Efficient knowledge-based optimization of expensive computational models using adaptive response correction

Computer simulation has become an indispensable tool in engineering design as they allow an accurate evaluation of the system performance. This is critical in order to carry out the design process in a reliable manner without costly prototyping and physical measurements. However, high-fidelity computer simulations are computationally expensive. This turns to be a fundamental bottleneck when it comes to design automation using numerical optimization techniques. In particular, direct optimization of simulation models, typically, requires a large number of model evaluations, which may be impractical or even infeasible in a reasonable timeframe. Possibly the most promising approach to alleviate this difficulty is surrogate-based optimization (SBO), where direct optimization of expensive models is replaced by an iterative enhancement and re-optimization of fast surrogate models. While a large variety of surrogate modeling and optimization are available, the methods exploiting the so-called physics-based surrogates seem to be the most efficient ones because the knowledge about the system of interest embedded in the underlying (often simulation-based) low-fidelity model ensures good generalization of the surrogate and a rapid convergence of the SBO algorithm. In this paper, we review a specific technique of this class, that is, the adaptive response correction (ARC). We discuss the formulation of the method, its limitations and generalizations, as well as illustrate its application for solving problems in various areas, including microwave engineering, antenna design, and aerodynamic shape optimization. (C) 2015 Elsevier B.V. All rights reserved.