Automated Aerodynamic Vehicle Shape Optimization Using Neural Networks and Evolutionary Optimization

摘要

The foremost aim of the work presented in this paper is to improve fuel economy and decrease CO_2 emissions by reducing the aerodynamic drag of passenger vehicles. In vehicle development, computer aided engineering (CAE) methods have become a development driver tool rather than a design assessment tool. Exploring and developing the capabilities of current CAE tools is therefore of great importance. An efficient method for vehicle shape optimization has been developed using recent years' advancements in neural networks and evolutionary optimization. The proposed method requires the definition of design variables as the only manual work. The optimization is performed on a solver approximation instead of the real solver, which considerably reduces computation time. A database is generated from simulations of sampled configurations within the pre-defined design space. The database is used to train an artificial neural network which acts as an approximation to the simulations. Finally an optimal vehicle shape is determined using the particle swarm optimization method. The method is solver independent and can handle multiple objectives. The method was incorporated in an optimization tool compatible with Volvo Car Corporation's aerodynamics computational fluid dynamics (CFD) process. The capabilities of the optimization tool were demonstrated on a simplified low-drag car model. An improved shape with a 13.0% lower C_D was achieved with a prediction error of 0.4%.