Abstract
<jats:p>Optimizing the calendering step in lithium-ion battery electrode manufacturing is vital to improve electronic and ionic transport and overall performance. However, Discrete Element Method (DEM) approaches of compaction remain computationally prohibitive for large scale analysis, while macroscopic modeling lacks analysis capabilities of the internal evolution of the electrode microstructure. This work introduces a novel dual-scale modeling framework that combines a macroscale finite element method (FEM) continuum model with a deep learning (DL) mesoscale surrogate model trained on prior DEM simulations. The parameters of the FEM model were calibrated via microindentation experimental data, successfully capturing macroscopic deformation and the characteristic spring-back effect of the electrode upon calendering. When evaluating heterogeneous electrode coatings with ±2 µm thickness variations, macroscale simulations revealed that a larger roll radius has more homogenous pressure distribution and thickness reduction compared to a smaller radius. To bridge length scales, macroscopic thickness irregularities were mapped into local variations of the applied calendering degree and passed to the DL surrogate. The DL model rapidly predicted localized microstructural features, yielding effective porosity and diffusivity metrics with errors below 10% relative to computationally expensive DEM benchmarks. This accelerated, multi-scale framework establishes a direct link between industrial process parameters and electrode microstructure, paving the way for real-time electrode optimization via digital twins of calendering.</jats:p>