Extending the Capability-Skill Paradigm to Digital Twins and Large Language Models

Abstract

Production workflows in Cyber-Physical Production Systems (CPPS) increasingly require functions beyond direct machine execution. Some must first be carried out on a virtual representation of the machine before physical execution. Others require cognitive functions such as understanding natural language and deriving results from it. These functions are increasingly automated through Digital Twins (DTs) and Large Language Models (LLMs), which differ from machine-based execution in how they are invoked, how their outputs are produced, and how those outputs can be verified. Orchestrating functions of all three kinds in a coherent workflow that realizes a business goal requires a uniform semantic description that abstracts from these technology-specific differences. The Capability, Skill, and Service (CSS) reference model provides such a description for machine functions, but does not yet cover virtual and cognitive ones. This paper extends CSS with a technology-driven classification of skill realizations and maps the resulting model onto RAMI 4.0 as the established reference architecture for Industry 4.0. A Knowledge Graph serves as the semantic layer that links heterogeneous implementations to shared capabilities.