With the increasing complexity of Industry 4.0 ecosystems, the need for intuitive yet powerful tools to model process behavior based on standardized digital representations — such as the Asset Administration Shell (AAS)— is becoming more pressing. While BPMN 2.0 offers a suitable language for modeling orchestrated and choreographed behavior among Digital Twins and humans, the creation of such models still requires domain expertise and manual effort.
Recent advances in Large Language Models (LLMs) such as GPT have shown promising capabilities in transforming natural language into structured models. This opens the door to co-pilot-style assistants that help domain experts generate executable process models from textual specifications.
Building upon an existing modeling taxonomy and BPMN tool that supports AAS-compliant I4.0 processes, this thesis proposes the design and development of an intelligent assistant that enables users to describe process logic in natural language. Based on these inputs, the assistant will suggest process structures composed of AAS SubmodelElements and corresponding BPMN elements.
The assistant should support tasks such as the automatic generation of BPMN diagrams, enrichment with semantic annotations (e.g., AAS references), and validation of model completeness. Ultimately, this approach aims to lower the entry barrier for modeling I4.0 processes and increase modeling speed and quality.
Key tasks include:
-
Integration of a Large Language Model (e.g., via API) into an existing BPMN modeling environment
-
Design of prompt strategies and modeling templates based on AAS SubmodelElements
-
Implementation of a co-pilot assistant that supports process model generation and refinement
-
Development of validation and suggestion mechanisms for semantically complete I4.0 processes
-
Evaluation of modeling accuracy, usability, and speed in comparison to manual modeling approaches
Supervisor: Bektas
