Shape Optimization of a Vehicle Crash-box using LS-OPT

The aim of this project is to optimize the geometry of a crash-box due to impact at low velocity impact. The optimization problem is solved in LS-OPT, using Neural Networks as meta-model. The Neural Networks meta-model has been evaluated on a small test example and it shows remarkable good approximation of the responses. The geometry was parameterized using HyperMorph. In addition to the geometry parameters, the sheet thickness and the material quality of the crash-box and the bumper-beam were also varied. The FE-model used is a passenger car from Saab Automobile. The objective is to minimize the mass of the crash-box subjected to two deformation constraints and a constraint on the maximum plastic strain in the main crash-rail, which is positioned behind the crash-box. During the optimization procedure, unfortunately, the crash-rail shown to be too weak and it need to be strengthening up using an extra component in the weak section of the crash-rail. Consequently no solution that fulfilled all constraints was found. However, LS-OPT reduced the mass of the component with 20 % and in the same time reduced the sum of all constraint violations with 50 %. Only the plastic strain constraint was violated after five iterations. The meta-modelling technique using Neural Networks showed good results with small surface approximation errors.