Methodology for selecting an information source for online diagnostics of cutting tool edge condition

The work focuses on the development of a method for selecting diagnostic features in the milling of materials with significantly different properties. The use of two very different materials allowed for better generalization of the conclusions. The research concentrated on steel and composite materials, defined as combinations of two or more materials with diverse chemical and mechanical properties. A method for assessing the usefulness of measured physical quantities for predicting tool wear using machine learning techniques was proposed. Both vibration signals and cutting forces were considered, but the proposed method is general. The study implements artificial intelligence methods through the practical deployment of multiple machine‑learning models trained on vibration‑ and force‑based features. The methodology is based on a measure of the significance of data partitioning into subsets created from forces and vibration accelerations, combined with a large set of supervised machine learning methods, including classification and regression algorithms with various hyperparameters. This approach enables a relatively objective statistical assessment of the usefulness of a given physical quantity in monitoring tool condition.