Research On Tool Wear Mechanism And Prediction Model Of Milling TC18 Titanium Alloy Based On Deep Learning

TC18(Ti-5Al-5Mo-5V-1Cr-1Fe)titanium alloy widely used in aerospace,deep-sea diving,biomedicine and other fields,but it also has the characteristics of being difficult to process.The rapid tool wear when milling TC18 titanium alloy will increase the cost and sharply decrease the quality of the surface of workpiece.Therefore,the research on the tool wear mechanism and tool wear prediction model of this process is particularly important.This article obtains 195 sets of tool wear data through a milling experiment at first,and at the same time collects milling force signals as the input of the prediction model.Analyzing the microscopic morphology of tool wear by scanning electron microscope and EDS energy spectrum analyzer.Based on the raw milling force data without any feature engineering processing,using convolutional bi-directional long short-term memory networks(CNN+BILSTM)and convolutional bi-directional gated recurrent unit(CNN+BIGRU)to establish a tool wear prediction model.In the tool wear prediction model,the experimental results show that compared with other single models,the prediction effect of the CNN+BILSTM model is the best.The effect of the CNN+BIGRU model is second,but its running time is shorter.At the same time,the CNN+BILSTM model and the CNN+BIGRU model converge faster,and the model loss function converges after the tenth iteration.In the tool wear state monitoring model,the average classification accuracy of CNN + BILSTM and CNN + BIGRU models for the three stages of the tool wear are 96.55% and 94.83% respectively,which can meet the actual needs.To effectively select the activation function,through comparative experiments on the sigmoid,tanh and Re LU activation functions which done on the CNN+BIGRU and CNN+BILSTM models,the results show that the model using the Re LU activation function has the shortest running time and the fastest convergence speed.In order to study the generalization ability of the prediction model in this paper,experiments are carried out on another data set of milling tool wear under different working conditions.The cross-validation results of three data sets show that the CNN+BILSTM and CNN+BIGRU models used in this article have good performance and generalization capabilities,providing a new and promising field for online tool wear prediction Methods.