This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

Crash analyses are always amongst the top priorities for improving vehicle safety and performance in the automotive industry. The evolution in the drive train technologies will not make these crash analyses obsolete any time soon as the safety of the battery pack introduced new challenges to the manufacturers. The historically existing crash analysis on the other hand have always the potential to be upcycled to gain knowledge before the design of battery packs in Electric Vehicle (EV) chassis. In this study, we are integrating AI-based surrogate models, and Multi-Fidelity Efficient Global Optimization (MF-EGO) algorithms within a cloud-native platform to streamline the crash-box design process, to achieve faster design cycles, reduced computational costs, and enhanced collaboration across organizational boundaries. In this work, we use a crash-box with parametrized material thicknesses in a full car simulation under different impact speeds. These high-fidelity transient simulations, run in Open Radioss. A frontloaded database of these simulations also serves as training dataset for a surrogate model. These surrogate models significantly reduce the computational load while preserving accuracy, enabling faster evaluations within optimization cycles. The parametric nature of the problem allows usage of Optimus for optimization of crash-box designs. Demanding optimization problems may end up in an excessive number of high-fidelity simulations for which the time and resources cannot always be justified. To address this trade-off between simulation accuracy and computational efficiency, the MF-EGO algorithm is used to orchestrate the usage of low-fidelity calculations like coarser mesh or surrogate models together with high-fidelity simulations for optimization process. In this specific problem the surrogate models generated by nvision are used for the low-fidelity calculations and the high-fidelity simulations will always be run on Open Radioss. The MF-EGO algorithm leverages these different fidelities in optimizing a design or system more efficiently than using only the highest fidelity level, which is typically the most computationally expensive. Central to this framework is a cloud-native collaborative engineering platform, id8, that serves as the backbone for integration, execution, and democratization of the workflow. This enables seamless orchestration of simulations and optimization tasks, leveraging cloud resources for scalability and cross-geographical collaboration. With its automated interface, users can define and execute complex engineering workflows through a single-click operation, thereby reducing process'™ complexity. In conclusion, this study showcases how integrating MF-EGO algorithms with embedded AI-driven surrogate models in a cloud-native environment can enhance the crash-box design process in the automotive industry. This not only improves efficiency and accuracy but also promotes accessibility and collaboration, setting a benchmark for future applications in engineering design.