- Research project
MBO-KISS
Evaluate and optimise AI control code
As part of the MBO-KISS research project, ITQ is working with industry and research partners to investigate how generative AI can be used productively to generate PLC control code. The aim is to achieve an end-to-end proof of concept – from the natural-language specification, through the generated Structured Text application (IEC 61131-3), to execution and testing in simulation and on a real plant.
A hybrid approach combines LLM-based code generation with contract-based orchestration and – depending on the level of granularity – evolutionary methods. Test and verification results are fed back iteratively into the next generation cycle. In this way, the consortium aims to counteract the shortage of skilled workers in the engineering of machinery and production facilities and to significantly reduce development times.
The consortium divides the work according to areas of expertise: Fraunhofer IKS is developing the generative AI module and the entire code-generation loop. ITQ is focusing on the automated quality analysis of the generated code – both statically and dynamically using a simulation environment – building on experience gained from the DigiPro research project on code analysis in control systems.
MAX STREICHER GmbH & Co. KGaA, as the application partner, contributes real-world specifications and test systems and constructs the physical setup. sepp.med GmbH is responsible for test analysis, design, implementation, automation and coverage as a further dimension of verification. The CODESYS Group (virtual controllers, development environments) and BOSCH Rexroth AG (automation platform, evaluation of project progress) are involved as associate partners. The project has been approved under the Bavarian Collaborative Research Programme (BayVFP), funding stream „Digitalisation“, for the period from January 2025 to December 2027.
Project partners
Tools and technologies used
FAQ
What is the aim of MBO-KISS?
The aim is to produce a working proof of concept that maps the entire life cycle of AI-assisted code generation – from the natural-language specification through to the tested and verified Structured Text programme for the actual plant.
What role does ITQ play in the project?
ITQ focuses on the automated analysis of the quality of AI-generated code: As there have so far been very few metrics for the static evaluation of ST code in the PLC environment, ITQ is developing suitable metrics and researching concepts which, with the aid of control architecture and simulation, also enable conclusions to be drawn about the dynamic behaviour of the generated code. In doing so, ITQ is building on experience gained from the DigiPro research project.
Who are the other project partners, and what do they bring to the table?
Fraunhofer IKS is developing the generative AI module and managing the entire code generation loop. MAX STREICHER GmbH & Co. KGaA is involved as an application partner, defining real-world specifications and test systems, and building the physical system. sepp.med GmbH is responsible for test analysis, design, implementation, automation and coverage. The CODESYS Group and BOSCH Rexroth AG are also involved in an associated capacity.
What challenges does the project address?
These include, amongst other things, the reliable extraction of structured code specifications from natural-language requirements, a viable approach to hybridising LLMs with planning and optimisation methods, the necessary extensions to simulation and system platforms, and the question of the extent to which generated code must be tested and verified in order to reliably meet quality and safety standards.
How long will the project run, and how is it funded?
MBO-KISS has applied for funding for the period from January 2025 to December 2027 under the Bavarian Collaborative Research Programme (BayVFP), under the „Digitalisation“ funding stream, with total costs of around 1.58 million euros and a funding requirement of around 939,150 euros.