Fast and Accurate Strategies for CFD-based Aerodynamic Shape Exploration in a System of Multi-Objective Evolutionary Algorithm

CFD-based stochastic optimizations involve time-consuming problems due to excessive test cases as a nature of the stochastic approach with randomness and the computational efforts of the CFD relying on the mesh sizes. Surrogate models can, in general, be an alternative way to reduce the computational time, they are however confronted with an accuracy issue based on the margin of error of predicted solutions. In this study, quasi-steady-state approximations with smooth transitions of solid surface are applied for accelerating CFD convergence and blocking fundamentally the error that the surrogate models produce. The CFD simulations are restarted from an existing database obtained during an optimization process, and a search algorithm explores the database to choose the closest dataset with the currently targeted candidate. While the dataset is smoothly transited to the targeted candidate, a transfinite interpolation method is implemented for regenerating the grid. The RAE 2822 airfoil, as a reference airfoil, is chosen to increase the maximum lift coefficient and the lift-to-drag ratio at specified Mach numbers and angles of attack conditions. The real-coded adaptive range multi-objective genetic algorithm with Pareto solutions is applied to move toward the max-max objectives, and the PARSEC airfoil parameterization method is used to determine the airfoil shape. Lastly, the computational efficiency of the current strategies is emphasized in terms of the total number of iterations of the CFD at every generation.