Machine Learning-Based Tool Wear Estimation for Milling Processes

Abstract

Replacing worn-out tools at the appropriate level of wear can prevent tool breakage, unnecessary maintenance efforts, and poor surface finish. A Tool Condition Monitoring (TCM) system can predict the wear and find the appropriate time for replacement. This can improve overall machining efficiency, maintain part quality, and reduce manufacturing costs. In this project, a TCM architecture is proposed to monitor tool conditions and predict tool wear using a combination of data acquisition and machine learning regression. A semi-automated data acquisition system is designed to operate milling processes under multiple cutting conditions. Signals are recorded from a 3-axis accelerometer and microphone attached to the tool head, and a microscope, also mounted to the milling machine, is used to characterize tool wear. Statistical features for use in the machine learning model are extracted from the sensor signals in the time domain and frequency domain. Cutting condition features, such as material removal, material removal rate, and cutting speed are also integrated into the machine learning model. The microscope images are processed systematically to evaluate the wear at 4 different positions and used to train the model output. Ultimately, wear estimation and prediction are carried out using a support vector machine learning model. Performance evaluation of the system using root mean square error indicates a reliable tool wear estimation, with low computational complexity.