Study On Tool Wear Monitoring Model Combining Visual Labels And Acoustic Emission Features

In conventional machining process,machine operators evaluate the tool wear state by their own experience,so as to judge whether the tool has reached its service life,thus making the decision whether to change the tool.This method is often difficult to determine the best time to replace the tool,which can easily lead to problems such as frequent machine downtime and increased tool cost.To accurately grasp the tool wear during the machining,the tool condition monitoring technology arises at the moment.It can reflect the tool wear state by considering the information related to the tool wear directly or indirectly,thus providing a scientific basis to estimate the service life of the tool.Therefore,the research on tool condition monitoring technology is an effective way to improve tool utilization rate and production efficiency,which has great significance.In this paper,a tool wear monitoring model is established based on the machine vision method in the direct method and the Acoustic Emission(AE)method in the indirect method,thus identifying the tool wear state,predicting the tool wear amount and tool life.On the one hand,the tool wear amount is obtained by using the machine vision method and used as a learning label.On the other hand,the feature value is obtained by the AE signals generated in milling.Then,the two are combined to form a data set for pattern recognition and regression analysis.The identification and prediction of tool wear state are realized by various models.The primary details of this thesis are as follows:(1)The wear form and wear process of milling tool during machining are expounded.The average wear width VBave of the tool flank wear zone B area is taken as the index.The tool dullness standard based on the material of the test is choosed.Then,the source of AE signal generated during milling are expounded.The relationship between AE signal and the tool wear are analyzed.Then,a tool wear monitoring platform based on machine vision and AE signals is designed.Finally,a tool wear monitoring platform is designed based on machine vision and AE signals.A test platform is built on this basis to collect tool images and AE signals during machining.(2)A machine vision module is designed to provide learning labels for tool wear monitoring models.An automatic image acquisition method based on structural similarity,an image acquisition and adaptive threshold segmentation method based on FAST feature point detection,and an image correction method using Hough transform meathod are proposed.Then,the wear value of zone B is accurately extracted by dividing the wear area.Finally,a milling experiment are carried out to verify the accuracy of the vision module.(3)The signal module is designed to provide feature values for the tool wear monitoring model.Firstly,AE signals samples in the machining process are established.Subsequently,AE features are extracted by preprocess,time domain,frequency domain and frequency domain analysis.Finally,feature selection is used to remove the features with no significant correlation with the wear quantity.Feature fusion is used to remove the highly redundant features,thus constructing AE feature vectors rich in tool wear related information.(4)The visual labels and AE feature vector are combined to construct pattern recognition and regression analysis datasets.The tool wear monitoring sub-models are established through BP neural network,Elman neural network,support vector machine and support vector regression machine.The combined model is established through artificial bee colony algorithm.Finally,the performance of each model was evaluated by statistical indexes.The optimal model is selected for the generalization test,thus realizing the identification of tool wear state,the prediction of tool wear value and tool life.