Aerodynamic designers rely on high-fidelity numerical models to approximate, within reasonable accuracy, the flow around complex aerodynamic shapes. The ability to improve the flow field behaviour through shape modifications has led to the use of optimization techniques. A significant challenge to the application of evolutionary algorithms for aerodynamic shape optimization is the often excessive number of expensive computational fluid dynamic evaluations required to identify optimal designs. The computational effort is intensified when considering multiple competing objectives, where a host of trade-off designs are possible. This research focuses on the development of control measures to improve efficiency and incorporate the domain knowledge and experience of the designer to facilitate the optimization process. A multi-objective particle swarm optimization framework is developed, which incorporates designer preferences to provide further guidance in the search. A reference point is projected on the objective landscape to guide the swarm towards solutions of interest. This point reflects the preferred compromise and is used to focus all computing effort on exploiting a preferred region of the Pareto front. Data mining tools are introduced to statistically extract information from the design space and confirm the relative influence of both variables and objectives to the preferred interests of the designer. The framework is assisted by the construction of time-adaptive Kriging models, for the management of high-fidelity problems restricted by a computational budget. A screening criterion to locally update the Kriging models in promising areas of the design space is developed, which ensures the swarm does not deviate from the preferred search trajectory. The successful integration of these design tools is facilitated through the specification of the reference point, which can ideally be based on an existing or target design. The over-arching goal of the developmental effort is to reduce the often prohibitive cost of multi-objective design to the level of practical affordability in aerospace problems. The superiority of the proposed framework over more conventional search methods is conclusively demonstrated via a series of experiments and aerodynamic design problems.
展开▼
