Automatic Shape Optimization of Aerodynamic Properties of Cars |

Automatic Shape Optimization of Aerodynamic Properties of Cars

Eysteinn Helgason, Haukar Elvar Hafsteinsson (Chalmers University of Technology)

Vehicle aerodynamic shape optimization is an important element in improving cars of the future taking into consideration the demands for more efficient and comfortable cars. The optimization process is always multi-objective and the objectives are often connected in a way that the improvement of one objective leads to the deterioration of another. When performing computer simulations huge computational power is required and proper pre-processing is necessary to keep the simulation time within reasonable limits. The aim of this work is to develop an automatic shape optimization procedure where simulations are run on a computer cluster. The user input is limited and the optimization is made automatic by connecting three commercial codes in a closed optimization loop, i.e. AVL Fire (mesh generation and CFD calculations), modeFRONTIER (optimization, process automation and post-processing) and Sculptor (mesh deformation). Optimization algorithm controls the deformation in order to minimize number of simulations needed to achieve the optimal design. The car model used is called the VRAK and is a 1:1 experimental model from Volvo Car Corporation.

Initial designs of experiments (DOE) are set or design parameters chosen by an optimization algorithm in modeFRONTIER. The design parameters are submitted to Sculptor which deforms the mesh accordingly. The deformed mesh is copied into a CFD case, often on another computer cluster, where AVL Fire maps the results from a previously fully developed flow onto this newly deformed mesh. The mapping is straightforward as no cells are created or removed during the deformation process. 

The simulation is then run and when finished the relevant values for forces are extracted from AVL Fire result files. The values are averaged, rewritten as dimensionless quantities and returned to modeFRONTIER. An optimization algorithm in modeFRONTIER chooses the next design parameters based on the results from previous simulations and the loop is closed. Both single objective and multi-objective optimization has been performed with one and two input variables. Two different optimization algorithms have been used, Simplex and Evolution Strategies. Comparison of identical optimization processes with two different turbulence models shows that results are highly dependent on the model used, even though in both cases a converged solution is obtained.