AAS Submodel Generator

Inferring Product Data from Natural Language

Abstract

Industry 4.0 concepts use the Asset Administration Shell (AAS) to represent physical and non-physical assets, facilitating  interoperability along the whole product lifecycle. Today, the AAS is barely established in brownfield plants, mainly due to the required manual effort and expert knowledge for setup. This study proposes automated AAS submodel generation, using Large Language Models (LLMs) in order to map data from diverse sources into AAS structures. Experimental results show high accuracy and efficiency. Future work will improve image preprocessing and expand template support.