Keynote

Software Architecture Challenges for LLM-based Control Code Generation

Speaker: Dr.-Ing. Heiko Koziolek, Corporate Research Fellow, ABB Corporate Research, Germany

Abstract

Generative AI now aids many software developers in programming tasks, often providing a significant productivity boost. In the manufacturing domain, control logic code is designed and implemented for industrial automation applications that manage power plants, petrochemical processes, or steel production. Popular large language models (LLM) can also synthesize low-level control code in programming notations according to the IEC 61131-3 standard programming notation used in industry, even without any custom fine-tuning. Integrating LLM-based control code generation into existing industrial programming environments however imposes several software architecture challenges. These include selecting appropriate technologies for integrating proprietary information, designing streamlined APIs to feed LLM output into programming environments, and processing large amounts of sophisticated requirements document in batch runs. This talk summarizes experiences and lessons learned while combining Generative AI and industrial programming from the perspective of a software architect.