Design knowledge extraction in multi-objective optimization problems

This work concerns the post-optimal analysis of the trade-off front of a multi-objective optimization problem to extract useful design knowledge pertaining to these high-performing solutions. The expected knowledge basically consists of statistically significant relationships between the objective functions and decision variables. These relationships are represented in an intuitive and easy-to-use mathematical form. Since a number of such relationships may exist, for efficiency it is desirable that they are obtained in a single knowledge extraction step. Further, problem knowledge can be explored at two levels: lower and higher. At the lower-level, our interest is in finding a subset of the trade-off solutions to which the obtained relationships are exclusive. The higher-level knowledge addresses the effect of varying the problem parameters (that are kept constant in one run) on the trade-off front and therefore on the relationships. These concepts are explained through different examples.