Otimização de fatoriais 2ᴷ obtidos por busca aleatória considerando mínima variância das estimativas

Abstract

Optimize factorial experiments is important, and its importance is justified by the fact that the process not always can neutralize unknown or undesirable factors which vary according to the nature of these experiments and can influence the response variable. In these terms, the purpose this paper was to perform a study of the optimization of the New Order (NO) designs from Adams, Cintas and Llabrés (2005) and Correa, Grima and Tort-Martorell (2012), generated by exhaustive method of random search and Dickinson's algorithm (1974) and adaptations. This study considered the exchange algorithm in the obtaining of A-optimal, Doptimal designs, and the search for neighboring designs using the simulated annealing algorithm. As a result, it was noted that, by the efficiency of the obtained optimal designs, there was an increase with regard to the values of the A-criteria and D-criteria. It was also noted that New Order designs inherited optimal properties with regard to the studied criteria and that both the New Order matrices and the A-optimal and D-optimal matrices, obtained from the New Order matrices, fairly stabilized the Factor Effect Estimates Variance (VEE F). New orders of execution were proposed for the 2 3 , 2 4 , 2 5 , 2 6 e 2 7 designs due to the results obtained with regard to the Maximum Bias Absolute Value (MBAV) and Variance of Factor Effect Estimates (VEEF) - measures that were directly linked in some cases.