AASbyLLM

generating digital twin models with large language models

This is a research prototype for generating digital twin models using large language models (LLMs). It receives technical properties of an asset as texts input from users and then generates a digital representation (Asset Administration Shell - AAS) in JSON format. The site also allows for downloading these models and provides tools for evaluating the generated information for research purposes.

Try the application and test View research evaluation process

Input raw texts regarding the technical properties of an asset:

Select from the samples provided above, or feel free to enter your own text to test the system's ability to handle diverse inputs.

0 /600

Choose the backend LLM:

Integrate external library (ECLASS) for AAS generation

Program debug information: system components diagram

The generated AAS serialized in JSON format:

Open the downloaded file by right-clicking and using AasxPackageExplorer , or open the file in ".json" format through the AasxPackageExplorer Windows application.

View screenshot images for instructions (Click to expand)

Open File (Method 1)	Open File (Method 2)	Final Result

Result evaluation section:

Name:

Conceptual Definition:

Value Type:

Data Value:

Unit:

Contextual Description:

Affordance:

LLM reasoning details (Click to expand)

Data Feature Name	Value	Best Semantic Annotation Selected from ECLASS	Searched Semantic Annotations from ECLASS (top 5)	AI Generated Annotation	Affordance	User Rating

Main Features and the Novelties of AASbyLLM

Technologies:

First research prototype applying generative LLM for generating AAS.
Apply generative LLMs for inferencing and embedding LLMs for similarity retrieval.

Benefits:

Wider range of input flexibility and format independent.
Generally applicable to different disciplinary domains.
Utilizes the knowledge learned in LLMs instead of relying on manually crafted mapping rules.
The system adds proper semantic annotation for disambiguating the concepts of the data property.
The system optimizes relevant details to enhance the quality of AAS.

Implications: