Application of multi-fidelity modeling technique on crash analyses

Frontal crashes constitute vast majority of all traffic accidents. In case of a frontal impact, bumper system is the most important energy absorbing structure of the vehicle. Absorption of crash energy by the bumper decrease the energy transmitted to passenger cabin. Therefore, bumper systems are a crucial automotive design element which helps to decrease passenger injuries in frontal crashes. In this study, crashworthiness of a bumper system whose bumper beam is filled with honeycomb structure under axial impact is investigating using multi-fidelity modeling technique. ECE R-29 regulation is a compulsory safety requirement for heavy commercial vehicles, which is enforced by United Nations Economic Commission of Europe. In order to evaluate crash performance of the bumper system, loading conditions of ECE R-29 safety regulation is used. Primarily, a rigid test fixture with three alternatives is designed to keep bumper system in place while impacting in front of pendulum test bench which complies ECE R-29 regulations. Then, numerical simulations are run on finite element models of three alternatives of test fixture and results are compared with old test fixture which is failed in previous tests. One of the alternatives are chosen according to rigidity, weight and cost, then used in simulations for bumper optimization. A typical bumper system consists of a bumper beam and two crashboxes behind the beam and all assembly is fixed to front end of the front rails of the vehicle. In the next step of the study, a bumper system is designed which has two pieces of honeycomb structures inside bumper beam in front of crashboxes. Crashbox length, honeycomb wall angle and honeycomb wall thickness values are sought for by using a surrogate-based multi-objective optimization approach to maximize a composite objective function that provides a compromise between crash force efficiency and specific energy absorption. Optimizations using two different fidelity finite element models compared with multi-fidelity optimizations according to their optimization performance and computation time. Finite element analyses are conducted by using commercial finite element code LS-DYNA. Quadratic response surface models are used to predict finite element model response at any point in design space. It is found that optimum crash force efficiency and specific energy absorption requires shorter crashboxes, high honeycomb wall angle and medium honeycomb wall thickness. It is also found that even though the low-fidelity model decreases computation costs, high-fidelity model provides better optimum design. Multi-fidelity optimization provides a compromise between accuracy of high-fidelity model and computational time advantage of low-fidelity model.