Automatic model generation for discrete-event simulation

Abstract

Automatic model generation for manufacturing- oriented discrete-event simulation (DES) aims to reduce the effort required to build and update material flow and production system models. Established automatic simulation model generation (ASMG) approaches often rely on structured engineering data, event logs, or predefined transformation logic. Recent progress in large language models creates new opportunities to use less formalized information, including textual planning notes, layout descriptions, process assumptions, and parameter tables. However, current LLM-assisted approaches still face limitations in reliability, transparency, reproducibility, and simulator integration. This paper derives a conceptual framework for LLM- assisted automatic model generation in manufacturing-oriented discrete-event simulation. The proposed workflow combines hybrid information extraction, a simulation-oriented intermediate representation, explicit validation, simulator-specific instantiation, and expert review. The LLM is positioned as a controlled component for semantic interpretation and structuring, not as an autonomous end-to-end model generator. Particular attention is given to behavioural and control-related aspects because these are required for executable and useful manufacturing simulation models. The paper concludes by outlining a phased prototype implementation and evaluation strategy using controlled manufacturing use cases and later validation with an industrial factory-planning partner.