Milling Tool Wear Estimation Using Machine Learning with Feature Extraction Approach

A cutting tool in any milling operation eventually becomes less efficient due to wear, which could result in reduction of quality and dimensional accuracy of the product during manufacturing. Precise and continuous monitoring is required to avoid these issues in machining. Manual inspection and arbitrary assessment have traditionally been used to identify faults and categorise machine tool wear. However, this method is time consuming and is prone to errors. The purpose of this study is to offer more effective, reliable, and accurate methods to predict tool wear using various data-driven machine learning models, which can then be used to determine Remaining Useful Life (RUL) of the milling tool. This paper implements the IEEE NUAA Ideahouse tool wear dataset containing various sensor readings and makes use of the TSFEL (Time Series Feature Extraction Library) for automatic feature extraction in the temporal, spectral and statistical domain, followed by the implementation of PCC (Pearson's Correlation Coefficient) for the selection of features that have a significant impact on tool wear. ML models such as K-Nearest Neighbors, Decision Tree, and Random Forest provides fairly good results with scaled data followed by feature extraction. All the models are evaluated based on MSE, MAPE and R
²
score, by passing unscaled and scaled selected features in temporal, spectral and statistical domains to find out the model best suited for accurate tool wear prediction in order to assist in predictive maintenance.