Tool wear classification based on convolutional neural network and time series images during high precision turning of copper

Abstract

Artificial intelligent techniques are critical in predicting tool wear in the machining process. Effective assessment of the rate and state of tool wear are important in the high-speed turning process and makes it possible to replace the tool before catastrophic wear occurs. It is well known that cutting forces are often used as an indicator to monitor tool wear during the machining process. However, how to use deep learning methods to incorporate cutting forces, accurately predicting the tool wear, remains a challenge. In this paper, a novel tool wear classification method based on time series images of cutting forces and convolutional neural networks was proposed. The one-dimensional cutting force signals are converted into two-dimensional time series images using the Gramina Angular Summation Fields method. Then, a modified AlexNet network is used to classify tool wear types based on the processed cutting force time series images. The prediction results show the overall prediction accuracy of the four tool wear categories can reach around 90%. In addition, by balancing the number of cases in the dataset, the prediction accuracy of each category can be improved while the overall prediction accuracy is maintained.