Multi-objective evolutionary optimization of unsupervised latent variables of turning process

Manufacturing process modeling and optimization is a challenging task due to the numerous objectives to be considered in the optimization. Generally, the optimization of these processes requires many objective optimization methods to deal with four or more objective functions. However, the correlation structure of the outputs cannot be disregarded. In this work, it is proposed the unsupervised learning of the outputs together with multi-objective evolutionary optimization of the turning process of AISI 4340 steel considering three scenarios varying the tool nose radius. A central composite design varying the process parameters is used to conduct the experimental tests. After tests and measurements of quality and productivity outputs the p correlated observed outputs are firstly transformed in m unobserved latent variables through factor analysis using principal axis extraction method and varimax rotation, with m < p. Subsequently, the relation between the process parameters and the scores of latent variables is modeled through response surface methodology. Multi-objective evolutionary optimization methods are applied in the reduced and uncorrelated set of response models of the transformed outputs. The multi-objective algorithms are compared through hypervolume metric and the pseudo-weights approach is used to decision making. The proposed method can also be applied in other multi-response processes with correlated outputs. (c) 2022 Elsevier B.V. All rights reserved.

Automated summary

This work proposes a method that combines unsupervised learning and multi-objective evolutionary optimization for optimizing the turning process in manufacturing. It utilizes factor analysis and response surface methodology to handle correlated outputs and achieve optimization goals.