Metodologia de análise e otimização de absorvedores de energia em automóveis.

Abstract

The development of automobiles in the present day takes different factors into consideration, such as performance, economic viability, safety standards and emissions. However, certain desired objectives often compete with each other. Improving the crashworthiness of a car can negatively affect its cost and handling. The use of optimization techniques is a good alternative when improving one metric without deteriorating other results linked to it. This work aims specifically at parametrically optimizing the midrail of light vehicles, which is a structural part responsible for mitigating occupant injuries during traffic collisions. The experiments are conducted exclusively by the finite element method, known for its flexibility and precision for this kind of phenomenon, and it's often employed during the initial stages of a project development. In spite of the component's simple geometry, the non-linear nature of the experiments, associated with the number of iterations required for optimization, greatly increases the computational cost for solving the problem. In order to reduce the time of the optimization process, there was studied the use of three different optimization algorithms. In parallel, there was analyzed the feasibility of the use of regression models called metamodels, whose main role is to reduce the overall number of experiments. The use of design of experiments was also explored. It is responsible both for feeding data into the metamodels, and for proposing initial projects for the optimizations. All optimizations were able to offer midrails with greater energy absorption per unit of volume, within a maximum reaction force constraint. However, none of them guaranteed that the midrail had the maximum possible performance within the analyzed domain. Despite limitations regarding the response surfaces' adhesion to certain physical phenomena, the combination of metamodels with optimization algorithms reduced the computational time by up to 75%, while generating midrails with more desirable characteristics, when compared to the results of optimizations without metamodels.