Rolling Bearing Fault Diagnosis Based on Improved Variational Mode Decomposition Combined with CNN-BiLSTM

To address the issue that the rolling bearing signal is highly susceptible to noise disturbing that poses challengs to identify weak fault features, this paper proposed a rolling bearing fault diagnosis method based on the combination of Gray Wolf optimized variational mode decomposition (GWO-VMD) and convolutional neural network integrated with a bidirectional long short-term memory network (CNN-BiLSTM). Firstly, Grey Wolf optimizer (GWO) was employed to optimize VMD parameters to realize the adaptive selection of two important parameters containing the number of decomposition layers and the secondary penalty factor. Secondly, the maximal kurtosis criterion was used to select the most optimial IMF of GWO-VMD decomposition. Subsequently, the features of the optimal IMF component were extracted and input into the CNN-BiLSTM network for training. The fault diagnosis experiments were conducted using he rolling bearing fault data set. Corresponding results indicate that the proposed method ahieves a 99.33% fault diagnosis accuracy, and a comparative analysis with the PSO-VMD-CNN-BiLSTM and VMD-CNN-BiLSTM methodes further demonstrates the superiority of this propsed strategy in fault diagnosis accuracy.