关键词: 刀具磨损预测, 特征选择, 最大信息系数, 集成学习, 高斯过程回归

Abstract: To accurately predict the state of tool wear and provide guidance for tool changing and parameter optimization in machining process,a tool wear prediction method based on Maximal Information Coefficient (MIC) and improved Bagging integrated Gaussian Process Regression (Bagging-GPR) was proposed,and a non-linear mapping relationship between cutting force signals and tool wear was established.The machining cutting force signal was collected,the features of the collected cutting force signal were extracted with time domain,wavelet packet decomposition and empirical mode decomposition,and the correlation between the features and tool wear was analyzed by using MIC to achieve feature selection and avoid the “curse of dimensionality” of the prediction model.To improve the precision of forecasting model,considering the difference and the accuracy of internal kernel function of Gaussian submodels,Bagging was used to combine Gaussian kernel functions randomly as the kernel functions of each submodel,and an improved Bagging-GPR model was constructed to predict tool wear values.The effectiveness and excellence of the proposed method were verified based on the milling experimental data.

Key words: tool wear prediction, feature selection, maximal information coefficient, ensemble learning, Gaussian process regression