Key Technologies On Data-Driven Real Time Prediction Of Tool Wear For NC Machining

Tool wear state is closely related to part quality,machining cost and machining efficiency during CNC machining.Accurate real time prediction of tool wear state during machining process is very important.Due to the variation of machining conditions,existing methods cannot realize accurate real time prediction of cutting tool wear under variation machining conditions.Data-driven methods have been deemed as an effective way for accurate cutting tool wear prediction,while there is an contradiction between the large-number requirement of labeled data and the difficult collection of tool wear values,which has becoming an limitation of data-driven method for application.In order to address the issue mentioned above,this paper conducts an in-depth study on data-driven real time tool wear prediction method of NC machining,and the research work is mainly developed in the aspects of automatic and accurate measurement of tool wear,monitoring signal processing of tool wear,and realtime prediction of tool wear.The main work and innovations of the thesis are as follows:(1)Aiming at the problem of accurate measurement of tool wear caused by the lack of the original boundary of tool blades after wear,an automatic and accurate measurement method of tool wear based on the probability of gray image is proposed.Different from existing methods where only the boundary information is considered for tool wear measurement,the variation law of the gray value of the blade wear area is discovered.The Bayesian formula is used to solve the curve of the maximum probability of the original boundary condition under the known wear boundary,and then the tool wear value is calculated accurately.Compared with the existing frequently-used automatically measurement methods of cutting tool wear error,the measurement error is reduced from more than 15% of existing methods to less than 5% by the proposed method.(2)Aiming at the problem of high dimension of cutting tool wear monitoring signal features under variable cutting conditions,a signal feature selection method based on entropy weight-grey correlation analysis and feature dimension reduction method based on manifold learning are proposed.The signal features are selected via the comprehensive correlation analysis based on the entropy weight-gray correlation analysis method by considering the comprehensive influence of tool wear state and cutting conditions on signal features.Space transformation and dimensional reduction of signal features are performed based on manifold learning.Compared with the existing signal processing methods,the accuracy of tool wear prediction is improved by 10% by using the same prediction model.(3)Aiming at the problem of accurate prediction of tool wear caused by continuous changes of cutting conditions during the machining process,a meta-learning method for accurate prediction method of tool wear based is proposed.The meta-LSTM(Long Short Term Memory)tool wear prediction model is established for specific machining features,and the essential change law of monitoring signals and tool wear under different cutting conditions are studied.The model can be finetuned and adapted by a small number of labeled samples for new cutting conditions in new prediction tasks.Aiming at the problem that a small number of samples are required to be fine-tuned under the new cutting conditions,a meta-learning method with machining features as samples is studied.Compared with existing methods,by using the same number of samples for model training,the prediction error of tool wear was reduced from 0.19 mm by existing methods to less than 0.06 mm by the proposed method.Based on the above research,the tool wear prediction system was developed,and it has been successfully applied and verified by typical aviation part.