[2O6-GS-13] AI application: Traffic

[2O6-GS-13-02] Interactive prediction of automotive aerodynamic drag coefficient using machine learning

〇Kei Akasaka¹, Fangge Chen¹, Teraguchi Takehito¹ (1. Nissan Motor Co., LTD)

Keywords:Aerodynamics, Automotive, 3D Convolutional neural network, Voxel

In order to evaluate the coefficient of drag (C_D) on aerodynamics, the wind-tunnel tests and Computational Fluid Dynamics (CFD) are generally used in the car development. However, these test and CFD require much cost and time. Therefore, the relationship between the car shape and C_D was learnt using a machine learning technology to replace CFD with a machine learning model. In this study, the prediction model of C_D was developed by using a 3-dimensional convolutional neural network and a voxel approximation of car shape. The developed prediction model shows lower cost and less time than the conventional CFD. Additionally, in order to predict C_D while deforming a car shape interactively, a tool for the C_D prediction with a graphical user interface was developed. This tool can help designers to solve the trade-off issues of aerodynamics, the package design of the car and the external styling.