A Fault Diagnosis Method for Rolling Bearings Based on CNN-Attention Time-Frequency Fusion

Conventional approaches for fault diagnosis in rolling bearings mainly concentrate on either time-domain or frequency-domain traits. This restricts the capacity to extract all-around features and undermines the accuracy of the model's diagnostic performance. To tackle this constraint, this current research puts forward an innovative fault detection technique for rolling bearings that incorporates CNN-Attention time-frequency integration. In this method, the Fast Fourier Transform (FFT) is employed to glean the frequency-domain signal from the time-series data. The original features from both the time and frequency domains are processed separately via Convolutional Neural Networks (CNN) and a Multi-Head Attention unit. These retrieved features are then joined together and fed into a fully connected layer for classification. To evaluate the efficacy of the proposed method, experiments were conducted using the Case Western Reserve University (CWRU) rolling bearing data collection. The experimental results disclose that the method attains an excellent accuracy of 99.13%. Compared with various existing methods, the proposed approach demonstrates a remarkable improvement in classification precision.