A machine learning-assisted multi-scale method for understanding reactive transport behaviors within electrode of redox flow battery

This work proposes a machine learning (ML)-assisted coupled numerical method spanning from quantum to device scales for redox flow battery (RFB) electrodes. Leveraging quantum-scale mechanisms provided by density functional theory calculations, kinetic Monte Carlo (KMC) method is employed to resolve the coupled interfacial reaction kinetics and ions transport within electric double layer, thereby fundamentally bridging RFB's reaction mechanisms and complex transport phenomena at microscopic scale. By collecting the dataset generated from KMC simulation cases, an ML model is used to predict the interfacial current/mass flux at complicated potential and species concentration conditions. Then, the trained ML model is integrated into a mesoscopic lattice Boltzmann (LB) model for establishing the pore-scale reactive transport model. Moreover, a bidirectional LB-finite element cross-scale computational framework is further built for linking mesoscopic and macroscopic numerical methods. This multi-scale framework investigates the reactive transport characteristics of negative-side electrode during all-vanadium RFB galvanostatic charging process. The numerical results reveal that: At the macroscopic scale, avoiding localized low-velocity dead zones is key to optimizing reaction and transport processes during practical RFB operation; At the mesoscopic scale, the disordered and heterogeneous pore structures of electrode induce an imbalance in reaction and transport behavior, thereby impacting the RFB's macroscopic performance. • Multi-scale numerical model is proposed to elucidate RFB's reactive transport. • Machine learning serves as a bridge numerical models across disparate scales. • Cross-scale framework is spanned from quantum to device scales. • This model delivers new insight into the complex behaviors of electrode.