Integration of Human Preference into Multidisciplinary Design Optimization

Abstract

In multidisciplinary design optimization (MDO), a designer is often faced with the challenge of selecting an optimal design in the presence of trade-offs between conflicting objectives. In this scenario, we must take into account the preferences of the designer in order to select a single most preferred solution (MPS) for the design problem. This thesis proposes an interactive system for MDO which converges to a MPS through the successive approximation and subsequent maximization of a designers value function. The system was designed and implemented in the form of an iterative loop which elicits designer preferences via pairwise comparison queries and provides feedback to the designer on the influence of his/her preference articulations. Simulation of this system in solving a three objective test problem validates its effectiveness in converging to the designers true value function and MPS. Subsequent analysis reveals the interesting effects of system parameter choices on the convergence behavior of the system. Looking forward, we hope to apply this work to the development of preference based optimization schemes which can more efficiently solve for the preferred region of the Pareto front.