AASia: An Intelligent Framework for Generating Asset Administration Shells from Natural Language

This dataset supports an exploratory comparative analysis of manual and automated Asset Administration Shell (AAS) modeling from unstructured textual descriptions of industrial assets. The underlying research objective is to examine whether large language model (LLM)-based automation can generate syntactically valid and structurally coherent AAS representations, and how such results compare to manually created AAS artifacts produced using conventional modeling tools.

The dataset is organized into two main parts: AASIA-Manual-AAS-Modeling-Study and AASIA-Automated-AAS-Modeling-Study. The manual modeling study includes three industrial test cases describing an electric motor (TC1), a centrifugal pump (TC2), and a filling and capping machine (TC3). These textual descriptions are provided as PDF files in the Test Cases folder. Three participants independently modeled AAS instances for each test case using the AASX Package Explorer. The resulting AAS artifacts are provided in both .aasx and equivalent JSON formats in the GeneratedAAS folder. The ParticipantResults folder contains individual evaluation spreadsheets completed by each participant, reporting estimated modeling time, perceived effort, and qualitative observations. The file Invitation – Master Students.pdf documents the instructions provided to participants prior to the modeling task.

The automated modeling study contains AAS artifacts generated using the AASia framework, which applies an LLM-based pipeline to transform the same textual test case descriptions into AAS-compliant structures. For each test case, the corresponding folder includes the generated AAS in JSON and AASX formats, execution logs, the original input description, and the intermediate textual interpretation produced by the system. For Test Case 3, two automated executions are included: the first uses the original textual description, while the second uses a revised version of the description with increased explicitness regarding quantitative ranges and units. The file Generic_Evaluation_Template_AASia.txt provides the generic evaluation template used to assess the generated AAS artifacts in a consistent and transparent manner.

The dataset allows inspection and comparison of structural modeling decisions, submodel organization, representation of quantitative information, and data typing strategies across manual and automated AAS artifacts. Differences observed among manually created AAS models reflect individual interpretation and modeling preferences, while the AASia-generated artifacts reflect the behavior of a fixed automated pipeline under different input conditions. The dataset is intended to support reproducibility of the reported analysis, facilitate independent inspection of the artifacts, and enable further research on automated AAS generation and comparative modeling approaches.