Research On Recognition Method Of Milling Cutter Wear State Based On Edge-labeling Graph Neural Network

During milling,the tool will wear gradually,and the wear degree will not only affect the accuracy of the parts processed,but also damage the products and machine tools in serious cases.Therefore,the research on tool wear condition monitoring technology becomes very important.This paper takes milling cutter as the research object and studies the monitoring method of tool wear state.The main contents of this paper are as follows:(1)Aiming at the disadvantage that the traditional deep learning method of image recognition relies on a large number of training samples,a tool wear classification method based on edge-labeling graph neural network(EGNN)is proposed in combination with the graph neural network theory.In this method,tool wear images are input into embedded network module to extract tool state related features,and then the full connection graph is established based on these features.Then the predicted value of the edge label can be obtained by iteratively updating the features of the nodes and edges in the full connection graph through the feature transformation network module.Finally,tool wear is identified by weighted voting based on the predicted results of sample label of support set and sample edge label of query set.The experimental results show that this method can effectively identify the type of tool wear with only a few images of tool wear.The recognition accuracy is better than that of CNN,AlexNet and ResNet.(2)Aiming at the disadvantage that most current tool condition monitoring methods based on sensing signal need to rely on signal processing technology and prior knowledge,a tool wear state classification method based on one-dimensional sensing signal imaging and improved edge-labeling graph neural network(IEGNN)is proposed.In this method,one-dimensional sensing signal data is coded into gray-scale recurrence plot(RP)and the gray-scale RP is input into the improved multi-scale embedded network module to extract features,which finally realizes the recognition of tool wear state.The experimental results show that this method can effectively identify tool wear state with less data.(3)In order to further improve the accuracy of tool wear state recognition based on sensor signals,a new method of IEGNN tool wear state classification based on multi-dimensional sensor signals is proposed.Based on the one-dimensional sensor signal IEGNN,each grey RP is aggregated into a color RP according to the data acquisition stage,and then the aggregated color RP is input into the IEGNN.The application of prognostics and health management(PHM)2010 milling cutter wear dataset and multi-dimensional sensing signal endmilling cutter condition monitoring experiment proves the feasibility and effectiveness of the proposed method,which can significantly improve the accuracy of tool wear status identification under small sample conditions.