Monitoring and forecasting of tool wear based on measurements of vibration accelerations during cast iron milling. (9th June 2023)

• Record Type: Journal Article
• Authors: Twardowski, Paweł; Czyżycki, Jakub; Felusiak-Czyryca, Agata; Tabaszewski, Maciej; Wiciak-Pikuła, Martyna
• Is Part Of: Journal of manufacturing processes. Volume 95(2023)
• Page Start: 342
• Page End: 350
• Publication Date: 2023-06-09
• Subjects: Machine learning -- Tool wear identification -- Milling -- Cast iron -- Vibration accelerations; Production management -- Data processing -- Periodicals; Manufacturing processes -- Periodicals; Procestechnologie; Productietechniek; Production -- Gestion -- Informatique -- Périodiques; Fabrication -- Périodiques; Manufacturing processes; Production management -- Data processing; Periodicals; 670.5
• DOI: 10.1016/j.jmapro.2023.04.036
• Languages: English
• ISSNs: 1526-6125

Abstract

Abstract: The research focused on the possibility of predicting the wear of cutting tools during milling of EN-GJL-250 cast iron by using machine learning in the form of two different classification trees. The measurement of vibration acceleration as one of the physical quantities tested during the cutting process can be measured in an accessible way (measurement on the workpiece or through the tool). Implementation of the measurement method presented in the research is possible in every field of industry with various methods of cutting. As input data for the models, measures based on vibration acceleration signals were selected. From the conducted experimental tests, the values of wear on the end mill flank VBc at the corner of the tool and vibration acceleration during machining at variable cutting speeds ( v c = 150–500 m/min) of cast iron castings were obtained, these values were correlated with each other. The machining was performed with the use of monolithic end mills made of cemented carbide. The test results show the possibility of predicting cutting edge wear using machine learning, obtaining an error of around 1 % compared to the regression models, which in the best variant reached an error of as much as 25 %.