Gaussian Process Surrogate Model for Levering Similar Trends Across Concepts.

The computational budget for designing and optimizing engineering systems has been tightened by the growth in the number of technology alternatives to satisfy the increasingly demanding goals in modern systems and the use of sophisticated, but also computationally burdensome, simulation tools to quantitatively choose the best technology alternative more accurately. This paper develops a new Gaussian process metamodel, which leverages trends in engineering objectives that are similar across derivative concepts, as opposed to current approximation methods that handle concepts independently. This metamodel is based on a muitifidelity approach that enhances a new concept prediction with observations of a previous concept with similar trends. The influence of the sizes of the new and the old concept training sets in the performance of the proposed metamodel is explored in this paper. The proposed multifidelity metamodel is tested against traditional independent surrogates in approximating the engine shaft horsepower of a UH-60A modification with a Fenestron tail, where observations from the UH-60A with a conventional tail are available. The proposed metamodels are up to two times as accurate as independent concept modeling for small new concept training sets; then, a multiobjective efficient global optimization algorithm is applied to the proposed metamodel to search for minimal UH-60A power consumption in hover and cruise. This results in Pareto-optimal solutions that are 10% more efficient than the UH-60A baseline. [ABSTRACT FROM AUTHOR]