Resumo:
Super duplex stainless steels are difficult-to-cut materials due to their high hardening rates, low thermal conductivity, high fracture resistence, high values of mechanical strength and ductility, high thermal expansion rate and high friction coefficient. These properties entail unstable chip formation, high cutting efforts, chatter, severe tool wear, low tool life, and inadequate surface finishing. In this work, helical milling is employed for hole-making of super duplex stainless steel UNS S32760. To get the best results regarding hole quality, cutting efforts, and productivity, a multi-objective robust evolutionary optimization approach for this multivariate process is proposed. A central composite design was defined, considering as control variables: axial feed per tooth, tangential feed per tooth, and cutting speed; and three noise variables: the tool overhang length, the hole measurement depth, and the lubri-cooling flow rate. This design combines control and noise variables and allows error propagation of the noise variables to achieve levels of the process variables that make the responses robust. Machining force components were measured during tests and roughness and geometrical error responses were measured in the machined holes. To consider the correlation structure of the outputs, factor analysis was applied. The factor analysis was performed considering the principal axis method and varimax rotation. Orthogonal factors or latent variables were obtained to enable dimensionality reduction and to represent the original outputs without correlation. Response models in the function of process and noise variables were obtained through the weighted least squares method. Mean and variance models in the function of process variables were obtained. Robust mean square error models were obtained to model the bias and variance of each latent variable. Finally, multi-objective evolutionary optimization was applied to get Pareto optimal solutions that optimize roughness, cutting forces, geometrical error, and material removal rate. Three evolutionary algorithms were applied and compared through hypervolume. The AGEMOEA multi objective evolutionary algorithm obtained the best performance and some solutions with high trade-off were selected through pseudo-weights to aid the decision-making.
