Bachelor-/Masterarbeit:
Industry 4.0 aims at the end-to-end digital networking of machines, plants, and IT systems. Cyber-physical systems (CPS) play a central role in this context, as they couple physical systems, also called assets, with digital models and services. The Asset Administration Shell (AAS) provides a standardized concept for this digital representation of assets and thus forms a key foundation for interoperability and scalability in Industry 4.0 applications [1].
However, developing and maintaining executable software based on such standardized descriptions still requires significant manual effort. In particular, the manual implementation of Cyber-Part for individual assets based on a configured AAS is time-intensive and error-prone.
This thesis investigates approaches to automatically derive the Cyber-Part from a given AAS such that only the core, domain-specific operation logic must be added manually. A spectrum of approaches should be explored, ranging from classical model-based code generators to methods involving Large Language Models (LLMs) and other AI-assisted tooling. In addition, the applicability of these approaches to retro-fitting scenarios, where code already exists and needs to be aligned with an AAS, should be evaluated.
An important part of this work is the functional testing of the code generation approaches. For practical evaluation, the generated Things should be integrated into the Smart System Service Infrastructure (S³I) [2, 3], a communication platform for cyber-physical systems. Furthermore, the overarching vision is to enable a workflow, in which the intended behavior of a Thing is described at a conceptual level and is subsequently transformed into directly executable code. In this context, integration options with an existing code generator that produces AI-generated, BPMN-compatible code skeletons based on user input and AAS operations, as well as with a companion thesis on AI-assisted AAS modeling, should be investigated.
Requirements
- Programming experience, preferably in Python, as relevant SDKs are implemented in this language
- Familiarity with code generation tools and APIs (optional but beneficial)
- Preknowledge or strong interest in model-driven engineering
- Experience with or strong interest in Large Language Models and AI-based code generation techniques
- Motivation to work with existing software components, AAS specifications, and integration infrastructures
Tasks
- Review existing approaches and tools for code generation from AAS-based models
- Design and implement a prototype for generating pre-configured Thing code from an AAS including functionality tests
- Investigate retro-fitting scenarios and integration into existing infrastructures
- Define evaluation criteria and representative usage scenarios
- Evaluate the approach using automated integration tests in the S³I
