Application of Deep Convolutional Network with Integrated Bilinear Pooling and Attention Mechanism in Bearing Fault Diagnosis

Bearing fault diagnosis is one of the most important tasks in machinery maintenance by industry, whereas some problems still exist relating to complex signal processing and model explanation. This paper presents a newly proposed deep bilinear convolutional neural network with attention, namely DBCNN-BA, for the solution. This is the model of DBCNN-BA with a multi-branch CNN structure integrating bilinear pooling and spatial-channel dual-attention mechanism. These experiments on the CWRU bearing fault dataset reveal that the DBCNN-BA model significantly outperforms traditional 1DCNN and 2DCNN approaches, having reached an accuracy of 99.04%, higher by 22.08 and 6.80 percentage points, respectively. In addition, the performance of this model is much more stable for different fault classes with good convergence characteristics. Furthermore, analytical tools such as the confusion matrix and t-SNE visuals confirm its efficiency in differentiating various classes. This is further enhancing the intelligence of fault diagnosis in industrial settings by developing models that are more accurate and robust, hence promising better results for real-world deployment. Further research will be undertaken in different industrial environments, integrating the model with transfer learning into cross-domain applications in the future.