Successful final conference of the AI research project KEEN has recently ended
What added values can the methods of artificial intelligence create for the process industry? After three years of project duration, the KEEN partners, among them INOSIM, presented their results at a final conference on May 22 and 23, 2023, in Frankfurt. As part of KEEN, about 40 use cases were processed, with selected examples being presented as tandem lectures from industry and science. An interested specialist audience received brand new information from research on industrially applied AI.
Within KEEN, INOSIM has collaborated with major industrial partners and universities to develop new methods and tools for industrial batch plants. In particular, the research contribution to AI-based scheduling in the process industry, jointly developed by Bayer and INOSIM, received positive feedback. Short-term access to optimal production schedules, for example following spontaneous changes, offers numerous advantages in terms of robustness, economy, and ultimately customer satisfaction, as delays are minimized. Deep Reinforcement Learning (DRL) for optimized planning in a typical batch production plant of the chemical industry was tested and evaluated here for the first time. The pilot study shows how DRL can be implemented with an approach based on discrete event simulation. It is expected that the application of DRL in the chemical industry is a promising research and innovation direction and that DRL can complement established methods such as process simulation and mathematical programming.
INOSIM employee Dominik Bleidorn is currently doing his PhD in this area at TU Dortmund under Prof. Dr.-Ing. Sergio Lucia (Laboratory of Process Automation Systems). Dominik also gave a lecture on ChatGPT and the underlying large language models at the KEEN closing conference as part of the doctoral program of the conference. ChatGPT is one of the biggest success stories in the field of AI in recent years and will also leave its mark in the field of engineering, simulation, and production support.
Other technical lectures covered hybrid models for predicting substance data of mixtures (TU Kaiserslautern-Landau), a use case that included the development of a column control using data-based AI models (Bayer AG and TU Dortmund) and the possibilities of systematic data management with the help of AI (TU Dresden). Another focus was the expansion of the KEEN data platform, which is based on Dataverse. In the final panel discussion titled “Are we still KEEN on AI?” participants discussed the value of data and the future of AI in the process industry. “KEEN gave me a realistic idea of what AI can and cannot do,” was a frequently expressed assessment.
A publication on the pilot study by Bayer and INOSIM on AI-based scheduling in the process industry is available here online and has also been published in a special issue of the professional journal Chemie Ingenieur Technik on the KEEN project.
Do you have any questions or would like to know more about this topic? Please contact us.