Machine learning para predição de rugosidades e desgastes de fresas com diferentes polimentos por arraste

Abstract

Tool edge preparation is considered one of the key aspects of modern and accurate cutting tools. It is a process that allows longer tool life, better cutting performance and surface quality of the machined material. In order to better understand the role of cutting edge preparation, it is necessary to deepen the understanding of the roughness of the machined material, tool wear and their relationships. This task, although challenging, as it is dependent on many configuration factors of the cutting process, is fundamental for process optimization. Predicting roughness and wear is a means of reducing production times and costs, as it reduces process operations and makes better use of the tool possible, where one seeks to use the tool until the end of its life, but without leaving reach a wear that compromises the quality of the surface machined by this tool. In order to explore these two fields of development, edge preparation and wear and roughness prediction, this work developed experiments to collect wear and roughness data obtained by tools with different edge preparations to understand the effects of this preparation and to train and evaluate prediction models. Through the evaluation of the experiments, it was confirmed a greater stability and homogeneity in milling with the tools with drag polishing when evaluating the roughness and a longer useful life of the tools with edge preparation with the media HSC 1/300 and H 4/ 400 of 18.62% and 20.49% higher, respectively, when compared to tool life without any edge preparation. For the predictions of wear and roughness, different methods were compared that achieved mean percentage absolute errors for wear, Ra, Rz and Rt around 0.1%, 2.3%, 10.4% and 23.8%, respectively. The comparison of new models brought new contributions to the literature testing new machine learning models, highlighting models based on decision trees such as Random Forest, Extra Tree and XGBoost potential for roughness predictions and NuSVR for wear prediction. [resumo fornecido pelo autor]