Solutions reproducibility in multiresponse optimization problems: A new desirability-based objective function

Abstract

Theoretical solutions for multiresponse problems may not yield the expected results when implemented in practice at the process and/or product level. Causes that have been overlooked and lead to such discrepancies in problems developed under the Response Surface Methodology framework are the magnitude of prediction errors in some regions of the solution space, unreplicated experimental runs, and responses' sensitivity to changes in the values of response model variables. That discrepancy value must be minimized and can be managed in the generation of optimal solutions. Therefore, to improve the reproducibility of theoretical solutions, a new desirability-based function is proposed. This objective function allows to balance the response's bias, predictions quality, robustness, and resilience according to the decision maker's preferences. Two case studies demonstrate its flexibility and usefulness.