Research of Tool Wear Monitoring and Tool Life Prediction Models

Tool wear is inevitable in manufacturing and affects the surface quality and geometric tolerance significantly. Detection of the tool wear and residual life can significantly improve the usage rate of machining and reduce the economic losses due to the tool breakage so as to optimize input-output ratio. In this paper, we propose the tool wear monitoring models based on XGBoost and the prediction model of tool remaining life based on short-term memory neural network. We choose cutting force, vibration and emission signal as the tool data, then extract the features of the tool data from the three aspects of time domain, frequency domain and time-frequency domain. Extract features that are more relevant to wear and combine them with Support Vector Regression (SVR), random forest and extreme gradient descent algorithm (XGBoost) to establish tool wear monitoring model. Through comparison, it is found that the prediction accuracy of model based on XGBoost is the highest, with MSE, Mae and R2 being 70.92, 6.01 and 94.56%, respectively. Based on the amount of wear after several times of cutting, optimized by Adam algorithm, the prediction model of tool remaining life based on LSTM can keep within the error of ten passes under normal circumstances.