Resumo:
Duplex stainless steel pertains to a class of materials with low machinability due to its right rate of hardening, low thermal conductivity and high ductility. This characteristic represents a significant challenge in the manufacture of components, especially in the end milling process. Optimization is a viable alternative to determine the best process parameters and obtain higher production values with sustainability and quality. The presence of noise variables is an additional complicating factor during material machining of this material, and their presence causes an increase in variability during the process, and their effect can be mitigated by employing robust modelling methods. This thesis presents the robust multivariate optimization in the end milling of duplex stainless steel UNS S32205. The tests were carried out using a central composite design combining the input variables (cutting speed, feed per tooth, milled width and depth of cut) and the noise variables (tool flank wear, fluid flow and overhang length). The concept of robust parameter design, response surface methodology, factor analysis, optimization of the multivariate mean square error for robust factors and the normal boundary intersection were applied. The combination of all these methodologies gave rise to the EQMMFR-NBI method. As a result of the factor analysis, the response variables were grouped into 3 latent variables, the first referring to the roughness Ra, Rq, Rt and Rz (quality indicator); the second to the electricity consumption and CO2 emissions (sustainability indicator) and the third to the material removal rate (productivity indicator). Multivariate robust optimization was performed considering sustainability and productivity indicators, while quality was used as a constraint to the nonlinear optimization problem. By applying the EQMMFR-NBI method, Pareto optimal solutions were obtained and an equispaced frontier was constructed. Confirmation tests were performed using Taguchi's L9 arrangement. The results showed that the optimal setups found were able to neutralize the influence of noise variables on the response variables, proving the good adequacy of proposal and the application of the method.