Optimization of the aerodynamic properties of high-speed train with CFD and response surface models

Paper i proceeding, 2007

The aerodynamic properties of high-speed trains have in past been optimized through trial and error design process. Such an approach relies on the experience and skills of the engineer to suggest changes in the design that will lead to an improvement of the aerodynamic performance of the train. Although this process leads to an improvement of the design there is no guarantee that the best design will be identified. A more rigorous numerical optimization methodology that allows the best design to be identified is required. The majority of numerical design optimization procedures in fluid machinery uses gradient-based search algorithms. These methods work iteratively through the design space until the optimal design is reached. Such an approach is impractical in optimization of vehicle aerodynamics due to computational effort required for such a large number of CFD simulations. The present work presents the use of a surrogate model in form of response surface approximation (RSA) for multi-objective optimization of train aerodynamics. The design problem of vehicle aerodynamics has multiple objectives, i.e. drag, lift force, cross-wind stability, aeroacoustics etc. An optimal solution of such a problem is called the Pareto optimal front and can help the designers to visualize the trade-offs between different objectives and select an compromise design. In this paper we use an example of optimization of aerodynamic properties of the front of a generic high-speed train to demonstrate an efficient multi-objective optimization procedure. Two object functions are chosen and the response surfaces are produced as a result of Reynolds-Averaged Navier-Stokes simulations (RANS) using simple two-equation turbulence model. The Pareto optimal front is obtained using an evolutionary algorithm (NSGA-II). The present work shows that our approach is very efficient in terms of optimization time and computational requirements. Instead of large number of CFD simulations (several hundreds ) required in traditional gradient-based search algorithms only small number of CFD simulations were required to find an optimal design of the front of the high-speed train.