Efficient Robust Aerodynamic Design Optimization Method with Unified Design and Uncertain Variables

Robust design optimization (RDO) can be used to obtain the optimal shape of an aircraft whose aerodynamic performance is less sensitive to uncertainties. However, the existing RDO methods are suffering from numerous expensive computational fluid dynamics (CFD) simulations required by the double-loop process, with an outer loop searching for the optimal design and an inner loop performing uncertainty quantification (UQ) of each candidate design. In this study, an efficient surrogate-based RDO framework with adaptive infill sampling is developed for a robust aerodynamic design considering geometric uncertainties. This framework consists of three main components. First, the geometric design and uncertain variables are unified into a single common surrogate model rather than two, and this surrogate model is built to assist both the optimization and UQ loops. Second, a combined infill-sampling method is proposed to adaptively select new samples to be evaluated by CFD, not only for exploring the global optimum but also for refining the common surrogate model to improve UQ accuracy. Third, a two-phase strategy is proposed to accelerate the convergence of a RDO when it is applied to aerodynamic problems in which the robust optimal solution is located near the deterministic optimum. The developed method is verified against analytical test cases and applied to the RDOs of three configurations, Research Aerodynamics Experiments (RAE) 2822 airfoil, ONERA M6 wing, and NASA Common Research Model (CRM) wing/body combination, which, respectively, use 18, 24, and 48 design/uncertain variables. In the results presented, the required number of expensive simulations was greatly reduced, and the developed method was significantly more efficient than conventional bilevel surrogate-based RDO.