Orchestrating Complex Engineering Workflows with Foundation Models

The paper presents a novel approach to engineering design—specifically for pedestrian protection systems—by leveraging foundation models as central orchestrators. Traditional design optimization often relies on computationally expensive simulations that are difficult to scale. This research proposes a surrogate-assisted workflow where foundation models manage the interaction between design parameters and predictive surrogate models.

By using a foundation model to orchestrate the process, the system can intelligently navigate the design space, reducing the number of high-fidelity simulations required. The foundation model acts as a reasoning engine, interpreting design requirements and guiding the surrogate model to identify optimal configurations for pedestrian safety. This approach shifts the paradigm from manual, iterative simulation to an automated, AI-driven design loop.

Surrogate-Assisted Optimization

The core technical contribution is the integration of surrogate models into a foundation-model-led pipeline. Surrogate models serve as fast, approximate proxies for complex physical simulations. The foundation model dynamically selects which design parameters to test, effectively 'learning' the landscape of the design space. This minimizes the reliance on brute-force computational methods, allowing for faster iteration cycles in safety-critical engineering applications. The methodology demonstrates that foundation models can effectively bridge the gap between high-level design intent and low-level numerical optimization, providing a scalable framework for complex product development.