Rolling Bearing Fault Diagnosis Based on HO-VMD and CNN

The objective of this paper is to discuss a series of problems in traditional methods, such as the nonlinearity and non-smoothness of the vibration signal of a rolling bearing, inaccurate extraction of fault features, and poor generalization ability when it faces a large dataset. It proposes a fault diagnostic model based on a Convolutional Neural Network (CNN) along with a Hippopotamus Optimization Algorithm (HO) optimized Variational Modal Decomposition (VMD) method. In this paper, the Hippopotamus Optimization Algorithm was adopted to optimize mode number (k) and regularization coefficient (a) of the VMD for enhancing accuracy and improving efficiency, in order to make variational modal decomposition adapt to needs of different signal processing better. Then, the extracted vibration signals are decomposed into a series of intrinsic mode functions by using VMD and selecting optimal mode components according to the minimum envelope entropy principle. This last step includes sets of sample features that feed into the Convolutional Neural Network, which will identify the types of faults. In this paper, compared with the traditional VMD-CNN fault diagnosis model, the VMD-CNN model optimized by Hippopotamus Optimization Algorithm was used, and then, it can be found that the prediction accuracy of this method reaches 97.4 % and 95.3% on the training set and the test set, respectively, which are well higher than that of the traditional VMD-CNN fault diagnosis model.