Rolling bearing fault diagnosis based on optimized VMD and CNN-BiLSTM

A parameter optimization-based variational modal decomposition-convolutional neural network-bidirectional long and short-term memory (VMD-CNN-BiLSTM) fault diagnosis model is proposed for the problem of three-phase motor bearing fault multi-level classification and identification. Firstly, the penalty factor and modal components of the variational mode decomposition (VMD) are optimized using the mixed-strategy optimization-based crayfish optimization algorithm (MSCOA); then, the time and spectrum features of the signals are extracted using the Convolutional Neural Network (CNN) and the features are fused; finally, the sequential modes of different faults are learned using the bi-directional long short-term memory network (BiLSTM) that accomplishes the task of multilevel classification of faults. It transpires that the VMD-CNN-BiLSTM model optimized based on MSCOA algorithm has obvious advantages in multi-level bearing fault identification when compared with CNN-LSTM, CNN-BiLSTM and VMD-CNN-LSTM models. The average recognition accuracy can reach 94.6667%, indicating that the model has excellent fault diagnosis performance, fast computation speed and excellent generalization ability, which provides an important application value for achieving efficient and reliable motor operation.