Fault Diagnosis Method for Bearing Based on Attention Mechanism and Multi-Scale Convolutional Neural Network

Abstract

Convolutional neural networks (CNNs) serve as powerful feature extraction tools capable of effectively extracting information from complex environments, thus improving the accuracy of fault identification for bearing data. In this paper, we present a method for diagnosing bearing faults using an attention mechanism and a multi-scale convolutional neural network (MSCNN). Firstly, truncate and sample the rolling bearing data, and use continuous wavelet transform to generate corresponding time-frequency images, which will be used as inputs to the neural network. Next, the MSCNN, which includes efficient convolutional modules with residual structures, is utilized to extract features from the input data while maximizing the retention of valuable information. The extracted data then undergoes feature selection through the employment of an Efficient Convolutional Module (ECM) with channel attention. Finally, after being mapped through fully connected layers, the features are fed into a softmax layer for fault category prediction. In this study, the model results were tested and verified using the Case Western Reserve University (CWRU) dataset and the bearing dataset of Jiangnan University(JNU). A comparison was made with the LeNet model, ResNet model, LSTM model, and WDCNN model. The results showed that the classification accuracy of the ten types of bearing signals at the same speed can reach 100%, and the classification accuracy of the thirty types of bearing signals at different speeds can reach over 99.4%, significantly higher than the other models. The proposed method achieves the recognition of different fault states of rolling bearings under complex conditions, including multiple operating conditions and variable operating conditions. It is capable of extracting the global characteristic information of bearing faults, resulting in high diagnostic accuracy and good generalization ability. This method can provide reference for the diagnosis of rolling bearing faults under corresponding operating conditions.

Fault diagnosis of rolling bearings based on dynamic convolution and dual-channel feature fusion under variable working conditions

“…To better demonstrate the outstanding advantages of this paper's model in the realm of fault diagnosis under varying operating conditions, we conducted a comprehensive comparison with several network models that classify image and fault diagnosis fields, such as CBAM-ResNet [26], MCAMDN [27], MSCNN [28], MANANR [29], and TLResNet34 [30]. These models exhibit less than 90% accuracy on A-C and C-A when subjected to fault diagnosis under variable working conditions.…”