Responses modeling and optimization criteria impact on the optimization of multiple quality characteristics

Responses modeling and optimization criteria impact on the optimization results were investigated. The Ordinary Least Squares and Seemingly Unrelated Regression techniques were illustrated in two examples from the literature and the performance of three optimization criteria evaluated. In contrast to the standard practice, compromise solutions were evaluated in terms of bias and robustness using optimization performance measures. The results of both examples show that responses modeling strongly impacts on the optimization results, while there is no significant difference between criteria performance. The Seemingly Unrelated Regression technique proved to be useful for modeling correlated responses. Otherwise, this technique can lead to results in close agreement to those obtained with models fitted with the OLS technique.

► Responses modeling strongly impacts on the optimization results.
► Seemingly Unrelated Regression technique usefulness was illustrated.
► Higher-order models led to better compromise solutions.
► Optimization criteria built on different approaches yielded similar results.
► Methods performance was evaluated based on bias and robustness measures.