Prediction of Tool Wear by Machine Learning Method in Turning Process

Abstract

An intelligent monitoring system has the benefit of being able to predicting tool condition, preventing a turning machine from catastrophic failure as well as to guarantee the accuracy of the worn-out cutting tool that can be replaced in a timely manner. The objective of this research is to predict the tool condition during turning based on the time series of vibration sensor signal by using the K Nearest Neighbor approach. Experiments have been carried out on turning of AISI 1045 carbon steel using carbide inserts. Time domain, frequency domain and time and frequency domain is extracted which is subsequently employed as an input for the output-corresponding classification model. Four different distance matrix types which are the "Euclidean," "Minkowski," "Chebychev," and "Cityblock" methods is used to classify the cutting tool condition. The result found that "Euclidean" technique yielded the lowest mean square error of 24.167% and the highest accuracy of 88.333% among the three models, according to the data.