Research On Prediction Of Tool Wear State Under Variable Conditions Based On Feature Transfer

In the metal processing process,tool wear not only affects the machining accuracy and surface quality of the workpiece,but also causes the workpiece to be scrapped and machine tool is damaged.Therefore,using online tool wear prediction technology to accurately identify tool wear states is of great significance to ensure the smooth progress of the machining process and product quality.Based on the similarity of specific machine tools,the same type of tools and workpieces and the same processing requirements,and fully mining the value of historical data,this paper proposes a tool wear prediction method based on feature transfer under variable conditions.It is used to realize the prediction of tool wear states under different working conditions,so as to reduce the repeated data analysis,modeling process and improve the generalization ability of tool wear prediction model.The main work is as follows:(1)Based on the principle of transfer learning,a scheme for predicting the tool wear states of the same tool and workpiece under different combinations of processing parameters is constructed,and the milling processing experiment in the life cycle of the tool under different combinations of processing parameters is established.The cutting force sensor signal data and tool wear data are collected to provide data support for subsequent research.(2)In order to reduce feature redundancy and obtain features with high correlation with tool wear.The time domain,frequency domain and wavelet domain features of cutting force signal are extracted,and the extreme learning machine model based on reserved cross validation is used to select the extracted features.(3)Analyze the correlation between the features(i.e.source domain features)selected during the entire machining process of the tool life cycle(recorded as historical tool)under a set of machining parameter combinations and the feature(i.e.target domain features)selected during the tool's(recorded as new tool)early machining process under another machining parameter combination,and establish a feature transfer model for source and target domain features.The source domain features are transferred to the target domain through the transfer model,so as to obtain the complete signal features of the whole process of the new tool life cycle.Then,the maximum mean difference(MMD)is used to evaluate the feature transfer results,and based on the evaluation results,the features with higher similarity in the source domain and target domain are selected as the best feature subset of the tool wear prediction model.(4)In order to further improve the prediction accuracy,a support vector machine(WOA-SVM)optimized by the whale optimization algorithm is used to construct a tool wear prediction model.The best feature subset of the whole process of the new tool life cycle obtained by feature transfer is used as input to predict the tool wear state of the new tool in the middle and late processing process.The experimental results show that the classification accuracy of the model is as high as 95%,which verifies the effectiveness of the proposed tool wear prediction scheme under variable working conditions.