Dassault Systèmes SIMULIA developed an OOTB workflow to optimize a vehicle concept structure for crashworthiness and applied it to design well-balanced energy-absorbing architecture within the SIMULIA and CATIA portfolio. We verified the workflow on an EV structure in collaborating with Hyundai Motor Group as a PoC project. This post recaps the presentation at the Concept Engineering Community Days events at Detroit and Long Beach on Oct 11th and Oct 18th respectively.
ABSTRACT
Optimizing a vehicle structure for crashworthiness in the early concept design phase is challenging because the design evolves frequently, and the simulation takes too long to follow the design changes simultaneously. SIMULIA proposed a workflow with a proper abstracted model and a meta-model that made the crash simulation in minutes. Hence, Hyundai Motor Group performed a PoC project collaborating with Dassault Systèmes SIMULIA to explore the workflow and optimize the structure of a product. We then verify the optimized design with the corresponding FE model and then optimize the structure further using the FE model with ML accelerated crash analysis, which makes the FE model run faster.
WORKFLOW
Crashworthiness optimization needs fast turnaround time. An abstracted model and a coarse mesh surrogate model run in a fraction of a complete crash FE model. The two models are back-upped by Machine Learning technology in connecting the crash KPIs and the corresponding design variables.
The abstracted model, the STICK model, evaluates the architecture design if it distributes the crash energy in a well-balanced manner and optimizes the load-carrying path and the energy distribution to satisfy the target KPIs. A meta-model eliminates the high-cost FE simulation for component crash and makes the architecture crash simulation in minutes on a laptop.
The meta-model defines the component's crash behavior (the KPIs) from the cross-section geometry (design variables). The ML algorithm provides an approximation using a good amount of training data. The STICK model uses the meta-model to compute the crash resistance from the given cross-section geometry or to determine the cross-section geometry from the optimized crash resistance.
The ML Coarse Mesh Surrogate model reduces the discretization error for a coarser mesh model and improves performance dramatically without sacrificing accuracy. The intermediate FE model runs in a fraction of a full FE model, so an iterative approach for optimization is also possible.
CONCLUSION
With the abstracted model and supporting the ML-based meta-model, we could propose a workflow that can quickly optimize the architecture or concept skeleton structure against crashworthiness. By reverse engineering, we can use the workflow to cascade target crash behavior to the components. It saves the design time and cost in the product design.
