Abstract

In automotive industry, optimisation on vehicle structure has been a constant focus in projects. The goal is to find the best compromise between mass, cost, and performances like crash, acoustics, overall stiffness... The structure of the vehicle body is made up of 200 metal parts mainly joined with thousands of spot-welds. Each performance is simulated with increasingly accurate finite element model to reduce the number of physical tests that are still time-consuming and expensive. Despite computational power are in constant progression, simulations remain time challenging to be used in optimisation studies. For example, crash simulation takes around ten hours on HPC. Groupe RENAULT has a rich experience in the use of design of experiments (DOE) for vehicle optimisation. Furthermore, DOE will be completed or replaced with Model Order Reduction (MOR) to improve efficiency, duration, and cost. “Classical” MOR allow to lead two-steps optimisation studies: first a lot of simulations to create an abacus, then uses it to quickly determine the results for new sets of parameters. However, our studies vary tens of parameters, so classical MOR is as costly as DOE. Therefore, RENAULT developed the Regression-CUR (ReCUR) method to reduce the number of simulations. The parametrized reduced order model gives an estimation of the high-fidelity result for a new set of parameters. The accuracy of the reduced model can be balanced with less simulations. Indeed, an overall improved design is searched instead of demonstrating the global optimality. The article will focus on a combinatorial optimisation problem by finding the right balance between assembly process and crash performance (with ReCUR). The first part treats about the optimisation of the main crash criterion from certification rules and media scoring derived from structural deformations. The second part deals with the optimisation of the spot-weld risk requirements used to avoid successive spot-welds failure by applying on a simple test case and on a side-impact industrial case.