A Generative Modeling Method for Digital Twin Shop Floor

Digital twin (DT) as a key enabling technology for achieving digitization, flexibility, and customization in shop floors has attracted significant attention. However, the shop floor involves diverse assets across multiple dimensions, scales, and interdisciplinary fields, making the modeling process complex. To address this issue, this article analyzes the construction process of ontology-based information models and proposes a generative modeling method for digital twin shop floor driven by large language models (LLMs). First, LLMs are utilized to analyze user intentions, acquiring the hierarchical object structure of DT models. Second, by combining an analysis–retrieval method to extract domain knowledge and generate dynamic prompts, LLMs are guided to realize the creation and fusion of objects and construct structured and semantically enriched DT models. Finally, the effectiveness of the proposed method is validated through examples of shop floor resource scheduling.