Minimizing wind resistance of vehicles with a parallel genetic algorithm

Abstract

We use a simple genetic algorithm with distributed fitness evaluation to optimize the simulated aerodynamic performance of idealized three-dimensional models of vehicles (car, truck) in a virtual wind-tunnel. The genetic algorithm manipulates part of the collection of numerical constants (genotype of candidate solutions) which describe the shape of said vehicle-models. For evaluation of fitness, the genotypes in a small population of 16-30 candidate solutions are translated into virtual vehicle-models (phenotypes) inside a mesh of specified density. The vehicle-models' wind-resistance which defines the inverse-proportional fitness of the associated genotypes is computed via the software package in parallel. After an optimization procedure over 100 generations when saturation of fitness improvement is seemingly achieved, we obtain (1) an optimized simple car-model which (up to practically necessary modifications) can be seen, in principle, on today's roads and (2), surprisingly, an optimized design for the driver's cabin of a truck and an air-shield on top of the latter which appears to be a new design. Both shapes found by the genetic algorithm described in this work significantly outperform elementary geometric shapes such as triangular or parabolic shapes in a simulation of aerodynamic resistance. The application of the findings in this work could yield to significant energy savings in the transport sector in the future. © 2014 Springer International Publishing.