SHAPE OPTIMIZATION OF CRASHWORTHY STRUCTURES

Crashworthiness problems, which are highly dynamic and nonlinear, do not lend themselves well to classical gradient optimization techniques. Evolutionary-based design approaches that employ a form of guided stochastic search algorithm have been successfully applied to these problems. While many design optimization approaches are limited to a small number of continuous design variables, the approach described here can productively search over hundreds at a time. The power of classical evolutionary algorithms can be increased by allowing flexible design variable decomposition and incorporating classical local optimization methods and/or by embedding them within adaptive agents, which communicate but work semi-independently on a common problem. The authors have developed a system that allows for flexible design variable decomposition while combining evolutionary algorithms with local optimization. Within this approach, autonomous agents break down a problem hierarchically, using problem-specific divide-and-conquer rules to organize design variables and design criteria into a set of highly decomposed, overlapped relationships. These agents simultaneously search a discretized design space at various levels of resolution and use different design variable representations, performance measures (combinations of objectives and constraints), and local search methods. The agents exchange information about the decomposed solution space with each other, helping them jointly to satisfy multiple constraints and objectives. This technology has been implemented into a software code called HEEDS (Hierarchical Evolutionary Engineering Design System), which can be run on a single processor or in a networked computing environment, including clusters of personal computers or simple networks of workstations. Using LS-DYNA explicit as the finite element solver within the HEEDS optimization environment, this process has been applied to several automotive lower compartment rail designs, resulting in significant gains in performance along with up to 20% reductions in mass compared to baseline rails designed by experienced engineers. An example application of this method is described herein.