The article, “LLM-Based Modelling of AAS-Compliant Digital Twins to Describe Capabilities in Manufacturing-as-a-Service”, authored by by Marc Leon Haller, Kym Watson, Felix Schöppenthau and Ljiljana Stojanovic has been published in Applied Sciences (MDPI).
Manufacturing-as-a-Service (MaaS) platforms fundamentally rely on interoperable descriptions of manufacturing capabilities to enable the scalable and efficient matchmaking of manufacturing requests and available resources across organizational boundaries. In this context, Asset Administration Shell (AAS)-compliant digital twins can provide the required standardized and semantic basis. However, modelling AAS-compliant digital twins remains challenging due to the complexity of the underlying specifications.
To address this challenge, the paper presents an extension to the FA³ST Ecosystem that enables the instantiation of the Capability Description SMT by combining:
- Large Language Models (LLMs)
- domain ontologies as structured knowledge bases
- Human-in-the-Loop validation
Abstract
Disruptions threaten supply chains, creating a need for more resilient manufacturing networks. Manufacturing-as-a-Service (MaaS) has emerged as a promising Industry 4.0 approach to address this challenge. Yet, its effectiveness relies on interoperable digital twins (DTs), enabling the standardized exchange of manufacturing capabilities across organizational boundaries. The Asset Administration Shell (AAS) standards can be used to meet this requirement. However, modeling AAS-compliant DTs is considered challenging due to the standard’s complexity. This paper, therefore, investigates the automatic generation of AAS-compliant DTs for representing manufacturing capabilities. Requirements from MaaS use cases in two research projects reveal limitations in current approaches. To address these limitations, this paper introduces an automated, LLM-supported generation process that leverages ontologies as a domain-specific knowledge base. The approach is operationalized in a modular software architecture and demonstrated through two use cases.
📄 Read the publication here: https://www.mdpi.com/2076-3417/16/10/5059