Tool Wear State Prediction Based on Two-branch Convolutional Neural Network and Transfer Learning
As the element with the most state changes in the cutting process,the tool's real-time health state directly affects machining accuracy and production quality.Under the same working condition,the model is trained under the histori-cal tool wear dataset,the model cannot accurately predict the tool wear state on the newly collected tool wear dataset due to the different data distribution,and thus the model cannot accurately predict the tool wear state.To address this phenome-non,a tool wear state prediction method based on a two-branch convolutional neural network and transfer learning is pro-posed,where the two-branch structure improves the convergence speed and accuracy of the network.Transfer learning im-proves the performance of the model on the target domain by aligning the edge distributions of the source and target domains using the maximum mean discrepancy method.After experimental verification,the prediction accuracy of the migrated model is not much different from the labeled transfer learning accuracy on the newly collected dataset,which proves that the meth-od is able to predict the tool wear state more accurately on the newly collected dataset.
tool weartwo-branch networktransfer learningmaximum mean discrepancy
万方数据
维普
