Research On Fault Diagnosis Method Of Rolling Bearing Of Printing Equipment Based On Deep Learning

Printing equipment is in the stage of transformation towards automation,intelligence,and greenery.Rolling bearings are the core components of printing equipment,and their fault can affect the normal operation of the equipment,so timely and effective fault diagnosis of rolling bearings can improve the service life and operational efficiency of printing equipment.Applying deep learning methods to rolling bearing fault diagnosis can improve the shortcomings of traditional fault diagnosis methods that require manual intervention,low recognition efficiency and accuracy,and help promote the development of intelligent manufacturing.The current bearing fault diagnosis methods based on deep learning have problems such as complex models,inadequate feature extraction,and poor generalization,to further improve the accuracy and efficiency of intelligent printing equipment rolling bearing fault detection,the relevant research works in this article are as follows:(1)To solve the problem of mean shift that the rolling bearing fault diagnosis method based on the modified linear unit(ReLU),which leads to inaccurate classification,a BCNN-LSTM fault diagnosis model based on the improved Bias Rectified Linear Unit(bReLU)combined with convolutional neural networks(CNN)and long short term memory(LSTM)neural networks was proposed.The model first used the bReLU activation function in the CNN network to complete the adaptive feature extraction,then added the Batch Normalization(BN)layer between the convolution layer and the activation function to accelerate the model convergence,finally learned the time-series features by the LSTM network.The accuracy of fault identification can reach over 99% under the fixed working condition dataset of Case Western Reserve University,and the training time was significantly reduced that the maximum reduction of 66.7% compared with other methods.(2)To improve the deficiency of insufficient feature extraction of the deep learning fault diagnosis model,the temporal convolution network(TCN)was applied to the rolling bearing fault diagnosis.In addition,to improve the overfitting problem of the TCN model,the temporal convolution with the average pooling network(TCAPN)model was proposed.Firstly,the huge receptive field of dilated causal convolution was used to mine the global fault features of bearings more fully,secondly,the convolution kernels of different sizes were used to obtain the bearing fault features of different scales,finally,the average pooling layer was used to reduce the redundant features and the training time.The experiment showed that the TCAPN model can converge quickly under fixed operating conditions,and the average accuracy was 98.73%,which was2.87% higher than that of the TCN model,proving that the model has high accuracy and robustness.(3)To improve the generalization of the fault diagnosis for rolling bearings under variable operating conditions,the BCNN-LSTM and TCAPN fault diagnosis models were respectively applied to fault diagnosis under variable operating conditions.And further,optimized the TCAPN fault diagnosis model.First,improved the network structure of the residual block and feature extraction layer in the residual module to avoid feature redundancy and too many parameters.Second,updated and trained the negative signal of the bearing by the bReLU activation function.Finally,used the residual connection to fuse the deep and shallow features,so that the model can extract the feature information of the same fault type under different operating conditions.The average accuracy of the BCNN-LSTM model under variable operating conditions reached 95.28%,the average accuracy of the TCAPN model was 97.57%,and the TCAPN model has stronger robustness compared with other methods.