Summary
Factories use digital models called Asset Administration Shells (AAS) to represent their machines and components, making it easier to manage and connect everything. Large language models like GPT can create these digital models from documents, but checking the quality of these AI-made models is hard because mistakes can happen and there aren't clear ways to measure errors. The authors tested different ways to measure quality by purposely adding errors and seeing which measurements best caught them. They found that checking the exact names of properties and how well the values matched worked best to spot problems, helping improve how AI tools are used in industrial settings.
Asset Administration ShellIndustry 4.0large language modelsquality metricsperturbation-based evaluationdigital transformationGPT-4o-minidata interoperabilityf1 scorevalue-based recall
Abstract
The rapid digital transformation of manufacturing, often referred to as Industry 4.0, relies on seamless interoperability between physical and software assets. A central enabler is the Asset Administration Shell (AAS), a standardized digital representation of such assets. Recent advances in large language models (LLMs) enable the generation of AAS submodels from unstructured sources such as product datasheets but raise challenges for quality assurance. In particular, unexpected errors, the lack of ground truth references, and the absence of standardized quality metrics hinder reliable adoption. In this work, we evaluate quality metrics for AI-generated AAS using a perturbation-based evaluation framework. By systematically degrading AAS generation along multiple dimensions, we assess how well different metrics reflect quality changes. Based on a dataset of 200 products from multiple manufacturers, we generate 6,400 AAS instances using GPT-4o-mini, Qwen3, and DeepSeek-R1. Our results show that metrics based on exact matching of property names and similarity-based soft matching of property values, in particular value-based recall and name-based F1 score, provide the most reliable indicators of quality degradation. Furthermore, we quantify the impact of different perturbation types and analyze differences across model families and product segments. These findings support the selection of suitable metrics, the tuning of LLM-based pipelines, and the integration of AI-generated AAS into industrial applications.