Multilevel parallelization strategy for optimization of aerodynamic shapes

Introduction There is a worldwide demand for efficient and robust software which implements aerodynamic optimization. This request is explained by the pivotal role of advanced aerodynamic design in the process of reducing costs and thus improving competetiveness of aircraft manufacturing. A conventional (“direct”) approach to the constrained design of aerodynamic shapes is oriented to the “trial and error” method. That is, an initial aerodynamic geometry is modified according to the experience of designers and to the aerodynamic data supplied by previous windtunnel experiment and Computational Fluid Dynamics (CFD) analysis. A new modification is tested (usually by means of time-consuming CFD methods) and the results are analyzed in order to supply considerations for a new iteration of the design process. Obviously (as confirmed by the cumulative experience in different aircraft industries) this process does not provide a desirable optimal configuration; at the best, it provides a certain improvement of the initial shape. Then, (after a number of modification cycles) the configuration is tested in the wind tunnel. It often happens that, due to the limitations of the optimization policy, as well as the insufficient accuracy of CFD analysis, the whole optimization loop is repeated at least twice. The time needed for 1 optimization loop is measured by months and the whole process may take even years. The above may occur even if the design treats only a limited part of the whole configuration, e.g. a nacelle/fuselage junction or a wing/body fairing. Thus the introduction of an automatic robust aerodynamic shape optimizer based on an accurate flow analysis may dramatically reduce the time devoted to testing and overall cost of design by reducing the number of optimization cycles to a single loop (one-shot optimization approach). On the other hand, the problem of multiobjective constrained optimization still remains stiff and open (especially in demanding engineering environment). Thus a robust and efficient solution of this problem is highly challenging. The problem of optimization of aerodynamic shapes is very time-consuming as it requires a huge amount of computational work. Each optimization step requires a number of heavy CFD runs, and a large number of such steps is needed to reach an optimum. Thus the construction of a computationally efficient algorithm is vital for the success of the method in engineering environment. To achieve this goal it is suggested to use a multilevel parallelization strategy. It includes parallelization of the multiblock full Navier-Stokes solver, parallel evaluation of objective function and, finally, parallelization of the optimization framework. The method was applied to the problem of transonic profile optimization with nonlinear constraints. The results demonstrated that the approach combines high accuracy of optimization (based on full Navier-Stokes computations) and efficient handling of various nonlinear constraints with high computational efficiency and robustness. A significant computational time-saving (in comparison with optimization tools fully based on Navier-Stokes computations) allowed the use of the method in a demanding engineering environment. The method retains high robustness of conventional GAs while keeping CFD