Automatic simulation and optimization of the energy absorption box driven by MO-SAC

Abstract This study proposes a reinforcement learning-based multi-objective optimization framework for automotive energy-absorbing box structures, aiming to improve specific energy absorption (SEA) and reduce peak crushing force (PCF). A parametric model with adjustable folding angle, layer number, and multi-wall thicknesses is developed and integrated with automated modeling, simulation, and evaluation. An improved MO-SAC algorithm enables joint optimization of W2W and W2C structures. The agent explores optimal designs based on simulation feedback. Results show that the optimal 9-layer W2W structure with a 122.46° folding angle and wall thicknesses of 1.35 mm, 1.29 mm, and 1.47 mm achieves a 21.28% increase in SEA and a 12.59% decrease in PCF. This method offers a data-efficient and interpretable solution for structural optimization.