Metodologia via redes neurais para a estimativa da rugosidade e do desgaste de ferramentas de corte no processo de fresamento frontal

Abstract

In milling processes where the dynamic behavior of the cutting tool/workpiece/machine tool system is particularly complex due to the discontinuity of the cutting operation and the large amount of variables involved, it becomes very difficult to establish a model correlating surface finish and tool wear to some of the main machining parameters. The present work proposes a neural network based procedure aiming the determination of an experimental relationship between surface finish (through the roughness Ra [mm]) and tool wear (through the maximum flank wear VBBmax [mm]), with some of the main cutting parameters: cutting speed, feed per tooth, depth of cut, hanging length of the cutter, power consumption, vibration level (measured both at the inferior bearing of the tool holder axis and at the work table of the machine), and position of the work table in relation to the milling tool. The choice of the neural network procedure was motivated by the satisfactory results showed by this technique when estimating and modeling nonlinear systems with many non-correlated variables. For the application and validation of the proposed methodology, face milling tests with ABNT 1045 steel bars and coated cemented carbide were carried out. The tests were used to train a neural network, and in the realization of a global sensitivity analysis to establish the influence of the studied parameters on the surface finish (Ra) and tool wear (VBBmax). The results showed that neural network is a promising technique to estimate the surface roughness and tool wear in face milling process.