Keywords: multifidelity method, domain-awareness, Bayesian optimization, active learning, cross-regime, aerodynamic optimization.

This paper proposes a domain-aware multifidelity scheme to speed-up the airfoil shape optimization in a cross-regime scenario where the aerodynamic domain evolves with the Mach number. Our strategy relies on a multifidelity Bayesian framework based on a surrogate model iteratively updated through a multifidelity acquisition function that selects the next design configuration and level of fidelity to query. We implement the multifidelity Gaussian process as the aerodynamic surrogate model and formulate an original domain-aware multifidelity acquisition function informed by the evolution of the fluid domain. This property allows to wisely select the level of fidelity of the aerodynamic model considering the compressibility and non-linear effects at higher speed regimes, improving the accuracy of the surrogate model. We validate our approach for the benchmark test-case of the constrained shape optimization problem of a RAE 2822 airfoil. The results suggest that our strategy outperforms popular multifidelity and single-fidelity methods reducing the drag coefficient of the optimized airfoil with an improvement of the 24% respect to the baseline airfoil with a limited computational budget.

download the full-text of this paper (PDF, 8 pages, 526 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description