The Design of a Wing via Surrogate-Based Multi-Fidelity Bayesian Optimization

Abstract The design optimization of wing geometry is treated as an initial step in the optimization of vertical and/or horizontal axis hydrokinetic turbines. The achievement of design optimization can be evaluated by its accuracy and computational cost, which are difficult to improve simultaneously. Presently, a meta-model is introduced to optimize a wing geometry by means of a multi-fidelity Bayesian optimization framework. The candidate points are selected using an acquisition function called maximum value entropy search. The selection of low- and high-fidelity observation methods is a crucial point in the present study. Accordingly, low-fidelity data generation depends on a 2D CFD analysis of the foil, while 3D Computational Fluid Dynamics (CFD) analysis of the wing is suggested as the high-fidelity method. Reynolds number, angle-of-attack, and the three additional geometric parameters are included in the process to achieve an optimal configuration that maximizes the lift-to-drag ratio. A hybrid optimization technique including a genetic algorithm and a gradient-based method is utilized to enhance the dataset at each step on the one hand and reach optimal configuration on the other.