Adaptive sampling criteria for multi-fidelity metamodels in CFD-based shape optimization

The paper presents a study on four adaptive sampling methods of a multi-fidelity global metamodel for expensive computer simulations. The multi-fidelity approximation is built as the sum of a low-fidelity-trained metamodel and the metamodel of the difference between high- and low-fidelity simulations. The multi-fidelity metamodel is trained selecting the fidelity to sample based on the prediction uncertainty and the computational cost ratio between the high- and low-fidelity evaluations. The adaptive sampling methods are applied to the CFD-shape optimization of a NACA hydrofoil. The performance of the sampling methods is assessed in terms of convergence of the maximum uncertainty and the minimum of the function.