A framework for fault diagnosis and classifications of rolling bearings and gears in rolling mills based on RGB principal component analysis with ShuffleNetV2 introducing dual attention module

Abstract Considering that previous deep learning-based fault diagnosis methods on rolling mills target a single component and the expressive power of the information is limited, resulting in low recognition accuracy. To address these issues, a ShuffleNetV2 based on RGB principal component analysis (RPCA) with the introduction of a dual attention module is proposed as a fault diagnosis and classifications framework for rolling bearings and gears in rolling mills. Firstly, the original one-dimensional vibration signal is converted into a two-dimensional picture by Gramian Angular Difference Fields (GADF). The picture data is downscaled as input by RGB principal component analysis (RPCA). Then the Convolutional Block Attention Module (CBAM) is introduced into the model to enhance the features. Transfer Learning (TL) is inserted to speed up the convergence of the model. The experimental results show that the effect of both rolling bearings and gears classification on the mill dataset is significant, reaching 98.77% and 98.47%, and the effect of both bearing and gear classification on the public dataset is superior to the previous model, respectively, which verifies the feasibility and validity of the proposed framework for the diagnosis of rolling bearings and gears faults of rolling mills.