Modeling of Electrode Particles in Lithium-Ion Batteries focuses on capturing the complex interplay between electrochemical reactions, ion transport, and microstructural characteristics of electrode materials. By representing how lithium ions diffuse within electrode particles and interact with the active material, these models provide insights into charge–discharge behavior, capacity fade, and degradation mechanisms. Such modeling serves as a bridge between material-scale physics and cell-level performance, enabling improved battery design, optimization, and lifetime prediction. 

This is a new neural network based training framework developed primarily my me with the guidance of my mentor (Anders) at CEA. In contrast to the Lagrangian  or  Hamiltonian  neural  networks—which  operate  in  the  full coordinate  space,  lack  separate  control  over  the  architectures  of  kinetic and potential energy networks, typically require an ODE solver within the training loop, and have often been demonstrated on unforced/autonomous systems—our approach first learns latent coordinates sufficient to capture the dynamics using an autoencoder, allows independent design of the kinetic and potential energy networks, leverages force supervision to eliminate the need for an ODE solver during training, and has been validated on several forced nonlinear systems.