A Learnable Convolutional Method With Wavelet Transform and Hybrid Attention Mechanism for Rolling Bearing Fault Diagnosis

Accurate diagnosis of rolling bearing faults is essential for ensuring the dependable operation and steady performance of rotating equipment. This study introduces an intelligent fault diagnosis approach that integrates convolutional neural networks, convolutional block attention modules, and global attention mechanisms, combined with wavelet transform and signal-to-noise ratio processing to achieve effective feature extraction. The proposed approach employs a learnable Conv1D layer to enhance feature extraction capability, utilizes the convolutional block attention module to adaptively focus on important channel and spatial features, and integrates an additional attention mechanism to further strengthen global information interaction. The bearing dataset from Case Western Reserve University was used to conduct experiments, which revealed that the suggested approach greatly outperforms traditional methods in classification accuracy. Additionally, t-distributed stochastic neighbor embedding visualization analysis further validates the capability of the model to distinguish different fault categories.