Aerodynamic drag optimization of a high-speed train

Abstract

This paper considers the optimization of the nose shape of a high-speed train to minimize the drag coefficient in zero-yaw-angle conditions. The optimization is performed using genetic algorithms (GA) and is based on the Aerodynamic Train Model (ATM) as the reference geometry. Since the GA requires the parameterization of each optimal candidate, 25 design variables are used to define the shape of the train nose and, in particular, to reproduce that of the ATM. The computational cost associated to the GA is reduced by introducing a surrogate model in the optimization workflow so that it evaluates each optimal candidate in a more efficient way. This surrogate model is built from a large set of simulations defined in a Latin Hypercube Sampling design of experiments, and its accuracy is improved each optimization iteration (online optimization). In this paper we detail the whole optimization process, ending with an extense analysis of results, both statistical (analysis of variance (ANOVA) to identify the most significant variables and clustering using Self-Organized Maps (SOM)), and aerodynamic. The latter is performed running two accurate simulations using Scale-Adaptive Simulation (SAS) turbulence model. The optimal design reduces the drag coefficient a 32.5% of the reference geometry.