A multi-scale model for local polarization prediction in flow batteries based on deep neural network

The side reaction of the flow battery will consume electrons, reduce efficiency, and eventually cause safety problems. Regarding gas management and removal, carefully controlling local over-polarization is a vital issue. However, among the multi-scale models that can predict local polarization, the lack of dimensionality inside the electrode makes it impractical to predict three-dimensional variations. In this work, we propose a three-dimensional multi-scale model that allows the rapid prediction of local polarization, which cooperates with deep neural networks, the pore network model, and the three-dimensional continuum model to have both the advantages of accuracy and extensibility. Parameters such as the porosity, permeability, and specific surface area of the electrode are calculated from 500 randomly generated microstructures, and the training samples for the deep neural network are calculated by the cell-scale model and pore-scale model. Through the developed model, we explore the effects of the interdigitated flow field, variable flow rate optimization strategies, and diverse operating conditions on local polarization. The results show that the proposed model can accurately predict local polarization. The research directions of future work include the collaborative optimization of the electrode's microstructure, the flow field, and the flow rate to ultimately improve the local polarization uniformity.