Statistical Benchmarking of Surrogate-Based and Other Optimization Methods Constrained by Fixed Computational Budget

The aerodynamic design of an asymmetric oversized payload fairing subject to stability constraints was used as an example of a derivative-free, expensive black box function to benchmark the relative performance of 16 different optimization methods, ranging from gradient-based to simulated annealing and genetic algorithms/evolution strategies, including four methods with surrogate-based accelerators. The focus of the present paper is on the practical attainability of getting an acceptable solution quickly. The various algorithms are compared using performance benchmarking in a statistical sense, yielding an “efficient frontier” with special emphasis on the case when designers are confronted with small computational budgets.