Benchmarking constrained surrogate-based optimization on low speed airfoil design problems.

2019 
In this paper, we investigated and compared the performance of various constrained surrogate-based optimization (SBO) techniques on solving low-fidelity, low-speed airfoil design problems. We aim to better understand the strengths and weakness of various constrained SBO algorithms on handling non-algebraic real-world problems. Low-fidelity problems are chosen since they allow us to perform multiple independent runs of optimization algorithms, but still in the domain of non-algebraic real-world problems. To be specific, the optimization methods that we compared are Kriging-based efficient global optimization (EGO) with the probability of feasibility (PoF), ConstrLMSRBF, COBRA, and COBYLA. Results on four airfoil design problems show that ConstrLMSRBF is the most robust method in terms of convergence and consistency of performance. On the other hand, EGO-PoF found high-quality solutions on two airfoil problems, but its robustness decreases as the dimensionality increases. We also observe that COBRA is significantly better than EGO-PoF on one high-dimensionality problem, but its general performance is not as good as that of ConstrLMSRBF. Finally, COBYLA is the worst performer, which implies that methods based on linear interpolation are not accurate for the problems considered in this paper.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    28
    References
    4
    Citations
    NaN
    KQI
    []