Surrogate Assisted Pedestrian Protection Design via a Foundation Model Orchestrated Workflow

The article introduces a novel foundation model-orchestrated workflow for crash safety design, specifically targeting pedestrian protection. This workflow integrates a surrogate model trained on CAE crash simulations, achieving an average R² of 0.87, multiobjective evolutionary search (NSGA-II), a morphing-based geometry generator, and a natural-language interface utilizing an LLM. By significantly reducing evaluation time from hours to seconds and generating 35 safety-compliant design alternatives in a single run, this approach demonstrates the potential of foundation models to enhance AI-driven engineering in safety-critical applications.