Comparison of particle swarm optimization and Bayesian optimization algorithms for the shape optimization of the star crash box

Abstract

Automobiles use a crash box as an energy-absorbing component to enhance their passive safety during a frontal collision. The existing literature extensively explores gradient-based optimization and population-based optimization of the crash box, but it hasn’t explored probabilistic models. This study compares particle swarm optimization, a heuristic model, and Bayesian optimization, a probabilistic model, to improve the shape of the crash box so it can better absorb energy during impacts. The optimization framework uses specific energy absorption (SEA) as the objective function. Python scripts create geometric modeling, grid generation, and simulation input files. LS-DYNA is used as an explicit dynamics solver to simulate crashing. The geometric shape parameters are height, width, x-intrusion, y-intrusion, and thickness. Mild steel serves as the reference material for each simulation. A 250 kg rigid impactor with a speed of 15 m/s is used for crushing. The baseline geometry is created by taking average values of design space. The baseline parameters are a = 90 mm, b = 90 mm, u = 15 mm, v = 15 mm, and t = 1.85 mm. Both optimization algorithms achieve higher SEA than baseline geometry. Among the optimization algorithms, Bayesian optimization performed better than particle swarm optimization in terms of SEA, computational time, and other crash safety measures like peak crush force (PCF), mean crush force (MCF), and crush force efficiency (CFE).