Navigating the Use of Generative AI in Modeling and Simulation: A Case-Study-Driven Methodology for Agentic Workflows in Synthetic Data Generation for Surrogate Modeling

Generative Artificial Intelligence (GenAI), particularly Large Language Models with natural language interfaces, is gaining traction in Modeling and Simulation (M&S) research for its potential to simplify user interaction and support workflow automation. However, integrating GenAI into complex M&S studies poses inherent risks, including ambiguities, hallucinations, and reductions in model fidelity when automation is not adequately controlled. Surrogate Modeling (SMing) builds computationally efficient approximations of high-fidelity simulations from data. This paper investigates the use of GenAI as a middleware-based workflow component to facilitate interaction between modelers and specialized tools for SMing of continuous dynamical systems. The proposed approach is derived through a case-study-driven methodology development. The focus is on supporting automation while preserving the reliability of established SMing tools. We investigate two key challenges: enabling GenAI-assisted automation without compromising the quality of surrogate models produced by specialized SMing tools and integrating such capabilities into practical and reusable workflow for industrial applications. ... mehrTo this end, we develop the Surrogate Modeling Agentic Workflow (SMAW), as a case-study-driven instantiation of the proposed approach, that orchestrates SMing tools and supports synthetic data generation through structured natural language interaction.