Research On Bearing Fault Diagnosis Method Under Complex Working Conditions Based On Convolutional Neural Network

As the core component of the rotating machinery system,the bearing plays a role in carrying and transferring loads.Its failure or abnormal working conditions will directly affect the working performance of the mechanical system.In industrial production,due to the influence of various complex factors such as environmental changes and load impacts,the bearing will inevitably fail.The fault signal has the characteristics of random and non-stationary.Therefore,it is of great significance to diagnose the fault of the bearing under complex working conditions.This dissertation takes the fault diagnosis of bearings under different working conditions and noise interference scenarios as the application background,and conducts in-depth research on the key technologies of convolutional neutral network in bearing fault diagnosis,which effectively improve the bearing fault diagnosis effect under complex working conditions.The main research contents of this dissertation are as follows:When the current CNN model is used to solve the problem of bearing fault diagnosis,each convolutional layer can only extract features on a single scale.And it is impossible to determine which time-scale features are sensitive to faults,which resulting in the network not being able to capture the effective characteristics of the vibration signal under variable working conditions.Therefore,this dissertation proposes a bearing fault diagnosis method based on adaptive weighted multi-scale convolutional neural network.The method uses channel grouping convolution to learn fault characteristics on multiple time scales.It combines adaptive weighting algorithm to ensure the correlation between multiple packets,which improves the feature extraction capability and generalization performance of the model under variable working conditions.The bearing fault diagnosis experiment proves that the method has high fault identification accuracy under the conditions of variable speed and variable load.Because the traditional CNN model does not consider the importance of the time domain vibration signal pulse segment and the relationship between the feature vectors of different channels when extracting the characteristics of the bearing vibration signal,this dissertation proposes a bearing fault diagnosis method based on dual attention convolutional neural network.The method can adaptively assign different weights to the time series characteristics and channel characteristics of the vibration signal at the same time,selectively learn more important feature information,and improve the feature learning ability and domain adaptability of the model.Thereby improving the accuracy of bearing fault identification under complex working conditions.The bearing fault diagnosis experiment proves that the method has high fault diagnosis accuracy and strong domain adaptability under strong noise and variable working conditions.At the same time,the effectiveness of the channel attention module and the signal attention module in optimizing the extracted features is analyzed through the visualization feature vectors.Aiming at the problem of unlabeled bearing fault diagnosis,this dissertation proposes a transfer learning method for unlabeled bearing fault diagnosis based on subdomain adaptation and dual attention convolutional neural network.The method uses dual attention convolutional neural network to adaptively extract features,introduces the concept of subdomain adaptation,and uses local maximum mean differences to align the distribution of related subdomains of the same category,which reduces the diagnosis error caused by the difference in distribution and improves model’s diagnostic effect on unlabeled faults.The fault diagnosis experiment of cross domain unlabeled bearing proves that the method can effectively complete the unlabeled fault diagnosis task and has excellent domain self adaptability.