Hierarchical Gray Wolf Optimizer-Tuned Flexible Residual Neural Network With Parallel Attention Module for Bearing Fault Diagnosis

Rolling bearings are vital components of rotating machinery, and their regular operation directly affects the machine lifespan and operating status. Aimed at improving the accuracy of fault diagnosis for rolling bearings, a hierarchical gray wolf optimizer (HGWO)-tuned flexible residual neural network (FResNet) with parallel attention module (PAM) is proposed. Specifically, a CNN-based flexible residual module is designed to form the FResNet, which allows changing the numbers of convolution layers and convolution kernels as an optimizer iterates. Optimal model structure and parameters are configured by the HGWO with a nonlinear convergence factor and hierarchical position update strategy. On the other hand, the PAM with convolutional layers is designed to fuse the output weights of channel and spatial attention. As a result, the integration of HGWO, PAM, and FResNet forms an effective model for bearing fault diagnosis, named HGWO-PAM-FResNet. Finally, the viability and efficacy of the proposed HGWO-PAM-FResNet model are verified using the bearing fault dataset from Case Western Reserve University (CWRU), and the higher accuracy of the proposed model compared to other intelligent models is demonstrated under different noise and variable load conditions.