Research And Implementation Of Tool Life Prediction Based On Industrial Gig Data

In the process of cutting,the cutting tool will gradually wear,light will affect the quality of processing,heavy will scrap the workpiece,machine damage.Therefore,the prediction of the remaining life of the tool becomes very important.With the development of sensor and communication technology,a large amount of monitoring data is produced in the cutting process of cutting tools.If we can mine the value based on these data and accurately predict the remaining life of the cutting tool,it will help to improve the production efficiency and have a strong practical value.In this paper,the tool wear monitoring and residual life prediction are studied.Based on the data of cutting force,vibration and transmitting signal,the features of basic data are extracted from time domain,frequency domain and time-frequency domain,and the features with high correlation of wear amount are obtained by using the correlation method to screen the features.Support vector regression(SVR),random forest(random forest)and extreme gradient descent algorithm(xgboost)are established Tool wear monitoring model based on forest and.Design experiments,using MSE,Mae and R2 as evaluation indexes,through comparison,it is found that the prediction accuracy of model based on xgboost is higher,MSE,Mae and R2 are 70.92,6.01 and 94.56% respectively.According to the amount of wear after several times of cutting,a prediction model of tool remaining life based on short-term memory neural network(LSTM)is established,and the hidden layer contains 11 neurons through parameter optimization.Then the LSTM model is optimized by Adam algorithm,which improves the convergence speed and accuracy of the model.Finally,the experiment is designed to evaluate the difference between the number of iterations in convergence and the number of iterations in scrapping and the number of real tool cuts.The number of iterations in convergence of the LSTM model optimized by Adam algorithm is 36,and the difference between the LSTM model optimized by Adam algorithm and the real value is within 10 when the number of tool cuts is more than 100,which is better than the model based on RNN.This is because LSTM is more suitable for predicting the interval and delay in time series events for.