Prediction of tool wear in the milling process using Neuro-Fuzzy inference systems and Feature Selection

Abstract

Milling is a vital manufacturing process in today’s industry. To meet demands and remain competitive, it is common to use the machine continuously for manufacturing, but eventually the cutting tool will wear out mechanically, resulting in negative effects on the part produced. Tool replacement can be made more efficient by using an indirect tool condition monitoring system. Several studies have explored artificial intelligence models to predict tool wear, including the neuro-fuzzy inference system. However, there is still a gap in the exploration of alternative feature selection methods in conjunction with this model. This study proposes the use of principal component analysis and Pearson’s correlation coefficient as methods for selecting features based on sensor signals for the neuro-fuzzy system for predicting tool wear in milling. From the results obtained, a mean absolute percentage error of up to 18.99% and a coefficient of determination (R2) of up to 0.76 were achieved from the training carried out on an experimental database considering different levels of wear. The limited number of samples and the complexity of the conditions were limiting factors for the performance of the proposed approach, but the model was robust in its predictions because it only used features extracted from sensor signals in the milling process.