Research On Fault Diagnosis Method Of Rolling Bearings Based On Attention-enhance Convolutional Neaural Network

Bearings are one of the most important parts of modern machinery and equipment,but because bearings work in the complex and harsh environment of high speed and high load for a long time,bearing failures occur frequently,which leads to mechanical equipment damage or even shutdown,brings economic losses to the society,and even serious harm to human health.Therefore,the research on rolling bearing fault diagnosis method is of great significance to the safe operation of equipment.Due to the harsh working environment of bearings,vibration signals contain a large amount of noise,showing strong non-stationary,nonlinear and nonGaussian characteristics,etc.,convolutional neural network fault diagnosis method is often used to extract global features,often ignoring the important invisible features,and domestic and foreign experts and scholars are committed to developing deeply complex network models to meet the accuracy requirements,but often aggravate the overfitting phenomenon.To solve the above problems,this paper proposes a fault diagnosis method for rolling bearings based on attention enhanced convolutional neural network.This method uses the attention module CBAM to extract the important features of channel dimension and spatial dimension,and uses the incentive mechanism of Q-learning to motivate the weight coefficient of attention diagram to achieve the optimal classification strategy.The main research contents and achievements of this paper are as follows:(1)The CBAM-CNN network model is proposed to solve the problem that bearing signal fault samples are few,easy to be interfered by environmental noise,and the characteristic information is difficult to be fully mined.This method is improved on the basis of convolutional neural network,and the dual-channel attention module CBAM is embedded into the convolutional neural network model.CBAM module can independently learn the main features of channel dimension and spatial dimension,so that the model can extract deep fault features in multiple dimensions,reduce the influence of redundant data,and improve the fault diagnosis and recognition rate of network model.The proposed method is verified by using the rolling bearing data set of Cass Western Reserve University.The results show that the recognition rate of the proposed method on the bearing data set can reach 99.2%,and the error is 0.0021.(2)In the CBAM-CNN network model,the weight coefficient of attention graph is not clearly selected,and the development of a deeply complex network model is blindly pursued to meet the accuracy requirements.A CBAM-CNN-QL neural network model is proposed to combine the autonomous decision-making ability of reinforcement learning with the deep perception ability of deep learning.The network model,as an agent,constructs the environment state space with fault samples,obtains the maximum Q function value and reward value through continuous interactive learning between agents and environment,updates the weight coefficient of attention diagram by gradient descent method,and finds the optimal strategy to realize fault diagnosis.The results show that the recognition rate of this method is 99.02% on the multicondition simulation data set.(3)In order to verify the performance of the CBAM-CNN-QL fusion model in bearing fault data,the CBAM-CNN-QL fusion model was applied to the experimental data of laboratory rolling bearings.The results show that the identification accuracy of the proposed method on the experimental data set of rolling bearings is 99.5%,which is 3.9%,1.8% and 1.0% higher than that of the CNN,CBAM-CNN and CNN-QL models.The experimental results show that compared with other network models,the proposed method has higher recognition accuracy in different load fault data sets,with an average accuracy of 99.60%.