Aplicação de redes neurais artificiais como preditor de rugosidade em processo de torneamento

Abstract

The machine tools are found in various metal and mechanical industries as well as garages. They are operated by three phase induction motors, which are subject to problems related to the applied load on their rotors and disturbances concerning to the quality of electrical supply network. Within the machining process the parameters related to the machine tool, the properties of the workpiece material, geometry and material tool and the process itself, may affect the surface completion of the machined parts. The roughness is considered one of the main indexes of the final product quality in machining processes which may produce changes in the electromagnetic torque on the motor shaft. The purpose of this work is to present a predictor of superficial roughness of parts based on the dynamics of the effective current that feeds the induction motor in the turning process using artificial neural networks to analyze the roughness actions according to the machining conditions employed (speed cutting feed and range of the tool tip). Simulation results are presented and show the performance of the Artificial Neural Network (ANN) proposed several operating situations with imbalances of tensions (between +10% to -10%) and load torque steps (25 steps in 1 1 Nm) with mean relative error (MRE) of 0.0120%. Experimental results depending on the method of (RNA) proposed for various situations power grid: balanced, phase loss and voltage imbalances (overvoltage and undervoltage), show high ability to approximate the behavior of the output variable (roughness Ra) with respect to input values (RMS currents of phases a, B and C). The greatest mean relative error of 0.001754% was observed.