Pareto front spacing with differential geometry in multidisciplinary systems

摘要

Multidisciplinary Design Optimization, also known as MDO, deals with the optimization of complex but typically single-objective design problems. As such, it would be valuable to combine it with Multi-Objective Optimization (MOO). Within MOO, Weighted Sums (WS) is a standard solution method, but it may not produce a good spacing of solution points along the Pareto Front (PF). Using our differential geometry MDO framework, we present a technique for intelligently spacing out Pareto solutions produced with WS in a multidisciplinary MOO formulation. Following an initial demonstration, we present modifications to our method to handle ill-conditioning better, to produce fewer dominated points in PF refinement, and to use MOO solution methods other than WS.