Adaptive and engineering knowledge based metamodeling in multidisciplinary design optimization of aircraft wing structures

Metamodeling methods have been widely studied to alleviate high computation costs on system analysis, such as finite element analysis and computational fluid dynamical simulations, in engineering design optimization. However, the curse of dimensionality is still an obstacle for problems with large numbers of design variables. In this paper, a knowledge-based metamodeling method with adaptive improvement is studied. First, engineering knowledge is shown to be used to reduce computational effort in construction of metamodels. With engineering knowledge, better choice of inputs and outputs to be approximated can be made. Physically meaningful terms, which represent structural mechanical properties are generated algebraically and act as a bridge between design variables and system responses. Second, an expected improvement and possibility of feasibility based infill criterion is used to refine the metamodel iteratively. By doing this, the accuracy of metamodels can be improved while limiting the number of additional sample points. The proposed method is applied to reduce computation effort in an aircraft wingbox structural optimization problem.