Perpustakaan Digital ITB

Economic growth inevitably increases the number of vehicles and, consequently, accidents, making it essential to enhance vehicle safety by improving crashworthiness. The principle of biomimetics has been applied to the crash box structure and gave good results. This research designs a lattice crash box, optimized using machine learning and validated through LS-DYNA simulations and additive manufacturing experiments. The crash box structure will use twisted octet lattice structure. The lattice structure modeling is made with a height of 7.5 cm with a 2x2x2 configuration. The comparison between experimental validation and numerical results with a maximum error value of 6.05% with the load displacement, mean crushing force, and deformation shows a very similar result. The numerical simulations on the baseline model shows a SEA value of 17.52 kJ. Artificial neural network (ANN) was used as optimization processes with 100 samples and produced a mean absolute error value of 1.15%. Global sensitivity analysis (GSA) using the SHapley Additive exPlanations (SHAP) method reveals that the parameters with the greatest influence on SEA are the structure’s ratio. The Non-dominated Sorting Genetic Algorithm-II (NSGA-II) identified an optimal structure with a Ti-6Al-4V material, a ratio of 2, and a diameter of 3.25 mm. This design was predicted to achieve an SEA of 46.05 kJ/kg. Verification through simulation yielded an SEA of 47.47 kJ/kg, corresponding to a relative error of 3%. Compared to the baseline, the optimization achieved a 170.95% improvement in SEA.