), followed by the selection of IMFs using a fusion index combining adaptive weighted sample entropy (AWSE) and correlation coefficient (CC) to retain the fault-related components. The selected signals are further processed by VMD and their parameters are optimized using the improved white shark optimization (IWSO) algorithm to suppress residual noise. Then, statistical features are extracted and input into the BiLSTM network, while IWSO also optimizes the parameters of the BP neural network to improve classification efficiency. Experiments on the CWRU and HUST datasets show that compared with traditional methods, this method effectively enhances noise robustness and achieves higher diagnostic accuracy.