Topology Optimization of Thin-Walled Square Tubes Under Axial and Oblique Crushing Using Deep Neural Networks

Crashworthiness topology optimization of energy absorber components faces numerous difficulties due to the complexities of nonlinear, large plastic deformation of such essential structural elements under subjected loads. To overcome the aforementioned difficulties, this study presents a novel algorithm based on the deep learning process for the topology optimization of thin-walled energy absorbers. In this way, different random patterns of holes have been considered on the walls of a square tube, and the force-displacement responses under axial crushing have been calculated numerically. Considering such results as input data for a deep learning process, the convolutional neural network (CNN) models for the force-displacement response have been extracted. The obtained deep neural network has been used in a multi-objective optimization process considering the pattern of holes as the design variables and crashworthiness criteria as objective functions. After the optimization process, several optimal points were identified as the Pareto front. To investigate the behavior of the obtained optimal thin-walled energy absorbers under oblique loads, different oblique load cases have been considered, and their response has been calculated numerically. Finally, using multi-criteria decision-making methods, one thin-walled energy absorber with a specific pattern of holes has been chosen, considering crashworthiness criteria in the oblique loading. In one of the optimized topologies, the specific mean crushing force and peak crushing force reached 139% and 62% of the baseline values, respectively, under axial loading, with improvements also observed under oblique angles.