A New Method for Efficient Global Optimization of Large Systems Using Sub-models HEEDS COMPOSE Demonstrated on a Crash Optimization Problem

Executing full vehicle finite element simulations can be very time consuming and expensive. Compound this with the large number of evaluations required for crash optimization problems due to their complicated design landscapes, and optimization of full vehicle crash simulations becomes very computationally expensive and difficult. An ideal solution is to reduce the full vehicle down to a manageable sub-model which runs significantly quicker while maintaining the boundary constraints as if utilizing the full vehicle model. The optimization process can then be managed with this sub-model while achieving significant improvements in computational requirements. This paper will demonstrate how to optimize the a-pillar, b-pillar, roof rail, rocker, front header, and roof bow components of a car for roof crush, utilizing Ultra High Strength materials. In addition to the gauge of the individual structural components, a soft zone trigger and its location within the b-pillar are introduced as design variables. LS-DYNA® is utilized as the simulation tool with the new COMPOSE (COMPonent Optimization in a System Environment) feature within the HEEDS-MDO optimization software utilized to perform the optimization. COMPOSE is a new module that enables the use of a sub-model for optimization in a way that significantly reduces the overall optimization time while encouraging the interactions of the optimal subsystem with the global model to be consistently maintained. This creates a design that gives a similar high performance in the global model as was found in the sub-model. It is shown here that the use of COMPOSE can significantly decrease the design time for finding high performing designs for the roof crush optimization, over a traditional global optimization approach. In addition, it is shown that performing an optimization on the sub-system level by using the original boundary conditions from the global model is not a robust approach for this optimization. The study shows that to take advantage of the reduced runtime in the sub-system model, the COMPOSE technology provides a robust solution for efficient optimization of the system.