In the transition toward a circular chemical industry, many processes have to be (re)designed. To facilitate this process, we are developing computer-aided methods for conceptualizing chemical and biotechnological processes. These methods comprehend modeling, simulation, and optimization, which are closely interlocked. Recent advances in machine learning and artificial intelligence led to great opportunities in computer-aided conceptual design. We are currently engaged in the following projects:
In collaboration with the Grimm group, we have pioneered using reinforcement learning to design entire chemical processes from scratch. We have developed robust process simulators to evaluate a large set of process ideas. Artificial intelligence in the form of a reinforcement agent is trained to suggest processes in the simulator that are economically optimal. In a competitive two-player game that encourages the agent to explore novel designs, the agent is trained from zero knowledge by playing many games against itself. The developed method can develop complex designs, including processes with recycles and azeotropic distillation sequences. The developed algorithms can be employed outside process engineering for literally any planning problems. DFG finances the project as part of the SPP 2331 Machine Learning in Chemical Engineering.