Project Details
Projekt
Autoregressive Deep Learning for Accelerated Multiscale Simulation of Electrochemical Flow Reactors
Subject Area
Chemical and Thermal Process Engineering
Theoretical Chemistry: Molecules, Materials, Surfaces
Theoretical Chemistry: Molecules, Materials, Surfaces
Term
since 2024
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 553293694
Flexible processes are a prerequisite for the transition towards renewable energy sources and feedstock. Electrochemical processes that utilize electrochemical flow reactors offer promising solutions due to the direct coupling between electrical energy and chemical conversion as well as the good controllability. However, the kinetics of the processes are often sluggish, and the selectivity is rather low. Therefore, a systematic optimization of catalysts and reactor design as well as their operation is required. This requires the use of detailed models that take into account the interaction of transport processes in the fluid and reactions at the interfaces. In particular, this requires multi-scale simulations with low computational cost that can take interface processes and their degradation into account. In this project, this is to be achieved by developing new methods based on machine learning. Molecular surface models will be embedded by surrogate models through the development of approaches based on autoregressive deep learning. In contrast to currently available approaches, details of the interface state are preserved and propagated over longer time scales. Embeddings are also used to integrate elementary steps and kinetic parameters, which improves the extrapolation capability of the models. As part of the project, methods and suitable numerical solvers are being developed and their accuracy and performance investigated. The method opens up new solution procedures in the field of chemical process engineering and new problems in the field of machine learning.
DFG Programme
Priority Programmes
