Adaptive multi‐surrogate‐based constrained optimization method and its application

Abstract This article presents an adaptive multi‐surrogate constrained optimization method (AMSCOM) that can automatically determine the appropriate metamodel for each black‐box function in the constrained optimization problem (COP) and concurrently find the optimum. In AMSCOM, each black‐box function is approximated initially by several different types of candidate surrogates. Then, as optimization progresses, the poorly performing candidate surrogates of each black‐box function are gradually eliminated until the appropriate surrogate is found. Meanwhile, as more than one candidate surrogate exists for each unknown function in the optimization process, multiple approximate optimization problems (AOPs) can be constructed, and new samples can be obtained by solving these AOPs. Additionally, we employ the genetic operator and the local‐linear approximation–Voronoi method to generate new samples. To verify the effectiveness and investigate several properties of AMSCOM, the proposed method is tested on 12 benchmark COPs and compared with several single surrogate‐based methods. Furthermore, AMSCOM is compared with several published surrogate‐based constrained optimization methods, and the results further prove the superior performance of AMSCOM. The proposed method is then employed to optimize the shaft‐clinching process of wheel‐hub‐bearing units, and a desirable result is achieved.