Adaptive sampling techniques of surrogate optimization for marine propeller design

Abstract

Abstract In this paper, two sampling techniques are proposed to improve the accuracy of surrogate models used to solve optimization problems involving computationally expensive objective functions. Both techniques depend on random walks and the acceptance criterion of the simulated annealing algorithm to ensure a balance between exploration and exploitation. Moreover, a genetic algorithm is used to locate the global optimal point of the so-far constructed surrogate model. This is to further enhance the search capabilities of our proposed techniques. A simple cubic radial basis function is used as a surrogate model in our experiments. The proposed surrogate-based optimization approach is applied to the marine propeller design problem, where the objective function is an expensive, black-box function. The design variables are the chord lengths and thicknesses for every blade, while the design objective is to achieve the maximum efficiency of the marine propeller. The obtained results improve on those previously reported in literature.