As part of the ongoing digitalization of the chemical process industry, there is currently strong interest in exploring how artificial intelligence (AI) methods can be used to optimize industrial production processes. Particularly promising in this context are novel approaches, such as world models, causal AI, and reinforcement learning, which make it possible to predict, understand, and actively influence dynamic process states. Within a collaboration between the Laboratory of Engineering Thermodynamics and BASF, these methods will be developed and tested for real chemical production processes. A central focus of the research project is the development of robust, interpretable, and transferable digital twins for chemical processes. To this end, data-driven models will be combined with physical process knowledge and causal relationships. The resulting hybrid models should not only describe stationary operating points, but in particular also capture dynamic and non-stationary situations, such as start-up, shutdown, and load-change operations. Building on this, AI agents will be developed that use reinforcement learning to propose suitable operating strategies and contribute to the optimization of process operation. The work will be carried out in close collaboration industrial collaboration partners. This will allow measurement data from industrial production plants, real operational constraints, and application-oriented optimization objectives to be directly incorporated into model development. The overall goal is to contribute to the next generation of intelligent, physics-informed, and industrially applicable methods for model-based process optimization.

[back]