Airfoil design generally involves high dimensionality and nonlinear mappings, leaving the preferred infill sampling strategies in the framework of Bayesian optimization uncertain across instances and changeable in the sequential optimization stages. Knottily, no single-infill sampler guarantees the optimal results for airfoil design for its changeless intrinsic selection mechanism, whereas parallel strategies consume multifold computing resources for high-fidelity simulations. Therefore, this paper proposes an airfoil optimization framework using portfolios that integrates several single-infill samplers and assimilates a suggestion from them under the instruction of a meta-criterion. Two entropy-based portfolios, entropy search portfolio and max-value entropy search portfolio, are used as illustration for alternative scenarios. Comparative results between the proposed framework and frameworks using individual constituent samplers of the portfolios on both aerodynamic design and aerodynamic-stealth design show that the portfolio-based framework automatically authorizes the most suitable constituent at each iteration and guides fast convergences to optimal airfoils of high confidence without being influenced by underperforming constituents across instances of different complexity. It is demonstrated that the proposed portfolio-based airfoil optimization framework is robust and reliable for airfoil design instances without priors for options.
Portfolio-Based Bayesian Optimization for Airfoil Design
AIAA Journal ; 59 , 6 ; 1975-1989
2021-06-01
Article (Journal)
Electronic Resource
English
Airfoil design by optimization
AIAA | 1977
|Inverse airfoil design using design optimization
AIAA | 1996
|AIAA | 1984
|AIAA | 1986
|