This paper presents an end-to-end generative artificial intelligence (Gen-AI) framework for automating the generation, adaptation, and evaluation of discrete-event simulation (DES) models in manufacturing. The approach integrates multiple large language models (LLMs) with a structured blueprint model and targeted human-in-the-loop controls to create executable simulation models from heterogeneous production data, implement targeted modifications, and interpret simulation outcomes. The workflow incorporates prompt engineering, zero- and one-shot implementations, and evaluator–optimizer loops. 21 experimental runs on two industrial case studies from a Swedish automotive manufacturer demonstrate that LLMs can support DES model generation and scenario exploration through a hybrid approach combining automation with human oversight. The results underline both the potential and current limitations of LLM-driven simulation, particularly regarding output consistency and generalizability. Future research should extend the method to more complex manufacturing systems and investigate the role of emerging autonomous Gen-AI tools in simulation-based decision support. • LLM-driven workflow automating discrete-event simulation (DES) model generation, adaptation, and evaluation. • Integrates multiple LLMs with a structured blueprint model and human-in-the-loop oversight. • Systematically tested across 21 runs in two industrial manufacturing case studies. • Demonstrates feasibility for scenario exploration while highlighting challenges in output consistency and reliability.
Schmitt et al. (Mon,) studied this question.