Efficiency of the variable fidelity design optimization method with variable complexity models

Abstract

The goal of this investigation is to provide a deeper understanding about a variable fidelity optimization algorithm and some scaling methods through three test problems. The first two problems are analytic, and the third one is a structural optimization problem. The test problems have been specifically constructed to look for insights regarding the use in the algorithm of models (high fidelity and low fidelity models) with different degree of nonlinearity and different number of design variables. Performance of the variable fidelity framework for first order and second order scaling methods (multiplicative and additive), is compared to a standard sequential quadratic programming optimization performed on the high fidelity model. The main contributions of this investigation are the insights gained with the specifically constructed test problems, which can be extended to other problems, about the use of a trust region variable fidelity framework, and the choice of the most suitable scaling methods depending on the case study at hand. In addition, results show how a reduction in the design cycle time can be obtained, by reducing the number of high fidelity function calls while achieving convergence.

Keywords

• Scaling methods
• Different degree of nonlinearity
• Different number of design variables
• Variable fidelity