Benefits of incorporating designer preferences within a multi-objective airfoil design framework

High-fidelity aerodynamic design problems are not easily managed, and identifying all Pareto-optimal design candidates is often unnecessary and computational exhaustive. We propose a variant of a multi-objective particle swarm optimization algorithm that draws on the domain knowledge of the designer to obtain solutions of interest. The swarm is guided by a reference point, which is viewed as an intuitive means of expressing the designer's preferred level of compromise that can ideally be based on some existing or target design. This hybrid methodology enhances the convergence proficiency and exploitation characteristics of the optimizer, due to the locally focused search effort. The algorithm attempts to identify a partial spread of Pareto-optimal designs that provide the most resemblance to the reference-point compromise. The algorithm is applied to a typical transonic airfoil design scenario in which the PARSEC parameterization model is used to represent candidate airfoil geometry, and a Reynolds-averaged Navier-Stokes solver is used to compute the aerodynamic coefficients. A hypervolume performance metric is applied to monitor convergence and solution spread. A data-mining technique using self-organizing maps is introduced and applied for postoptimization tradeoff analyses. Sample design problems are presented that depict the computational efficiency of the optimization framework and have the flexibility to adapt to disparate design philosophies.