Fault diagnosis and classifications of rolling mill bearing-gear based on GADF-TL-ShuffleNet-V2

Abstract

Considering that deep learning could only perform fault diagnosis for a single component at this stage, a deep learning framework based on GADF-TL-ShuffleNet-V2 for mill fault diagnosis was proposed and used for the first time in multi-component fault diagnosis of bearings and gears in a laboratory mill. First, Gramian Angular Difference Field (GADF) was used to transform one-dimensional mill vibration data into two-dimensional image data recognized by convolutional neural network (CNN), and then transfer learning (TL) was used to make the model converge quickly, and to implement and accelerate the training of ShuffleNet-V2. Finally, the effectiveness of the model was verified using the public dataset and laboratory mill bearing and gear vibration data, respectively, and more satisfactory results were achieved. The diagnostic accuracy of the public dataset is close to 100%, and the accuracy of the mill experimental platform dataset is as high as 98.84%. This would provide theoretical and data support for accumulating and forming the fault diagnosis database of rolling mill bearings and gears, and further applying it to the field rolling mill fault diagnosis.