Rolling bearing fault diagnosis based on parameter optimized VMD and improved GoogLeNet

Abstract

ObjectiveThe application of deep learning methods in the field of rolling bearing fault diagnosis is very effective, but traditional neural networks cannot extract features at multiple scales due to the use of a single scale convolution kernel, and do not consider the importance of different features in fault diagnosis. Therefore, it is difficult to extract fault features of rolling bearing signals under noise interference. A rolling bearing fault diagnosis method based on parameter-optimized variational mode decomposition (VMD) noise reduction and GoogLeNet network with improved attention mechanism was proposed.MethodsThe local minimal envelope entropy was used as the fitness function, and the sparrow search algorithm (SSA) was used to optimize the VMD parameter combination; the optimized VMD algorithm was used to decompose the bearing vibration signals to obtain several modal components, and the signals were reconstructed by filtering the modal components with rich fault features according to the envelope entropy and kurtosis; the reconstructed signals were used to construct the feature matrix and input into the improved GoogLeNet network to complete the diagnosis.ResultsThe test results show that the diagnostic accuracy of the method is 95.5% to 99.8% under different noise backgrounds, which is better than other methods in terms of noise robustness.

Keywords

• Rolling bearing
• Variational mode decomposition
• Sparrow search algorithm
• Convolutional neural network
• Fault diagnosis
• Attention mechanism