A rolling bearing fault diagnosis method based on an improved parallel one-dimensional convolutional neural network

As a critical component of industrial equipment, the fault diagnosis of rolling bearings is essential for reducing unplanned downtime and improving equipment reliability. Existing methods achieve an accuracy of no more than 92% in low signal-to-noise ratio environments. To address this issue, this paper proposes an improved parallel one-dimensional convolutional neural network model, which integrates a parallel dual-channel convolutional kernel, a gated recurrent unit, and an attention mechanism. The classification is performed using a global max-pooling layer followed by a Softmax layer. This dual-channel configuration captures both global and local features, decreases parameter redundancy, and reduces overfitting risk. Meanwhile, the GRU addresses the vanishing gradient issue and models long-term dependencies. Additionally, the attention mechanism emphasizes crucial features dynamically, improving feature selection and generalization. The global max-pooling layer replaces the fully connected layer, reducing the number of parameters, improving computational efficiency, and lowering the risk of overfitting. Experimental results demonstrate that the proposed model achieves superior performance in fault diagnosis, attaining an accuracy of 99.62%, significantly outperforming traditional CNNs and other benchmark methods.