Dual Kriging assisted efficient global optimization of expensive problems with evaluation failures

The Kriging-based efficient global optimization (EGO) method has been applied successfully in many aerospace engineering optimization problems. However, in those practical applications, the evaluation for the infill point obtained by expected improvement (EI) using computational expensive simulation tool may fail sometimes. And such evaluation failures are critical to the sequential EGO method as it leads to a premature halt of the optimization process due to the impossibility to update the Kriging model. In this work, a dual Kriging assisted EGO-based method, termed as EGO-Kriging, is proposed to deal with the existence of evaluation failures in computational expensive simulation-based optimization. By introducing one more Kriging model, which can be updated regardless of the evaluation status of the newly selected infill point, to approximate the evaluation success possibility, the iterative search process of EGO will not halt prematurely if an evaluation failure of infill point happens. With the available evaluation success possibility, new infill criteria based on EI are developed to ensure the effectiveness and efficiency of the proposed method. Numerical experiments are conducted over analytic problems and a practical axial flow compressor tandem profile optimization problem to study the performance of the EGO-Kriging method. The results show that EGO-Kriging is better than the compared methods over all the analytic problems as well as the tandem profile optimization problem.