Unsupervised online prediction of tool wear values using force model coefficients in milling

Abstract

Tool wear prediction is an important research in metal cutting, which aims to improve machining accuracy and production efficiency, maximize tool utilization, and reduce machining cost. However, due to high complexity and nonlinearity of tool wear process, it is difficult to establish a general tool wear prediction model, which limits its application in industrial production. To solve this problem, an unsupervised online prediction method for tool wear values is proposed. In the method, a milling force model considering tool wear is established by using analytical method, and parameters varying with tool wear in the force model are integrated into five force model coefficients. The coefficients are solved and updated continuously using the least square estimation method according to the cutting force signals collected in real time. Based on study of relationship between the coefficients and tool wear, a tool flank wear value estimation model is constructed, combined with a time series analysis model, to achieve prediction of tool flank wear values. Experiments are conducted to test the prediction accuracy of tool wear values using the proposed method, and the results show that the average online prediction accuracy reached 72.0%, without supervision. The method has the advantages of low cost and strong adaptability, and can be used for online prediction of tool wear in machining industry.