Classifier-assisted frameworks for computationally expensive optimization problems

The modern engineering design optimization process often replaces laboratory experiments with computer simulations. This setup formulates an optimization problem of a black-box function, namely, which has no analytic expression and which is computationally expensive to evaluate. However, computer simulations introduce an additional challenge into the design optimization process: often there will exist candidate designs which cause the simulation to fail, which implies that a large portion of the optimization resources may be wasted, and this can lead to a poor final result. To effectively handle such problems, this chapter describes two optimization frameworks which incorporate a classifier into the search. The classifier's role is to predict which solutions are expected to crash the simulation, and this prediction is then used to bias the search towards solutions for which the simulation is expected to succeed. A baseline framework is described which incorporates a single classifier, followed by a more elaborate framework which selects an optimal type of classifier out of a family candidates. Performance analysis using a representative simulation-driven engineering problem of airfoil shape optimization shows the effectiveness of the proposed frameworks and highlights the merit of incorporating a classifier into the search.