Vortrag: Machine learning-driven design of experiments and new chemical reactors

Reactor design and optimization are crucial aspects of chemical and biochemical engineering. With the advent of additive manufacturing, advanced reactor geometries are now possible, offering improved operational efficiency and cost-effectiveness. However, due to this extra flexibility, optimizing over these designs is even more challenging. In this talk, we discuss work that integrates computational fluid dynamics (CFD) simulations with a multi-fidelity Bayesian optimization. We introduce an approach that not only recommends optimal reactor configurations and operating conditions but also determines fidelity (level of accuracy of the CFD simulator) levels based on statistical likelihoods and information content, optimizing accuracy and computational efficiency.

The methodology discussed focuses on plug-flow reactors but can be extended to various reactor types. By maximizing plug-flow performance, we identify crucial design characteristics and validate two novel geometries through 3D printing and experimental validation. Through this data-driven optimization of highly parameterized reactors, we aim to establish a framework for next-generation reactors, highlighting how machine learning and advanced manufacturing processes can revolutionize the performance and sustainability of future chemical processes.