As a key equipment in the underground coal mining transportation system, the scraper conveyor's bearings operate for extended periods under extreme conditions such as high temperature, high impact loads, dust pollution, and continuous operation, resulting in a high risk of failure. Existing fault diagnosis methods still face numerous challenges in feature extraction and classification accuracy when dealing with complex working conditions and multiple failure modes. To address this, a bearing fault diagnosis method based on a dual-branch convolutional neural network was proposed. Firstly, one-dimensional vibration signals were converted into two-dimensional time-frequency images using continuous wavelet transform to enhance the time-frequency feature representation capability of the input data. Subsequently, a dual-branch convolutional neural network architecture was constructed, in which one branch employed a dilated convolutional residual module to strengthen local feature extraction, while the other branch introduced a global attention mechanism to optimize global feature learning, and feature fusion was applied to improve the model's diagnostic capability. Finally, experimental validation was conducted on the CWRU and PU bearing datasets. The results show that the proposed method can achieve fault diagnosis accuracy rates of 99.78% and 97.94% on the two datasets, respectively, and outperforms comparison models such as DCRB-CNN, GAM-CNN and GAM-ResNet in overall performance, demonstrating significant diagnostic capability and strong generalization ability. Moreover, bearing fault diagnosis experiments were carried out using real-world coal mine field data, further validating the method's practicality and potential for engineering deployment under strong noise and non-stationary operating conditions. The research results provide an efficient and reliable deep learning solution for the intelligent fault diagnosis of complex coal mining equipment such as scraper conveyors.

Rolling bearing  /  Fault diagnosis  /  Neural network  /  Attention mechanism