Answers are grounded in this paper and cite their sources.
In recent years, deep learning-based fault diagnosis methods have been widely applied in rolling bearing fault classification and achieved good results. However, many existing methods still struggle with insufficient feature learning capability under varying operating conditions, leading to reduced fault recognition accuracy. Herein, this paper proposes a fault diagnosis method for rolling bearings under variable working conditions, integrating an improved activation function with a Multistage Residual Connection Network incorporating SE Attention (MRCN-SE). First, we modify the traditional ReLU function and propose a Feature-Enhanced ReLU (FEReLU) activation function, which retains responses within a limited negative-input interval and increases the gain for positive inputs to modify feature propagation. Second, multiple residual connections are employed to fully fuse deep and shallow features, boosting the network’s fault feature extraction capability and mitigating network degradation. Then, SE attention is introduced into the model to focus on more critical fault features. Finally, the proposed method is tested by using multiple bearing fault datasets. The experimental results show that the configurations using FEReLU achieve higher classification accuracies than those using the selected traditional activation functions. The proposed model obtains average accuracies of 99.39% in the six CWRU fixed-load transfer tasks and 99.23% in the PU noise-transfer tasks based on operating condition 1, showing its classification performance under the tested load conditions and Gaussian noise levels.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: