Application of machine learning algorithms for recognizing the wear of the cutting tool during precision milling of hardened tool steel

The paper presents extensive research on tool wear and the analysis of diagnostic measures for different cutting speeds (vc). The work is divided into two parts. The first part involves conducting an experiment on a machining center, measuring the tool wear index, and recording vibration acceleration signals, followed by analyzing the obtained results. In the second part, the authors focus on determining appropriate diagnostic signal measures and their selection and applying various machine learning methods. The machine learning pertains to classifying the tool condition as operational or non-operational. The best of the tested classifiers achieved an accuracy of 0.999. Thanks to the comparative analysis, it was possible to propose a condition monitoring method that is based only on vibration acceleration without taking into account the cutting speed parameter. Vibration measurement can be performed on the spindle. In this case, the weighted accuracy value determined on the test set was 0.993. The F1 coefficient characterizing both precision and accuracy was 0.982. The authors consider this result to be satisfactory in industrial conditions. Measurement on the spindle without the need to take into account the cutting speed is easy to use in industrial practice