Tool Wear State Recognition And RUL Prediction Of High-precision Milling Of Cell Phone

During the high-precision machining process,various quality problems occur in the workpiece due to the tool wear.Taking the case of chamfering and high-light processing of the mobile phone as an example,there are phenomena such as drawing,chipping,and fogging cause the workpiece to be scrapped.An effective tool wear condition monitoring model can reduce scrap rate,increase production efficiency,and contribute to the realization of automated production.In the high-precision machining process,the tool wear is so small that the features which are sensitive to tool wear are not easy to extract.In this paper,a multi-sensor signal feature extraction technology based on CNN were proposed,and XGBoost,Stacked-LSTM-PF models were used to monitoring the tool wear state.According to the actual finishing tool wear characteristics and the common abnormal quality forms,the experiment of condition monitoring using multi-sensor fusion technology of spindle vibration,workpiece vibration and cutting force signal was determined.The short-term energy method was used to intercept the processed signal and wavelet was used to reduce noises.A feature extraction method for time-frequency domain feature extraction and using CNN for secondary extraction is proposed.The feature selection method based on distance evaluation technique and information gain is used to obtain the sensitive features of tool wear.SVM,decision tree and XGBoost are used to identify the tool wear state,and classification result of the time-frequency domain features,the original signal directly extracted by CNN and the feature extracted by the proposed mathord are compared.It shows the superiority of the proposed feature extraction method and the efficiency of the XGBoost algorithm which achieved the accuracy rate to 97.52%.The tool RUL prediction model based on Stacked-LSTM-PF is proposed including he prediction of tool wear value and the PF parameter identification of the physical model of tool wear.By comparing the influence of different LSTM hidden layer nodes and LSTM stacking layer on the prediction error of tool wear value,the RUL prediction model of the optimal network structure with 36 nodes in hidden layer and 3 layers in LSTM is established.The predicted MSE loss of wear value is only 3.8968(10^-3)and the predicted RUL is within5 of the actual RUL.The effectiveness of the feature extraction method,the XGBoost tool wear condition recognition model and the Stacked-LSTM-PF RUL prediction model was verified by application in a certain enterprise.