Research On Bearing Fault Diagnosis Method Based On Gated Loop Unit Network

Rolling bearing is an essential part of mechanical equipment and the basic condition to ensure the safe production and operation of the equipment.If there is a slight failure of the bearing at work,it may cause the decline of the working performance of the equipment or the shutdown of the equipment,resulting in serious economic losses and casualties.Thus,the effective bearing fault diagnosis method has attracted much attention.The traditional fault diagnosis method of rolling bearing is limited by signal feature extraction,and the diagnosis process of this method is tedious.It is extremely difficult to rely on manual feature extraction under massive data.However,deep learning has the strength of automatic learning signal characteristics and good data processing ability,it can offer the possibility for accurate and efficient fault diagnosis under complicated workplaces.This paper is mainly conducive to the depth learning algorithm for bearing fault diagnosis,focusing on the following algorithms:(1)When using machine learning algorithm for fault diagnosis,it is usually necessary to process the vibration signal to get the main characteristic signal,and then use the extracted information for diagnosis.This method is easy to cause the problem of data loss and low accuracy of diagnosis.Therefore,a bearing fault diagnosis method based on bi-directional gated cycle unit with direct input of original data is proposed to ensure that the vibration data information will not be destroyed.Tests were conducted on Case Western Reserve University(CWRU),the diagnosis accuracy is 99.25%,under the condition of 0Hp load,but the generalization of the model is poor.(2)In order to solve the problems of weak generalization ability and long training time of single depth neural network model,a bearing fault diagnosis algorithm based on one-dimensional convolution,attention mechanism and gated cycle network is proposed.The one-dimensional convolution network is used to extract the feature signal and speed up the training,and the maximum pooling layer is used instead of the full connection layer as the connection between the 1DCNN layer and the BIGRU layer to avoid the loss of these features.By adding the attention mechanism layer,the problem of low recognition rate of long sequence data is solved.Using case West Storage bearing data set to verify,The diagnostic accuracy is more than 99% under different load conditions,it can enhance the generalization ability.(3)In the actual production,the operation condition of rolling bearing is complex and changeable,and the bearing fault data has the problems of small amount of data and uneven sample distribution.However,transfer learning can realize the fault diagnosis of data under different independent and identical distribution.Thus,a bearing fault diagnosis approach based on deep learning and transfer learning is proposed.First of all,make use of sufficient source domain data for network training,and then freeze the underlying structure of the trained network.Finally,we obtain the final model by fine-tuning its top-level structure using less target domain data.From the experimental results,under small samples and variable working conditions,the diagnostic accuracy of the model is more than 97% and transfer learning can improve the diagnostic ability of the model.