Research On Tool Wear State Monitoring Based On Support Vector Machine In Milling Under Variable Parameters

Tool wear state monitoring has important practical significance for guaranteeingthe quality of workpiece and improving the efficiency of machining process. Toolwear state monitoring process is a pattern recognition process. There are three stepsduring the monitoring process. Firstly, the sensor signals in the cutting process mustbe collected and then the effective features are extracted from the sensor signals.Secondly, the relationship model between the features and tool wear state is built.Thirdly, in order to monitor the state of the tool wear, a pattern recognition model isused to classify and identify the unknown sample. Currently, most research of toolwear state monitoring is being carried out only for fixed parameter, while the variablecutting parameters are widely used in the real machining process. In this case, boththe cutting parameters and tool wear state have an effect on the sensor signal andfeatures, which will reduce the accuracy and robustness of tool wear state monitoringgreatly. Therefore, how to realize the monitoring of tool wear state under variableparameters becomes a problem need to be solved urgently.The original time domain features from the cutting force are vulnerable tochanges of cutting parameters. So it is not suitable for tool wear state monitoringunder variable parameters. In this paper, a new set of dimensionless time domainfeatures including normalized cutting force indicator (NCF), variation coefficient (Cv)and peak force ratio (MFR) are proposed to solve this problem. To verify theeffectiveness of the new features for tool wear state monitoring under varyingparameters, an end milling experiment of TC4titanium alloy was carried out. Cuttingforces corresponding to four kinds of tool wear state under every cutting parameterwere collected. Then the new features and original features were extracted based onthe same force signal sample respectively. Due to the excellent learning andgeneralization ability of support vector machine (SVM) under small training samples,the new features and the original features are input SVM to identify the tool wear staterespectively. The results of analysis and comparison show that no matter what kind offeatures are firstly input SVM to optimize the model parameters, the classificationaccurancy of tool wear state based on new features is higher than the original features. The research results show that the SVM model based on the dimensionless timedomain features proposed in this paper has a higher classification accurancy and abetter robustness. So it is very suitable for tool wear state monitoring under variableparameters, which will cast some new lights on improving the surface quality ofworkpiece and machining efficiency in real machining process.