Multivariate variational mode decomposition and 1D residual neural network for subtle feature recognition of rolling bearings

Rolling bearings are the critical components of rotating mechanical equipment, and it is more important to fault diagnosis and recognition of the rolling bearings. Multivariate variational mode decomposition (MVMD) with one dimensional residual network (1D ResNet) is proposed to fault diagnosis and subtle feature recognition of the rolling bearings. Intrinsic modal components are extracted to further signal process under different operational conditions to the segmentation of signal components and the feature reconstruction. The average success accuracy rate for the ten types of rolling bearing faults (normal, ball fault, inner race fault, outer race fault with different damage degree) exceeds 99.32 %. MVMD-1D ResNet with the advantage of fault recognition of rolling bearings is validated by comparing to other algorithms (1D ResNet, 1D CNN and KNN). MVMD-1D ResNet model has great potential to condition monitoring and subtle feature recognition with limited sample sizes of the rolling bearings.