Intelligent fault diagnosis of rolling bearings under imbalanced data conditions using attention-based deep learning method

Data-driven intelligent method has been widely used in fault diagnostics. However, it is observed that previous research studies focusing on imbalanced datasets for fault diagnosis have a limitation, that is, the number of normal and fault samples is assumed to be same or similar in the diagnosis process. This hypothesis decreases the accuracy and stability of fault diagnosis model for imbalanced datasets in practical working conditions. In this paper, a rolling bearing fault diagnosis model which combines Dual-stage Attention-based Recurrent Neural Network (DA-RNN) and Convolutional Block Attention Module (CBAM) is proposed. Firstly, the DA-RNN model is used to extend imbalanced datasets in real fault diagnosis cases. Secondly, an image processing method is designed to convert vibration signal into image by using vibration acceleration signal and its corresponding integrated velocity and displacement signals. Finally, the Convolution Neural Network (CNN) model with embedded CBAM structure is used for fault classification. Two datasets of vibration data from rolling bearings are used to evaluate the performance of the proposed methodology for fault diagnosis. Results show that the proposed DARNN-CBAM-CNN method improved fault diagnosis accuracy in 10.90%, 7.56%, 2.73% and 1.90% the performance metrics compared to a neural network based method, a machine learning based method, a deep learning based method, and a DARNN-CNN based method without using CBAM when the imbalance ratio of the dataset is 100:50. Diagnosis accuracy results of datasets with four different imbalance ratios show that the proposed method has the best performance compared to other six intelligent fault diagnosis methods, indicating that the proposed method is a promising potential for rolling bearings under imbalanced data conditions.