Research On Tool Wear State Monitoring Based On Support Vector Machine In Ultrasonic Milling

Ultrasonic milling is a kind of precision machining.It is extremely sensitive to the wear state of the tool.There must be tool wear in the milling process.The tool wear will affect the quality and dimensional accuracy of the processed surface.The workpiece will be scrapped in serious cases.It may even cause damage to processing equipment.In the past,machine tool operators mainly judged the wear status of tools based on experience.But tool monitoring urgently needs to get rid of the dependence on operators and realize online automatic monitoring of machine tools with the development of intelligent and unmanned manufacturing.In this paper,the real-time monitoring of the tool wear state in ultrasonic milling was realized based on the method of support vector machine and the fusion of milling force and temperature signal.Firstly,the milling cutter wear test of ultrasonic milling TC4 titanium alloy was carried out,the milling force signal and temperature signal during the milling process were collected,and the milling cutter rear face wear value was measured.The influence of ultrasound on the milling force signal,temperature signal and the variation of the two signals with tool wear are analyzed.Secondly,through the analysis of milling force and temperature signals in the time domain,frequency domain,and time-frequency domain,135 signal characteristic values were extracted from them.The binary bat algorithm was improved.The improved binary bat algorithm was used for feature selection of two signals of milling force and temperature.From the 135-dimensional feature vector,11-dimensional feature vectors were effective for the selected tool wear prediction model.Then compared with standard BBA and BPSO,the results showed that the improved binary bat algorithm had higher search accuracy and greatly reduced the dimension of the feature vector.The reduction can be up to 91.85%.Greatly reduced the training time of the prediction model.Finally,a tool wear prediction model was established based on a nonlinear regression support vector machine.The model parameters were optimized by a combination of grid search and cross-validation.The results showed that the mean square error of the prediction model is0.0028 and the square correlation coefficient is 0.9756.The prediction accuracy of the model is high.And it could realize online real-time monitoring of tool wear status in ultrasonic milling.At the same time,the effects of different monitoring signals and different dimensionality reduction methods on the accuracy of the prediction model were studied.Besides,this article also designed a tool wear status monitoring system.And explained the interface,background,and functions of the system.