Solving multi-objective optimization problems that involve computationally expensive functions is a normal part of many engineering applications. Runtime issues are magnified when one moves from a single discipline to multiple disciplines in a multidisciplinary design optimization setting. The tools selected to perform a design space exploration within this setting must be chosen with care.

更多

Normal-Boundary Intersection (NBI) is a robust general purpose multi-objective opti-mization method for design problems. In the setting of computationally expensive func-tions, running NBI directly using the simulation requires a prohibitive amount of compu-tational cost. Alternatively, running on a surrogate model approximation to the simulation fails to produce sufficiently accurate solutions. Our approach combines the use of mod-

els and simulations in a way similar to the general surrogate management framework by iteratively using both the models and the simulations. While this approach can be ex-ecuted with the push of a button, it is best used, especially at the beginning of a new problem, as one step in a larger design space exploration process. The process includes

the design, execution and analysis of computer experiments, surrogate model development, single objective optimization and problem refinement.

This paper outlines the surrogate model management framework approach to solving multi-objective optimization problems within the context of a general design exploration process. The importance of the use of data visualization techniques will be highlighted

in the discussion. We conclude with a representative aircraft design problem from the aerospace industry.

收起