Efficient Design Optimization Assisted by Sequential Surrogate Models

Abstract

The paper proposes a global optimization algorithm employing surrogate modeling and adaptive infill criteria. The surrogates are exploited to screen the design space and provide lower-fidelity predictions across it; on the other hand, specific criteria are designed to suggest new points for high-fidelity evaluation so as to enrich the optimizer database. Both Kriging and radial basis function network are used as surrogates with different training strategies. Sequential design is achieved by introducing several infill criteria according to the realization of the exploration-exploitation trade-off. Optimization results are provided both for scalable and analytical test functions and for a practical aerodynamic shape optimization problem.