Fault Diagnosis of Rolling Bearing Based on Parameter Optimized VMD and CNN-BiLSTM

The rolling bearing fault vibration signal contains multiple frequency components, and the diagnosis recognition rate is low due to the lack of detailed fault signal decomposition and inaccurate fault feature screening. A feature extraction method based on parameter optimization VMD and composite indicator WEP is proposed, and a CNN-BiLSTM network is used for fault diagnosis. First, in order to overcome the shortcomings of the variational modal decomposition (VMD) which needs to rely on manual experience to set the key parameters, an adaptation function based on the correlation coefficient and mutual information is proposed, and the crested porcupine optimizer (CPO) is used to determine the optimal parameter combinations of the VMD, and the optimal parameter decomposition of the original vibration signals is used to obtain the intrinsic modal function (IMF); then, the weighted envelope spectral peak factor (WEP) composite screening index is constructed by combining fuzzy entropy and correlation coefficient, and the modal components with fault characteristic information are screened according to the index; finally, the screened fault IMFs are inputted into the convolutional bi-directional long and short-term memory (CNNBiLSTM) network as the feature vectors to classify and recognize the faults. Simulation experiments show that the bearing fault diagnosis method proposed in this paper can distinguish fault categories more effectively.