Uma nova técnica de otimização multiobjetivo de modelos probabilísticos multivariados de um processo de soldagem MIG em tubos de alumínio AA6063

Abstract

To assist in solving the problem of process improvement, restrictions and better welding operation conditions, this work applies the Design of Experiments (DoE), Multiobjective Optimization and Multivariate Statistics methodologies together to provide the necessary support in the management of the production process of MIG welding (Metal Inert Gas), of anti-corona protection rings, manufactured with tubes aluminum alloy 6063 (Aluminum Alloy 6063 - AA6063), T4, 100 mm in diameter and 2 mm thick. This type of process can be controlled by a relatively small number of input variables, that is, the wire feed rate (WF), voltage (V), welding speed (Fr) and the distance from the contact tip to the part of work (Cf). In addition, many outputs can be evaluated and optimized simultaneously. In the present work, the variables of yield (Y), dilution (D), reinforcement index (IR) and penetration index (PI) were investigated. To consider the multivariate nature of the problem, techniques such as Factor Analysis and Bonferroni's simultaneous confidence intervals were applied combined with elliptical constraints. The response variables were modeled mathematically using Poisson regression and the results obtained were satisfactory, since accurate models were achieved. The normal bound intersection method (NBI) produced a set of viable configurations for the input variables that allows the experimenter to find the best configuration of the system in relation to the level of importance of each response. The application demonstrated the optimal parameter solution for the welding process in AA6063 and presented characteristics of minimizing the weld bead geometry to contribute to the better efficiency and effectiveness of the productive management of the welding process. An experimental confirmation procedure was successfully performed to validate the theoretical results obtained in the prediction model.